Original Research

The annual mortality in patients with mildly symptomatic heart failure ranges from 5% to 10%. However, as symptoms progress in severity, annual mortality may reach 30% to 40%.¹ Mortality remains high in patients maintained on standard therapy (including an angiotensin-converting enzyme [ACE] inhibitor^1,2) in large part because of the impact of sudden cardiac death in patients with heart failure.³ Additional therapeutic agents are needed to improve survival for this patient population.^1,3

Heart failure may be associated with either systolic or diastolic dysfunction. Patients with systolic dysfunction have poor left ventricular wall motion and an ejection fraction of less than 40%. Patients with diastolic dysfunction have a normal ejection fraction but a noncompliant left ventricular wall that impairs diastolic filling. The etiology of heart failure may be ischemic or nonischemic.¹ Coronary artery disease,¹⁴ and hypertension are common etiologic factors.¹

Beta-blocker therapy results in an acute negative inotropic effect.⁵ This acute effect led to the conventional belief that b-blockers were contraindicated in patients with heart failure.^2,4 However, clinical trials have demonstrated beneficial effects of long-term b-blockade in patients with heart failure, including improvements in clinical status,⁶ heart failure symptoms,¹ ventricular function,⁵ disease progression,^7,8 and hospitalization.^8-12 As a result of the identification of these beneficial effects, clinical trials have been conducted to assess the impact of b-blocker therapy on mortality.^3,4,9,11-15 Our goal for this systematic review was to synthesize available data and determine whether b-blocker therapy reduces the risk of mortality in patients with systolic dysfunction.

Methods

Literature Search

Two independent searches of MEDLINE (1966 to February 2000) were conducted to identify published randomized double-blind controlled trials of b-blockers in patients with chronic heart failure. We used both Medical Subject Heading (MeSH) terms and key words to provide a broad search. The results were limited to English language articles involving human subjects and the publication type “clinical trial.” In addition to MEDLINE, we searched several other databases for relevant literature, including International Pharmaceutical Abstracts, Iowa Drug Information Service, Current Contents, and The Cochrane Library. We examined references from published clinical trials and reviewed articles to identify further potential articles for inclusion in our evaluation. Finally, the manufacturers of metoprolol (Novartis Pharmaceuticals Corporation, East Hanover, NJ), carvedilol (SmithKline Beecham Pharmaceuticals, Philadelphia, Pa), and bisoprolol (Lederle Laboratories, Philadelphia, Pa) were contacted in an effort to ensure that pertinent articles were identified and included.

We determined the criteria for inclusion before conducting the literature searches. Clinical trials were included if they were randomized, double-blinded, and controlled; if the patients had a diagnosis of systolic heart failure; and if they assessed mortality as a primary or secondary end point. They were excluded if the study duration was shorter than 3 months or if publication occurred before 1975.

We examined 280 articles or abstracts for our review. Although several clinical trials were not included because of study design issues (eg, lack of randomization or blinding), the majority (approximately 95%) were excluded because they lacked a mortality end point. None of the identified trials used an active comparator; therefore, all trials were placebo controlled. Ultimately, we identified 6 clinical trials^3,9,11-14 meeting the criteria and used them in our analysis.

Quality Assessment

Each study was evaluated using the instrument designed by Jadad and colleagues¹⁶ that rates study quality from 0 (worst) to 5 (best). Two of the 6 clinical trials identified received a score of^5,11,14 while the remaining 4 received a score of^4,3,9,12,13 The reasons for lower scores included a lack of description of the randomization process¹³ or of which patients withdrew from the trials.^3,9,12

Analysis of Data

We critically analyzed the mortality end points of each study. Two independent reviewers calculated relative risk, relative risk reduction (RRR), absolute risk reduction, and number needed to treat (NNT) for total mortality. The NNT is a numerical representation of the number of patients that must be treated to prevent 1 adverse outcome (in this case death) in the population studied for the duration of the trial. Direct comparisons of NNT between studies should be avoided; since the benefit cannot be assumed to have occurred in even increments throughout the trial, the NNT serves only as an alternate form of data presentation. The results of these 4 calculations are provided in an effort to allow for additional assessment of the clinical trials evaluated.

Homogeneity of effect was assessed by calculating a Q statistic, where the number of degrees of freedom is equal to the number of trials included in the analysis minus 1. A P value of less than .05 was considered statistically significant. The pooled estimate of effect was calculated using the Mantel-Haenszel method, which assumes a fixed effect model. The variance of the individual risk assessments, the weight of each study, and ultimately the summary odds ratio (OR) and 95% confidence interval (CI) were calculated for mortality end points when studies were homogenous using this method as previously described.¹⁷

Results

Study Characteristics

Of those identified, 2 trials each evaluated bisoprolol (the Cardiac Insufficiency Bisoprolol Study [CIBIS]¹¹ and the Cardiac Insufficiency Bisoprolol Study II [CIBIS-II]¹²), carvedilol (the Australia/New Zealand Heart Failure Research Collaborative Trial [ANZ]⁹ and the US Carvedilol Heart Failure Program [US Carvedilol]¹³), and metoprolol (the Metoprolol CR/XL Randomised Intervention Trial in Congestive Heart Failure [MERIT-HF]3 and the Metoprolol in Dilated Cardiomyopathy [MDC]¹⁴ trial). A total of 9335 patients were enrolled in the 6 trials.^3,9,11-14 With an average 7.5% withdrawal rate during the active run-in phases,^9,13,14 9171 patients were ultimately randomized to receive either b-blocker or placebo therapy. The average withdrawal rate after randomization was 15.8% for patients receiving active treatment compared with 17.2% for patients receiving placebo (P >.05 for each individual trial).^3,9,11-14

Patient Characteristics

There were no statistically significant differences between the baseline characteristics of the patients receiving active treatment versus placebo. The mean age of patients was 60 years, with an average of 78% men. All patients were receiving pharmacologic treatment for their heart failure before study enrollment. Although the studies used various ejection fraction requirements, the average ejection fraction ranged from 22% to 29%.^3,9,11-14 In all studies the majority of patients had New York Heart Association (NYHA) class II or III heart failure.^3,9,11-14Table 1 shows the study characteristics for each included clinical trial and the baseline characteristics of enrolled patients.

End Points

Mortality offers an objective and clear end point for clinical trials. It supersedes other safety and efficacy end points.¹Table 2 provides a summary of selected mortality end points and the notation of the statistical significance of the end point in each trial. The calculated risk reduction demonstrated for total mortality in trials generating a statistically significant mortality benefit is given.

A statistically and clinically significant decrease in mortality was reported in CIBIS-II, US Carvedilol, and MERIT-HF. These clinical trials included a total of 7732 patients receiving 1 of 3 different b-blockers.^3,12,13 A nonsignificant reduction in total mortality was noted in patients receiving b-blocker therapy in CIBIS-I and ANZ,^9,11 while a nonsignificant increase in total mortality was reported in patients receiving b-blocker therapy in the MDC trial.¹⁴ These clinical trials included a total of 1439 patients and were not powered to assess this end point.^9,11,14

Analysis revealed that the 6 clinical trials are homogenous in total mortality (Q=9.3). This indicates that the trials are measuring an effect of the same size and that differences in issues such as study design and patient population would not be expected to have an appreciable impact on the outcome of meta-analysis. When the results of all 6 trials were pooled, a statistically significant reduction in the risk of total mortality was noted (ORMH=0.66; 95% CI, 0.58-0.75). The Figure 1 is a representation of the effect of b-blocker therapy on total mortality.

Five of the 6 trials reported mortality caused by sudden death.^3,11-14 Both MERIT-HF and CIBIS-II reported significant decreases in sudden death in patients receiving b-blocker therapy.^3,12 Of the remaining 3 studies, the MDC trial showed a nonsignificant increase in sudden death,¹⁴ and CIBIS and US Carvedilo¹ showed a nonsignificant decrease in sudden death.^11,13 US Carvedilol had a larger RRR with respect to sudden death than MERIT-HF or CIBIS-II; however, statistical significance was not reached.¹³ Analysis revealed that these 5 trials were homogenous (Q=7.0) with respect to sudden death. When their results were pooled, a statistically significant reduction in the risk of sudden death was noted (ORMH=0.61; 95% CI, 0.5-0.75). All the included trials evaluated cardiovascular mortality as an end point; however, analysis revealed that the trials are heterogenous with respect to cardiovascular mortality (Q=12.7), and therefore a summary statistic would be misleading.

With analyses of the impact of disease severity and etiology of heart failure within the clinical trials, we attempted to determine the response of various subgroups to b-blocker therapy. Patients with NYHA class IV heart failure represented 3.6% and 17% of the MERIT-HF and CIBIS-II study populations, respectively. Subgroup analyses in these studies indicated that patients with NYHA class III heart failure had a nonsignificant but greater RRR in mortality than did those of NYHA class IV^3,12 and II heart failure.³ US Carvedilol did not categorize patients according to NYHA class but rather stratified them on the basis of exercise test performance into mild, moderate, and severe categories. There was no difference in mortality benefit with b-blocker therapy between these 3 defined groups.¹³

The results of CIBIS-II, MERIT-HF, and US Carvedilol stratified by heart failure etiology indicated a decrease in mortality in patients with both ischemic and nonischemic heart failure. All 3 studies noted a significantly reduced mortality in patients with ischemic heart failure.^3,12,13 US Carvedilol demonstrated a statistically significant decrease in mortality in patients with nonischemic heart failure receiving b-blocker therapy,¹³ while MERIT-HF and CIBIS-II did not.^3,12

Discussion

Our findings indicate that there is a significant reduction in the risk of total mortality in patients receiving b-blocker therapy. US Carvedilol reported the largest RRR in mortality with b-blocker therapy in patients with heart failure¹³ compared with the other large trials that found a significant decrease in the risk of death.^3,12 The RRR noted with carvedilol therapy approaches twice that reported with bisoprolol or extended-release metoprolol, although the absolute risk reductions are similar. The shorter study period employed in US Carvedilol compared with the other trials may provide some explanation for the large benefit noted, especially if the benefit of b-blocker therapy in this patient population is maximized early in the course of therapy. Also, the active run-in phase resulted in the withdrawal of 8.6% of the patients before randomization. Eliminating patients intolerant to carvedilol may have created the appearance of an elevated response to therapy compared with that seen in trials that did not employ an active run-in phase. Moreover, the patients included in US Carvedilol had a lower mean ejection fraction and a greater prevalence of nonischemic heart failure than the other clinical trials reviewed. The most intriguing potential explanation for the apparent greater mortality benefit with carvedilol compared with bisoprolol or metoprolol is the additional pharmacologic effects associated with carvedilol. In addition to b1-adrenergic blockade, carvedilol blocks a1- and b2-adrenergic receptors and has antioxidant effects.¹⁸ The contribution of these pharmacologic properties to the mortality benefit demonstrated with carvedilol therapy is currently unknown.

The limited number of patients with NYHA class IV heart failure enrolled in MERIT-HF make comparative benefit speculations unreliable. Also, the disease severity stratification used in US Carvedilol is not standardized, and the sample size representing the most severe patients was less than 10% of the study population. Based on the data currently available, it is not known if the magnitude of benefit with b-blocker therapy is related to disease severity.

The number of patients with nonischemic heart failure receiving b-blocker therapy was largest in MERIT-HF,³ followed by US Carvedilol¹³ and CIBIS-II.¹² Of these, US Carvedilol was the only trial to demonstrate a statistically significant decrease in mortality in patients with nonischemic heart failure receiving b-blocker therapy.¹³ This may indicate that greater benefit can be derived with carvedilol than with bisoprolol or metoprolol therapy in patients with nonischemic heart failure. Further data are needed to fully assess the impact of heart failure etiology on mortality following b-blocker therapy.

Sudden death, which is often attributed to ventricular tachycardia or fibrillation,¹⁹ is a major cause of mortality in patients with heart failure. Although large trials evaluating metoprolol and carvedilol reported a significant reduction in sudden death,^3,12 the use of concomitant antiarrhythmic medications varied between each trial evaluated. For example, approximately 15% of patients in CIBIS-II were receiving therapy with amiodarone,¹² which has been shown to decrease the occurrence of sudden death in patients with heart failure.²⁰ However, amiodarone therapy was evenly distributed between patients receiving bisoprolol and placebo1² and would not be expected to have a untoward effect on the assessment of sudden death.

Relative contraindications to b-blocker use are reflected in the exclusion criteria used in the trials discussed and include clinical instability,^3,11-13,14 second- or third-degree heart block in the absence of an implantable pacemaker,^3,9,13 low blood pressure,^3,9,12 low heart rate,^9,12,13 or treatment-requiring obstructive airway disease.^9,12,14 All the trials we reviewed indicated that b-blocker therapy in patients with heart failure does not result in a clinically significant decrease in systolic or diastolic blood pressure. As expected, however, a decrease in heart rate of approximately 10 to 15 beats per minute is likely to occur.^3,9,11-14 It is important to note that in US Carvedilol, increased dizziness in the carvediloltreated patients was noted compared with placebo. Dizziness was most pronounced after a dosage increase and dissipated with continued use.¹³ Target doses reached with b-blocker and placebo therapies were similar,^9,13,14 further confirming the tolerability of these agents with progressive dose titration.

Patients with Heart Failure

On the basis of studies reviewed, it is reasonable to anticipate a 15% to 20% withdrawal rate from b-blocker therapy in patients with systolic heart failure. The reasons for withdrawal will most commonly be worsening heart failure or side effects, such as fatigue and dyspnea. It is reassuring that the rates of withdrawal in the trials were similar for placebo and active therapies, suggesting that b-blocker therapy was not the primary cause of withdrawal.

Questions remain concerning the effects of b-blockade in patients with heart failure. The Carvedilol Prospective Randomized Cumulative Survival Trial was designed to evaluate the mortality effect of carvedilol versus placebo in patients with severe heart failure.^4,21,22 Although not published at the time of this review, the trial was discontinued early as a result of a large and consistent decrease in mortality in patients receiving carvedilol.²³ The Carvedilol ACE Inhibitors Remodeling Mild Heart Failure Evaluation trial is designed to evaluate carvedilol in asymptomatic patients with left ventricular dysfunction.^12,21 Also, the Betablocker Evaluation Survival Trial evaluated bucindilol versus placebo in patients with moderate to severe heart failure.^4,15,21 This trial was discontinued early because of lack of mortality benefit.²⁴

Heart failure primarily affects the elderly population older than 65 years.¹ Currently published trials have not included large numbers of elderly patients,^12,21,22 and therefore, extrapolation of the data to this patient population should be done with caution. Also, the effects of race and ethnicity have yet to be systematically addressed.²¹ Studies designed to assess the impact of b-blockade in patients with heart failure depending on the etiology of their disease would allow for further maximization of benefit. The Carvedilol Post-Infarction Survival Control in Left Ventricular Dysfunction trial²² was designed to evaluate patients with heart failure following an acute myocardial infarction.^12,22 Clinical trials are needed to determine optimal dosing regimens.³

Comparative clinical trials with a mortality end point have not been conducted. Efficacy comparisons between selective and nonselective b-blockers are necessary to quantify the survival benefit.³ The Betaxolol Versus Carvedilol in Chronic Heart Failure study is designed to compare betaxolol with carvedilol in patients with NYHA class II or III heart failure.⁴ The comparative efficacy of carvedilol with the b1-selective agent metoprolol is being assessed in the Carvedilol and Metoprolol European Trial.^4,21,22

Conclusions

All stable patients with NYHA class II and III heart failure should be considered for b-blocker therapy. Therapy should be initiated only in stable patients using a low dose of a b-blocker. This dose may be subsequently gradually titrated upward by doubling the current dose every 1 to 2 weeks on the basis of clinical response and patient tolerability.

Acknowledgments

We would like to acknowledge the assistance of Deborah S. Carson, PharmD, BCPS, and Ché Jordan, PharmD.

Related Resources

• American Heart Association-congestive heart failure www.americanheart.org/chf
• WebMD-Heart Disease Center www.my.webmd.com/condition_center/cvd
• Heart Failure Society of America www.hfsa.org Organizers describe the site as “a forum for all those interested in heart function, heart failure, and congestive heart failure (CHF) research and patient care.”

The annual mortality in patients with mildly symptomatic heart failure ranges from 5% to 10%. However, as symptoms progress in severity, annual mortality may reach 30% to 40%.¹ Mortality remains high in patients maintained on standard therapy (including an angiotensin-converting enzyme [ACE] inhibitor^1,2) in large part because of the impact of sudden cardiac death in patients with heart failure.³ Additional therapeutic agents are needed to improve survival for this patient population.^1,3

Heart failure may be associated with either systolic or diastolic dysfunction. Patients with systolic dysfunction have poor left ventricular wall motion and an ejection fraction of less than 40%. Patients with diastolic dysfunction have a normal ejection fraction but a noncompliant left ventricular wall that impairs diastolic filling. The etiology of heart failure may be ischemic or nonischemic.¹ Coronary artery disease,¹⁴ and hypertension are common etiologic factors.¹

Beta-blocker therapy results in an acute negative inotropic effect.⁵ This acute effect led to the conventional belief that b-blockers were contraindicated in patients with heart failure.^2,4 However, clinical trials have demonstrated beneficial effects of long-term b-blockade in patients with heart failure, including improvements in clinical status,⁶ heart failure symptoms,¹ ventricular function,⁵ disease progression,^7,8 and hospitalization.^8-12 As a result of the identification of these beneficial effects, clinical trials have been conducted to assess the impact of b-blocker therapy on mortality.^3,4,9,11-15 Our goal for this systematic review was to synthesize available data and determine whether b-blocker therapy reduces the risk of mortality in patients with systolic dysfunction.

Methods

Literature Search

Two independent searches of MEDLINE (1966 to February 2000) were conducted to identify published randomized double-blind controlled trials of b-blockers in patients with chronic heart failure. We used both Medical Subject Heading (MeSH) terms and key words to provide a broad search. The results were limited to English language articles involving human subjects and the publication type “clinical trial.” In addition to MEDLINE, we searched several other databases for relevant literature, including International Pharmaceutical Abstracts, Iowa Drug Information Service, Current Contents, and The Cochrane Library. We examined references from published clinical trials and reviewed articles to identify further potential articles for inclusion in our evaluation. Finally, the manufacturers of metoprolol (Novartis Pharmaceuticals Corporation, East Hanover, NJ), carvedilol (SmithKline Beecham Pharmaceuticals, Philadelphia, Pa), and bisoprolol (Lederle Laboratories, Philadelphia, Pa) were contacted in an effort to ensure that pertinent articles were identified and included.

We determined the criteria for inclusion before conducting the literature searches. Clinical trials were included if they were randomized, double-blinded, and controlled; if the patients had a diagnosis of systolic heart failure; and if they assessed mortality as a primary or secondary end point. They were excluded if the study duration was shorter than 3 months or if publication occurred before 1975.

We examined 280 articles or abstracts for our review. Although several clinical trials were not included because of study design issues (eg, lack of randomization or blinding), the majority (approximately 95%) were excluded because they lacked a mortality end point. None of the identified trials used an active comparator; therefore, all trials were placebo controlled. Ultimately, we identified 6 clinical trials^3,9,11-14 meeting the criteria and used them in our analysis.

Quality Assessment

Each study was evaluated using the instrument designed by Jadad and colleagues¹⁶ that rates study quality from 0 (worst) to 5 (best). Two of the 6 clinical trials identified received a score of^5,11,14 while the remaining 4 received a score of^4,3,9,12,13 The reasons for lower scores included a lack of description of the randomization process¹³ or of which patients withdrew from the trials.^3,9,12

Analysis of Data

We critically analyzed the mortality end points of each study. Two independent reviewers calculated relative risk, relative risk reduction (RRR), absolute risk reduction, and number needed to treat (NNT) for total mortality. The NNT is a numerical representation of the number of patients that must be treated to prevent 1 adverse outcome (in this case death) in the population studied for the duration of the trial. Direct comparisons of NNT between studies should be avoided; since the benefit cannot be assumed to have occurred in even increments throughout the trial, the NNT serves only as an alternate form of data presentation. The results of these 4 calculations are provided in an effort to allow for additional assessment of the clinical trials evaluated.

Homogeneity of effect was assessed by calculating a Q statistic, where the number of degrees of freedom is equal to the number of trials included in the analysis minus 1. A P value of less than .05 was considered statistically significant. The pooled estimate of effect was calculated using the Mantel-Haenszel method, which assumes a fixed effect model. The variance of the individual risk assessments, the weight of each study, and ultimately the summary odds ratio (OR) and 95% confidence interval (CI) were calculated for mortality end points when studies were homogenous using this method as previously described.¹⁷

Results

Study Characteristics

Of those identified, 2 trials each evaluated bisoprolol (the Cardiac Insufficiency Bisoprolol Study [CIBIS]¹¹ and the Cardiac Insufficiency Bisoprolol Study II [CIBIS-II]¹²), carvedilol (the Australia/New Zealand Heart Failure Research Collaborative Trial [ANZ]⁹ and the US Carvedilol Heart Failure Program [US Carvedilol]¹³), and metoprolol (the Metoprolol CR/XL Randomised Intervention Trial in Congestive Heart Failure [MERIT-HF]3 and the Metoprolol in Dilated Cardiomyopathy [MDC]¹⁴ trial). A total of 9335 patients were enrolled in the 6 trials.^3,9,11-14 With an average 7.5% withdrawal rate during the active run-in phases,^9,13,14 9171 patients were ultimately randomized to receive either b-blocker or placebo therapy. The average withdrawal rate after randomization was 15.8% for patients receiving active treatment compared with 17.2% for patients receiving placebo (P >.05 for each individual trial).^3,9,11-14

Patient Characteristics

There were no statistically significant differences between the baseline characteristics of the patients receiving active treatment versus placebo. The mean age of patients was 60 years, with an average of 78% men. All patients were receiving pharmacologic treatment for their heart failure before study enrollment. Although the studies used various ejection fraction requirements, the average ejection fraction ranged from 22% to 29%.^3,9,11-14 In all studies the majority of patients had New York Heart Association (NYHA) class II or III heart failure.^3,9,11-14Table 1 shows the study characteristics for each included clinical trial and the baseline characteristics of enrolled patients.

End Points

Mortality offers an objective and clear end point for clinical trials. It supersedes other safety and efficacy end points.¹Table 2 provides a summary of selected mortality end points and the notation of the statistical significance of the end point in each trial. The calculated risk reduction demonstrated for total mortality in trials generating a statistically significant mortality benefit is given.

A statistically and clinically significant decrease in mortality was reported in CIBIS-II, US Carvedilol, and MERIT-HF. These clinical trials included a total of 7732 patients receiving 1 of 3 different b-blockers.^3,12,13 A nonsignificant reduction in total mortality was noted in patients receiving b-blocker therapy in CIBIS-I and ANZ,^9,11 while a nonsignificant increase in total mortality was reported in patients receiving b-blocker therapy in the MDC trial.¹⁴ These clinical trials included a total of 1439 patients and were not powered to assess this end point.^9,11,14

Analysis revealed that the 6 clinical trials are homogenous in total mortality (Q=9.3). This indicates that the trials are measuring an effect of the same size and that differences in issues such as study design and patient population would not be expected to have an appreciable impact on the outcome of meta-analysis. When the results of all 6 trials were pooled, a statistically significant reduction in the risk of total mortality was noted (ORMH=0.66; 95% CI, 0.58-0.75). The Figure 1 is a representation of the effect of b-blocker therapy on total mortality.

Five of the 6 trials reported mortality caused by sudden death.^3,11-14 Both MERIT-HF and CIBIS-II reported significant decreases in sudden death in patients receiving b-blocker therapy.^3,12 Of the remaining 3 studies, the MDC trial showed a nonsignificant increase in sudden death,¹⁴ and CIBIS and US Carvedilo¹ showed a nonsignificant decrease in sudden death.^11,13 US Carvedilol had a larger RRR with respect to sudden death than MERIT-HF or CIBIS-II; however, statistical significance was not reached.¹³ Analysis revealed that these 5 trials were homogenous (Q=7.0) with respect to sudden death. When their results were pooled, a statistically significant reduction in the risk of sudden death was noted (ORMH=0.61; 95% CI, 0.5-0.75). All the included trials evaluated cardiovascular mortality as an end point; however, analysis revealed that the trials are heterogenous with respect to cardiovascular mortality (Q=12.7), and therefore a summary statistic would be misleading.

With analyses of the impact of disease severity and etiology of heart failure within the clinical trials, we attempted to determine the response of various subgroups to b-blocker therapy. Patients with NYHA class IV heart failure represented 3.6% and 17% of the MERIT-HF and CIBIS-II study populations, respectively. Subgroup analyses in these studies indicated that patients with NYHA class III heart failure had a nonsignificant but greater RRR in mortality than did those of NYHA class IV^3,12 and II heart failure.³ US Carvedilol did not categorize patients according to NYHA class but rather stratified them on the basis of exercise test performance into mild, moderate, and severe categories. There was no difference in mortality benefit with b-blocker therapy between these 3 defined groups.¹³

The results of CIBIS-II, MERIT-HF, and US Carvedilol stratified by heart failure etiology indicated a decrease in mortality in patients with both ischemic and nonischemic heart failure. All 3 studies noted a significantly reduced mortality in patients with ischemic heart failure.^3,12,13 US Carvedilol demonstrated a statistically significant decrease in mortality in patients with nonischemic heart failure receiving b-blocker therapy,¹³ while MERIT-HF and CIBIS-II did not.^3,12

Discussion

Our findings indicate that there is a significant reduction in the risk of total mortality in patients receiving b-blocker therapy. US Carvedilol reported the largest RRR in mortality with b-blocker therapy in patients with heart failure¹³ compared with the other large trials that found a significant decrease in the risk of death.^3,12 The RRR noted with carvedilol therapy approaches twice that reported with bisoprolol or extended-release metoprolol, although the absolute risk reductions are similar. The shorter study period employed in US Carvedilol compared with the other trials may provide some explanation for the large benefit noted, especially if the benefit of b-blocker therapy in this patient population is maximized early in the course of therapy. Also, the active run-in phase resulted in the withdrawal of 8.6% of the patients before randomization. Eliminating patients intolerant to carvedilol may have created the appearance of an elevated response to therapy compared with that seen in trials that did not employ an active run-in phase. Moreover, the patients included in US Carvedilol had a lower mean ejection fraction and a greater prevalence of nonischemic heart failure than the other clinical trials reviewed. The most intriguing potential explanation for the apparent greater mortality benefit with carvedilol compared with bisoprolol or metoprolol is the additional pharmacologic effects associated with carvedilol. In addition to b1-adrenergic blockade, carvedilol blocks a1- and b2-adrenergic receptors and has antioxidant effects.¹⁸ The contribution of these pharmacologic properties to the mortality benefit demonstrated with carvedilol therapy is currently unknown.

The limited number of patients with NYHA class IV heart failure enrolled in MERIT-HF make comparative benefit speculations unreliable. Also, the disease severity stratification used in US Carvedilol is not standardized, and the sample size representing the most severe patients was less than 10% of the study population. Based on the data currently available, it is not known if the magnitude of benefit with b-blocker therapy is related to disease severity.

The number of patients with nonischemic heart failure receiving b-blocker therapy was largest in MERIT-HF,³ followed by US Carvedilol¹³ and CIBIS-II.¹² Of these, US Carvedilol was the only trial to demonstrate a statistically significant decrease in mortality in patients with nonischemic heart failure receiving b-blocker therapy.¹³ This may indicate that greater benefit can be derived with carvedilol than with bisoprolol or metoprolol therapy in patients with nonischemic heart failure. Further data are needed to fully assess the impact of heart failure etiology on mortality following b-blocker therapy.

Sudden death, which is often attributed to ventricular tachycardia or fibrillation,¹⁹ is a major cause of mortality in patients with heart failure. Although large trials evaluating metoprolol and carvedilol reported a significant reduction in sudden death,^3,12 the use of concomitant antiarrhythmic medications varied between each trial evaluated. For example, approximately 15% of patients in CIBIS-II were receiving therapy with amiodarone,¹² which has been shown to decrease the occurrence of sudden death in patients with heart failure.²⁰ However, amiodarone therapy was evenly distributed between patients receiving bisoprolol and placebo1² and would not be expected to have a untoward effect on the assessment of sudden death.

Relative contraindications to b-blocker use are reflected in the exclusion criteria used in the trials discussed and include clinical instability,^3,11-13,14 second- or third-degree heart block in the absence of an implantable pacemaker,^3,9,13 low blood pressure,^3,9,12 low heart rate,^9,12,13 or treatment-requiring obstructive airway disease.^9,12,14 All the trials we reviewed indicated that b-blocker therapy in patients with heart failure does not result in a clinically significant decrease in systolic or diastolic blood pressure. As expected, however, a decrease in heart rate of approximately 10 to 15 beats per minute is likely to occur.^3,9,11-14 It is important to note that in US Carvedilol, increased dizziness in the carvediloltreated patients was noted compared with placebo. Dizziness was most pronounced after a dosage increase and dissipated with continued use.¹³ Target doses reached with b-blocker and placebo therapies were similar,^9,13,14 further confirming the tolerability of these agents with progressive dose titration.

Patients with Heart Failure

On the basis of studies reviewed, it is reasonable to anticipate a 15% to 20% withdrawal rate from b-blocker therapy in patients with systolic heart failure. The reasons for withdrawal will most commonly be worsening heart failure or side effects, such as fatigue and dyspnea. It is reassuring that the rates of withdrawal in the trials were similar for placebo and active therapies, suggesting that b-blocker therapy was not the primary cause of withdrawal.

Questions remain concerning the effects of b-blockade in patients with heart failure. The Carvedilol Prospective Randomized Cumulative Survival Trial was designed to evaluate the mortality effect of carvedilol versus placebo in patients with severe heart failure.^4,21,22 Although not published at the time of this review, the trial was discontinued early as a result of a large and consistent decrease in mortality in patients receiving carvedilol.²³ The Carvedilol ACE Inhibitors Remodeling Mild Heart Failure Evaluation trial is designed to evaluate carvedilol in asymptomatic patients with left ventricular dysfunction.^12,21 Also, the Betablocker Evaluation Survival Trial evaluated bucindilol versus placebo in patients with moderate to severe heart failure.^4,15,21 This trial was discontinued early because of lack of mortality benefit.²⁴

Heart failure primarily affects the elderly population older than 65 years.¹ Currently published trials have not included large numbers of elderly patients,^12,21,22 and therefore, extrapolation of the data to this patient population should be done with caution. Also, the effects of race and ethnicity have yet to be systematically addressed.²¹ Studies designed to assess the impact of b-blockade in patients with heart failure depending on the etiology of their disease would allow for further maximization of benefit. The Carvedilol Post-Infarction Survival Control in Left Ventricular Dysfunction trial²² was designed to evaluate patients with heart failure following an acute myocardial infarction.^12,22 Clinical trials are needed to determine optimal dosing regimens.³

Comparative clinical trials with a mortality end point have not been conducted. Efficacy comparisons between selective and nonselective b-blockers are necessary to quantify the survival benefit.³ The Betaxolol Versus Carvedilol in Chronic Heart Failure study is designed to compare betaxolol with carvedilol in patients with NYHA class II or III heart failure.⁴ The comparative efficacy of carvedilol with the b1-selective agent metoprolol is being assessed in the Carvedilol and Metoprolol European Trial.^4,21,22

Conclusions

All stable patients with NYHA class II and III heart failure should be considered for b-blocker therapy. Therapy should be initiated only in stable patients using a low dose of a b-blocker. This dose may be subsequently gradually titrated upward by doubling the current dose every 1 to 2 weeks on the basis of clinical response and patient tolerability.

Acknowledgments

We would like to acknowledge the assistance of Deborah S. Carson, PharmD, BCPS, and Ché Jordan, PharmD.

Related Resources

• American Heart Association-congestive heart failure www.americanheart.org/chf
• WebMD-Heart Disease Center www.my.webmd.com/condition_center/cvd
• Heart Failure Society of America www.hfsa.org Organizers describe the site as “a forum for all those interested in heart function, heart failure, and congestive heart failure (CHF) research and patient care.”

The annual mortality in patients with mildly symptomatic heart failure ranges from 5% to 10%. However, as symptoms progress in severity, annual mortality may reach 30% to 40%.¹ Mortality remains high in patients maintained on standard therapy (including an angiotensin-converting enzyme [ACE] inhibitor^1,2) in large part because of the impact of sudden cardiac death in patients with heart failure.³ Additional therapeutic agents are needed to improve survival for this patient population.^1,3

Heart failure may be associated with either systolic or diastolic dysfunction. Patients with systolic dysfunction have poor left ventricular wall motion and an ejection fraction of less than 40%. Patients with diastolic dysfunction have a normal ejection fraction but a noncompliant left ventricular wall that impairs diastolic filling. The etiology of heart failure may be ischemic or nonischemic.¹ Coronary artery disease,¹⁴ and hypertension are common etiologic factors.¹

Beta-blocker therapy results in an acute negative inotropic effect.⁵ This acute effect led to the conventional belief that b-blockers were contraindicated in patients with heart failure.^2,4 However, clinical trials have demonstrated beneficial effects of long-term b-blockade in patients with heart failure, including improvements in clinical status,⁶ heart failure symptoms,¹ ventricular function,⁵ disease progression,^7,8 and hospitalization.^8-12 As a result of the identification of these beneficial effects, clinical trials have been conducted to assess the impact of b-blocker therapy on mortality.^3,4,9,11-15 Our goal for this systematic review was to synthesize available data and determine whether b-blocker therapy reduces the risk of mortality in patients with systolic dysfunction.

Methods

Literature Search

Two independent searches of MEDLINE (1966 to February 2000) were conducted to identify published randomized double-blind controlled trials of b-blockers in patients with chronic heart failure. We used both Medical Subject Heading (MeSH) terms and key words to provide a broad search. The results were limited to English language articles involving human subjects and the publication type “clinical trial.” In addition to MEDLINE, we searched several other databases for relevant literature, including International Pharmaceutical Abstracts, Iowa Drug Information Service, Current Contents, and The Cochrane Library. We examined references from published clinical trials and reviewed articles to identify further potential articles for inclusion in our evaluation. Finally, the manufacturers of metoprolol (Novartis Pharmaceuticals Corporation, East Hanover, NJ), carvedilol (SmithKline Beecham Pharmaceuticals, Philadelphia, Pa), and bisoprolol (Lederle Laboratories, Philadelphia, Pa) were contacted in an effort to ensure that pertinent articles were identified and included.

We determined the criteria for inclusion before conducting the literature searches. Clinical trials were included if they were randomized, double-blinded, and controlled; if the patients had a diagnosis of systolic heart failure; and if they assessed mortality as a primary or secondary end point. They were excluded if the study duration was shorter than 3 months or if publication occurred before 1975.

We examined 280 articles or abstracts for our review. Although several clinical trials were not included because of study design issues (eg, lack of randomization or blinding), the majority (approximately 95%) were excluded because they lacked a mortality end point. None of the identified trials used an active comparator; therefore, all trials were placebo controlled. Ultimately, we identified 6 clinical trials^3,9,11-14 meeting the criteria and used them in our analysis.

Quality Assessment

Each study was evaluated using the instrument designed by Jadad and colleagues¹⁶ that rates study quality from 0 (worst) to 5 (best). Two of the 6 clinical trials identified received a score of^5,11,14 while the remaining 4 received a score of^4,3,9,12,13 The reasons for lower scores included a lack of description of the randomization process¹³ or of which patients withdrew from the trials.^3,9,12

Analysis of Data

We critically analyzed the mortality end points of each study. Two independent reviewers calculated relative risk, relative risk reduction (RRR), absolute risk reduction, and number needed to treat (NNT) for total mortality. The NNT is a numerical representation of the number of patients that must be treated to prevent 1 adverse outcome (in this case death) in the population studied for the duration of the trial. Direct comparisons of NNT between studies should be avoided; since the benefit cannot be assumed to have occurred in even increments throughout the trial, the NNT serves only as an alternate form of data presentation. The results of these 4 calculations are provided in an effort to allow for additional assessment of the clinical trials evaluated.

Homogeneity of effect was assessed by calculating a Q statistic, where the number of degrees of freedom is equal to the number of trials included in the analysis minus 1. A P value of less than .05 was considered statistically significant. The pooled estimate of effect was calculated using the Mantel-Haenszel method, which assumes a fixed effect model. The variance of the individual risk assessments, the weight of each study, and ultimately the summary odds ratio (OR) and 95% confidence interval (CI) were calculated for mortality end points when studies were homogenous using this method as previously described.¹⁷

Results

Study Characteristics

Of those identified, 2 trials each evaluated bisoprolol (the Cardiac Insufficiency Bisoprolol Study [CIBIS]¹¹ and the Cardiac Insufficiency Bisoprolol Study II [CIBIS-II]¹²), carvedilol (the Australia/New Zealand Heart Failure Research Collaborative Trial [ANZ]⁹ and the US Carvedilol Heart Failure Program [US Carvedilol]¹³), and metoprolol (the Metoprolol CR/XL Randomised Intervention Trial in Congestive Heart Failure [MERIT-HF]3 and the Metoprolol in Dilated Cardiomyopathy [MDC]¹⁴ trial). A total of 9335 patients were enrolled in the 6 trials.^3,9,11-14 With an average 7.5% withdrawal rate during the active run-in phases,^9,13,14 9171 patients were ultimately randomized to receive either b-blocker or placebo therapy. The average withdrawal rate after randomization was 15.8% for patients receiving active treatment compared with 17.2% for patients receiving placebo (P >.05 for each individual trial).^3,9,11-14

Patient Characteristics

There were no statistically significant differences between the baseline characteristics of the patients receiving active treatment versus placebo. The mean age of patients was 60 years, with an average of 78% men. All patients were receiving pharmacologic treatment for their heart failure before study enrollment. Although the studies used various ejection fraction requirements, the average ejection fraction ranged from 22% to 29%.^3,9,11-14 In all studies the majority of patients had New York Heart Association (NYHA) class II or III heart failure.^3,9,11-14Table 1 shows the study characteristics for each included clinical trial and the baseline characteristics of enrolled patients.

End Points

Mortality offers an objective and clear end point for clinical trials. It supersedes other safety and efficacy end points.¹Table 2 provides a summary of selected mortality end points and the notation of the statistical significance of the end point in each trial. The calculated risk reduction demonstrated for total mortality in trials generating a statistically significant mortality benefit is given.

A statistically and clinically significant decrease in mortality was reported in CIBIS-II, US Carvedilol, and MERIT-HF. These clinical trials included a total of 7732 patients receiving 1 of 3 different b-blockers.^3,12,13 A nonsignificant reduction in total mortality was noted in patients receiving b-blocker therapy in CIBIS-I and ANZ,^9,11 while a nonsignificant increase in total mortality was reported in patients receiving b-blocker therapy in the MDC trial.¹⁴ These clinical trials included a total of 1439 patients and were not powered to assess this end point.^9,11,14

Analysis revealed that the 6 clinical trials are homogenous in total mortality (Q=9.3). This indicates that the trials are measuring an effect of the same size and that differences in issues such as study design and patient population would not be expected to have an appreciable impact on the outcome of meta-analysis. When the results of all 6 trials were pooled, a statistically significant reduction in the risk of total mortality was noted (ORMH=0.66; 95% CI, 0.58-0.75). The Figure 1 is a representation of the effect of b-blocker therapy on total mortality.

Five of the 6 trials reported mortality caused by sudden death.^3,11-14 Both MERIT-HF and CIBIS-II reported significant decreases in sudden death in patients receiving b-blocker therapy.^3,12 Of the remaining 3 studies, the MDC trial showed a nonsignificant increase in sudden death,¹⁴ and CIBIS and US Carvedilo¹ showed a nonsignificant decrease in sudden death.^11,13 US Carvedilol had a larger RRR with respect to sudden death than MERIT-HF or CIBIS-II; however, statistical significance was not reached.¹³ Analysis revealed that these 5 trials were homogenous (Q=7.0) with respect to sudden death. When their results were pooled, a statistically significant reduction in the risk of sudden death was noted (ORMH=0.61; 95% CI, 0.5-0.75). All the included trials evaluated cardiovascular mortality as an end point; however, analysis revealed that the trials are heterogenous with respect to cardiovascular mortality (Q=12.7), and therefore a summary statistic would be misleading.

With analyses of the impact of disease severity and etiology of heart failure within the clinical trials, we attempted to determine the response of various subgroups to b-blocker therapy. Patients with NYHA class IV heart failure represented 3.6% and 17% of the MERIT-HF and CIBIS-II study populations, respectively. Subgroup analyses in these studies indicated that patients with NYHA class III heart failure had a nonsignificant but greater RRR in mortality than did those of NYHA class IV^3,12 and II heart failure.³ US Carvedilol did not categorize patients according to NYHA class but rather stratified them on the basis of exercise test performance into mild, moderate, and severe categories. There was no difference in mortality benefit with b-blocker therapy between these 3 defined groups.¹³

The results of CIBIS-II, MERIT-HF, and US Carvedilol stratified by heart failure etiology indicated a decrease in mortality in patients with both ischemic and nonischemic heart failure. All 3 studies noted a significantly reduced mortality in patients with ischemic heart failure.^3,12,13 US Carvedilol demonstrated a statistically significant decrease in mortality in patients with nonischemic heart failure receiving b-blocker therapy,¹³ while MERIT-HF and CIBIS-II did not.^3,12

Discussion

Our findings indicate that there is a significant reduction in the risk of total mortality in patients receiving b-blocker therapy. US Carvedilol reported the largest RRR in mortality with b-blocker therapy in patients with heart failure¹³ compared with the other large trials that found a significant decrease in the risk of death.^3,12 The RRR noted with carvedilol therapy approaches twice that reported with bisoprolol or extended-release metoprolol, although the absolute risk reductions are similar. The shorter study period employed in US Carvedilol compared with the other trials may provide some explanation for the large benefit noted, especially if the benefit of b-blocker therapy in this patient population is maximized early in the course of therapy. Also, the active run-in phase resulted in the withdrawal of 8.6% of the patients before randomization. Eliminating patients intolerant to carvedilol may have created the appearance of an elevated response to therapy compared with that seen in trials that did not employ an active run-in phase. Moreover, the patients included in US Carvedilol had a lower mean ejection fraction and a greater prevalence of nonischemic heart failure than the other clinical trials reviewed. The most intriguing potential explanation for the apparent greater mortality benefit with carvedilol compared with bisoprolol or metoprolol is the additional pharmacologic effects associated with carvedilol. In addition to b1-adrenergic blockade, carvedilol blocks a1- and b2-adrenergic receptors and has antioxidant effects.¹⁸ The contribution of these pharmacologic properties to the mortality benefit demonstrated with carvedilol therapy is currently unknown.

The limited number of patients with NYHA class IV heart failure enrolled in MERIT-HF make comparative benefit speculations unreliable. Also, the disease severity stratification used in US Carvedilol is not standardized, and the sample size representing the most severe patients was less than 10% of the study population. Based on the data currently available, it is not known if the magnitude of benefit with b-blocker therapy is related to disease severity.

The number of patients with nonischemic heart failure receiving b-blocker therapy was largest in MERIT-HF,³ followed by US Carvedilol¹³ and CIBIS-II.¹² Of these, US Carvedilol was the only trial to demonstrate a statistically significant decrease in mortality in patients with nonischemic heart failure receiving b-blocker therapy.¹³ This may indicate that greater benefit can be derived with carvedilol than with bisoprolol or metoprolol therapy in patients with nonischemic heart failure. Further data are needed to fully assess the impact of heart failure etiology on mortality following b-blocker therapy.

Sudden death, which is often attributed to ventricular tachycardia or fibrillation,¹⁹ is a major cause of mortality in patients with heart failure. Although large trials evaluating metoprolol and carvedilol reported a significant reduction in sudden death,^3,12 the use of concomitant antiarrhythmic medications varied between each trial evaluated. For example, approximately 15% of patients in CIBIS-II were receiving therapy with amiodarone,¹² which has been shown to decrease the occurrence of sudden death in patients with heart failure.²⁰ However, amiodarone therapy was evenly distributed between patients receiving bisoprolol and placebo1² and would not be expected to have a untoward effect on the assessment of sudden death.

Relative contraindications to b-blocker use are reflected in the exclusion criteria used in the trials discussed and include clinical instability,^3,11-13,14 second- or third-degree heart block in the absence of an implantable pacemaker,^3,9,13 low blood pressure,^3,9,12 low heart rate,^9,12,13 or treatment-requiring obstructive airway disease.^9,12,14 All the trials we reviewed indicated that b-blocker therapy in patients with heart failure does not result in a clinically significant decrease in systolic or diastolic blood pressure. As expected, however, a decrease in heart rate of approximately 10 to 15 beats per minute is likely to occur.^3,9,11-14 It is important to note that in US Carvedilol, increased dizziness in the carvediloltreated patients was noted compared with placebo. Dizziness was most pronounced after a dosage increase and dissipated with continued use.¹³ Target doses reached with b-blocker and placebo therapies were similar,^9,13,14 further confirming the tolerability of these agents with progressive dose titration.

Patients with Heart Failure

On the basis of studies reviewed, it is reasonable to anticipate a 15% to 20% withdrawal rate from b-blocker therapy in patients with systolic heart failure. The reasons for withdrawal will most commonly be worsening heart failure or side effects, such as fatigue and dyspnea. It is reassuring that the rates of withdrawal in the trials were similar for placebo and active therapies, suggesting that b-blocker therapy was not the primary cause of withdrawal.

Questions remain concerning the effects of b-blockade in patients with heart failure. The Carvedilol Prospective Randomized Cumulative Survival Trial was designed to evaluate the mortality effect of carvedilol versus placebo in patients with severe heart failure.^4,21,22 Although not published at the time of this review, the trial was discontinued early as a result of a large and consistent decrease in mortality in patients receiving carvedilol.²³ The Carvedilol ACE Inhibitors Remodeling Mild Heart Failure Evaluation trial is designed to evaluate carvedilol in asymptomatic patients with left ventricular dysfunction.^12,21 Also, the Betablocker Evaluation Survival Trial evaluated bucindilol versus placebo in patients with moderate to severe heart failure.^4,15,21 This trial was discontinued early because of lack of mortality benefit.²⁴

Heart failure primarily affects the elderly population older than 65 years.¹ Currently published trials have not included large numbers of elderly patients,^12,21,22 and therefore, extrapolation of the data to this patient population should be done with caution. Also, the effects of race and ethnicity have yet to be systematically addressed.²¹ Studies designed to assess the impact of b-blockade in patients with heart failure depending on the etiology of their disease would allow for further maximization of benefit. The Carvedilol Post-Infarction Survival Control in Left Ventricular Dysfunction trial²² was designed to evaluate patients with heart failure following an acute myocardial infarction.^12,22 Clinical trials are needed to determine optimal dosing regimens.³

Comparative clinical trials with a mortality end point have not been conducted. Efficacy comparisons between selective and nonselective b-blockers are necessary to quantify the survival benefit.³ The Betaxolol Versus Carvedilol in Chronic Heart Failure study is designed to compare betaxolol with carvedilol in patients with NYHA class II or III heart failure.⁴ The comparative efficacy of carvedilol with the b1-selective agent metoprolol is being assessed in the Carvedilol and Metoprolol European Trial.^4,21,22

Conclusions

All stable patients with NYHA class II and III heart failure should be considered for b-blocker therapy. Therapy should be initiated only in stable patients using a low dose of a b-blocker. This dose may be subsequently gradually titrated upward by doubling the current dose every 1 to 2 weeks on the basis of clinical response and patient tolerability.

Acknowledgments

We would like to acknowledge the assistance of Deborah S. Carson, PharmD, BCPS, and Ché Jordan, PharmD.

Related Resources

• American Heart Association-congestive heart failure www.americanheart.org/chf
• WebMD-Heart Disease Center www.my.webmd.com/condition_center/cvd
• Heart Failure Society of America www.hfsa.org Organizers describe the site as “a forum for all those interested in heart function, heart failure, and congestive heart failure (CHF) research and patient care.”

Research into the efficacy and effectiveness of treatments for depression has grown exponentially during the past several decades. Numerous studies show that disorders like major depression and dysthymia can be treated successfully with antidepressant medication and/or psychotherapy.¹ Therapeutic effect sizes from efficacy trials are 18% for antidepressants compared with placebo and 26% for psychotherapy compared with no treatment.^2-4 However, when these interventions are administered in real world settings, much lower response rates have been found.^5-8

The evidence from these studies indicates that in controlled settings and with highly selected patients, current methods for treating depression are efficacious. However, there are still many people who do not respond to current guideline-based treatment and many segments of the population that have not been included in clinical studies. We briefly review the research and discuss which populations respond to current treatment, which do not, and which require further study.

Definitions of treatment and treatment response

For the purposes of this paper we defined “treatment” as antidepressant medication or psychotherapy. Because the majority of psychotherapy research has focused on cognitive-behavioral and interpersonal therapies, we use psychotherapy as a generic term for these 2 types of treatment. Most depression treatment studies have defined “response” as no longer meeting Diagnostic and Statistical Manual of Mental Disorders criteria for a disorder and exhibiting a statistically significant change on a symptom severity scale (usually a decrement of at least 50%). For the purpose of this paper, we will consider response to mean a significant change in depression severity.

Who responds to treatment?

The authors of several studies have examined patient traits linked to treatment response.⁹ For the most part, patients who are educated, are experiencing uncomplicated depression, have had 2 or fewer previous episodes of depression, and have faith in their treatment typically respond to guideline-level treatment.¹⁰ Depression experts once believed that only patients of this type were likely to show a treatment response and that those who were older, a member of an ethnic minority, or from lower socioeconomic groups were less likely to respond to guideline treatment. Recent research shows that people from low socioeconomic backgrounds can respond to existing treatment, provided they have access to quality care.^11,12 Several studies specifically about treating depression in older adults have found positive effects, both in university and primary care settings.^4,6 Although research on ethnic minorities is scarce and focused primarily on Latinos and African Americans, the literature indicates that members of these ethnic groups do respond to psychotherapy. With respect to medication treatment, research on the pharmacokinetics of antidepressants in African Americans, Asians, and certain Latino groups indicates that dosages may need to be altered to reduce side effects.¹³

Current research also indicates that patients with complicated psychiatric presentations can respond to guideline-level treatment. For example, although patients more severe depressive symptoms may not respond to monotherapies as well as patients with milder symptoms,¹⁴ they generally respond well to combination treatments.⁸ The presence of Axis II features and comorbid anxiety or substance abuse does not necessarily have an impact on treatment outcome, although much of the data focus primarily on acute care of depression.¹⁴ Finally, even patients with cognitive impairment can respond to both medication and psychotherapy for depression.¹⁵

Who does not respond?

Growing evidence suggests that while effective treatments for depression do exist, they are not helpful for everyone. Treatment nonresponders fall into 2 categories: those who are treatment resistant and those who simply resist treatment.

Patients who are treatment resistant have been given an adequate course of either antidepressant medication or psychotherapy and have either no response or a limited response to treatment. Research investigating predictors of treatment failure indicates that several psychiatric and psychosocial variables are related to treatment resistance. Patients with more psychosocial stressors and less social support are more likely to show a limited response to treatment,⁹ as are patients with a greater number of previous depressive episodes.¹⁶ This may be due in part to increased feelings of hopelessness¹⁷ or lack of faith in treatment,¹⁸ both found to contribute to treatment resistance. Comorbid Axis II features, such as borderline and dependent personality traits tend to predict a decreased treatment response, in part because of the poor psychologic resources these patients have to cope with their symptoms.¹⁵ Such patients may benefit from additional interventions to alleviate their symptoms, such as case management, longer courses of psychotherapy, and multiple medications.

Patients who resist treatment include those who despite being identified as depressed and offered treatment, never follow through with the treatment plan. The primary reasons why patients do not adhere to treatment for depression include stigma concerns¹⁹ and the belief that depressive symptoms are not significant enough to treat.²⁰ Other factors, such as cognitive impairment, using multiple medications,²¹ comorbid medical illnesses, sensitivity to side effects,¹³ cost of mental health services,²² location of mental health services, and cross-cultural issues²³ may also have an impact on patient willingness to accept treatment. Once in treatment, psychologic factors such as self-efficacy²⁴ and readiness for change²⁵ can influence whether a patient will adhere to a treatment plan. There is early evidence that educational interventions or treatment management programs may benefit patients with acceptance or adherence issues.²⁶

Who do we need to study?

Several subgroups of patients have typically been excluded from treatment research. In particular, patients with coexisting Axis I disorders are routinely excluded from many treatment studies because of the complications concerning the management of separate conditions. However, the National Comorbidity Survey²⁷ has shown that depression often co-occurs with many other disorders, including substance abuse, psychosis, and anxiety disorders. Although past studies have included patients with comorbid symptoms of substance use and anxiety, little is known about the impact these interventions have when full-blown comorbid disorders are present.

Samples included in recent studies of depression treatment are becoming more diverse with respect to age and minority representation. However, little is known regarding the specific response to treatment in these populations or how response rates compare with those found with more traditional study populations. This is important work to undertake, given that certain age and minority groups have been found to have varying responses to existing treatments. For example, given the pharmacokinetic complications that have been associated with antidepressant medications in ethnic minority populations, investigating the effectiveness of existing interventions in these populations is also important.¹³

Along similar lines, preliminary research suggests that older people take longer to respond to antidepressant therapies and require smaller doses to prevent toxic effects.⁴ Other age groups, such as children and adolescents are rarely studied, though this may change as the result of new National Institutes of Health guidelines on the inclusion of children as research subjects. Also, people seeking treatment in medical organizations other than primary care medicine or psychiatry have not been systematically studied. For example, the rates for depression in women seen in obstetrics/gynecology are quite high, but there are no published treatment studies with this population.²⁸ Finally, patients who live in areas where care is hard to access (ie, rural populations) are currently being studied with promising results, yet to date there are no published outcomes.

Conclusions

The current literature shows that depression can be treated in many patients, but treatment response largely depends on the chronicity of the illness and the level of psychosocial stress faced by the patient. Future research should focus on how to best treat patients who tend not to respond to or accept existing treatment and should also examine the effectiveness of existing interventions for special populations who have not been included in past research. Thus far the evidence regarding the effectiveness of depression treatment is very promising, and the results of previous research will be useful in informing future work.

Research into the efficacy and effectiveness of treatments for depression has grown exponentially during the past several decades. Numerous studies show that disorders like major depression and dysthymia can be treated successfully with antidepressant medication and/or psychotherapy.¹ Therapeutic effect sizes from efficacy trials are 18% for antidepressants compared with placebo and 26% for psychotherapy compared with no treatment.^2-4 However, when these interventions are administered in real world settings, much lower response rates have been found.^5-8

The evidence from these studies indicates that in controlled settings and with highly selected patients, current methods for treating depression are efficacious. However, there are still many people who do not respond to current guideline-based treatment and many segments of the population that have not been included in clinical studies. We briefly review the research and discuss which populations respond to current treatment, which do not, and which require further study.

Definitions of treatment and treatment response

For the purposes of this paper we defined “treatment” as antidepressant medication or psychotherapy. Because the majority of psychotherapy research has focused on cognitive-behavioral and interpersonal therapies, we use psychotherapy as a generic term for these 2 types of treatment. Most depression treatment studies have defined “response” as no longer meeting Diagnostic and Statistical Manual of Mental Disorders criteria for a disorder and exhibiting a statistically significant change on a symptom severity scale (usually a decrement of at least 50%). For the purpose of this paper, we will consider response to mean a significant change in depression severity.

Who responds to treatment?

The authors of several studies have examined patient traits linked to treatment response.⁹ For the most part, patients who are educated, are experiencing uncomplicated depression, have had 2 or fewer previous episodes of depression, and have faith in their treatment typically respond to guideline-level treatment.¹⁰ Depression experts once believed that only patients of this type were likely to show a treatment response and that those who were older, a member of an ethnic minority, or from lower socioeconomic groups were less likely to respond to guideline treatment. Recent research shows that people from low socioeconomic backgrounds can respond to existing treatment, provided they have access to quality care.^11,12 Several studies specifically about treating depression in older adults have found positive effects, both in university and primary care settings.^4,6 Although research on ethnic minorities is scarce and focused primarily on Latinos and African Americans, the literature indicates that members of these ethnic groups do respond to psychotherapy. With respect to medication treatment, research on the pharmacokinetics of antidepressants in African Americans, Asians, and certain Latino groups indicates that dosages may need to be altered to reduce side effects.¹³

Current research also indicates that patients with complicated psychiatric presentations can respond to guideline-level treatment. For example, although patients more severe depressive symptoms may not respond to monotherapies as well as patients with milder symptoms,¹⁴ they generally respond well to combination treatments.⁸ The presence of Axis II features and comorbid anxiety or substance abuse does not necessarily have an impact on treatment outcome, although much of the data focus primarily on acute care of depression.¹⁴ Finally, even patients with cognitive impairment can respond to both medication and psychotherapy for depression.¹⁵

Who does not respond?

Growing evidence suggests that while effective treatments for depression do exist, they are not helpful for everyone. Treatment nonresponders fall into 2 categories: those who are treatment resistant and those who simply resist treatment.

Patients who are treatment resistant have been given an adequate course of either antidepressant medication or psychotherapy and have either no response or a limited response to treatment. Research investigating predictors of treatment failure indicates that several psychiatric and psychosocial variables are related to treatment resistance. Patients with more psychosocial stressors and less social support are more likely to show a limited response to treatment,⁹ as are patients with a greater number of previous depressive episodes.¹⁶ This may be due in part to increased feelings of hopelessness¹⁷ or lack of faith in treatment,¹⁸ both found to contribute to treatment resistance. Comorbid Axis II features, such as borderline and dependent personality traits tend to predict a decreased treatment response, in part because of the poor psychologic resources these patients have to cope with their symptoms.¹⁵ Such patients may benefit from additional interventions to alleviate their symptoms, such as case management, longer courses of psychotherapy, and multiple medications.

Patients who resist treatment include those who despite being identified as depressed and offered treatment, never follow through with the treatment plan. The primary reasons why patients do not adhere to treatment for depression include stigma concerns¹⁹ and the belief that depressive symptoms are not significant enough to treat.²⁰ Other factors, such as cognitive impairment, using multiple medications,²¹ comorbid medical illnesses, sensitivity to side effects,¹³ cost of mental health services,²² location of mental health services, and cross-cultural issues²³ may also have an impact on patient willingness to accept treatment. Once in treatment, psychologic factors such as self-efficacy²⁴ and readiness for change²⁵ can influence whether a patient will adhere to a treatment plan. There is early evidence that educational interventions or treatment management programs may benefit patients with acceptance or adherence issues.²⁶

Who do we need to study?

Several subgroups of patients have typically been excluded from treatment research. In particular, patients with coexisting Axis I disorders are routinely excluded from many treatment studies because of the complications concerning the management of separate conditions. However, the National Comorbidity Survey²⁷ has shown that depression often co-occurs with many other disorders, including substance abuse, psychosis, and anxiety disorders. Although past studies have included patients with comorbid symptoms of substance use and anxiety, little is known about the impact these interventions have when full-blown comorbid disorders are present.

Samples included in recent studies of depression treatment are becoming more diverse with respect to age and minority representation. However, little is known regarding the specific response to treatment in these populations or how response rates compare with those found with more traditional study populations. This is important work to undertake, given that certain age and minority groups have been found to have varying responses to existing treatments. For example, given the pharmacokinetic complications that have been associated with antidepressant medications in ethnic minority populations, investigating the effectiveness of existing interventions in these populations is also important.¹³

Along similar lines, preliminary research suggests that older people take longer to respond to antidepressant therapies and require smaller doses to prevent toxic effects.⁴ Other age groups, such as children and adolescents are rarely studied, though this may change as the result of new National Institutes of Health guidelines on the inclusion of children as research subjects. Also, people seeking treatment in medical organizations other than primary care medicine or psychiatry have not been systematically studied. For example, the rates for depression in women seen in obstetrics/gynecology are quite high, but there are no published treatment studies with this population.²⁸ Finally, patients who live in areas where care is hard to access (ie, rural populations) are currently being studied with promising results, yet to date there are no published outcomes.

Conclusions

The current literature shows that depression can be treated in many patients, but treatment response largely depends on the chronicity of the illness and the level of psychosocial stress faced by the patient. Future research should focus on how to best treat patients who tend not to respond to or accept existing treatment and should also examine the effectiveness of existing interventions for special populations who have not been included in past research. Thus far the evidence regarding the effectiveness of depression treatment is very promising, and the results of previous research will be useful in informing future work.

Research into the efficacy and effectiveness of treatments for depression has grown exponentially during the past several decades. Numerous studies show that disorders like major depression and dysthymia can be treated successfully with antidepressant medication and/or psychotherapy.¹ Therapeutic effect sizes from efficacy trials are 18% for antidepressants compared with placebo and 26% for psychotherapy compared with no treatment.^2-4 However, when these interventions are administered in real world settings, much lower response rates have been found.^5-8

The evidence from these studies indicates that in controlled settings and with highly selected patients, current methods for treating depression are efficacious. However, there are still many people who do not respond to current guideline-based treatment and many segments of the population that have not been included in clinical studies. We briefly review the research and discuss which populations respond to current treatment, which do not, and which require further study.

Definitions of treatment and treatment response

For the purposes of this paper we defined “treatment” as antidepressant medication or psychotherapy. Because the majority of psychotherapy research has focused on cognitive-behavioral and interpersonal therapies, we use psychotherapy as a generic term for these 2 types of treatment. Most depression treatment studies have defined “response” as no longer meeting Diagnostic and Statistical Manual of Mental Disorders criteria for a disorder and exhibiting a statistically significant change on a symptom severity scale (usually a decrement of at least 50%). For the purpose of this paper, we will consider response to mean a significant change in depression severity.

Who responds to treatment?

The authors of several studies have examined patient traits linked to treatment response.⁹ For the most part, patients who are educated, are experiencing uncomplicated depression, have had 2 or fewer previous episodes of depression, and have faith in their treatment typically respond to guideline-level treatment.¹⁰ Depression experts once believed that only patients of this type were likely to show a treatment response and that those who were older, a member of an ethnic minority, or from lower socioeconomic groups were less likely to respond to guideline treatment. Recent research shows that people from low socioeconomic backgrounds can respond to existing treatment, provided they have access to quality care.^11,12 Several studies specifically about treating depression in older adults have found positive effects, both in university and primary care settings.^4,6 Although research on ethnic minorities is scarce and focused primarily on Latinos and African Americans, the literature indicates that members of these ethnic groups do respond to psychotherapy. With respect to medication treatment, research on the pharmacokinetics of antidepressants in African Americans, Asians, and certain Latino groups indicates that dosages may need to be altered to reduce side effects.¹³

Current research also indicates that patients with complicated psychiatric presentations can respond to guideline-level treatment. For example, although patients more severe depressive symptoms may not respond to monotherapies as well as patients with milder symptoms,¹⁴ they generally respond well to combination treatments.⁸ The presence of Axis II features and comorbid anxiety or substance abuse does not necessarily have an impact on treatment outcome, although much of the data focus primarily on acute care of depression.¹⁴ Finally, even patients with cognitive impairment can respond to both medication and psychotherapy for depression.¹⁵

Who does not respond?

Growing evidence suggests that while effective treatments for depression do exist, they are not helpful for everyone. Treatment nonresponders fall into 2 categories: those who are treatment resistant and those who simply resist treatment.

Patients who are treatment resistant have been given an adequate course of either antidepressant medication or psychotherapy and have either no response or a limited response to treatment. Research investigating predictors of treatment failure indicates that several psychiatric and psychosocial variables are related to treatment resistance. Patients with more psychosocial stressors and less social support are more likely to show a limited response to treatment,⁹ as are patients with a greater number of previous depressive episodes.¹⁶ This may be due in part to increased feelings of hopelessness¹⁷ or lack of faith in treatment,¹⁸ both found to contribute to treatment resistance. Comorbid Axis II features, such as borderline and dependent personality traits tend to predict a decreased treatment response, in part because of the poor psychologic resources these patients have to cope with their symptoms.¹⁵ Such patients may benefit from additional interventions to alleviate their symptoms, such as case management, longer courses of psychotherapy, and multiple medications.

Patients who resist treatment include those who despite being identified as depressed and offered treatment, never follow through with the treatment plan. The primary reasons why patients do not adhere to treatment for depression include stigma concerns¹⁹ and the belief that depressive symptoms are not significant enough to treat.²⁰ Other factors, such as cognitive impairment, using multiple medications,²¹ comorbid medical illnesses, sensitivity to side effects,¹³ cost of mental health services,²² location of mental health services, and cross-cultural issues²³ may also have an impact on patient willingness to accept treatment. Once in treatment, psychologic factors such as self-efficacy²⁴ and readiness for change²⁵ can influence whether a patient will adhere to a treatment plan. There is early evidence that educational interventions or treatment management programs may benefit patients with acceptance or adherence issues.²⁶

Who do we need to study?

Several subgroups of patients have typically been excluded from treatment research. In particular, patients with coexisting Axis I disorders are routinely excluded from many treatment studies because of the complications concerning the management of separate conditions. However, the National Comorbidity Survey²⁷ has shown that depression often co-occurs with many other disorders, including substance abuse, psychosis, and anxiety disorders. Although past studies have included patients with comorbid symptoms of substance use and anxiety, little is known about the impact these interventions have when full-blown comorbid disorders are present.

Samples included in recent studies of depression treatment are becoming more diverse with respect to age and minority representation. However, little is known regarding the specific response to treatment in these populations or how response rates compare with those found with more traditional study populations. This is important work to undertake, given that certain age and minority groups have been found to have varying responses to existing treatments. For example, given the pharmacokinetic complications that have been associated with antidepressant medications in ethnic minority populations, investigating the effectiveness of existing interventions in these populations is also important.¹³

Along similar lines, preliminary research suggests that older people take longer to respond to antidepressant therapies and require smaller doses to prevent toxic effects.⁴ Other age groups, such as children and adolescents are rarely studied, though this may change as the result of new National Institutes of Health guidelines on the inclusion of children as research subjects. Also, people seeking treatment in medical organizations other than primary care medicine or psychiatry have not been systematically studied. For example, the rates for depression in women seen in obstetrics/gynecology are quite high, but there are no published treatment studies with this population.²⁸ Finally, patients who live in areas where care is hard to access (ie, rural populations) are currently being studied with promising results, yet to date there are no published outcomes.

Conclusions

The current literature shows that depression can be treated in many patients, but treatment response largely depends on the chronicity of the illness and the level of psychosocial stress faced by the patient. Future research should focus on how to best treat patients who tend not to respond to or accept existing treatment and should also examine the effectiveness of existing interventions for special populations who have not been included in past research. Thus far the evidence regarding the effectiveness of depression treatment is very promising, and the results of previous research will be useful in informing future work.

The barriers to improving care of depressive illness are well known. Patients resist mental disorder diagnoses, are not ready to accept treatment, or fail to follow through on prescribed treatments. Primary care physicians fail to recognize depression in their patients, fail to prescribe an adequate treatment regimen, or fail to follow-up with patients once treatment is initiated. Psychiatrists and other mental health professionals are not accessible to many depressed individuals (elderly, rural, medically ill, and economically disadvantaged populations). Health care systems often fail to organize mental health consultation services to support the work of primary care physicians who treat the majority of depressed patients. Insurers and employers resist adequate insurance benefits for mental health services. Given the extent and complexity of the barriers to improved care of depressive illness, it is not surprising that little progress has been made in reducing the burden of depressive illness on a population basis, despite the availability of effective treatments. The problem is not lack of effective treatments for depression but deficiencies in the organization and delivery of health and mental health services.

Although understanding the barriers to improved care is important, focusing on barriers alone can be contagious and counterproductive. The litany of barriers can easily become a rationale for inaction. Bringing potential solutions to light provides an invitation to experiment, try things out, and take action.

At our conference on improving care for depression in organized health care systems, current experimental research was presented in which possible approaches to improving care of depressive illness were tested and effects on patient outcomes were assessed. After the research presentations the participants (eg, leading researchers and persons responsible for improving the quality of care for mental disorders in systems serving of millions of people) identified possible solutions to the well-known barriers to improved depression care.

Barriers and solutions are listed in the Table 1 using the framework of the Model for Improving Chronic Illness Care described elsewhere.¹ In each area we enumerate barriers and potential solutions identified by the conference participants. In the final analysis, the most significant barrier to improving the quality of care for depressive illness may be inaction, because all other barriers are insurmountable in the absence of effort to produce change.

Research needs

The conference participants also considered research needs that have not been adequately addressed by the current generation of depression research.

Case Management

The current generation of depression care studies has tested different forms of case management with generally promising results. Critical unresolved questions focus on whom case management services are needed for and how long it should be sustained. Some research suggests that case management services may need to be continued over long periods of time, but outcome data beyond 1 year are lacking. Additional research is needed to clarify the benefits of having specialist-consultants both supervise the work of case managers and provide services targeted to patients who do not achieve a favorable outcome with case management services alone. Although case managers have been used most frequently in support of pharmacotherapy, it remains unclear to what extent patients benefit from the behavioral and supportive interventions they provide. To what extent do case management services benefit patients through mobilization of hope and behavioral activation versus improved adherence to treatment regimens? Should these services be delivered by depression case managers who follow a large caseload of depressed primary care patients, or should depression be one of many chronic conditions such as diabetes, hypertension, and asthma that are managed by a generalist case manager working as part of the primary care team? Also, there is a need for development and testing of new modes of delivering case management services in addition to in-person and telephonic services (eg, telemedicine services or the Internet). New approaches are needed to increase the feasibility of sustained case management and to reduce costs.

Research in New Populations

A logical next step is to test the care models proven useful for depressed patients in diverse patient populations. Research on care of other common psychiatric illnesses such as anxiety disorders, somatoform disorders, and bipolar disorder is needed. Adapting the new care models and testing their effectiveness in the care of rural, economically disadvantaged, and elderly populations would also be useful. Future research might test provision of case management and specialist consultation services through telemedicine connections for patient populations lacking direct access to such services in their primary care setting (eg, rural practices, network model practices). Finally, the management of treatment-resistant patients was identified as a critically important issue that has not been resolved. Will treatment-resistant patients benefit more from referral for specialty mental health care, or can they be effectively managed in the primary care setting with effective organization of treatment and support services? Surprisingly little is known about the care of depression among patients with comorbid medical disease (eg, diabetes, heart disease, chronic obstructive pulmonary disease). There is now substantial evidence that depression is associated with increased physical symptoms, increased disability, increased use of general medical services, and increased likelihood of comorbid medical illness.² Enrolling patients with a specific chronic disease such as diabetes or coronary artery disease would enable researchers to more precisely delineate the effect of improved depression care on biologic measures of disease severity as well as physical symptoms, disability, and use of health care services. The impact of improved depression care on the ability of patients to manage a comorbid chronic disease is of considerable interest.

Stepped Care and Relapse Prevention

An emerging theme in the current generation of depression care research is the use of sequential or stepped care management strategies. In stepped care interventions, patient outcomes are monitored, and modifications in the care plan and/or more intensive management are targeted toward patients who do not have a favorable outcome by a defined time point (eg, 2 months after the initiation of treatment).^3,4 It is hoped that stepped care models will enhance the cost-effectiveness of depression care programs by reserving the use of case management and specialist consultation services for those patients who cannot be effectively managed by the primary care physician alone. In general, there is a need for new research (and analyses from completed studies) that identify ways of using limited specialist and case management services to greatest effect in improving the long-term outcomes per unit cost. Effectiveness studies now need to develop and test interventions that follow patients for continuation and maintenance phases to assess their ability to prevent relapse and maximize patient functioning over extended periods of time.

Societal Benefits of Improved Depression Care

There was a sense of urgency about the need for new research that more adequately assesses the effects of treating depression on labor force participation, market and nonmarket productivity, work absenteeism, family functioning, and time off work for travel to mental health treatments. The need to evaluate the effects of treating depression on both the depressed individual and members of their families was recognized. Research in these important areas has been hampered by the lack of reliable and valid measures. In particular, the development of reliable and valid measures of work productivity and family burden were seen as critically important.

Expert panels have recommended that alternative treatments be compared using cost utility methods that explicitly incorporate patient outcome preferences.⁵ The evaluation of health state preferences remains uncommon in depression clinical trials because of measurement problems. Different research groups have recently attempted to indirectly measure how patient utilities vary with different dimensions of health quality of life using the 12-item Medical Outcomes Study Short Form or the 36-item Medical Outcomes Study Short Form.^6-9 However, the validity of these alternative approaches is unclear. Resolution of these uncertainties was considered to be an important research area.

Multisite Trials and Meta-Analytic Approaches

As depression research increasingly focuses on assessing effects on societal costs and benefits of improved depression care, the large variance of policy-relevant outcome measures such as disability days, unemployment, and health care costs is of increasing concern. Effectiveness studies enrolling even 200 to 300 patients are underpowered to detect clinically significant differences in many of these outcomes,¹⁰ even though even modest beneficial effects of treatment could have substantial social significance on a population basis. This suggests the need for large-scale multisite trials of depression care programs. The effects of depression care programs on these outcomes might also be assessed through meta-analyses of completed trials or new intervention studies that are designed to vary components of the intervention strategy.

Conclusions

Completed trials suggest that depression care programs integrated into the primary care setting can improve depression and disability outcomes of patients with major and possibly minor depression. The completed research increases the public health imperative for refining understanding of how to provide depression care most effectively and cost-effectively and for determining the extent to which these interventions can benefit new patient populations and the societal benefits of such care programs. The completed trials have established a stronger basis for organized efforts to improve the quality of depression care in health care systems. They have also set the stage for both larger trials and meta-analyses of completed trials that are designed to answer key questions about the societal benefits of improved depression care.

The barriers to improving care of depressive illness are well known. Patients resist mental disorder diagnoses, are not ready to accept treatment, or fail to follow through on prescribed treatments. Primary care physicians fail to recognize depression in their patients, fail to prescribe an adequate treatment regimen, or fail to follow-up with patients once treatment is initiated. Psychiatrists and other mental health professionals are not accessible to many depressed individuals (elderly, rural, medically ill, and economically disadvantaged populations). Health care systems often fail to organize mental health consultation services to support the work of primary care physicians who treat the majority of depressed patients. Insurers and employers resist adequate insurance benefits for mental health services. Given the extent and complexity of the barriers to improved care of depressive illness, it is not surprising that little progress has been made in reducing the burden of depressive illness on a population basis, despite the availability of effective treatments. The problem is not lack of effective treatments for depression but deficiencies in the organization and delivery of health and mental health services.

Although understanding the barriers to improved care is important, focusing on barriers alone can be contagious and counterproductive. The litany of barriers can easily become a rationale for inaction. Bringing potential solutions to light provides an invitation to experiment, try things out, and take action.

At our conference on improving care for depression in organized health care systems, current experimental research was presented in which possible approaches to improving care of depressive illness were tested and effects on patient outcomes were assessed. After the research presentations the participants (eg, leading researchers and persons responsible for improving the quality of care for mental disorders in systems serving of millions of people) identified possible solutions to the well-known barriers to improved depression care.

Barriers and solutions are listed in the Table 1 using the framework of the Model for Improving Chronic Illness Care described elsewhere.¹ In each area we enumerate barriers and potential solutions identified by the conference participants. In the final analysis, the most significant barrier to improving the quality of care for depressive illness may be inaction, because all other barriers are insurmountable in the absence of effort to produce change.

Research needs

The conference participants also considered research needs that have not been adequately addressed by the current generation of depression research.

Case Management

The current generation of depression care studies has tested different forms of case management with generally promising results. Critical unresolved questions focus on whom case management services are needed for and how long it should be sustained. Some research suggests that case management services may need to be continued over long periods of time, but outcome data beyond 1 year are lacking. Additional research is needed to clarify the benefits of having specialist-consultants both supervise the work of case managers and provide services targeted to patients who do not achieve a favorable outcome with case management services alone. Although case managers have been used most frequently in support of pharmacotherapy, it remains unclear to what extent patients benefit from the behavioral and supportive interventions they provide. To what extent do case management services benefit patients through mobilization of hope and behavioral activation versus improved adherence to treatment regimens? Should these services be delivered by depression case managers who follow a large caseload of depressed primary care patients, or should depression be one of many chronic conditions such as diabetes, hypertension, and asthma that are managed by a generalist case manager working as part of the primary care team? Also, there is a need for development and testing of new modes of delivering case management services in addition to in-person and telephonic services (eg, telemedicine services or the Internet). New approaches are needed to increase the feasibility of sustained case management and to reduce costs.

Research in New Populations

A logical next step is to test the care models proven useful for depressed patients in diverse patient populations. Research on care of other common psychiatric illnesses such as anxiety disorders, somatoform disorders, and bipolar disorder is needed. Adapting the new care models and testing their effectiveness in the care of rural, economically disadvantaged, and elderly populations would also be useful. Future research might test provision of case management and specialist consultation services through telemedicine connections for patient populations lacking direct access to such services in their primary care setting (eg, rural practices, network model practices). Finally, the management of treatment-resistant patients was identified as a critically important issue that has not been resolved. Will treatment-resistant patients benefit more from referral for specialty mental health care, or can they be effectively managed in the primary care setting with effective organization of treatment and support services? Surprisingly little is known about the care of depression among patients with comorbid medical disease (eg, diabetes, heart disease, chronic obstructive pulmonary disease). There is now substantial evidence that depression is associated with increased physical symptoms, increased disability, increased use of general medical services, and increased likelihood of comorbid medical illness.² Enrolling patients with a specific chronic disease such as diabetes or coronary artery disease would enable researchers to more precisely delineate the effect of improved depression care on biologic measures of disease severity as well as physical symptoms, disability, and use of health care services. The impact of improved depression care on the ability of patients to manage a comorbid chronic disease is of considerable interest.

Stepped Care and Relapse Prevention

An emerging theme in the current generation of depression care research is the use of sequential or stepped care management strategies. In stepped care interventions, patient outcomes are monitored, and modifications in the care plan and/or more intensive management are targeted toward patients who do not have a favorable outcome by a defined time point (eg, 2 months after the initiation of treatment).^3,4 It is hoped that stepped care models will enhance the cost-effectiveness of depression care programs by reserving the use of case management and specialist consultation services for those patients who cannot be effectively managed by the primary care physician alone. In general, there is a need for new research (and analyses from completed studies) that identify ways of using limited specialist and case management services to greatest effect in improving the long-term outcomes per unit cost. Effectiveness studies now need to develop and test interventions that follow patients for continuation and maintenance phases to assess their ability to prevent relapse and maximize patient functioning over extended periods of time.

Societal Benefits of Improved Depression Care

There was a sense of urgency about the need for new research that more adequately assesses the effects of treating depression on labor force participation, market and nonmarket productivity, work absenteeism, family functioning, and time off work for travel to mental health treatments. The need to evaluate the effects of treating depression on both the depressed individual and members of their families was recognized. Research in these important areas has been hampered by the lack of reliable and valid measures. In particular, the development of reliable and valid measures of work productivity and family burden were seen as critically important.

Expert panels have recommended that alternative treatments be compared using cost utility methods that explicitly incorporate patient outcome preferences.⁵ The evaluation of health state preferences remains uncommon in depression clinical trials because of measurement problems. Different research groups have recently attempted to indirectly measure how patient utilities vary with different dimensions of health quality of life using the 12-item Medical Outcomes Study Short Form or the 36-item Medical Outcomes Study Short Form.^6-9 However, the validity of these alternative approaches is unclear. Resolution of these uncertainties was considered to be an important research area.

Multisite Trials and Meta-Analytic Approaches

As depression research increasingly focuses on assessing effects on societal costs and benefits of improved depression care, the large variance of policy-relevant outcome measures such as disability days, unemployment, and health care costs is of increasing concern. Effectiveness studies enrolling even 200 to 300 patients are underpowered to detect clinically significant differences in many of these outcomes,¹⁰ even though even modest beneficial effects of treatment could have substantial social significance on a population basis. This suggests the need for large-scale multisite trials of depression care programs. The effects of depression care programs on these outcomes might also be assessed through meta-analyses of completed trials or new intervention studies that are designed to vary components of the intervention strategy.

Conclusions

Completed trials suggest that depression care programs integrated into the primary care setting can improve depression and disability outcomes of patients with major and possibly minor depression. The completed research increases the public health imperative for refining understanding of how to provide depression care most effectively and cost-effectively and for determining the extent to which these interventions can benefit new patient populations and the societal benefits of such care programs. The completed trials have established a stronger basis for organized efforts to improve the quality of depression care in health care systems. They have also set the stage for both larger trials and meta-analyses of completed trials that are designed to answer key questions about the societal benefits of improved depression care.

The barriers to improving care of depressive illness are well known. Patients resist mental disorder diagnoses, are not ready to accept treatment, or fail to follow through on prescribed treatments. Primary care physicians fail to recognize depression in their patients, fail to prescribe an adequate treatment regimen, or fail to follow-up with patients once treatment is initiated. Psychiatrists and other mental health professionals are not accessible to many depressed individuals (elderly, rural, medically ill, and economically disadvantaged populations). Health care systems often fail to organize mental health consultation services to support the work of primary care physicians who treat the majority of depressed patients. Insurers and employers resist adequate insurance benefits for mental health services. Given the extent and complexity of the barriers to improved care of depressive illness, it is not surprising that little progress has been made in reducing the burden of depressive illness on a population basis, despite the availability of effective treatments. The problem is not lack of effective treatments for depression but deficiencies in the organization and delivery of health and mental health services.

Although understanding the barriers to improved care is important, focusing on barriers alone can be contagious and counterproductive. The litany of barriers can easily become a rationale for inaction. Bringing potential solutions to light provides an invitation to experiment, try things out, and take action.

At our conference on improving care for depression in organized health care systems, current experimental research was presented in which possible approaches to improving care of depressive illness were tested and effects on patient outcomes were assessed. After the research presentations the participants (eg, leading researchers and persons responsible for improving the quality of care for mental disorders in systems serving of millions of people) identified possible solutions to the well-known barriers to improved depression care.

Barriers and solutions are listed in the Table 1 using the framework of the Model for Improving Chronic Illness Care described elsewhere.¹ In each area we enumerate barriers and potential solutions identified by the conference participants. In the final analysis, the most significant barrier to improving the quality of care for depressive illness may be inaction, because all other barriers are insurmountable in the absence of effort to produce change.

Research needs

The conference participants also considered research needs that have not been adequately addressed by the current generation of depression research.

Case Management

The current generation of depression care studies has tested different forms of case management with generally promising results. Critical unresolved questions focus on whom case management services are needed for and how long it should be sustained. Some research suggests that case management services may need to be continued over long periods of time, but outcome data beyond 1 year are lacking. Additional research is needed to clarify the benefits of having specialist-consultants both supervise the work of case managers and provide services targeted to patients who do not achieve a favorable outcome with case management services alone. Although case managers have been used most frequently in support of pharmacotherapy, it remains unclear to what extent patients benefit from the behavioral and supportive interventions they provide. To what extent do case management services benefit patients through mobilization of hope and behavioral activation versus improved adherence to treatment regimens? Should these services be delivered by depression case managers who follow a large caseload of depressed primary care patients, or should depression be one of many chronic conditions such as diabetes, hypertension, and asthma that are managed by a generalist case manager working as part of the primary care team? Also, there is a need for development and testing of new modes of delivering case management services in addition to in-person and telephonic services (eg, telemedicine services or the Internet). New approaches are needed to increase the feasibility of sustained case management and to reduce costs.

Research in New Populations

A logical next step is to test the care models proven useful for depressed patients in diverse patient populations. Research on care of other common psychiatric illnesses such as anxiety disorders, somatoform disorders, and bipolar disorder is needed. Adapting the new care models and testing their effectiveness in the care of rural, economically disadvantaged, and elderly populations would also be useful. Future research might test provision of case management and specialist consultation services through telemedicine connections for patient populations lacking direct access to such services in their primary care setting (eg, rural practices, network model practices). Finally, the management of treatment-resistant patients was identified as a critically important issue that has not been resolved. Will treatment-resistant patients benefit more from referral for specialty mental health care, or can they be effectively managed in the primary care setting with effective organization of treatment and support services? Surprisingly little is known about the care of depression among patients with comorbid medical disease (eg, diabetes, heart disease, chronic obstructive pulmonary disease). There is now substantial evidence that depression is associated with increased physical symptoms, increased disability, increased use of general medical services, and increased likelihood of comorbid medical illness.² Enrolling patients with a specific chronic disease such as diabetes or coronary artery disease would enable researchers to more precisely delineate the effect of improved depression care on biologic measures of disease severity as well as physical symptoms, disability, and use of health care services. The impact of improved depression care on the ability of patients to manage a comorbid chronic disease is of considerable interest.

Stepped Care and Relapse Prevention

An emerging theme in the current generation of depression care research is the use of sequential or stepped care management strategies. In stepped care interventions, patient outcomes are monitored, and modifications in the care plan and/or more intensive management are targeted toward patients who do not have a favorable outcome by a defined time point (eg, 2 months after the initiation of treatment).^3,4 It is hoped that stepped care models will enhance the cost-effectiveness of depression care programs by reserving the use of case management and specialist consultation services for those patients who cannot be effectively managed by the primary care physician alone. In general, there is a need for new research (and analyses from completed studies) that identify ways of using limited specialist and case management services to greatest effect in improving the long-term outcomes per unit cost. Effectiveness studies now need to develop and test interventions that follow patients for continuation and maintenance phases to assess their ability to prevent relapse and maximize patient functioning over extended periods of time.

Societal Benefits of Improved Depression Care

There was a sense of urgency about the need for new research that more adequately assesses the effects of treating depression on labor force participation, market and nonmarket productivity, work absenteeism, family functioning, and time off work for travel to mental health treatments. The need to evaluate the effects of treating depression on both the depressed individual and members of their families was recognized. Research in these important areas has been hampered by the lack of reliable and valid measures. In particular, the development of reliable and valid measures of work productivity and family burden were seen as critically important.

Expert panels have recommended that alternative treatments be compared using cost utility methods that explicitly incorporate patient outcome preferences.⁵ The evaluation of health state preferences remains uncommon in depression clinical trials because of measurement problems. Different research groups have recently attempted to indirectly measure how patient utilities vary with different dimensions of health quality of life using the 12-item Medical Outcomes Study Short Form or the 36-item Medical Outcomes Study Short Form.^6-9 However, the validity of these alternative approaches is unclear. Resolution of these uncertainties was considered to be an important research area.

Multisite Trials and Meta-Analytic Approaches

As depression research increasingly focuses on assessing effects on societal costs and benefits of improved depression care, the large variance of policy-relevant outcome measures such as disability days, unemployment, and health care costs is of increasing concern. Effectiveness studies enrolling even 200 to 300 patients are underpowered to detect clinically significant differences in many of these outcomes,¹⁰ even though even modest beneficial effects of treatment could have substantial social significance on a population basis. This suggests the need for large-scale multisite trials of depression care programs. The effects of depression care programs on these outcomes might also be assessed through meta-analyses of completed trials or new intervention studies that are designed to vary components of the intervention strategy.

Conclusions

Completed trials suggest that depression care programs integrated into the primary care setting can improve depression and disability outcomes of patients with major and possibly minor depression. The completed research increases the public health imperative for refining understanding of how to provide depression care most effectively and cost-effectively and for determining the extent to which these interventions can benefit new patient populations and the societal benefits of such care programs. The completed trials have established a stronger basis for organized efforts to improve the quality of depression care in health care systems. They have also set the stage for both larger trials and meta-analyses of completed trials that are designed to answer key questions about the societal benefits of improved depression care.

Do any of the following situations sound familiar?

1. You want to prescribe 25 mg of a drug but are uncertain whether the drug is supplied as a 25-mg or 50-mg tablet.
2. An elderly patient comes in for an update on her multiple problems. At the end of a long visit, she asks to have all 17 of her prescriptions refilled.
3. You need to determine whether there is a possible drug interaction between existing medications and an upcoming medicine.
4. You would like to be able to generate a list of all your patients taking a recently recalled drug.

Now imagine a hand-held prescribing device the size of a prescription pad that turns on instantly, knows the date, and through a connection to your computer, knows your patient list for the day. Using this device you could quickly select the patient’s name from a drop-down list and see his or her insurance plan, age, and other demographic information. You could select medications from an alphabetic scrolling list that includes information specific to the patient’s insurance plan. You could check allergies and drug interactions against the patient’s list of previously prescribed drugs and known medication allergies, alerting you only if a clinically significant problem is detected. After selecting the medication, you could choose from a list of available dosages and specify the quantity to be dispensed Figure 1, and the prescription could be digitally signed and sent electronically to the patient’s pharmacy or to your printer. Chart documentation could be produced concurrently on a sticky-backed label that can be affixed to the patient’s medical record. The prescription would be saved for generating rapid renewals in the future. Drug information and insurance formularies would be updated regularly from central computers, and new drugs and dosage forms could appear on the hand-held devices before they show up on the pharmacists’ shelves.

Does all this sound far-fetched? It’s not. This type of technology is currently available. Systems range from simple prescription writers—a hand-held device with initial manual entry of patient data and prescription printing from an infrared-equipped printer—to full electronic medical records systems. Prospective users have to decide which features provide sufficient benefit to merit the investment of time necessary to learn them. In this article, 3 systems (PocketScript, ePhysician ePad, and AllScripts;)Table 1, Table 1a are reviewed by physicians who have successfully incorporated electronic prescribing into their practices.

A system that does nothing other than produce legible prescriptions in the same time it would take to write one by hand can benefit many physicians. One that provides an accurate record of previous prescriptions that allows the rapid renewal of multiple prescriptions is even more valuable. Another important potential benefit of an electronic prescribing system is fewer errors, including those caused by illegible handwriting, incorrect dosage selections, drug-drug or drug-disease interactions, and drug allergies.

These systems still have some limitations, however. Most do not automatically suggest a dosage adjustment for age, renal function, hepatic function, and alcohol and tobacco use (although drug monographs may be available on the hand-held device itself). Similarly, weight-based pediatric prescription information may be available in monographs on the hand-held device but is not built into the prescribing process. Some states allow electronic transmission of prescriptions, some do not, and some have made no ruling; specific rules vary. Fortunately, each of the vendors reviewed in this article made certain that their system is installed in a way that conforms to state laws.

Most primary care physicians have not made the leap to electronic medical records; package delivery services and supermarket checkouts use more technology at the point of service than the average physician does. These devices may change that by providing a relatively low-cost low-risk improvement with less impact on work processes than a full electronic medical record.

Common Features

Each of the systems we reviewed is “wired” somewhat differently Figure 2. PocketScript and AllScripts use a wireless local area network in the physician’s office to move information from the hand-held device to a desktop computer server in the physician’s office. This server is in turn connected to the vendor’s computer through a modem or other high-speed access device to get drug information updates. The office computer also sends prescriptions directly to pharmacies either as an electronic message or by fax.

A system using ePhysician includes a personal computer in the physician’s office to load patient demographic information onto the hand-held using a docking cradle. Prescriptions can be sent directly from the hand-held device to a central computer server at the vendor’s offices using a wireless modem. Alternately, they physician can dock the hand-held device in a cradle attached to the computer, which is connected by a modem to the vendor’s central computer. The vendor’s computer server then sends prescriptions to the selected pharmacy, either electronically or by fax. All systems use sophisticated encryption to protect the confidentiality of data as it is sent between computers.

The process of generating a prescription follows that of using a paper prescription pad. Prescribers may override the predetermined choices and enter personal choices for quantity, instructions, and refills. The physician then views the completed prescription information, confirms the pharmacy information, and sends it through the wireless network and the computer system to the pharmacy or prints it in the office. All products support some type of list to facilitate rapid generation of custom prescriptions. Prescriptions produced by these systems benefit from perfect legibility, fraud and error reduction, automatic drug interactions checking (except ePad), and improved formulary compliance.

Each system can connect to front office systems and move patient demographic data into the prescription writing system. A real time interface updates both applications simultaneously, so information entered in the registration system is immediately available in the prescription writing system. With a bridge program, information entered into the registration system is unavailable in the prescription-writing program until the program is run and transfers the data to the prescription writing system.

PocketScript
Reviewed by Eric Weidmann, MD

PocketScript is unique, because the user can select voice recognition, screen touch, or the keyboard to quickly compose and send prescriptions. Patients’ records may be added or edited from the hand-held PC or the physician’s server. Hand-held to server communication is fast enough to provide high-speed Internet connectivity from the examination room, which is useful for patient care, business, and personal uses. Drug-drug interactions are checked automatically when choosing a drug and before sending the prescription. Patient allergy and diagnosis fields are promised for future releases.

PocketScript does not have to be linked to the practice’s registration system. If not linked, each patient must be entered manually the first time only. Patient data (eg, weight, insurance plan, and so on) can be changed using the Patient Maintenance function on the hand-held device. A scrollable Favorites list can be created of the physician’s most commonly used drugs; it uses PocketScript’s standard instructions for the signature. There is a scrollable, pull-down menu called Macros containing the physician’s custom-named drugs with physician-generated customized instructions and quantities (eg, special creams or a corticosteroid taper). Internet access on the wireless hand-held provides references for specialized disease and interaction information. I use my hand-held to access my E-mail as I move between patients. The keyboard allows me to use the hand-held to reply easily. We also use the E-mail system as an intra-office memo system for routine discussion and reminders.

Most of our prescriptions are faxed from the hand-held through the computer in our office to local pharmacies with the graphic of our log-in signature. Printed prescriptions can be sent to the printer in our office, which we then manually sign. PocketScript can provide digital signing when that complies with a state’s board of pharmacy regulations. Internet-enabled pharmacies can receive prescription data electronically using an E-mail–like system. There is no option for chart documentation other than printing a second script.

PocketScript is our choice because of its strong physician leadership, ethics, design, and serious commitment to the prescription process. It is usable by even the most computer phobic provider and does not attempt to replace anything but the paper prescription pad, phone, and wasted time—at least for now. Also, the vendor does not sell profiles of our prescribing habits. Security of the patient data and prescription process are excellent and are maintained even if the hand-held device is stolen. PocketScript is intuitive, simple, and efficient, with no tolerance for errors.

The design of the PocketScript server-based system eliminates the need to dock hand-held devices, avoiding time-consuming and frustrating tasks for the physician. This system is ideal for physicians who do most of their work in one or a few sites in which wireless nodes can be installed. We have changed office processes for refill and pharmacy calls, saving an estimated 1 hour per day per nurse and 30 minutes per day per file clerk. Patients and pharmacists frequently praise the clarity, accuracy, and convenience of the PocketScript system. PocketScript is a serious medical care partner and the perfect first step into e-medicine for our group practice.

ePhysician ePad
Reviewed by Daniel E. Diamond, MD

The ePhysician system uses a wireless wide area network (WAN) rather than a wireless local area network in the physician’s office. Prescriptions can be transmitted directly to the vendor’s server through an OmniSky (or equivalent) wireless cellular-WAN modem. The physicians who do not want to use the wireless interface can use the docking cradle attached to a computer in the office for synchronization. The prescription is then securely sent through the Internet to the vendor’s computer (using the cradle takes approximately 30 seconds per synchronization). The wireless method is more efficient and allows the physician to write and send the prescription without leaving the examination room. It also allows the physician to write and send prescriptions when away from the office.

ePad has several other unique features. Because patient demographics are entered into the physician’s existing practice management system and transmitted to the hand-held when it is docked, the physician does not have to enter any patient data. To write a prescription for a patient who has not previously been entered into the system, the user docks the hand-held in the cradle for synchronization. Also, all prescriptions are transmitted from the vendor’s central computer servers either electronically or by fax to any pharmacy in the United States. Users can also print from the hand-held to a local printer. To generate chart documentation, the user can configure ePad to automatically print a copy of the prescription on the local computer when it transmits to the pharmacy. ePad currently comes bundled with ePhysician Superbill for charge capture and is integrated with Drug Facts and Comparisons. It also includes a Web interface that allows my staff to efficiently process prescription renewals, and a scheduling module that makes it possible to schedule patients on the computer. This schedule is viewable on the hand-held device.

ePad by ePhysician with the OmniSky wireless service is simple, efficient, and powerful. Its interface was designed with a focus on the physician user. Although there is a subscription charge for ePad, the fee pales in comparison with the cost of lost physician productivity due to the inefficient technology of printing prescriptions locally (or writing them by hand). Using the ePad system in our office has resulted in very positive patient feedback and increased efficiency and job satisfaction among staff.

AllScripts
Reviewed by Azar A. Korbey, MD

At the installation stage, AllScripts’ on-site team realizes that the use of these systems require change. They take the time to train the staff and make them part of the process. AllScripts differs from the other systems, because after the selection of the patient name, users select their most frequently coded diseases listed by common name and International Classification of Diseases-ninth revision code. This list is prescriber-specific, not patient-specific. Users can also access the patient’s prescription and diagnosis histories, and the AllScripts system automatically checks for previous adverse drug reactions. Next, AllScripts shows the medicines that the user most frequently prescribes for a disease. AllScripts learns each user’s prescribing preferences, so users do not need to enter their favorite prescriptions. The list also shows preferred instructions for any given medication, and if AllScripts’ office dispensing system is used, it will tell users whether they have the drug in inventory. To produce a prescription, the prescriber may modify a prescription from a frequently used list or choose from any other drug in AllScripts’ database.

Users have access to all patients in the practice through the network, regardless of provider or schedule. Patients not registered with the practice can be manually added into the hand-held device or desktop computer server. Unlike the other systems, AllScripts automatically produces a mailing address-sized label for chart documentation that includes all details of the prescription. The vendor also sells a turnkey pharmacy service (FirstFill) for in-office dispensing. It is only for common medications, and is similar to providing samples. The local wireless network is fast; high-speed access to the Internet is proving very useful.

AllScripts is complete and does what it promises. It is a well-funded, publicly traded company that is adding applications and acquiring additional companies that will allow dictation and charge-capture systems to be added. I have had long-term experience with the product and its exemplary phone support, and I believe they will both be around in the future. They are even willing to come on site immediately, if required. I strongly recommend this system.

Editor’s Note: We had hoped to include the iScribe 3000 system (www.iscribe.com) in this review but could not obtain the product and complete the review in time for publication. We plan to publish a software review of iScribe in a future issue of JFP.

Do any of the following situations sound familiar?

1. You want to prescribe 25 mg of a drug but are uncertain whether the drug is supplied as a 25-mg or 50-mg tablet.
2. An elderly patient comes in for an update on her multiple problems. At the end of a long visit, she asks to have all 17 of her prescriptions refilled.
3. You need to determine whether there is a possible drug interaction between existing medications and an upcoming medicine.
4. You would like to be able to generate a list of all your patients taking a recently recalled drug.

Now imagine a hand-held prescribing device the size of a prescription pad that turns on instantly, knows the date, and through a connection to your computer, knows your patient list for the day. Using this device you could quickly select the patient’s name from a drop-down list and see his or her insurance plan, age, and other demographic information. You could select medications from an alphabetic scrolling list that includes information specific to the patient’s insurance plan. You could check allergies and drug interactions against the patient’s list of previously prescribed drugs and known medication allergies, alerting you only if a clinically significant problem is detected. After selecting the medication, you could choose from a list of available dosages and specify the quantity to be dispensed Figure 1, and the prescription could be digitally signed and sent electronically to the patient’s pharmacy or to your printer. Chart documentation could be produced concurrently on a sticky-backed label that can be affixed to the patient’s medical record. The prescription would be saved for generating rapid renewals in the future. Drug information and insurance formularies would be updated regularly from central computers, and new drugs and dosage forms could appear on the hand-held devices before they show up on the pharmacists’ shelves.

Does all this sound far-fetched? It’s not. This type of technology is currently available. Systems range from simple prescription writers—a hand-held device with initial manual entry of patient data and prescription printing from an infrared-equipped printer—to full electronic medical records systems. Prospective users have to decide which features provide sufficient benefit to merit the investment of time necessary to learn them. In this article, 3 systems (PocketScript, ePhysician ePad, and AllScripts;)Table 1, Table 1a are reviewed by physicians who have successfully incorporated electronic prescribing into their practices.

A system that does nothing other than produce legible prescriptions in the same time it would take to write one by hand can benefit many physicians. One that provides an accurate record of previous prescriptions that allows the rapid renewal of multiple prescriptions is even more valuable. Another important potential benefit of an electronic prescribing system is fewer errors, including those caused by illegible handwriting, incorrect dosage selections, drug-drug or drug-disease interactions, and drug allergies.

These systems still have some limitations, however. Most do not automatically suggest a dosage adjustment for age, renal function, hepatic function, and alcohol and tobacco use (although drug monographs may be available on the hand-held device itself). Similarly, weight-based pediatric prescription information may be available in monographs on the hand-held device but is not built into the prescribing process. Some states allow electronic transmission of prescriptions, some do not, and some have made no ruling; specific rules vary. Fortunately, each of the vendors reviewed in this article made certain that their system is installed in a way that conforms to state laws.

Most primary care physicians have not made the leap to electronic medical records; package delivery services and supermarket checkouts use more technology at the point of service than the average physician does. These devices may change that by providing a relatively low-cost low-risk improvement with less impact on work processes than a full electronic medical record.

Common Features

Each of the systems we reviewed is “wired” somewhat differently Figure 2. PocketScript and AllScripts use a wireless local area network in the physician’s office to move information from the hand-held device to a desktop computer server in the physician’s office. This server is in turn connected to the vendor’s computer through a modem or other high-speed access device to get drug information updates. The office computer also sends prescriptions directly to pharmacies either as an electronic message or by fax.

A system using ePhysician includes a personal computer in the physician’s office to load patient demographic information onto the hand-held using a docking cradle. Prescriptions can be sent directly from the hand-held device to a central computer server at the vendor’s offices using a wireless modem. Alternately, they physician can dock the hand-held device in a cradle attached to the computer, which is connected by a modem to the vendor’s central computer. The vendor’s computer server then sends prescriptions to the selected pharmacy, either electronically or by fax. All systems use sophisticated encryption to protect the confidentiality of data as it is sent between computers.

The process of generating a prescription follows that of using a paper prescription pad. Prescribers may override the predetermined choices and enter personal choices for quantity, instructions, and refills. The physician then views the completed prescription information, confirms the pharmacy information, and sends it through the wireless network and the computer system to the pharmacy or prints it in the office. All products support some type of list to facilitate rapid generation of custom prescriptions. Prescriptions produced by these systems benefit from perfect legibility, fraud and error reduction, automatic drug interactions checking (except ePad), and improved formulary compliance.

Each system can connect to front office systems and move patient demographic data into the prescription writing system. A real time interface updates both applications simultaneously, so information entered in the registration system is immediately available in the prescription writing system. With a bridge program, information entered into the registration system is unavailable in the prescription-writing program until the program is run and transfers the data to the prescription writing system.

PocketScript
Reviewed by Eric Weidmann, MD

PocketScript is unique, because the user can select voice recognition, screen touch, or the keyboard to quickly compose and send prescriptions. Patients’ records may be added or edited from the hand-held PC or the physician’s server. Hand-held to server communication is fast enough to provide high-speed Internet connectivity from the examination room, which is useful for patient care, business, and personal uses. Drug-drug interactions are checked automatically when choosing a drug and before sending the prescription. Patient allergy and diagnosis fields are promised for future releases.

PocketScript does not have to be linked to the practice’s registration system. If not linked, each patient must be entered manually the first time only. Patient data (eg, weight, insurance plan, and so on) can be changed using the Patient Maintenance function on the hand-held device. A scrollable Favorites list can be created of the physician’s most commonly used drugs; it uses PocketScript’s standard instructions for the signature. There is a scrollable, pull-down menu called Macros containing the physician’s custom-named drugs with physician-generated customized instructions and quantities (eg, special creams or a corticosteroid taper). Internet access on the wireless hand-held provides references for specialized disease and interaction information. I use my hand-held to access my E-mail as I move between patients. The keyboard allows me to use the hand-held to reply easily. We also use the E-mail system as an intra-office memo system for routine discussion and reminders.

Most of our prescriptions are faxed from the hand-held through the computer in our office to local pharmacies with the graphic of our log-in signature. Printed prescriptions can be sent to the printer in our office, which we then manually sign. PocketScript can provide digital signing when that complies with a state’s board of pharmacy regulations. Internet-enabled pharmacies can receive prescription data electronically using an E-mail–like system. There is no option for chart documentation other than printing a second script.

PocketScript is our choice because of its strong physician leadership, ethics, design, and serious commitment to the prescription process. It is usable by even the most computer phobic provider and does not attempt to replace anything but the paper prescription pad, phone, and wasted time—at least for now. Also, the vendor does not sell profiles of our prescribing habits. Security of the patient data and prescription process are excellent and are maintained even if the hand-held device is stolen. PocketScript is intuitive, simple, and efficient, with no tolerance for errors.

The design of the PocketScript server-based system eliminates the need to dock hand-held devices, avoiding time-consuming and frustrating tasks for the physician. This system is ideal for physicians who do most of their work in one or a few sites in which wireless nodes can be installed. We have changed office processes for refill and pharmacy calls, saving an estimated 1 hour per day per nurse and 30 minutes per day per file clerk. Patients and pharmacists frequently praise the clarity, accuracy, and convenience of the PocketScript system. PocketScript is a serious medical care partner and the perfect first step into e-medicine for our group practice.

ePhysician ePad
Reviewed by Daniel E. Diamond, MD

The ePhysician system uses a wireless wide area network (WAN) rather than a wireless local area network in the physician’s office. Prescriptions can be transmitted directly to the vendor’s server through an OmniSky (or equivalent) wireless cellular-WAN modem. The physicians who do not want to use the wireless interface can use the docking cradle attached to a computer in the office for synchronization. The prescription is then securely sent through the Internet to the vendor’s computer (using the cradle takes approximately 30 seconds per synchronization). The wireless method is more efficient and allows the physician to write and send the prescription without leaving the examination room. It also allows the physician to write and send prescriptions when away from the office.

ePad has several other unique features. Because patient demographics are entered into the physician’s existing practice management system and transmitted to the hand-held when it is docked, the physician does not have to enter any patient data. To write a prescription for a patient who has not previously been entered into the system, the user docks the hand-held in the cradle for synchronization. Also, all prescriptions are transmitted from the vendor’s central computer servers either electronically or by fax to any pharmacy in the United States. Users can also print from the hand-held to a local printer. To generate chart documentation, the user can configure ePad to automatically print a copy of the prescription on the local computer when it transmits to the pharmacy. ePad currently comes bundled with ePhysician Superbill for charge capture and is integrated with Drug Facts and Comparisons. It also includes a Web interface that allows my staff to efficiently process prescription renewals, and a scheduling module that makes it possible to schedule patients on the computer. This schedule is viewable on the hand-held device.

ePad by ePhysician with the OmniSky wireless service is simple, efficient, and powerful. Its interface was designed with a focus on the physician user. Although there is a subscription charge for ePad, the fee pales in comparison with the cost of lost physician productivity due to the inefficient technology of printing prescriptions locally (or writing them by hand). Using the ePad system in our office has resulted in very positive patient feedback and increased efficiency and job satisfaction among staff.

AllScripts
Reviewed by Azar A. Korbey, MD

At the installation stage, AllScripts’ on-site team realizes that the use of these systems require change. They take the time to train the staff and make them part of the process. AllScripts differs from the other systems, because after the selection of the patient name, users select their most frequently coded diseases listed by common name and International Classification of Diseases-ninth revision code. This list is prescriber-specific, not patient-specific. Users can also access the patient’s prescription and diagnosis histories, and the AllScripts system automatically checks for previous adverse drug reactions. Next, AllScripts shows the medicines that the user most frequently prescribes for a disease. AllScripts learns each user’s prescribing preferences, so users do not need to enter their favorite prescriptions. The list also shows preferred instructions for any given medication, and if AllScripts’ office dispensing system is used, it will tell users whether they have the drug in inventory. To produce a prescription, the prescriber may modify a prescription from a frequently used list or choose from any other drug in AllScripts’ database.

Users have access to all patients in the practice through the network, regardless of provider or schedule. Patients not registered with the practice can be manually added into the hand-held device or desktop computer server. Unlike the other systems, AllScripts automatically produces a mailing address-sized label for chart documentation that includes all details of the prescription. The vendor also sells a turnkey pharmacy service (FirstFill) for in-office dispensing. It is only for common medications, and is similar to providing samples. The local wireless network is fast; high-speed access to the Internet is proving very useful.

AllScripts is complete and does what it promises. It is a well-funded, publicly traded company that is adding applications and acquiring additional companies that will allow dictation and charge-capture systems to be added. I have had long-term experience with the product and its exemplary phone support, and I believe they will both be around in the future. They are even willing to come on site immediately, if required. I strongly recommend this system.

Editor’s Note: We had hoped to include the iScribe 3000 system (www.iscribe.com) in this review but could not obtain the product and complete the review in time for publication. We plan to publish a software review of iScribe in a future issue of JFP.

Do any of the following situations sound familiar?

1. You want to prescribe 25 mg of a drug but are uncertain whether the drug is supplied as a 25-mg or 50-mg tablet.
2. An elderly patient comes in for an update on her multiple problems. At the end of a long visit, she asks to have all 17 of her prescriptions refilled.
3. You need to determine whether there is a possible drug interaction between existing medications and an upcoming medicine.
4. You would like to be able to generate a list of all your patients taking a recently recalled drug.

Now imagine a hand-held prescribing device the size of a prescription pad that turns on instantly, knows the date, and through a connection to your computer, knows your patient list for the day. Using this device you could quickly select the patient’s name from a drop-down list and see his or her insurance plan, age, and other demographic information. You could select medications from an alphabetic scrolling list that includes information specific to the patient’s insurance plan. You could check allergies and drug interactions against the patient’s list of previously prescribed drugs and known medication allergies, alerting you only if a clinically significant problem is detected. After selecting the medication, you could choose from a list of available dosages and specify the quantity to be dispensed Figure 1, and the prescription could be digitally signed and sent electronically to the patient’s pharmacy or to your printer. Chart documentation could be produced concurrently on a sticky-backed label that can be affixed to the patient’s medical record. The prescription would be saved for generating rapid renewals in the future. Drug information and insurance formularies would be updated regularly from central computers, and new drugs and dosage forms could appear on the hand-held devices before they show up on the pharmacists’ shelves.

Does all this sound far-fetched? It’s not. This type of technology is currently available. Systems range from simple prescription writers—a hand-held device with initial manual entry of patient data and prescription printing from an infrared-equipped printer—to full electronic medical records systems. Prospective users have to decide which features provide sufficient benefit to merit the investment of time necessary to learn them. In this article, 3 systems (PocketScript, ePhysician ePad, and AllScripts;)Table 1, Table 1a are reviewed by physicians who have successfully incorporated electronic prescribing into their practices.

A system that does nothing other than produce legible prescriptions in the same time it would take to write one by hand can benefit many physicians. One that provides an accurate record of previous prescriptions that allows the rapid renewal of multiple prescriptions is even more valuable. Another important potential benefit of an electronic prescribing system is fewer errors, including those caused by illegible handwriting, incorrect dosage selections, drug-drug or drug-disease interactions, and drug allergies.

These systems still have some limitations, however. Most do not automatically suggest a dosage adjustment for age, renal function, hepatic function, and alcohol and tobacco use (although drug monographs may be available on the hand-held device itself). Similarly, weight-based pediatric prescription information may be available in monographs on the hand-held device but is not built into the prescribing process. Some states allow electronic transmission of prescriptions, some do not, and some have made no ruling; specific rules vary. Fortunately, each of the vendors reviewed in this article made certain that their system is installed in a way that conforms to state laws.

Most primary care physicians have not made the leap to electronic medical records; package delivery services and supermarket checkouts use more technology at the point of service than the average physician does. These devices may change that by providing a relatively low-cost low-risk improvement with less impact on work processes than a full electronic medical record.

Common Features

Each of the systems we reviewed is “wired” somewhat differently Figure 2. PocketScript and AllScripts use a wireless local area network in the physician’s office to move information from the hand-held device to a desktop computer server in the physician’s office. This server is in turn connected to the vendor’s computer through a modem or other high-speed access device to get drug information updates. The office computer also sends prescriptions directly to pharmacies either as an electronic message or by fax.

A system using ePhysician includes a personal computer in the physician’s office to load patient demographic information onto the hand-held using a docking cradle. Prescriptions can be sent directly from the hand-held device to a central computer server at the vendor’s offices using a wireless modem. Alternately, they physician can dock the hand-held device in a cradle attached to the computer, which is connected by a modem to the vendor’s central computer. The vendor’s computer server then sends prescriptions to the selected pharmacy, either electronically or by fax. All systems use sophisticated encryption to protect the confidentiality of data as it is sent between computers.

The process of generating a prescription follows that of using a paper prescription pad. Prescribers may override the predetermined choices and enter personal choices for quantity, instructions, and refills. The physician then views the completed prescription information, confirms the pharmacy information, and sends it through the wireless network and the computer system to the pharmacy or prints it in the office. All products support some type of list to facilitate rapid generation of custom prescriptions. Prescriptions produced by these systems benefit from perfect legibility, fraud and error reduction, automatic drug interactions checking (except ePad), and improved formulary compliance.

Each system can connect to front office systems and move patient demographic data into the prescription writing system. A real time interface updates both applications simultaneously, so information entered in the registration system is immediately available in the prescription writing system. With a bridge program, information entered into the registration system is unavailable in the prescription-writing program until the program is run and transfers the data to the prescription writing system.

PocketScript
Reviewed by Eric Weidmann, MD

PocketScript is unique, because the user can select voice recognition, screen touch, or the keyboard to quickly compose and send prescriptions. Patients’ records may be added or edited from the hand-held PC or the physician’s server. Hand-held to server communication is fast enough to provide high-speed Internet connectivity from the examination room, which is useful for patient care, business, and personal uses. Drug-drug interactions are checked automatically when choosing a drug and before sending the prescription. Patient allergy and diagnosis fields are promised for future releases.

PocketScript does not have to be linked to the practice’s registration system. If not linked, each patient must be entered manually the first time only. Patient data (eg, weight, insurance plan, and so on) can be changed using the Patient Maintenance function on the hand-held device. A scrollable Favorites list can be created of the physician’s most commonly used drugs; it uses PocketScript’s standard instructions for the signature. There is a scrollable, pull-down menu called Macros containing the physician’s custom-named drugs with physician-generated customized instructions and quantities (eg, special creams or a corticosteroid taper). Internet access on the wireless hand-held provides references for specialized disease and interaction information. I use my hand-held to access my E-mail as I move between patients. The keyboard allows me to use the hand-held to reply easily. We also use the E-mail system as an intra-office memo system for routine discussion and reminders.

Most of our prescriptions are faxed from the hand-held through the computer in our office to local pharmacies with the graphic of our log-in signature. Printed prescriptions can be sent to the printer in our office, which we then manually sign. PocketScript can provide digital signing when that complies with a state’s board of pharmacy regulations. Internet-enabled pharmacies can receive prescription data electronically using an E-mail–like system. There is no option for chart documentation other than printing a second script.

PocketScript is our choice because of its strong physician leadership, ethics, design, and serious commitment to the prescription process. It is usable by even the most computer phobic provider and does not attempt to replace anything but the paper prescription pad, phone, and wasted time—at least for now. Also, the vendor does not sell profiles of our prescribing habits. Security of the patient data and prescription process are excellent and are maintained even if the hand-held device is stolen. PocketScript is intuitive, simple, and efficient, with no tolerance for errors.

The design of the PocketScript server-based system eliminates the need to dock hand-held devices, avoiding time-consuming and frustrating tasks for the physician. This system is ideal for physicians who do most of their work in one or a few sites in which wireless nodes can be installed. We have changed office processes for refill and pharmacy calls, saving an estimated 1 hour per day per nurse and 30 minutes per day per file clerk. Patients and pharmacists frequently praise the clarity, accuracy, and convenience of the PocketScript system. PocketScript is a serious medical care partner and the perfect first step into e-medicine for our group practice.

ePhysician ePad
Reviewed by Daniel E. Diamond, MD

The ePhysician system uses a wireless wide area network (WAN) rather than a wireless local area network in the physician’s office. Prescriptions can be transmitted directly to the vendor’s server through an OmniSky (or equivalent) wireless cellular-WAN modem. The physicians who do not want to use the wireless interface can use the docking cradle attached to a computer in the office for synchronization. The prescription is then securely sent through the Internet to the vendor’s computer (using the cradle takes approximately 30 seconds per synchronization). The wireless method is more efficient and allows the physician to write and send the prescription without leaving the examination room. It also allows the physician to write and send prescriptions when away from the office.

ePad has several other unique features. Because patient demographics are entered into the physician’s existing practice management system and transmitted to the hand-held when it is docked, the physician does not have to enter any patient data. To write a prescription for a patient who has not previously been entered into the system, the user docks the hand-held in the cradle for synchronization. Also, all prescriptions are transmitted from the vendor’s central computer servers either electronically or by fax to any pharmacy in the United States. Users can also print from the hand-held to a local printer. To generate chart documentation, the user can configure ePad to automatically print a copy of the prescription on the local computer when it transmits to the pharmacy. ePad currently comes bundled with ePhysician Superbill for charge capture and is integrated with Drug Facts and Comparisons. It also includes a Web interface that allows my staff to efficiently process prescription renewals, and a scheduling module that makes it possible to schedule patients on the computer. This schedule is viewable on the hand-held device.

ePad by ePhysician with the OmniSky wireless service is simple, efficient, and powerful. Its interface was designed with a focus on the physician user. Although there is a subscription charge for ePad, the fee pales in comparison with the cost of lost physician productivity due to the inefficient technology of printing prescriptions locally (or writing them by hand). Using the ePad system in our office has resulted in very positive patient feedback and increased efficiency and job satisfaction among staff.

AllScripts
Reviewed by Azar A. Korbey, MD

At the installation stage, AllScripts’ on-site team realizes that the use of these systems require change. They take the time to train the staff and make them part of the process. AllScripts differs from the other systems, because after the selection of the patient name, users select their most frequently coded diseases listed by common name and International Classification of Diseases-ninth revision code. This list is prescriber-specific, not patient-specific. Users can also access the patient’s prescription and diagnosis histories, and the AllScripts system automatically checks for previous adverse drug reactions. Next, AllScripts shows the medicines that the user most frequently prescribes for a disease. AllScripts learns each user’s prescribing preferences, so users do not need to enter their favorite prescriptions. The list also shows preferred instructions for any given medication, and if AllScripts’ office dispensing system is used, it will tell users whether they have the drug in inventory. To produce a prescription, the prescriber may modify a prescription from a frequently used list or choose from any other drug in AllScripts’ database.

Users have access to all patients in the practice through the network, regardless of provider or schedule. Patients not registered with the practice can be manually added into the hand-held device or desktop computer server. Unlike the other systems, AllScripts automatically produces a mailing address-sized label for chart documentation that includes all details of the prescription. The vendor also sells a turnkey pharmacy service (FirstFill) for in-office dispensing. It is only for common medications, and is similar to providing samples. The local wireless network is fast; high-speed access to the Internet is proving very useful.

AllScripts is complete and does what it promises. It is a well-funded, publicly traded company that is adding applications and acquiring additional companies that will allow dictation and charge-capture systems to be added. I have had long-term experience with the product and its exemplary phone support, and I believe they will both be around in the future. They are even willing to come on site immediately, if required. I strongly recommend this system.

Editor’s Note: We had hoped to include the iScribe 3000 system (www.iscribe.com) in this review but could not obtain the product and complete the review in time for publication. We plan to publish a software review of iScribe in a future issue of JFP.

The extent to which the physician intervenes with a hospitalized patient who has an alcohol use disorder correlates with the patient’s reported change in alcohol use after discharge.¹ Primary care physicians may be called on to help manage hospitalized patients with alcohol use disorders, but exactly what they should do to help these patients is not always clear.

Alcohol abuse and trauma are common and related clinical problems.² A dose-response relationship has been observed between alcohol consumption and the risk of fatal injury.³ Traumatic injury is a major public health problem and a leading cause of morbidity and mortality in the United States. It ranks first in years of life lost, first in the utilization of hospital-days, second in disability-adjusted life-years, and fourth in overall mortality.⁴ Sims and colleagues⁵ found that violent trauma had a recurrence rate of 44% and a 5-year mortality rate of 20% and that 62% of these patients abused alcohol or drugs. Rivera and coworkers⁶ found that trauma victims who were intoxicated on presentation to a trauma center were 2.5 times more likely to be readmitted for another injury than those who were not intoxicated, and those with evidence of a chronic alcohol problem were 3.5 times more likely. Others have noted similar findings.^7,8 There is a unique opportunity to initiate treatment for patients with substance abuse disorders when they are hospitalized for a traumatic injury.⁹ Often this opportunity is missed.^10,11

It is not known how to intervene with victims of alcohol-related injuries to prevent subsequent injuries. Currently there are several options that could potentially improve the outcomes of hospitalized patients who have substance use disorders, such as brief advice, brief interventions, referral to a consultation team, and referral to a treatment center. At the very least, trauma victims with substance abuse problems should be given some brief advice from the surgeon. Although most surgeons appear willing to give this advice, many feel inadequately prepared to do it.¹² Thus, the burden of performing these interventions may fall to the patient’s primary care physician or the physician who is requested by the surgeon to provide consultation services.

A technique known as “brief intervention” consists of advice in a structured format that is given to patients with a substance use disorder.¹³ These interventions have been found to be effective in a number of clinical settings, including outpatient primary care.^14-16 In particular, brief interventions performed by a trained psychologist in a trauma center have been associated with a reduction in alcohol intake and a reduced risk of trauma recidivism.¹⁷ However, these interventions require significant physician training to implement. What is needed is a simple and practical method that can be used by primary care physicians that does not require extensive physician training.

Peer interventions have been used successfully in education.^18,19 This success is based in some part on what is known as the “attraction paradigm”. The attraction process purports that the more similar the members of a relationship are in experiences, the more likely they will respond to one another positively.²⁰ Peers have been used in some settings to augment treatment in primary care. In one study, trained peers who were recovered from depression were found to provide no additional improvement in clinical outcomes.²¹ In that study, one group of patients with depression who were treated with antidepressant drugs and emotional support provided by a nurse during 10 6-minute telephone calls over a 4-month period were compared with another group who also received peer support. The finding is not surprising, because peers cannot be expected to add much benefit to patients who are already receiving optimal treatment. Volunteers from Alcoholics Anonymous (AA) have been used to talk with alcoholic patients in a general hospital.²² Although impressions are that these peers are helpful, the outcomes of this procedure have not been well studied. This process can also be performed by a professional and has been called Twelve Step Facilitation.²³

At our institution, volunteers from the community who were active in AA were used to speak with patients who were admitted to the hospital with alcohol-related injuries following a brief intervention from a primary care physician. These peers appeared to produce favorable outcomes with our patients. The purpose of our study was to evaluate the effectiveness of this approach and to test the alternative hypothesis that those in a peer intervention group would demonstrate more favorable outcomes than those in a brief intervention group who, in turn, would demonstrate more favorable outcomes than those in a control group.

Methods

Setting

We conducted this study in a Level I trauma center located in the primary university teaching hospital that serves a metropolitan area of more than 1 million people in a 2-state area of the Midwest. The trauma service is staffed by 2 teams of attending and resident surgeons who alternate 24-hour shifts. An addiction medicine physician provides consultative services to these patients.

Study Population

A total of 2530 patients were admitted to the hospital trauma service for injuries between August 1, 1998, and March 31, 2000, and 957 (37.8%) of these had positive toxicology tests (351 alcohol only, 352 drugs only, and 254 alcohol and drugs) on admission to the hospital. Positive toxicology tests were defined as a blood alcohol concentration (BAC) of 4.34 mmol per L or greater (Ž20 mg/dL) and/or the detection of psychoactive drugs. Toxicology screening was not performed in approximately 20% of the patients. Also, 95 patients with negative toxicology screens were known to have an active alcohol use disorder on admission to the hospital. Thus, 1052 patients served as potential study subjects.

The flow of patients through the study is shown in Table 1. A total of 738 patients were excluded as potential subjects, 632 by block randomization and 106 for other reasons. Since patients with positive toxicology tests who did not have either alcohol abuse or alcohol dependence as defined by the Diagnostic and Statistical Manual of Mental Disorders, 4th edition²⁴ were excluded from the study, all of those who were ultimately eligible for follow-up had an alcohol use disorder.

Before the patients were contacted for follow-up, they were categorized into 1 of 3 groups: a usual care group (n=125), a brief intervention group (n=119), or a peer intervention group (n=70), according to the study methods. Volunteer availability often determined which patients received a peer intervention or a brief intervention. One patient had been originally assigned to the brief intervention group, but we learned at the time of the telephone follow-up interview that a family member had arranged for the patient’s AA sponsor to visit the patient on several occasions before and after hospital discharge. This patient was subsequently excluded from our study and is not included in the numeric values of Table 1.

Procedures

This was a retrospective nonrandomized intervention study evaluating the effectiveness of interventions used to encourage trauma patients to abstain from alcohol and to initiate substance abuse treatment or self-help. We obtained initial data retrospectively from the patient’s medical record, including the patient’s demographic characteristics (age, sex, race), the patient’s telephone numbers, and the telephone numbers of relatives or friends.

The follow-up telephone interviews were conducted at 2 different times. The university’s institutional review board approved both parts of our study. The main outcome measures were: (1) complete abstinence from alcohol during the first 6 months following discharge from the hospital, (2) abstinence from alcohol during the sixth month following hospital discharge (ie, those who drank initially after discharge but subsequently became abstinent), and (3) initiation of professional alcohol treatment or self-help. The patients were considered to have initiated treatment or self-help if within the first 6 months following hospital discharge they had either: (1) attended at least one AA meeting, (2) visited a mental health or substance abuse professional at least once, (3) attended at least 1 session at an outpatient alcohol treatment center, or (4) spent at least 1 day at an inpatient or residential alcohol treatment program.

The first part of the follow-up was conducted as a component of a quality improvement program and involved patients who were admitted to the hospital between August 1, 1998, and June 30, 1999. These patients were contacted between September 1999 and January 2000. One of the investigators attempted to contact the patient and/or a relative or friend identified from the medical record 6 to 12 months after the subject (n=86) was discharged from the hospital. A standard introduction was read to the respondent, and verbal consent was obtained. Following this, a series of open-ended questions was asked (eg, “Are you better?” and “In what way?”). Then some specific questions were asked. The responses were summarized according to the subject’s patterns of alcohol use since hospital discharge and whether the subject had initiated substance abuse treatment or a self-help program. During March 2000, the responses about drinking patterns and initiation of treatment or self-help during the first 6 months following hospital discharge were categorized and coded by 2 of the authors (S.B.R. and R.L.M.).

The second part of the follow-up was conducted as a medical student summer research project and involved patients who were admitted to the hospital between June 1, 1999, and March 31, 2000. These patients were contacted during June and July of 2000. There was a 1-month overlap in admission dates with the first part of the study because of a variation in hospital length of stay. Another investigator attempted to contact the patient and/or a relative or friend identified from the medical record 4 to 12 months after the study subject (n=228) was discharged from the hospital. A standard introduction was read to the respondent, and verbal consent was obtained. Following this, a series of structured questions was asked (eg, “During the first 6 [4 or 5 months in 7 cases] months following your discharge from the hospital did you try to cut down or quit drinking?”). The responses were coded according to the patient’s patterns of alcohol use during the 6 months following hospital discharge and whether the subject had initiated substance abuse treatment or a self-help program.

Interventions

Patients who received usual care served as the control group (n=125). There were 2 groups that received an intervention: a brief intervention group (n=119) and a peer intervention group (n=70).

Usual Physician Care. Patients in this group received care by the residents and attending surgeons of the trauma service only, because the addiction medicine consultant was not available to see them. The surgeons and the hospital’s social workers may or may not have specifically addressed the patients’ substance abuse problems before discharge. During the study period, the hospital nurses, social workers, and resident physicians were given a 1-hour educational conference (in 12 separate sessions) about alcohol detoxification, screening, and brief intervention based on a national standard.²⁵ Preprinted protocols for detoxification were available in the hospital.

Brief Intervention. Patients in this group received the services of an addiction medicine consultant as part of their overall hospital care. Before discharge, these patients were given brief (5 to 15 minutes) advice following a previously described method.²⁶

Peer Intervention. Patients in this group received the same type of brief physician advice of those in the brief intervention group, and a 30- to 60-minute visit from a peer who was active in AA. Patients had to agree to a peer visit, but refusals were rare. Volunteers were recruited through a local residential facility for individuals with substance abuse problems. There were separate facilities for men and women, but they were administered by the same organization and followed the same basic program based on the AA model. The volunteers, called assistant staff, had successfully completed the program at that facility. These volunteers attended 3 2-hour training workshops designed to increase their skills at carrying the message of AA to others. These training workshops included both didactic and role-playing sessions that were designed to help the volunteers follow a protocol based on the AA model. These peers visited with the patient in pairs before hospital discharge. They did not give advice or make treatment recommendations. Instead, they shared their personal stories and their “experience, strength, and hope” with the patient. The peers were always matched with the patients’ sex and usually with the patients’ race. There was a period of time (approximately 6 months) during the study when women volunteers were not available because the facility for women was being relocated.

Statistical Analysis

The data sets from the 2 parts of the study were combined by one of the authors who also performed the data analysis. We used the exact version of the Fisher-Freeman-Halton test²⁷ to compare the 3 treatment groups in terms of categorical baseline characteristics and follow-up rates. One-way analysis of variance was used for continuous baseline characteristics. The same analysis was performed to compare those patients for whom follow-up data could be obtained with those for whom such data could not be obtained.

We use the exact version of the Cochran-Armitage test for trend²⁸ to test the null hypothesis of no difference among the treatment groups against the alternative hypothesis that the true proportions of positive outcomes would be in the following order: control < brief intervention < peer intervention (ie, we hypothesized that the success rate for the peer intervention group would be greater than that for the brief intervention group and that the success rate for the brief intervention group would be greater than that for the control group). If a significant difference was found among the treatment groups, we used the Fisher exact test with a Bonferroni adjustment to determine which pairs of treatment groups differed from each other. Stratified analysis was used to adjust for the effect of any confounding variables.

A sample size of 28 per group was sufficient to achieve 80% power for detecting differences in success rates (as measured by initiation of treatment or self-help) among the groups of 5% in the control group, 10% in the brief intervention group, and 30% in the peer intervention group using a one-tailed significance level of 0.05. All computations for the study were performed using Epi Info Version 6.04c (USD Inc, Stone Mountain, Ga, 1999), StatXact 4.0.1 (CYTEL Software Corp, Cambridge, Mass, 1998), and SPSS software version 10.0 (SPSS, Inc, Chicago, Ill, 1999). Continuous variables were summarized as mean plus or minus the standard deviation.

Results

Of the 314 patients in the study 258 (82.2%) were men; 244 (77.7%) were white; and the mean age was 37.2 years plus or minus 12.5 years (range=18-80). The mean blood alcohol concentration on admission for these 314 patients was 35.8 mmol per L plus or minus 26.5 mmol per L (165 mg/dL±122 mg/dL ) with a range of 00.0 to 143.3 mmol per L (000-660 mg/dL).

Of the 314 patients in our study, 140 (44.6%) were contacted following hospital discharge through communication with the subject, the subjects’ relatives, or both. Among the members of the control group, the follow-up rate was 35.2% (44/125); among those who received a brief intervention it was 47.9% (57/119); and among those who received a peer intervention, it was 55.7% (39/70). This represents a statistically significant difference at the Bonferroni cutoff of 0.05 divided by 3 (0.0167) between the control and peer intervention groups (P=.003) but not between the control and brief intervention groups (P=.023), or the brief and peer intervention groups (P=.152) using the Fisher exact test Table 1.

Among the 140 patients in the study, follow-up data were obtained from the patient in 97 instances (69%), from a friend or family member in 38 (27%), and from other sources in 5 (4%). For the 44 members of the control group, follow-up data were obtained from the patient in 37 instances (84%), from a friend or family member in 6 (14%), and from other sources in 1 (2%). For the 57 patients who received a brief intervention, follow-up data were obtained from the patient in 35 instances (61%), from a friend or family member in 21 (37%), and from other sources in 1 (2%). For the 39 patients who received a peer intervention, follow-up data were obtained from the patient in 25 instances (64%), from a friend or family member in 11 (28%), and from other sources in 3 (8%). The Fisher-Freeman-Halton test indicates a significant difference between the control and brief intervention groups (P=.012) but no difference between the control group and the peer intervention group (P=.117) or between the brief and peer intervention groups (P=.341), using the Bonferroni criterion of 0.0167. Those patients for whom follow-up data could be obtained were compared with those for whom it could not be obtained in terms of age, race, sex, and BAC on admission. The only significant difference that was found was for race: Follow-up data were available for 49.2% of the white patients but for only 28.6% of the nonwhite patients (P=.003). In terms of sex, follow-up data were available for 42.2% of the men and 55.4% of the women (P=.051). The mean age of those for whom follow-up data were available was 38.1 years plus or minus 12.8, compared with 36.4 years plus or minus 12.3 for those lost to follow-up (P=.226). The mean BAC on admission was 38.0 mmol per L plus or minus 27.8 (175 mg/dL±128) for those we were able to follow up, compared with 34.1 mmol per L plus or minus 25.4 (157 mg/dL±117) for those lost to follow-up (P=.233).

Comparisons of the baseline characteristics of the 140 patients across the 3 treatment groups are shown in Table 2. No significant differences were found at baseline between the groups at the 0.05 level except for male sex (P=.003); however, BAC almost reached statistical significance (P=.054).

The results for the main outcome measures of the 3 groups are shown in Table 3. The data reflect the fact that 7 patients drank for several weeks following hospital discharge but then abstained from drinking. As hypothesized, the success rates were greatest in the peer intervention group, followed by the brief intervention and control groups. All 3 outcomes showed statistically significant differences across groups. In terms of pairwise comparisons, the comparison between the control group and the peer intervention group met the Bonferroni criterion of 0.0167 for both abstinence for 6 months following hospital discharge (P=.013) and abstinence during the sixth month following hospital discharge (P=.007). For initiation of treatment or self-help, the comparisons of the peer group with both the control group and the brief intervention group were significant using the Bonferroni criterion (P <.001 in both cases).

Stratifying by sex yielded results that were not materially different from those presented in the Table (P=.016 for 6 months of abstinence; P=.007 for abstinence at during the sixth month; and P <.001 for initiation of treatment or self-help). Stratifying by BAC also did not affect the P values in any material way (data not shown).

Because of inconsistencies between the data from the 2 parts of the study and because of missing or unrecorded data, we can only make qualitative statements about other outcomes. No patient who was completely abstinent for the entire 6 months following hospital discharge had began drinking again by the time of the telephone interview. Many patients in the intervention groups (approximately a third) drank after hospital discharge and continued to drink up to the time of the follow-up interview, although a few of these patients claimed to have cut down. Only a few patients initially abstained from alcohol but returned to drinking at the time of follow-up. Most of the follow-up information came from the patient, our preferred source for outcome data. No patient who claimed to be abstinent had a family member who contradicted that report. However, several patients admitted to drinking (or using drugs) who had a member of the family who reported that the patient was abstinent. In those cases in which a family member could be located but the patient could not, it was usually because the patient was still drinking, living on the streets, or had no telephone. It was rare that the family member reported a favorable outcome (ie, abstinence), and we could not confirm this directly with the patient.

Several patients in the peer intervention group expressed gratitude for the help they received with their drinking problems while in the hospital and especially for the visits by the peers. Some of these patients dramatically changed their lives. At least 3 patients in the peer intervention group went from being unemployed and homeless to full-time employment and having a permanent residence after they entered a treatment program and became involved in AA. They credited the peer intervention as being the most important factor that motivated them to seek help for their alcohol use disorder. At the time this manuscript was being prepared, one of these individuals was serving as a volunteer making visits to hospitalized patients with drinking problems.

Discussion

Previous studies have shown that brief interventions by professionals appear to help motivate patients to reduce drinking. Our study demonstrates that peers may help motivate patients to initiate treatment or self-help as well as promote abstinence. Brief physician advice followed by a visit with a volunteer from AA shows promise as a simple, practical, inexpensive, and effective intervention that may help to prevent patients from returning to alcohol use. This could lead to reductions in recurrent injuries for patients hospitalized with alcohol-related injuries.

Primary care physicians could use this approach to intervene with any patient hospitalized with alcohol-related problems. At our institution, peer volunteers are often called to visit patients with substance use disorders who are hospitalized by the surgery, medicine, family medicine, and psychiatry services. We used trauma patients, because there is a large volume of such patients at our institution who routinely have had toxicology tests performed on admission. Also, an existing trauma registry database facilitated the collection of patient data.

Many primary care physicians already possess the skills required to give patients brief advice about harmful lifestyles and are familiar with the use of community resources that can help their patients. Most communities that are large enough to have a hospital are large enough to support several AA groups. As part of the AA program, members are expected to carry the message of AA to alcoholics who are still drinking. They consider this Twelfth Step Work an essential part of the program that leads to personal progress in AA. Most physicians can easily identify patients who could benefit from hearing the message of AA. It is often not difficult to link up these 2 groups of individuals.²⁹ The local AA office can be called from the patient’s bedside telephone. After the physician explains the situation to the person who answers the call, the telephone can then be given to the patient. If the patient agrees, a member of AA may come to the hospital for a visit. These visits typically last 30 minutes to an hour. Sometimes the AA member may visit again during the patient’s hospital stay or at the time of discharge to escort the patient to an AA meeting. This service is provided without cost to the patient, the patient’s insurance carrier, or the hospital.

Limitations

Our study has many of the limitations of initial retrospective studies: a nonrandomized design, a study sample limited to a particular type of patient, limited follow-up data, variation in the interval from the time of the intervention to the time of follow-up data collection, reliance on self-report, and treatment groups that were not masked to the follow-up interviewers. The nonrandomized design might suggest that some of the favorable outcomes could be the result of selection bias. However, as indicated in Table 2, the baseline characteristics of the 3 groups were similar, except that women were under-represented in the peer intervention group. This finding is probably because of the limited availability of women peer volunteers during a 6-month period of time during the study. The trend towards a lower BAC in the control group suggests that patients with severe alcohol problems may be over-represented in the experimental groups. If anything, this would have biased the study results against the 2 intervention groups. However, the diagnosis of an alcohol use disorder was made using a chart audit for the control group (which did not always provide enough information to differentiate abuse from dependence), while an unstructured patient interview was used for the intervention group. Limited follow-up is a frequent problem in the patient populations used for alcohol use studies. Response rates of approximately 50% are typical. It is not clear why the follow-up rates for nonwhite patients were lower than for white subjects. We did seem to experience more problems with disconnected telephones in the nonwhite population, suggesting that there may be some economic differences between the 2 groups. We observed a significantly better follow-up rate for those in the peer intervention group. This is probably because we established contacts for follow-up directly from the patient in the intervention groups and could verify telephone numbers, but we had to rely on the medical record for the telephone numbers of the members of the control group. These numbers were not always correct. Also, we had significantly fewer family contacts in the control group. We did not record the exact timing of the follow-up after hospital discharge for each individual patient, and therefore could not compare the mean follow-up intervals between groups. However, for the reasons mentioned in the qualitative part of the results section, we do not believe that this problem would have influenced our results in any material manner. Our study was performed with trauma patients who may not be representative of other patient groups. Painful injuries and court appearances related to driving while intoxicated may be important factors that influence drinking behaviors. However, we have observed some nonsurgical patients who have benefited from peer interventions. We relied on patient self-report for outcomes and found a difference between the control and the 2 intervention groups. Although patients with alcohol use disorders may not accurately report their alcohol consumption, it is unlikely that those in the intervention groups would be more likely to report abstinence or to report initiation of treatment or self-help than those in the control group. We preferentially coded the poorest outcome information we obtained from the patient or the family member. Therefore, the source of the follow-up data had a minimal favorable impact on its accuracy. Although the follow-up interviewers knew to which group an individual patient belonged, they asked the interview questions from a printed script, to reduce observer bias to a minimum. Finally, although we obtained severity of disease data for the intervention groups (ie, abuse vs dependence) this information was not available for the control group.

Conclusions

The significant findings of our study suggest that the methods we employed should be evaluated in a well-funded rigorously designed prospective randomized study with more patients who would be objectively evaluated for the severity of their alcohol use disorder and with mechanisms to confirm and quantify the subjects’ self-reports of alcohol consumption and to ensure higher follow-up rates.

In the meantime, physicians can request that members of AA visit their hospitalized patients who have alcohol use disorders. Interventions by recovering alcoholics are not difficult to arrange, involve no costs, pose little patient risk, and might be of great benefit to some patients. We have continued to observe individual patients who were able to find sobriety following these interventions. These patients have expressed opinions that it was primarily the peers who motivated them to seek help for their problem drinking.

Acknowledgments

This work was supported, in part, by the University of Louisville Summer Research Scholarship Program and the University of Louisville Hospital Trauma Institute. We are indebted to the anonymous alcoholic members of a local self-help organization and to The Healing Place for assistance with locating volunteers to visit with our patients. We thank Karen Newton and Gail Wulfman for their assistance with the training of the volunteers. We thank Phillip Boaz, Janet Wallace, and Lance Hottman for their help with the data collection and Margaret M. Steptoe and Murphy Shields for their assistance in the preparation of this manuscript.

The extent to which the physician intervenes with a hospitalized patient who has an alcohol use disorder correlates with the patient’s reported change in alcohol use after discharge.¹ Primary care physicians may be called on to help manage hospitalized patients with alcohol use disorders, but exactly what they should do to help these patients is not always clear.

Alcohol abuse and trauma are common and related clinical problems.² A dose-response relationship has been observed between alcohol consumption and the risk of fatal injury.³ Traumatic injury is a major public health problem and a leading cause of morbidity and mortality in the United States. It ranks first in years of life lost, first in the utilization of hospital-days, second in disability-adjusted life-years, and fourth in overall mortality.⁴ Sims and colleagues⁵ found that violent trauma had a recurrence rate of 44% and a 5-year mortality rate of 20% and that 62% of these patients abused alcohol or drugs. Rivera and coworkers⁶ found that trauma victims who were intoxicated on presentation to a trauma center were 2.5 times more likely to be readmitted for another injury than those who were not intoxicated, and those with evidence of a chronic alcohol problem were 3.5 times more likely. Others have noted similar findings.^7,8 There is a unique opportunity to initiate treatment for patients with substance abuse disorders when they are hospitalized for a traumatic injury.⁹ Often this opportunity is missed.^10,11

It is not known how to intervene with victims of alcohol-related injuries to prevent subsequent injuries. Currently there are several options that could potentially improve the outcomes of hospitalized patients who have substance use disorders, such as brief advice, brief interventions, referral to a consultation team, and referral to a treatment center. At the very least, trauma victims with substance abuse problems should be given some brief advice from the surgeon. Although most surgeons appear willing to give this advice, many feel inadequately prepared to do it.¹² Thus, the burden of performing these interventions may fall to the patient’s primary care physician or the physician who is requested by the surgeon to provide consultation services.

A technique known as “brief intervention” consists of advice in a structured format that is given to patients with a substance use disorder.¹³ These interventions have been found to be effective in a number of clinical settings, including outpatient primary care.^14-16 In particular, brief interventions performed by a trained psychologist in a trauma center have been associated with a reduction in alcohol intake and a reduced risk of trauma recidivism.¹⁷ However, these interventions require significant physician training to implement. What is needed is a simple and practical method that can be used by primary care physicians that does not require extensive physician training.

Peer interventions have been used successfully in education.^18,19 This success is based in some part on what is known as the “attraction paradigm”. The attraction process purports that the more similar the members of a relationship are in experiences, the more likely they will respond to one another positively.²⁰ Peers have been used in some settings to augment treatment in primary care. In one study, trained peers who were recovered from depression were found to provide no additional improvement in clinical outcomes.²¹ In that study, one group of patients with depression who were treated with antidepressant drugs and emotional support provided by a nurse during 10 6-minute telephone calls over a 4-month period were compared with another group who also received peer support. The finding is not surprising, because peers cannot be expected to add much benefit to patients who are already receiving optimal treatment. Volunteers from Alcoholics Anonymous (AA) have been used to talk with alcoholic patients in a general hospital.²² Although impressions are that these peers are helpful, the outcomes of this procedure have not been well studied. This process can also be performed by a professional and has been called Twelve Step Facilitation.²³

At our institution, volunteers from the community who were active in AA were used to speak with patients who were admitted to the hospital with alcohol-related injuries following a brief intervention from a primary care physician. These peers appeared to produce favorable outcomes with our patients. The purpose of our study was to evaluate the effectiveness of this approach and to test the alternative hypothesis that those in a peer intervention group would demonstrate more favorable outcomes than those in a brief intervention group who, in turn, would demonstrate more favorable outcomes than those in a control group.

Methods

Setting

We conducted this study in a Level I trauma center located in the primary university teaching hospital that serves a metropolitan area of more than 1 million people in a 2-state area of the Midwest. The trauma service is staffed by 2 teams of attending and resident surgeons who alternate 24-hour shifts. An addiction medicine physician provides consultative services to these patients.

Study Population

A total of 2530 patients were admitted to the hospital trauma service for injuries between August 1, 1998, and March 31, 2000, and 957 (37.8%) of these had positive toxicology tests (351 alcohol only, 352 drugs only, and 254 alcohol and drugs) on admission to the hospital. Positive toxicology tests were defined as a blood alcohol concentration (BAC) of 4.34 mmol per L or greater (Ž20 mg/dL) and/or the detection of psychoactive drugs. Toxicology screening was not performed in approximately 20% of the patients. Also, 95 patients with negative toxicology screens were known to have an active alcohol use disorder on admission to the hospital. Thus, 1052 patients served as potential study subjects.

The flow of patients through the study is shown in Table 1. A total of 738 patients were excluded as potential subjects, 632 by block randomization and 106 for other reasons. Since patients with positive toxicology tests who did not have either alcohol abuse or alcohol dependence as defined by the Diagnostic and Statistical Manual of Mental Disorders, 4th edition²⁴ were excluded from the study, all of those who were ultimately eligible for follow-up had an alcohol use disorder.

Before the patients were contacted for follow-up, they were categorized into 1 of 3 groups: a usual care group (n=125), a brief intervention group (n=119), or a peer intervention group (n=70), according to the study methods. Volunteer availability often determined which patients received a peer intervention or a brief intervention. One patient had been originally assigned to the brief intervention group, but we learned at the time of the telephone follow-up interview that a family member had arranged for the patient’s AA sponsor to visit the patient on several occasions before and after hospital discharge. This patient was subsequently excluded from our study and is not included in the numeric values of Table 1.

Procedures

This was a retrospective nonrandomized intervention study evaluating the effectiveness of interventions used to encourage trauma patients to abstain from alcohol and to initiate substance abuse treatment or self-help. We obtained initial data retrospectively from the patient’s medical record, including the patient’s demographic characteristics (age, sex, race), the patient’s telephone numbers, and the telephone numbers of relatives or friends.

The follow-up telephone interviews were conducted at 2 different times. The university’s institutional review board approved both parts of our study. The main outcome measures were: (1) complete abstinence from alcohol during the first 6 months following discharge from the hospital, (2) abstinence from alcohol during the sixth month following hospital discharge (ie, those who drank initially after discharge but subsequently became abstinent), and (3) initiation of professional alcohol treatment or self-help. The patients were considered to have initiated treatment or self-help if within the first 6 months following hospital discharge they had either: (1) attended at least one AA meeting, (2) visited a mental health or substance abuse professional at least once, (3) attended at least 1 session at an outpatient alcohol treatment center, or (4) spent at least 1 day at an inpatient or residential alcohol treatment program.

The first part of the follow-up was conducted as a component of a quality improvement program and involved patients who were admitted to the hospital between August 1, 1998, and June 30, 1999. These patients were contacted between September 1999 and January 2000. One of the investigators attempted to contact the patient and/or a relative or friend identified from the medical record 6 to 12 months after the subject (n=86) was discharged from the hospital. A standard introduction was read to the respondent, and verbal consent was obtained. Following this, a series of open-ended questions was asked (eg, “Are you better?” and “In what way?”). Then some specific questions were asked. The responses were summarized according to the subject’s patterns of alcohol use since hospital discharge and whether the subject had initiated substance abuse treatment or a self-help program. During March 2000, the responses about drinking patterns and initiation of treatment or self-help during the first 6 months following hospital discharge were categorized and coded by 2 of the authors (S.B.R. and R.L.M.).

The second part of the follow-up was conducted as a medical student summer research project and involved patients who were admitted to the hospital between June 1, 1999, and March 31, 2000. These patients were contacted during June and July of 2000. There was a 1-month overlap in admission dates with the first part of the study because of a variation in hospital length of stay. Another investigator attempted to contact the patient and/or a relative or friend identified from the medical record 4 to 12 months after the study subject (n=228) was discharged from the hospital. A standard introduction was read to the respondent, and verbal consent was obtained. Following this, a series of structured questions was asked (eg, “During the first 6 [4 or 5 months in 7 cases] months following your discharge from the hospital did you try to cut down or quit drinking?”). The responses were coded according to the patient’s patterns of alcohol use during the 6 months following hospital discharge and whether the subject had initiated substance abuse treatment or a self-help program.

Interventions

Patients who received usual care served as the control group (n=125). There were 2 groups that received an intervention: a brief intervention group (n=119) and a peer intervention group (n=70).

Usual Physician Care. Patients in this group received care by the residents and attending surgeons of the trauma service only, because the addiction medicine consultant was not available to see them. The surgeons and the hospital’s social workers may or may not have specifically addressed the patients’ substance abuse problems before discharge. During the study period, the hospital nurses, social workers, and resident physicians were given a 1-hour educational conference (in 12 separate sessions) about alcohol detoxification, screening, and brief intervention based on a national standard.²⁵ Preprinted protocols for detoxification were available in the hospital.

Brief Intervention. Patients in this group received the services of an addiction medicine consultant as part of their overall hospital care. Before discharge, these patients were given brief (5 to 15 minutes) advice following a previously described method.²⁶

Peer Intervention. Patients in this group received the same type of brief physician advice of those in the brief intervention group, and a 30- to 60-minute visit from a peer who was active in AA. Patients had to agree to a peer visit, but refusals were rare. Volunteers were recruited through a local residential facility for individuals with substance abuse problems. There were separate facilities for men and women, but they were administered by the same organization and followed the same basic program based on the AA model. The volunteers, called assistant staff, had successfully completed the program at that facility. These volunteers attended 3 2-hour training workshops designed to increase their skills at carrying the message of AA to others. These training workshops included both didactic and role-playing sessions that were designed to help the volunteers follow a protocol based on the AA model. These peers visited with the patient in pairs before hospital discharge. They did not give advice or make treatment recommendations. Instead, they shared their personal stories and their “experience, strength, and hope” with the patient. The peers were always matched with the patients’ sex and usually with the patients’ race. There was a period of time (approximately 6 months) during the study when women volunteers were not available because the facility for women was being relocated.

Statistical Analysis

The data sets from the 2 parts of the study were combined by one of the authors who also performed the data analysis. We used the exact version of the Fisher-Freeman-Halton test²⁷ to compare the 3 treatment groups in terms of categorical baseline characteristics and follow-up rates. One-way analysis of variance was used for continuous baseline characteristics. The same analysis was performed to compare those patients for whom follow-up data could be obtained with those for whom such data could not be obtained.

We use the exact version of the Cochran-Armitage test for trend²⁸ to test the null hypothesis of no difference among the treatment groups against the alternative hypothesis that the true proportions of positive outcomes would be in the following order: control < brief intervention < peer intervention (ie, we hypothesized that the success rate for the peer intervention group would be greater than that for the brief intervention group and that the success rate for the brief intervention group would be greater than that for the control group). If a significant difference was found among the treatment groups, we used the Fisher exact test with a Bonferroni adjustment to determine which pairs of treatment groups differed from each other. Stratified analysis was used to adjust for the effect of any confounding variables.

A sample size of 28 per group was sufficient to achieve 80% power for detecting differences in success rates (as measured by initiation of treatment or self-help) among the groups of 5% in the control group, 10% in the brief intervention group, and 30% in the peer intervention group using a one-tailed significance level of 0.05. All computations for the study were performed using Epi Info Version 6.04c (USD Inc, Stone Mountain, Ga, 1999), StatXact 4.0.1 (CYTEL Software Corp, Cambridge, Mass, 1998), and SPSS software version 10.0 (SPSS, Inc, Chicago, Ill, 1999). Continuous variables were summarized as mean plus or minus the standard deviation.

Results

Of the 314 patients in the study 258 (82.2%) were men; 244 (77.7%) were white; and the mean age was 37.2 years plus or minus 12.5 years (range=18-80). The mean blood alcohol concentration on admission for these 314 patients was 35.8 mmol per L plus or minus 26.5 mmol per L (165 mg/dL±122 mg/dL ) with a range of 00.0 to 143.3 mmol per L (000-660 mg/dL).

Of the 314 patients in our study, 140 (44.6%) were contacted following hospital discharge through communication with the subject, the subjects’ relatives, or both. Among the members of the control group, the follow-up rate was 35.2% (44/125); among those who received a brief intervention it was 47.9% (57/119); and among those who received a peer intervention, it was 55.7% (39/70). This represents a statistically significant difference at the Bonferroni cutoff of 0.05 divided by 3 (0.0167) between the control and peer intervention groups (P=.003) but not between the control and brief intervention groups (P=.023), or the brief and peer intervention groups (P=.152) using the Fisher exact test Table 1.

Among the 140 patients in the study, follow-up data were obtained from the patient in 97 instances (69%), from a friend or family member in 38 (27%), and from other sources in 5 (4%). For the 44 members of the control group, follow-up data were obtained from the patient in 37 instances (84%), from a friend or family member in 6 (14%), and from other sources in 1 (2%). For the 57 patients who received a brief intervention, follow-up data were obtained from the patient in 35 instances (61%), from a friend or family member in 21 (37%), and from other sources in 1 (2%). For the 39 patients who received a peer intervention, follow-up data were obtained from the patient in 25 instances (64%), from a friend or family member in 11 (28%), and from other sources in 3 (8%). The Fisher-Freeman-Halton test indicates a significant difference between the control and brief intervention groups (P=.012) but no difference between the control group and the peer intervention group (P=.117) or between the brief and peer intervention groups (P=.341), using the Bonferroni criterion of 0.0167. Those patients for whom follow-up data could be obtained were compared with those for whom it could not be obtained in terms of age, race, sex, and BAC on admission. The only significant difference that was found was for race: Follow-up data were available for 49.2% of the white patients but for only 28.6% of the nonwhite patients (P=.003). In terms of sex, follow-up data were available for 42.2% of the men and 55.4% of the women (P=.051). The mean age of those for whom follow-up data were available was 38.1 years plus or minus 12.8, compared with 36.4 years plus or minus 12.3 for those lost to follow-up (P=.226). The mean BAC on admission was 38.0 mmol per L plus or minus 27.8 (175 mg/dL±128) for those we were able to follow up, compared with 34.1 mmol per L plus or minus 25.4 (157 mg/dL±117) for those lost to follow-up (P=.233).

Comparisons of the baseline characteristics of the 140 patients across the 3 treatment groups are shown in Table 2. No significant differences were found at baseline between the groups at the 0.05 level except for male sex (P=.003); however, BAC almost reached statistical significance (P=.054).

The results for the main outcome measures of the 3 groups are shown in Table 3. The data reflect the fact that 7 patients drank for several weeks following hospital discharge but then abstained from drinking. As hypothesized, the success rates were greatest in the peer intervention group, followed by the brief intervention and control groups. All 3 outcomes showed statistically significant differences across groups. In terms of pairwise comparisons, the comparison between the control group and the peer intervention group met the Bonferroni criterion of 0.0167 for both abstinence for 6 months following hospital discharge (P=.013) and abstinence during the sixth month following hospital discharge (P=.007). For initiation of treatment or self-help, the comparisons of the peer group with both the control group and the brief intervention group were significant using the Bonferroni criterion (P <.001 in both cases).

Stratifying by sex yielded results that were not materially different from those presented in the Table (P=.016 for 6 months of abstinence; P=.007 for abstinence at during the sixth month; and P <.001 for initiation of treatment or self-help). Stratifying by BAC also did not affect the P values in any material way (data not shown).

Because of inconsistencies between the data from the 2 parts of the study and because of missing or unrecorded data, we can only make qualitative statements about other outcomes. No patient who was completely abstinent for the entire 6 months following hospital discharge had began drinking again by the time of the telephone interview. Many patients in the intervention groups (approximately a third) drank after hospital discharge and continued to drink up to the time of the follow-up interview, although a few of these patients claimed to have cut down. Only a few patients initially abstained from alcohol but returned to drinking at the time of follow-up. Most of the follow-up information came from the patient, our preferred source for outcome data. No patient who claimed to be abstinent had a family member who contradicted that report. However, several patients admitted to drinking (or using drugs) who had a member of the family who reported that the patient was abstinent. In those cases in which a family member could be located but the patient could not, it was usually because the patient was still drinking, living on the streets, or had no telephone. It was rare that the family member reported a favorable outcome (ie, abstinence), and we could not confirm this directly with the patient.

Several patients in the peer intervention group expressed gratitude for the help they received with their drinking problems while in the hospital and especially for the visits by the peers. Some of these patients dramatically changed their lives. At least 3 patients in the peer intervention group went from being unemployed and homeless to full-time employment and having a permanent residence after they entered a treatment program and became involved in AA. They credited the peer intervention as being the most important factor that motivated them to seek help for their alcohol use disorder. At the time this manuscript was being prepared, one of these individuals was serving as a volunteer making visits to hospitalized patients with drinking problems.

Discussion

Previous studies have shown that brief interventions by professionals appear to help motivate patients to reduce drinking. Our study demonstrates that peers may help motivate patients to initiate treatment or self-help as well as promote abstinence. Brief physician advice followed by a visit with a volunteer from AA shows promise as a simple, practical, inexpensive, and effective intervention that may help to prevent patients from returning to alcohol use. This could lead to reductions in recurrent injuries for patients hospitalized with alcohol-related injuries.

Primary care physicians could use this approach to intervene with any patient hospitalized with alcohol-related problems. At our institution, peer volunteers are often called to visit patients with substance use disorders who are hospitalized by the surgery, medicine, family medicine, and psychiatry services. We used trauma patients, because there is a large volume of such patients at our institution who routinely have had toxicology tests performed on admission. Also, an existing trauma registry database facilitated the collection of patient data.

Many primary care physicians already possess the skills required to give patients brief advice about harmful lifestyles and are familiar with the use of community resources that can help their patients. Most communities that are large enough to have a hospital are large enough to support several AA groups. As part of the AA program, members are expected to carry the message of AA to alcoholics who are still drinking. They consider this Twelfth Step Work an essential part of the program that leads to personal progress in AA. Most physicians can easily identify patients who could benefit from hearing the message of AA. It is often not difficult to link up these 2 groups of individuals.²⁹ The local AA office can be called from the patient’s bedside telephone. After the physician explains the situation to the person who answers the call, the telephone can then be given to the patient. If the patient agrees, a member of AA may come to the hospital for a visit. These visits typically last 30 minutes to an hour. Sometimes the AA member may visit again during the patient’s hospital stay or at the time of discharge to escort the patient to an AA meeting. This service is provided without cost to the patient, the patient’s insurance carrier, or the hospital.

Limitations

Our study has many of the limitations of initial retrospective studies: a nonrandomized design, a study sample limited to a particular type of patient, limited follow-up data, variation in the interval from the time of the intervention to the time of follow-up data collection, reliance on self-report, and treatment groups that were not masked to the follow-up interviewers. The nonrandomized design might suggest that some of the favorable outcomes could be the result of selection bias. However, as indicated in Table 2, the baseline characteristics of the 3 groups were similar, except that women were under-represented in the peer intervention group. This finding is probably because of the limited availability of women peer volunteers during a 6-month period of time during the study. The trend towards a lower BAC in the control group suggests that patients with severe alcohol problems may be over-represented in the experimental groups. If anything, this would have biased the study results against the 2 intervention groups. However, the diagnosis of an alcohol use disorder was made using a chart audit for the control group (which did not always provide enough information to differentiate abuse from dependence), while an unstructured patient interview was used for the intervention group. Limited follow-up is a frequent problem in the patient populations used for alcohol use studies. Response rates of approximately 50% are typical. It is not clear why the follow-up rates for nonwhite patients were lower than for white subjects. We did seem to experience more problems with disconnected telephones in the nonwhite population, suggesting that there may be some economic differences between the 2 groups. We observed a significantly better follow-up rate for those in the peer intervention group. This is probably because we established contacts for follow-up directly from the patient in the intervention groups and could verify telephone numbers, but we had to rely on the medical record for the telephone numbers of the members of the control group. These numbers were not always correct. Also, we had significantly fewer family contacts in the control group. We did not record the exact timing of the follow-up after hospital discharge for each individual patient, and therefore could not compare the mean follow-up intervals between groups. However, for the reasons mentioned in the qualitative part of the results section, we do not believe that this problem would have influenced our results in any material manner. Our study was performed with trauma patients who may not be representative of other patient groups. Painful injuries and court appearances related to driving while intoxicated may be important factors that influence drinking behaviors. However, we have observed some nonsurgical patients who have benefited from peer interventions. We relied on patient self-report for outcomes and found a difference between the control and the 2 intervention groups. Although patients with alcohol use disorders may not accurately report their alcohol consumption, it is unlikely that those in the intervention groups would be more likely to report abstinence or to report initiation of treatment or self-help than those in the control group. We preferentially coded the poorest outcome information we obtained from the patient or the family member. Therefore, the source of the follow-up data had a minimal favorable impact on its accuracy. Although the follow-up interviewers knew to which group an individual patient belonged, they asked the interview questions from a printed script, to reduce observer bias to a minimum. Finally, although we obtained severity of disease data for the intervention groups (ie, abuse vs dependence) this information was not available for the control group.

Conclusions

The significant findings of our study suggest that the methods we employed should be evaluated in a well-funded rigorously designed prospective randomized study with more patients who would be objectively evaluated for the severity of their alcohol use disorder and with mechanisms to confirm and quantify the subjects’ self-reports of alcohol consumption and to ensure higher follow-up rates.

In the meantime, physicians can request that members of AA visit their hospitalized patients who have alcohol use disorders. Interventions by recovering alcoholics are not difficult to arrange, involve no costs, pose little patient risk, and might be of great benefit to some patients. We have continued to observe individual patients who were able to find sobriety following these interventions. These patients have expressed opinions that it was primarily the peers who motivated them to seek help for their problem drinking.

Acknowledgments

This work was supported, in part, by the University of Louisville Summer Research Scholarship Program and the University of Louisville Hospital Trauma Institute. We are indebted to the anonymous alcoholic members of a local self-help organization and to The Healing Place for assistance with locating volunteers to visit with our patients. We thank Karen Newton and Gail Wulfman for their assistance with the training of the volunteers. We thank Phillip Boaz, Janet Wallace, and Lance Hottman for their help with the data collection and Margaret M. Steptoe and Murphy Shields for their assistance in the preparation of this manuscript.

The extent to which the physician intervenes with a hospitalized patient who has an alcohol use disorder correlates with the patient’s reported change in alcohol use after discharge.¹ Primary care physicians may be called on to help manage hospitalized patients with alcohol use disorders, but exactly what they should do to help these patients is not always clear.

Alcohol abuse and trauma are common and related clinical problems.² A dose-response relationship has been observed between alcohol consumption and the risk of fatal injury.³ Traumatic injury is a major public health problem and a leading cause of morbidity and mortality in the United States. It ranks first in years of life lost, first in the utilization of hospital-days, second in disability-adjusted life-years, and fourth in overall mortality.⁴ Sims and colleagues⁵ found that violent trauma had a recurrence rate of 44% and a 5-year mortality rate of 20% and that 62% of these patients abused alcohol or drugs. Rivera and coworkers⁶ found that trauma victims who were intoxicated on presentation to a trauma center were 2.5 times more likely to be readmitted for another injury than those who were not intoxicated, and those with evidence of a chronic alcohol problem were 3.5 times more likely. Others have noted similar findings.^7,8 There is a unique opportunity to initiate treatment for patients with substance abuse disorders when they are hospitalized for a traumatic injury.⁹ Often this opportunity is missed.^10,11

It is not known how to intervene with victims of alcohol-related injuries to prevent subsequent injuries. Currently there are several options that could potentially improve the outcomes of hospitalized patients who have substance use disorders, such as brief advice, brief interventions, referral to a consultation team, and referral to a treatment center. At the very least, trauma victims with substance abuse problems should be given some brief advice from the surgeon. Although most surgeons appear willing to give this advice, many feel inadequately prepared to do it.¹² Thus, the burden of performing these interventions may fall to the patient’s primary care physician or the physician who is requested by the surgeon to provide consultation services.

A technique known as “brief intervention” consists of advice in a structured format that is given to patients with a substance use disorder.¹³ These interventions have been found to be effective in a number of clinical settings, including outpatient primary care.^14-16 In particular, brief interventions performed by a trained psychologist in a trauma center have been associated with a reduction in alcohol intake and a reduced risk of trauma recidivism.¹⁷ However, these interventions require significant physician training to implement. What is needed is a simple and practical method that can be used by primary care physicians that does not require extensive physician training.

Peer interventions have been used successfully in education.^18,19 This success is based in some part on what is known as the “attraction paradigm”. The attraction process purports that the more similar the members of a relationship are in experiences, the more likely they will respond to one another positively.²⁰ Peers have been used in some settings to augment treatment in primary care. In one study, trained peers who were recovered from depression were found to provide no additional improvement in clinical outcomes.²¹ In that study, one group of patients with depression who were treated with antidepressant drugs and emotional support provided by a nurse during 10 6-minute telephone calls over a 4-month period were compared with another group who also received peer support. The finding is not surprising, because peers cannot be expected to add much benefit to patients who are already receiving optimal treatment. Volunteers from Alcoholics Anonymous (AA) have been used to talk with alcoholic patients in a general hospital.²² Although impressions are that these peers are helpful, the outcomes of this procedure have not been well studied. This process can also be performed by a professional and has been called Twelve Step Facilitation.²³

At our institution, volunteers from the community who were active in AA were used to speak with patients who were admitted to the hospital with alcohol-related injuries following a brief intervention from a primary care physician. These peers appeared to produce favorable outcomes with our patients. The purpose of our study was to evaluate the effectiveness of this approach and to test the alternative hypothesis that those in a peer intervention group would demonstrate more favorable outcomes than those in a brief intervention group who, in turn, would demonstrate more favorable outcomes than those in a control group.

Methods

Setting

We conducted this study in a Level I trauma center located in the primary university teaching hospital that serves a metropolitan area of more than 1 million people in a 2-state area of the Midwest. The trauma service is staffed by 2 teams of attending and resident surgeons who alternate 24-hour shifts. An addiction medicine physician provides consultative services to these patients.

Study Population

A total of 2530 patients were admitted to the hospital trauma service for injuries between August 1, 1998, and March 31, 2000, and 957 (37.8%) of these had positive toxicology tests (351 alcohol only, 352 drugs only, and 254 alcohol and drugs) on admission to the hospital. Positive toxicology tests were defined as a blood alcohol concentration (BAC) of 4.34 mmol per L or greater (Ž20 mg/dL) and/or the detection of psychoactive drugs. Toxicology screening was not performed in approximately 20% of the patients. Also, 95 patients with negative toxicology screens were known to have an active alcohol use disorder on admission to the hospital. Thus, 1052 patients served as potential study subjects.

The flow of patients through the study is shown in Table 1. A total of 738 patients were excluded as potential subjects, 632 by block randomization and 106 for other reasons. Since patients with positive toxicology tests who did not have either alcohol abuse or alcohol dependence as defined by the Diagnostic and Statistical Manual of Mental Disorders, 4th edition²⁴ were excluded from the study, all of those who were ultimately eligible for follow-up had an alcohol use disorder.

Before the patients were contacted for follow-up, they were categorized into 1 of 3 groups: a usual care group (n=125), a brief intervention group (n=119), or a peer intervention group (n=70), according to the study methods. Volunteer availability often determined which patients received a peer intervention or a brief intervention. One patient had been originally assigned to the brief intervention group, but we learned at the time of the telephone follow-up interview that a family member had arranged for the patient’s AA sponsor to visit the patient on several occasions before and after hospital discharge. This patient was subsequently excluded from our study and is not included in the numeric values of Table 1.

Procedures

This was a retrospective nonrandomized intervention study evaluating the effectiveness of interventions used to encourage trauma patients to abstain from alcohol and to initiate substance abuse treatment or self-help. We obtained initial data retrospectively from the patient’s medical record, including the patient’s demographic characteristics (age, sex, race), the patient’s telephone numbers, and the telephone numbers of relatives or friends.

The follow-up telephone interviews were conducted at 2 different times. The university’s institutional review board approved both parts of our study. The main outcome measures were: (1) complete abstinence from alcohol during the first 6 months following discharge from the hospital, (2) abstinence from alcohol during the sixth month following hospital discharge (ie, those who drank initially after discharge but subsequently became abstinent), and (3) initiation of professional alcohol treatment or self-help. The patients were considered to have initiated treatment or self-help if within the first 6 months following hospital discharge they had either: (1) attended at least one AA meeting, (2) visited a mental health or substance abuse professional at least once, (3) attended at least 1 session at an outpatient alcohol treatment center, or (4) spent at least 1 day at an inpatient or residential alcohol treatment program.

The first part of the follow-up was conducted as a component of a quality improvement program and involved patients who were admitted to the hospital between August 1, 1998, and June 30, 1999. These patients were contacted between September 1999 and January 2000. One of the investigators attempted to contact the patient and/or a relative or friend identified from the medical record 6 to 12 months after the subject (n=86) was discharged from the hospital. A standard introduction was read to the respondent, and verbal consent was obtained. Following this, a series of open-ended questions was asked (eg, “Are you better?” and “In what way?”). Then some specific questions were asked. The responses were summarized according to the subject’s patterns of alcohol use since hospital discharge and whether the subject had initiated substance abuse treatment or a self-help program. During March 2000, the responses about drinking patterns and initiation of treatment or self-help during the first 6 months following hospital discharge were categorized and coded by 2 of the authors (S.B.R. and R.L.M.).

The second part of the follow-up was conducted as a medical student summer research project and involved patients who were admitted to the hospital between June 1, 1999, and March 31, 2000. These patients were contacted during June and July of 2000. There was a 1-month overlap in admission dates with the first part of the study because of a variation in hospital length of stay. Another investigator attempted to contact the patient and/or a relative or friend identified from the medical record 4 to 12 months after the study subject (n=228) was discharged from the hospital. A standard introduction was read to the respondent, and verbal consent was obtained. Following this, a series of structured questions was asked (eg, “During the first 6 [4 or 5 months in 7 cases] months following your discharge from the hospital did you try to cut down or quit drinking?”). The responses were coded according to the patient’s patterns of alcohol use during the 6 months following hospital discharge and whether the subject had initiated substance abuse treatment or a self-help program.

Interventions

Patients who received usual care served as the control group (n=125). There were 2 groups that received an intervention: a brief intervention group (n=119) and a peer intervention group (n=70).

Usual Physician Care. Patients in this group received care by the residents and attending surgeons of the trauma service only, because the addiction medicine consultant was not available to see them. The surgeons and the hospital’s social workers may or may not have specifically addressed the patients’ substance abuse problems before discharge. During the study period, the hospital nurses, social workers, and resident physicians were given a 1-hour educational conference (in 12 separate sessions) about alcohol detoxification, screening, and brief intervention based on a national standard.²⁵ Preprinted protocols for detoxification were available in the hospital.

Brief Intervention. Patients in this group received the services of an addiction medicine consultant as part of their overall hospital care. Before discharge, these patients were given brief (5 to 15 minutes) advice following a previously described method.²⁶

Peer Intervention. Patients in this group received the same type of brief physician advice of those in the brief intervention group, and a 30- to 60-minute visit from a peer who was active in AA. Patients had to agree to a peer visit, but refusals were rare. Volunteers were recruited through a local residential facility for individuals with substance abuse problems. There were separate facilities for men and women, but they were administered by the same organization and followed the same basic program based on the AA model. The volunteers, called assistant staff, had successfully completed the program at that facility. These volunteers attended 3 2-hour training workshops designed to increase their skills at carrying the message of AA to others. These training workshops included both didactic and role-playing sessions that were designed to help the volunteers follow a protocol based on the AA model. These peers visited with the patient in pairs before hospital discharge. They did not give advice or make treatment recommendations. Instead, they shared their personal stories and their “experience, strength, and hope” with the patient. The peers were always matched with the patients’ sex and usually with the patients’ race. There was a period of time (approximately 6 months) during the study when women volunteers were not available because the facility for women was being relocated.

Statistical Analysis

The data sets from the 2 parts of the study were combined by one of the authors who also performed the data analysis. We used the exact version of the Fisher-Freeman-Halton test²⁷ to compare the 3 treatment groups in terms of categorical baseline characteristics and follow-up rates. One-way analysis of variance was used for continuous baseline characteristics. The same analysis was performed to compare those patients for whom follow-up data could be obtained with those for whom such data could not be obtained.

We use the exact version of the Cochran-Armitage test for trend²⁸ to test the null hypothesis of no difference among the treatment groups against the alternative hypothesis that the true proportions of positive outcomes would be in the following order: control < brief intervention < peer intervention (ie, we hypothesized that the success rate for the peer intervention group would be greater than that for the brief intervention group and that the success rate for the brief intervention group would be greater than that for the control group). If a significant difference was found among the treatment groups, we used the Fisher exact test with a Bonferroni adjustment to determine which pairs of treatment groups differed from each other. Stratified analysis was used to adjust for the effect of any confounding variables.

A sample size of 28 per group was sufficient to achieve 80% power for detecting differences in success rates (as measured by initiation of treatment or self-help) among the groups of 5% in the control group, 10% in the brief intervention group, and 30% in the peer intervention group using a one-tailed significance level of 0.05. All computations for the study were performed using Epi Info Version 6.04c (USD Inc, Stone Mountain, Ga, 1999), StatXact 4.0.1 (CYTEL Software Corp, Cambridge, Mass, 1998), and SPSS software version 10.0 (SPSS, Inc, Chicago, Ill, 1999). Continuous variables were summarized as mean plus or minus the standard deviation.

Results

Of the 314 patients in the study 258 (82.2%) were men; 244 (77.7%) were white; and the mean age was 37.2 years plus or minus 12.5 years (range=18-80). The mean blood alcohol concentration on admission for these 314 patients was 35.8 mmol per L plus or minus 26.5 mmol per L (165 mg/dL±122 mg/dL ) with a range of 00.0 to 143.3 mmol per L (000-660 mg/dL).

Of the 314 patients in our study, 140 (44.6%) were contacted following hospital discharge through communication with the subject, the subjects’ relatives, or both. Among the members of the control group, the follow-up rate was 35.2% (44/125); among those who received a brief intervention it was 47.9% (57/119); and among those who received a peer intervention, it was 55.7% (39/70). This represents a statistically significant difference at the Bonferroni cutoff of 0.05 divided by 3 (0.0167) between the control and peer intervention groups (P=.003) but not between the control and brief intervention groups (P=.023), or the brief and peer intervention groups (P=.152) using the Fisher exact test Table 1.

Among the 140 patients in the study, follow-up data were obtained from the patient in 97 instances (69%), from a friend or family member in 38 (27%), and from other sources in 5 (4%). For the 44 members of the control group, follow-up data were obtained from the patient in 37 instances (84%), from a friend or family member in 6 (14%), and from other sources in 1 (2%). For the 57 patients who received a brief intervention, follow-up data were obtained from the patient in 35 instances (61%), from a friend or family member in 21 (37%), and from other sources in 1 (2%). For the 39 patients who received a peer intervention, follow-up data were obtained from the patient in 25 instances (64%), from a friend or family member in 11 (28%), and from other sources in 3 (8%). The Fisher-Freeman-Halton test indicates a significant difference between the control and brief intervention groups (P=.012) but no difference between the control group and the peer intervention group (P=.117) or between the brief and peer intervention groups (P=.341), using the Bonferroni criterion of 0.0167. Those patients for whom follow-up data could be obtained were compared with those for whom it could not be obtained in terms of age, race, sex, and BAC on admission. The only significant difference that was found was for race: Follow-up data were available for 49.2% of the white patients but for only 28.6% of the nonwhite patients (P=.003). In terms of sex, follow-up data were available for 42.2% of the men and 55.4% of the women (P=.051). The mean age of those for whom follow-up data were available was 38.1 years plus or minus 12.8, compared with 36.4 years plus or minus 12.3 for those lost to follow-up (P=.226). The mean BAC on admission was 38.0 mmol per L plus or minus 27.8 (175 mg/dL±128) for those we were able to follow up, compared with 34.1 mmol per L plus or minus 25.4 (157 mg/dL±117) for those lost to follow-up (P=.233).

Comparisons of the baseline characteristics of the 140 patients across the 3 treatment groups are shown in Table 2. No significant differences were found at baseline between the groups at the 0.05 level except for male sex (P=.003); however, BAC almost reached statistical significance (P=.054).

The results for the main outcome measures of the 3 groups are shown in Table 3. The data reflect the fact that 7 patients drank for several weeks following hospital discharge but then abstained from drinking. As hypothesized, the success rates were greatest in the peer intervention group, followed by the brief intervention and control groups. All 3 outcomes showed statistically significant differences across groups. In terms of pairwise comparisons, the comparison between the control group and the peer intervention group met the Bonferroni criterion of 0.0167 for both abstinence for 6 months following hospital discharge (P=.013) and abstinence during the sixth month following hospital discharge (P=.007). For initiation of treatment or self-help, the comparisons of the peer group with both the control group and the brief intervention group were significant using the Bonferroni criterion (P <.001 in both cases).

Stratifying by sex yielded results that were not materially different from those presented in the Table (P=.016 for 6 months of abstinence; P=.007 for abstinence at during the sixth month; and P <.001 for initiation of treatment or self-help). Stratifying by BAC also did not affect the P values in any material way (data not shown).

Because of inconsistencies between the data from the 2 parts of the study and because of missing or unrecorded data, we can only make qualitative statements about other outcomes. No patient who was completely abstinent for the entire 6 months following hospital discharge had began drinking again by the time of the telephone interview. Many patients in the intervention groups (approximately a third) drank after hospital discharge and continued to drink up to the time of the follow-up interview, although a few of these patients claimed to have cut down. Only a few patients initially abstained from alcohol but returned to drinking at the time of follow-up. Most of the follow-up information came from the patient, our preferred source for outcome data. No patient who claimed to be abstinent had a family member who contradicted that report. However, several patients admitted to drinking (or using drugs) who had a member of the family who reported that the patient was abstinent. In those cases in which a family member could be located but the patient could not, it was usually because the patient was still drinking, living on the streets, or had no telephone. It was rare that the family member reported a favorable outcome (ie, abstinence), and we could not confirm this directly with the patient.

Several patients in the peer intervention group expressed gratitude for the help they received with their drinking problems while in the hospital and especially for the visits by the peers. Some of these patients dramatically changed their lives. At least 3 patients in the peer intervention group went from being unemployed and homeless to full-time employment and having a permanent residence after they entered a treatment program and became involved in AA. They credited the peer intervention as being the most important factor that motivated them to seek help for their alcohol use disorder. At the time this manuscript was being prepared, one of these individuals was serving as a volunteer making visits to hospitalized patients with drinking problems.

Discussion

Previous studies have shown that brief interventions by professionals appear to help motivate patients to reduce drinking. Our study demonstrates that peers may help motivate patients to initiate treatment or self-help as well as promote abstinence. Brief physician advice followed by a visit with a volunteer from AA shows promise as a simple, practical, inexpensive, and effective intervention that may help to prevent patients from returning to alcohol use. This could lead to reductions in recurrent injuries for patients hospitalized with alcohol-related injuries.

Primary care physicians could use this approach to intervene with any patient hospitalized with alcohol-related problems. At our institution, peer volunteers are often called to visit patients with substance use disorders who are hospitalized by the surgery, medicine, family medicine, and psychiatry services. We used trauma patients, because there is a large volume of such patients at our institution who routinely have had toxicology tests performed on admission. Also, an existing trauma registry database facilitated the collection of patient data.

Many primary care physicians already possess the skills required to give patients brief advice about harmful lifestyles and are familiar with the use of community resources that can help their patients. Most communities that are large enough to have a hospital are large enough to support several AA groups. As part of the AA program, members are expected to carry the message of AA to alcoholics who are still drinking. They consider this Twelfth Step Work an essential part of the program that leads to personal progress in AA. Most physicians can easily identify patients who could benefit from hearing the message of AA. It is often not difficult to link up these 2 groups of individuals.²⁹ The local AA office can be called from the patient’s bedside telephone. After the physician explains the situation to the person who answers the call, the telephone can then be given to the patient. If the patient agrees, a member of AA may come to the hospital for a visit. These visits typically last 30 minutes to an hour. Sometimes the AA member may visit again during the patient’s hospital stay or at the time of discharge to escort the patient to an AA meeting. This service is provided without cost to the patient, the patient’s insurance carrier, or the hospital.

Limitations

Our study has many of the limitations of initial retrospective studies: a nonrandomized design, a study sample limited to a particular type of patient, limited follow-up data, variation in the interval from the time of the intervention to the time of follow-up data collection, reliance on self-report, and treatment groups that were not masked to the follow-up interviewers. The nonrandomized design might suggest that some of the favorable outcomes could be the result of selection bias. However, as indicated in Table 2, the baseline characteristics of the 3 groups were similar, except that women were under-represented in the peer intervention group. This finding is probably because of the limited availability of women peer volunteers during a 6-month period of time during the study. The trend towards a lower BAC in the control group suggests that patients with severe alcohol problems may be over-represented in the experimental groups. If anything, this would have biased the study results against the 2 intervention groups. However, the diagnosis of an alcohol use disorder was made using a chart audit for the control group (which did not always provide enough information to differentiate abuse from dependence), while an unstructured patient interview was used for the intervention group. Limited follow-up is a frequent problem in the patient populations used for alcohol use studies. Response rates of approximately 50% are typical. It is not clear why the follow-up rates for nonwhite patients were lower than for white subjects. We did seem to experience more problems with disconnected telephones in the nonwhite population, suggesting that there may be some economic differences between the 2 groups. We observed a significantly better follow-up rate for those in the peer intervention group. This is probably because we established contacts for follow-up directly from the patient in the intervention groups and could verify telephone numbers, but we had to rely on the medical record for the telephone numbers of the members of the control group. These numbers were not always correct. Also, we had significantly fewer family contacts in the control group. We did not record the exact timing of the follow-up after hospital discharge for each individual patient, and therefore could not compare the mean follow-up intervals between groups. However, for the reasons mentioned in the qualitative part of the results section, we do not believe that this problem would have influenced our results in any material manner. Our study was performed with trauma patients who may not be representative of other patient groups. Painful injuries and court appearances related to driving while intoxicated may be important factors that influence drinking behaviors. However, we have observed some nonsurgical patients who have benefited from peer interventions. We relied on patient self-report for outcomes and found a difference between the control and the 2 intervention groups. Although patients with alcohol use disorders may not accurately report their alcohol consumption, it is unlikely that those in the intervention groups would be more likely to report abstinence or to report initiation of treatment or self-help than those in the control group. We preferentially coded the poorest outcome information we obtained from the patient or the family member. Therefore, the source of the follow-up data had a minimal favorable impact on its accuracy. Although the follow-up interviewers knew to which group an individual patient belonged, they asked the interview questions from a printed script, to reduce observer bias to a minimum. Finally, although we obtained severity of disease data for the intervention groups (ie, abuse vs dependence) this information was not available for the control group.

Conclusions

The significant findings of our study suggest that the methods we employed should be evaluated in a well-funded rigorously designed prospective randomized study with more patients who would be objectively evaluated for the severity of their alcohol use disorder and with mechanisms to confirm and quantify the subjects’ self-reports of alcohol consumption and to ensure higher follow-up rates.

In the meantime, physicians can request that members of AA visit their hospitalized patients who have alcohol use disorders. Interventions by recovering alcoholics are not difficult to arrange, involve no costs, pose little patient risk, and might be of great benefit to some patients. We have continued to observe individual patients who were able to find sobriety following these interventions. These patients have expressed opinions that it was primarily the peers who motivated them to seek help for their problem drinking.

Acknowledgments

This work was supported, in part, by the University of Louisville Summer Research Scholarship Program and the University of Louisville Hospital Trauma Institute. We are indebted to the anonymous alcoholic members of a local self-help organization and to The Healing Place for assistance with locating volunteers to visit with our patients. We thank Karen Newton and Gail Wulfman for their assistance with the training of the volunteers. We thank Phillip Boaz, Janet Wallace, and Lance Hottman for their help with the data collection and Margaret M. Steptoe and Murphy Shields for their assistance in the preparation of this manuscript.

Primary care providers know that the most common site of infection during pregnancy is the urinary tract.^1-6 Pregnant women at the highest risk for urinary tract infection (UTI) include those with a history of UTI, high frequency of sexual activity, high parity, functional urinary tract abnormalities, sickle cell trait, and diabetes mellitus.^1,3,5,7 The well-documented consequences of UTIs include pyelonephritis in pregnant women and preterm labor and low birth weight in the infants.^4,8-14 Fetal death has also been associated with maternal UTI. Leviton and Gilles^15-17 conducted autopsies on fetuses with clinical reports of maternal UTI and found endotoxin-damaged glial cells in the maturing forebrain. Glial cells (destined to become oligodendroglia and lay down myelin) are either destroyed (causing necrosis) or transformed (resulting in hypertrophic astrocytes and damaged glia). The end result of this process can be perinatal leukoencephalopathy and death.^17,18 The relationship between maternal UTI and deficits in child development has also been explored; there has not been a consensus, however, about this association. Researchers for the National Collaborative Perinatal Project (NCPP) reported a 2.38-point decrease in the intelligence quotient (IQ) score in white boys and no significant variation in IQ scores in girls or black boys.^19,20 There are reports of an association between UTI and delayed motor performance at the age of 8 months and an increased risk for cerebral palsy.^21,22 Others found no relationship between maternal UTI and subsequent psychomotor impairments.²³

We analyzed the relative risk for mental retardation or developmental delay following UTI, taking into account the trimester of infection and the impact of treatment. Medicaid reimbursement files were used to analyze the association for more than 41,000 mother-child pairs for the period 1994 to 1996. The proportion of women with a presumably untreated UTI who had a child with mental retardation or developmental delay was 35% higher than the unexposed group and 24% higher than the group that had a UTI and had prescriptions filled.²⁴ To further elucidate the relationships between fetal exposure to UTI and subsequent mental retardation or developmental delay, we compared the analysis of the NCPP data set with our previous analysis of the Medicaid data. We also used survival analysis to explore the potential relationship between maternal UTI and intrauterine fetal demise. This was done since the maternal UTI and death associations reported by Leviton and Gilles,^15-17 was based on autopsy studies from the NCPP, and the comparison was between dead infants whose mothers did and did not have a reported UTI. We used the retrospective cohort design to compare the risk for fetal death for infants with and without maternal UTI exposure.

Methods

The data used for these analyses were the research variables from the NCPP. Those women were recruited from 12 urban university medical centers throughout the United States between 1959 and 1974. The NCPP data for the outcome of mental retardation or developmental delay were compared with the South Carolina Medicaid data set from 1994 to 1996, using the methods previously described in the literature.²⁴

The NCPP was a longitudinal study of the outcomes of pregnancies of primarily urban poor women. A total of 53,043 pregnancies (including 7522 repeat pregnancies) were followed up, with 64% of the participants residing in the Northeast. The data were collected on the pregnancies during the prenatal visits, at admission for delivery, and during scheduled follow-up visits for 8 years. Psychologic evaluations of the children were performed at 8 months, 4 years, and 7 years. It should be noted that not all children were evaluated, since there was a 25% loss to follow-up for the 4-year examination. Of these, 9.6% died before age 4 years; 2.3% were tested for IQ, but no scores were obtained. The remaining families were either not located or refused to return for testing and examination.^19,20,25

We analyzed 41,692 NCPP mother-child pairs for whom information about prenatal care and child outcomes were available. The data on exposure to maternal UTIs were recorded in the medical record and coded (month and year). The clinical diagnosis was supported by laboratory tests and recorded by the attending physician. The diagnoses of mental retardation or developmental delay were based on standardized scores on the Stanford-Binet Intelligence Scale Form L-M, which was administered to children aged 4 years in the NCPP data set.²⁰ Children with scores lower than 70 were classified as having mental retardation.

Information about fetal deaths was available for the NCPP; however, these data were limited by the late entry into the study. Only a small number of women entered the NCPP study in the first trimester, so first trimester terminations were not available. The case definition for fetal death was death occurring before and up to birth. We included neonatal deaths (deaths during the first 28 days of life) in our analysis; some infections occurred late in pregnancy, and the deaths did not occur until the postnatal period. A 35-day critical period following the date of infection was used to allow for detection of either a fetal death at the monthly prenatal examination or spontaneous fetal loss.

We analyzed the data with chi-square tests, logistic regression modeling, and survival analysis procedures using SAS software (SAS Institute, Cary, NC). Chi-square tests were used to compare the distribution of independent variables to the 2 outcome variables (death and mental retardation or developmental delay). The woman’s age when the infant was born, infant birth weight, infant sex, maternal education, gestational age at study entry, and race were considered potential confounders. When the final logistic regression models were developed to measure the impact of exposure to maternal UTI on the relative risk for mental retardation or developmental delay compared with survivors without mental retardation or developmental delay, the control variables were woman’s age at the time of the birth of the infant, infant birth weight, and race. This was based on standard data-based variable selection procedures. We conducted survival analysis using Cox proportional hazard models by applying Lifetest and PHreg Procedures from SAS.

Results

Demographics and other baseline characteristics of the NCPP mother-child pairs are shown in Table 1. The mother and child characteristics reflect the study entry criteria of an equal proportion of black and white participants from poor economic environments. More than half the women had less than a high school education, and 13.4% of the infants were born weighing less than 2500 g. The fetal death rate was 1.9%, and the UTI rate was 15.6%.

The overall risk for mental retardation in the children of pregnant women with UTI was 16% higher than for women without a UTI Table 2. The only trimester that indicated a statistically significant increased risk was the third trimester (relative risk [RR]=1.40; 95% confidence interval [CI], 1.01-1.95). Because of late entry into the NCPP study, there were only 8 mothers with UTI in the first trimester who had infants with MR; thus, this estimate of risk was unstable. Also, we did not have treatment data for the NCPP group.

Table 3 shows the death risk associated with UTI exposure. There was a two-fold increased overall risk for death (RR=2.02; 95% CI, 1.32-3.07) when the fetus was exposed to a maternal UTI. Survival analysis was used to predict the risk for death 35 days after exposure to maternal UTI when taking into account the interaction of time and the UTI exposure. When the actual time of exposure is taken into account, the relative hazard was 1.41 (95% CI, 1.07-1.76) in the second trimester and 2.23 (95% CI, 1.40-3.55) in the third trimester.

Discussion

Our analyses suggest that maternal UTIs are associated with MR and fetal death in the third trimester. The NCPP study recruited poor women, and 56% had less than a high school education. The infant birth weight proportions are similar to those reported nationally in 1970, with the low birth weight proportion 7.9% for all races and 13.9% for blacks.²⁶ Analyses of more highly educated women with different access to care would be required to ascertain whether this result is reproducible in other population groups. Support for the observed associations are strengthened, however, by the consistency between these data and the more recent analysis of women and children funded by Medicaid in South Carolina during 1994 to 1996.²⁴

We analyzed the relationship between maternal UTI and mental retardation or developmental delay using logistic regression models for both the NCPP and Medicaid. The Medicaid data included information about whether a prescription for antibiotics was filled following the diagnosis of a UTI. Thus, in the Medicaid data set we were able to identify women who probably did and did not have treatment following the diagnosis. In the NCPP and the Medicaid models we determined the risk for mental retardation or developmental delay in children after controlling for gestational age at entry into the study, maternal age, maternal race, and birth weight. These confounders control for the effect of factors already known to be associated with both the exposure and the outcome, and are not believed to be in the causal pathway. The Medicaid data revealed an increased relative risks for mental retardation or developmental delay of 1.47 (95% CI, 1.08-2.01) in the first trimester and an RR of 1.42 (95% CI, 1.12-1.81) in the third trimester, when there was no documentation of an antibiotic prescription being filled. When medication prescriptions were filled there were no increased risks for mental retardation or developmental delay.

As described by Leviton and Gilles, UTI was also associated with an increased risk for death. In both the NCPP and Medicaid data sets, more than 1.9% of the pregnancies resulted in a fetal death. For the NCPP analysis of death, the second trimester results are made on the basis of maternal UTI in the second trimester and fetal death in the second or third trimester. The third trimester results include some live-born infants, because the 35-day critical period following UTI occasionally extended into the postnatal period for some of the infants. Thus, for the infants who had postnatal deaths, prematurity and low birth weight could be an intermediate variable, since the relationship between UTI and prematurity has been established.^16-18 The biologic explanation of fetal death associated with maternal UTI implicated endotoxins from gram-negative bacteria. It is likely that a similar mechanism is responsible for the brain damage (mental retardation or developmental delay) associated with maternal UTI.

Limitations

Our analyses and the comparison with the Medicaid data have several important limitations. First, there were issues related to the exposure variable. The case definition for UTI from the 2 data sources differed, since the NCPP data relied on a physician diagnosis and a date of occurrence and the Medicaid data relied on a physician diagnosis or a urine culture followed by a prescription for antibiotics within 14 days of diagnosis. The UTI rate was 20.9% for the Medicaid group and 15.6% for the NCPP group. Also, we did not have data on the specific symptoms or reason for the urine culture and the organism identified on culture. It is possible that only febrile cases of bacteriuria or specific bacterial species were associated with the adverse outcomes, but this could not be identified in these data sets. We could not determine how long the symptoms, if any, were present before treatment or the efficacy of treatment as measured by test-of-cure follow-up cultures later in pregnancy. Women with a positive urine culture or urine analysis who did not fill an antibiotic claim within 14 days following the laboratory test were not cases in the Medicaid analysis. This misclassification would result in some women with an untreated bacteriuria remaining in the comparison group and biasing the results toward the null hypothesis of not finding a difference between those in the case group and those in the control group.

Second, the case definition for mental retardation or developmental delay differed for the 2 data sets. For the NCPP we had actual scores on a test of cognitive functioning, and 4.5% scored in the mental retardation range (IQ 69). The Medicaid mental retardation or developmental delay diagnoses were identified for 7.0% using International Classification of Diseases-ninth revision-Clinical Modification codes 315 (specific delays in development), 317 (mild mental retardation), 318 (other specified mental retardation), or 319 (unspecified mental retardation).

We calculated the risk for death and mental retardation without regard to antibiotic prescription status for the NCPP, because the medication data were not coded with a date. The Medicaid data were useful in this regard, since we had the actual date the prescription was filled. Women with a filled prescription did not necessarily take the medication, and women without a filled prescription might have received samples of the antibiotic from their physicians. Thus, we do not know the actual compliance rate for treatment. It must be noted that there could be a difference in some other unidentified characteristic in these analyses of the women who filled their prescription compared with the women who did not. This could be referred to as a healthy patient effect. Finally, for the NCPP data set, children with mental retardation were more likely to be lost to follow-up before their fourth-year checkup than the children with normal cognitive functioning since out-of-home and institutional placement was still recommended for children with special needs before 1975. These factors, which could not be controlled for in this secondary data analysis, might bias the results toward the null hypothesis of no difference between the groups and therefore dilute the magnitude of our findings.

Further Research

Additional longitudinal studies are needed to evaluate the association of the time of infection with presenting symptoms and organisms. Also, animal models are needed to understand the mechanisms of injury to the fetal brain.

Conclusions

Our findings support an association between third trimester maternal UTI and fetal death, mental retardation, or developmental delay. Women with asymptomatic bacteriuria early in pregnancy may be at higher risk for UTI during the latter half of pregnancy,⁷ and more aggressive screening techniques may be appropriate for this population. The Medicaid data suggest there is no difference in mental retardation or developmental delay outcomes between the treated women and women without UTIs. Some physicians may want to advise women of the potential risks for infection of their fetus when a UTI is diagnosed, in an attempt to increase compliance with the medication regimen.

Primary care providers know that the most common site of infection during pregnancy is the urinary tract.^1-6 Pregnant women at the highest risk for urinary tract infection (UTI) include those with a history of UTI, high frequency of sexual activity, high parity, functional urinary tract abnormalities, sickle cell trait, and diabetes mellitus.^1,3,5,7 The well-documented consequences of UTIs include pyelonephritis in pregnant women and preterm labor and low birth weight in the infants.^4,8-14 Fetal death has also been associated with maternal UTI. Leviton and Gilles^15-17 conducted autopsies on fetuses with clinical reports of maternal UTI and found endotoxin-damaged glial cells in the maturing forebrain. Glial cells (destined to become oligodendroglia and lay down myelin) are either destroyed (causing necrosis) or transformed (resulting in hypertrophic astrocytes and damaged glia). The end result of this process can be perinatal leukoencephalopathy and death.^17,18 The relationship between maternal UTI and deficits in child development has also been explored; there has not been a consensus, however, about this association. Researchers for the National Collaborative Perinatal Project (NCPP) reported a 2.38-point decrease in the intelligence quotient (IQ) score in white boys and no significant variation in IQ scores in girls or black boys.^19,20 There are reports of an association between UTI and delayed motor performance at the age of 8 months and an increased risk for cerebral palsy.^21,22 Others found no relationship between maternal UTI and subsequent psychomotor impairments.²³

We analyzed the relative risk for mental retardation or developmental delay following UTI, taking into account the trimester of infection and the impact of treatment. Medicaid reimbursement files were used to analyze the association for more than 41,000 mother-child pairs for the period 1994 to 1996. The proportion of women with a presumably untreated UTI who had a child with mental retardation or developmental delay was 35% higher than the unexposed group and 24% higher than the group that had a UTI and had prescriptions filled.²⁴ To further elucidate the relationships between fetal exposure to UTI and subsequent mental retardation or developmental delay, we compared the analysis of the NCPP data set with our previous analysis of the Medicaid data. We also used survival analysis to explore the potential relationship between maternal UTI and intrauterine fetal demise. This was done since the maternal UTI and death associations reported by Leviton and Gilles,^15-17 was based on autopsy studies from the NCPP, and the comparison was between dead infants whose mothers did and did not have a reported UTI. We used the retrospective cohort design to compare the risk for fetal death for infants with and without maternal UTI exposure.

Methods

The data used for these analyses were the research variables from the NCPP. Those women were recruited from 12 urban university medical centers throughout the United States between 1959 and 1974. The NCPP data for the outcome of mental retardation or developmental delay were compared with the South Carolina Medicaid data set from 1994 to 1996, using the methods previously described in the literature.²⁴

The NCPP was a longitudinal study of the outcomes of pregnancies of primarily urban poor women. A total of 53,043 pregnancies (including 7522 repeat pregnancies) were followed up, with 64% of the participants residing in the Northeast. The data were collected on the pregnancies during the prenatal visits, at admission for delivery, and during scheduled follow-up visits for 8 years. Psychologic evaluations of the children were performed at 8 months, 4 years, and 7 years. It should be noted that not all children were evaluated, since there was a 25% loss to follow-up for the 4-year examination. Of these, 9.6% died before age 4 years; 2.3% were tested for IQ, but no scores were obtained. The remaining families were either not located or refused to return for testing and examination.^19,20,25

We analyzed 41,692 NCPP mother-child pairs for whom information about prenatal care and child outcomes were available. The data on exposure to maternal UTIs were recorded in the medical record and coded (month and year). The clinical diagnosis was supported by laboratory tests and recorded by the attending physician. The diagnoses of mental retardation or developmental delay were based on standardized scores on the Stanford-Binet Intelligence Scale Form L-M, which was administered to children aged 4 years in the NCPP data set.²⁰ Children with scores lower than 70 were classified as having mental retardation.

Information about fetal deaths was available for the NCPP; however, these data were limited by the late entry into the study. Only a small number of women entered the NCPP study in the first trimester, so first trimester terminations were not available. The case definition for fetal death was death occurring before and up to birth. We included neonatal deaths (deaths during the first 28 days of life) in our analysis; some infections occurred late in pregnancy, and the deaths did not occur until the postnatal period. A 35-day critical period following the date of infection was used to allow for detection of either a fetal death at the monthly prenatal examination or spontaneous fetal loss.

We analyzed the data with chi-square tests, logistic regression modeling, and survival analysis procedures using SAS software (SAS Institute, Cary, NC). Chi-square tests were used to compare the distribution of independent variables to the 2 outcome variables (death and mental retardation or developmental delay). The woman’s age when the infant was born, infant birth weight, infant sex, maternal education, gestational age at study entry, and race were considered potential confounders. When the final logistic regression models were developed to measure the impact of exposure to maternal UTI on the relative risk for mental retardation or developmental delay compared with survivors without mental retardation or developmental delay, the control variables were woman’s age at the time of the birth of the infant, infant birth weight, and race. This was based on standard data-based variable selection procedures. We conducted survival analysis using Cox proportional hazard models by applying Lifetest and PHreg Procedures from SAS.

Results

Demographics and other baseline characteristics of the NCPP mother-child pairs are shown in Table 1. The mother and child characteristics reflect the study entry criteria of an equal proportion of black and white participants from poor economic environments. More than half the women had less than a high school education, and 13.4% of the infants were born weighing less than 2500 g. The fetal death rate was 1.9%, and the UTI rate was 15.6%.

The overall risk for mental retardation in the children of pregnant women with UTI was 16% higher than for women without a UTI Table 2. The only trimester that indicated a statistically significant increased risk was the third trimester (relative risk [RR]=1.40; 95% confidence interval [CI], 1.01-1.95). Because of late entry into the NCPP study, there were only 8 mothers with UTI in the first trimester who had infants with MR; thus, this estimate of risk was unstable. Also, we did not have treatment data for the NCPP group.

Table 3 shows the death risk associated with UTI exposure. There was a two-fold increased overall risk for death (RR=2.02; 95% CI, 1.32-3.07) when the fetus was exposed to a maternal UTI. Survival analysis was used to predict the risk for death 35 days after exposure to maternal UTI when taking into account the interaction of time and the UTI exposure. When the actual time of exposure is taken into account, the relative hazard was 1.41 (95% CI, 1.07-1.76) in the second trimester and 2.23 (95% CI, 1.40-3.55) in the third trimester.

Discussion

Our analyses suggest that maternal UTIs are associated with MR and fetal death in the third trimester. The NCPP study recruited poor women, and 56% had less than a high school education. The infant birth weight proportions are similar to those reported nationally in 1970, with the low birth weight proportion 7.9% for all races and 13.9% for blacks.²⁶ Analyses of more highly educated women with different access to care would be required to ascertain whether this result is reproducible in other population groups. Support for the observed associations are strengthened, however, by the consistency between these data and the more recent analysis of women and children funded by Medicaid in South Carolina during 1994 to 1996.²⁴

We analyzed the relationship between maternal UTI and mental retardation or developmental delay using logistic regression models for both the NCPP and Medicaid. The Medicaid data included information about whether a prescription for antibiotics was filled following the diagnosis of a UTI. Thus, in the Medicaid data set we were able to identify women who probably did and did not have treatment following the diagnosis. In the NCPP and the Medicaid models we determined the risk for mental retardation or developmental delay in children after controlling for gestational age at entry into the study, maternal age, maternal race, and birth weight. These confounders control for the effect of factors already known to be associated with both the exposure and the outcome, and are not believed to be in the causal pathway. The Medicaid data revealed an increased relative risks for mental retardation or developmental delay of 1.47 (95% CI, 1.08-2.01) in the first trimester and an RR of 1.42 (95% CI, 1.12-1.81) in the third trimester, when there was no documentation of an antibiotic prescription being filled. When medication prescriptions were filled there were no increased risks for mental retardation or developmental delay.

As described by Leviton and Gilles, UTI was also associated with an increased risk for death. In both the NCPP and Medicaid data sets, more than 1.9% of the pregnancies resulted in a fetal death. For the NCPP analysis of death, the second trimester results are made on the basis of maternal UTI in the second trimester and fetal death in the second or third trimester. The third trimester results include some live-born infants, because the 35-day critical period following UTI occasionally extended into the postnatal period for some of the infants. Thus, for the infants who had postnatal deaths, prematurity and low birth weight could be an intermediate variable, since the relationship between UTI and prematurity has been established.^16-18 The biologic explanation of fetal death associated with maternal UTI implicated endotoxins from gram-negative bacteria. It is likely that a similar mechanism is responsible for the brain damage (mental retardation or developmental delay) associated with maternal UTI.

Limitations

Our analyses and the comparison with the Medicaid data have several important limitations. First, there were issues related to the exposure variable. The case definition for UTI from the 2 data sources differed, since the NCPP data relied on a physician diagnosis and a date of occurrence and the Medicaid data relied on a physician diagnosis or a urine culture followed by a prescription for antibiotics within 14 days of diagnosis. The UTI rate was 20.9% for the Medicaid group and 15.6% for the NCPP group. Also, we did not have data on the specific symptoms or reason for the urine culture and the organism identified on culture. It is possible that only febrile cases of bacteriuria or specific bacterial species were associated with the adverse outcomes, but this could not be identified in these data sets. We could not determine how long the symptoms, if any, were present before treatment or the efficacy of treatment as measured by test-of-cure follow-up cultures later in pregnancy. Women with a positive urine culture or urine analysis who did not fill an antibiotic claim within 14 days following the laboratory test were not cases in the Medicaid analysis. This misclassification would result in some women with an untreated bacteriuria remaining in the comparison group and biasing the results toward the null hypothesis of not finding a difference between those in the case group and those in the control group.

Second, the case definition for mental retardation or developmental delay differed for the 2 data sets. For the NCPP we had actual scores on a test of cognitive functioning, and 4.5% scored in the mental retardation range (IQ 69). The Medicaid mental retardation or developmental delay diagnoses were identified for 7.0% using International Classification of Diseases-ninth revision-Clinical Modification codes 315 (specific delays in development), 317 (mild mental retardation), 318 (other specified mental retardation), or 319 (unspecified mental retardation).

We calculated the risk for death and mental retardation without regard to antibiotic prescription status for the NCPP, because the medication data were not coded with a date. The Medicaid data were useful in this regard, since we had the actual date the prescription was filled. Women with a filled prescription did not necessarily take the medication, and women without a filled prescription might have received samples of the antibiotic from their physicians. Thus, we do not know the actual compliance rate for treatment. It must be noted that there could be a difference in some other unidentified characteristic in these analyses of the women who filled their prescription compared with the women who did not. This could be referred to as a healthy patient effect. Finally, for the NCPP data set, children with mental retardation were more likely to be lost to follow-up before their fourth-year checkup than the children with normal cognitive functioning since out-of-home and institutional placement was still recommended for children with special needs before 1975. These factors, which could not be controlled for in this secondary data analysis, might bias the results toward the null hypothesis of no difference between the groups and therefore dilute the magnitude of our findings.

Further Research

Additional longitudinal studies are needed to evaluate the association of the time of infection with presenting symptoms and organisms. Also, animal models are needed to understand the mechanisms of injury to the fetal brain.

Conclusions

Our findings support an association between third trimester maternal UTI and fetal death, mental retardation, or developmental delay. Women with asymptomatic bacteriuria early in pregnancy may be at higher risk for UTI during the latter half of pregnancy,⁷ and more aggressive screening techniques may be appropriate for this population. The Medicaid data suggest there is no difference in mental retardation or developmental delay outcomes between the treated women and women without UTIs. Some physicians may want to advise women of the potential risks for infection of their fetus when a UTI is diagnosed, in an attempt to increase compliance with the medication regimen.

Primary care providers know that the most common site of infection during pregnancy is the urinary tract.^1-6 Pregnant women at the highest risk for urinary tract infection (UTI) include those with a history of UTI, high frequency of sexual activity, high parity, functional urinary tract abnormalities, sickle cell trait, and diabetes mellitus.^1,3,5,7 The well-documented consequences of UTIs include pyelonephritis in pregnant women and preterm labor and low birth weight in the infants.^4,8-14 Fetal death has also been associated with maternal UTI. Leviton and Gilles^15-17 conducted autopsies on fetuses with clinical reports of maternal UTI and found endotoxin-damaged glial cells in the maturing forebrain. Glial cells (destined to become oligodendroglia and lay down myelin) are either destroyed (causing necrosis) or transformed (resulting in hypertrophic astrocytes and damaged glia). The end result of this process can be perinatal leukoencephalopathy and death.^17,18 The relationship between maternal UTI and deficits in child development has also been explored; there has not been a consensus, however, about this association. Researchers for the National Collaborative Perinatal Project (NCPP) reported a 2.38-point decrease in the intelligence quotient (IQ) score in white boys and no significant variation in IQ scores in girls or black boys.^19,20 There are reports of an association between UTI and delayed motor performance at the age of 8 months and an increased risk for cerebral palsy.^21,22 Others found no relationship between maternal UTI and subsequent psychomotor impairments.²³

We analyzed the relative risk for mental retardation or developmental delay following UTI, taking into account the trimester of infection and the impact of treatment. Medicaid reimbursement files were used to analyze the association for more than 41,000 mother-child pairs for the period 1994 to 1996. The proportion of women with a presumably untreated UTI who had a child with mental retardation or developmental delay was 35% higher than the unexposed group and 24% higher than the group that had a UTI and had prescriptions filled.²⁴ To further elucidate the relationships between fetal exposure to UTI and subsequent mental retardation or developmental delay, we compared the analysis of the NCPP data set with our previous analysis of the Medicaid data. We also used survival analysis to explore the potential relationship between maternal UTI and intrauterine fetal demise. This was done since the maternal UTI and death associations reported by Leviton and Gilles,^15-17 was based on autopsy studies from the NCPP, and the comparison was between dead infants whose mothers did and did not have a reported UTI. We used the retrospective cohort design to compare the risk for fetal death for infants with and without maternal UTI exposure.

Methods

The data used for these analyses were the research variables from the NCPP. Those women were recruited from 12 urban university medical centers throughout the United States between 1959 and 1974. The NCPP data for the outcome of mental retardation or developmental delay were compared with the South Carolina Medicaid data set from 1994 to 1996, using the methods previously described in the literature.²⁴

The NCPP was a longitudinal study of the outcomes of pregnancies of primarily urban poor women. A total of 53,043 pregnancies (including 7522 repeat pregnancies) were followed up, with 64% of the participants residing in the Northeast. The data were collected on the pregnancies during the prenatal visits, at admission for delivery, and during scheduled follow-up visits for 8 years. Psychologic evaluations of the children were performed at 8 months, 4 years, and 7 years. It should be noted that not all children were evaluated, since there was a 25% loss to follow-up for the 4-year examination. Of these, 9.6% died before age 4 years; 2.3% were tested for IQ, but no scores were obtained. The remaining families were either not located or refused to return for testing and examination.^19,20,25

We analyzed 41,692 NCPP mother-child pairs for whom information about prenatal care and child outcomes were available. The data on exposure to maternal UTIs were recorded in the medical record and coded (month and year). The clinical diagnosis was supported by laboratory tests and recorded by the attending physician. The diagnoses of mental retardation or developmental delay were based on standardized scores on the Stanford-Binet Intelligence Scale Form L-M, which was administered to children aged 4 years in the NCPP data set.²⁰ Children with scores lower than 70 were classified as having mental retardation.

Information about fetal deaths was available for the NCPP; however, these data were limited by the late entry into the study. Only a small number of women entered the NCPP study in the first trimester, so first trimester terminations were not available. The case definition for fetal death was death occurring before and up to birth. We included neonatal deaths (deaths during the first 28 days of life) in our analysis; some infections occurred late in pregnancy, and the deaths did not occur until the postnatal period. A 35-day critical period following the date of infection was used to allow for detection of either a fetal death at the monthly prenatal examination or spontaneous fetal loss.

We analyzed the data with chi-square tests, logistic regression modeling, and survival analysis procedures using SAS software (SAS Institute, Cary, NC). Chi-square tests were used to compare the distribution of independent variables to the 2 outcome variables (death and mental retardation or developmental delay). The woman’s age when the infant was born, infant birth weight, infant sex, maternal education, gestational age at study entry, and race were considered potential confounders. When the final logistic regression models were developed to measure the impact of exposure to maternal UTI on the relative risk for mental retardation or developmental delay compared with survivors without mental retardation or developmental delay, the control variables were woman’s age at the time of the birth of the infant, infant birth weight, and race. This was based on standard data-based variable selection procedures. We conducted survival analysis using Cox proportional hazard models by applying Lifetest and PHreg Procedures from SAS.

Results

Demographics and other baseline characteristics of the NCPP mother-child pairs are shown in Table 1. The mother and child characteristics reflect the study entry criteria of an equal proportion of black and white participants from poor economic environments. More than half the women had less than a high school education, and 13.4% of the infants were born weighing less than 2500 g. The fetal death rate was 1.9%, and the UTI rate was 15.6%.

The overall risk for mental retardation in the children of pregnant women with UTI was 16% higher than for women without a UTI Table 2. The only trimester that indicated a statistically significant increased risk was the third trimester (relative risk [RR]=1.40; 95% confidence interval [CI], 1.01-1.95). Because of late entry into the NCPP study, there were only 8 mothers with UTI in the first trimester who had infants with MR; thus, this estimate of risk was unstable. Also, we did not have treatment data for the NCPP group.

Table 3 shows the death risk associated with UTI exposure. There was a two-fold increased overall risk for death (RR=2.02; 95% CI, 1.32-3.07) when the fetus was exposed to a maternal UTI. Survival analysis was used to predict the risk for death 35 days after exposure to maternal UTI when taking into account the interaction of time and the UTI exposure. When the actual time of exposure is taken into account, the relative hazard was 1.41 (95% CI, 1.07-1.76) in the second trimester and 2.23 (95% CI, 1.40-3.55) in the third trimester.

Discussion

Our analyses suggest that maternal UTIs are associated with MR and fetal death in the third trimester. The NCPP study recruited poor women, and 56% had less than a high school education. The infant birth weight proportions are similar to those reported nationally in 1970, with the low birth weight proportion 7.9% for all races and 13.9% for blacks.²⁶ Analyses of more highly educated women with different access to care would be required to ascertain whether this result is reproducible in other population groups. Support for the observed associations are strengthened, however, by the consistency between these data and the more recent analysis of women and children funded by Medicaid in South Carolina during 1994 to 1996.²⁴

We analyzed the relationship between maternal UTI and mental retardation or developmental delay using logistic regression models for both the NCPP and Medicaid. The Medicaid data included information about whether a prescription for antibiotics was filled following the diagnosis of a UTI. Thus, in the Medicaid data set we were able to identify women who probably did and did not have treatment following the diagnosis. In the NCPP and the Medicaid models we determined the risk for mental retardation or developmental delay in children after controlling for gestational age at entry into the study, maternal age, maternal race, and birth weight. These confounders control for the effect of factors already known to be associated with both the exposure and the outcome, and are not believed to be in the causal pathway. The Medicaid data revealed an increased relative risks for mental retardation or developmental delay of 1.47 (95% CI, 1.08-2.01) in the first trimester and an RR of 1.42 (95% CI, 1.12-1.81) in the third trimester, when there was no documentation of an antibiotic prescription being filled. When medication prescriptions were filled there were no increased risks for mental retardation or developmental delay.

As described by Leviton and Gilles, UTI was also associated with an increased risk for death. In both the NCPP and Medicaid data sets, more than 1.9% of the pregnancies resulted in a fetal death. For the NCPP analysis of death, the second trimester results are made on the basis of maternal UTI in the second trimester and fetal death in the second or third trimester. The third trimester results include some live-born infants, because the 35-day critical period following UTI occasionally extended into the postnatal period for some of the infants. Thus, for the infants who had postnatal deaths, prematurity and low birth weight could be an intermediate variable, since the relationship between UTI and prematurity has been established.^16-18 The biologic explanation of fetal death associated with maternal UTI implicated endotoxins from gram-negative bacteria. It is likely that a similar mechanism is responsible for the brain damage (mental retardation or developmental delay) associated with maternal UTI.

Limitations

Our analyses and the comparison with the Medicaid data have several important limitations. First, there were issues related to the exposure variable. The case definition for UTI from the 2 data sources differed, since the NCPP data relied on a physician diagnosis and a date of occurrence and the Medicaid data relied on a physician diagnosis or a urine culture followed by a prescription for antibiotics within 14 days of diagnosis. The UTI rate was 20.9% for the Medicaid group and 15.6% for the NCPP group. Also, we did not have data on the specific symptoms or reason for the urine culture and the organism identified on culture. It is possible that only febrile cases of bacteriuria or specific bacterial species were associated with the adverse outcomes, but this could not be identified in these data sets. We could not determine how long the symptoms, if any, were present before treatment or the efficacy of treatment as measured by test-of-cure follow-up cultures later in pregnancy. Women with a positive urine culture or urine analysis who did not fill an antibiotic claim within 14 days following the laboratory test were not cases in the Medicaid analysis. This misclassification would result in some women with an untreated bacteriuria remaining in the comparison group and biasing the results toward the null hypothesis of not finding a difference between those in the case group and those in the control group.

Second, the case definition for mental retardation or developmental delay differed for the 2 data sets. For the NCPP we had actual scores on a test of cognitive functioning, and 4.5% scored in the mental retardation range (IQ 69). The Medicaid mental retardation or developmental delay diagnoses were identified for 7.0% using International Classification of Diseases-ninth revision-Clinical Modification codes 315 (specific delays in development), 317 (mild mental retardation), 318 (other specified mental retardation), or 319 (unspecified mental retardation).

We calculated the risk for death and mental retardation without regard to antibiotic prescription status for the NCPP, because the medication data were not coded with a date. The Medicaid data were useful in this regard, since we had the actual date the prescription was filled. Women with a filled prescription did not necessarily take the medication, and women without a filled prescription might have received samples of the antibiotic from their physicians. Thus, we do not know the actual compliance rate for treatment. It must be noted that there could be a difference in some other unidentified characteristic in these analyses of the women who filled their prescription compared with the women who did not. This could be referred to as a healthy patient effect. Finally, for the NCPP data set, children with mental retardation were more likely to be lost to follow-up before their fourth-year checkup than the children with normal cognitive functioning since out-of-home and institutional placement was still recommended for children with special needs before 1975. These factors, which could not be controlled for in this secondary data analysis, might bias the results toward the null hypothesis of no difference between the groups and therefore dilute the magnitude of our findings.

Further Research

Additional longitudinal studies are needed to evaluate the association of the time of infection with presenting symptoms and organisms. Also, animal models are needed to understand the mechanisms of injury to the fetal brain.

Conclusions

Our findings support an association between third trimester maternal UTI and fetal death, mental retardation, or developmental delay. Women with asymptomatic bacteriuria early in pregnancy may be at higher risk for UTI during the latter half of pregnancy,⁷ and more aggressive screening techniques may be appropriate for this population. The Medicaid data suggest there is no difference in mental retardation or developmental delay outcomes between the treated women and women without UTIs. Some physicians may want to advise women of the potential risks for infection of their fetus when a UTI is diagnosed, in an attempt to increase compliance with the medication regimen.

Primary care physicians make specialty referrals to obtain advice for clinically uncertain diagnostic evaluations or treatment plans, to obtain a specialized service that falls outside their scope of practice, because of patient or third-party requests, or because of a combination of these reasons.¹ The clinical reasons for these referral decisions include characteristics of the presenting health problem, the burden and severity of comorbidities, and patient preferences for various treatments and outcomes.

Previous research has shown that certain features ascribed to morbidities influence the likelihood of specialty referral. The type of diagnosis is the most obvious determinant. In one study,² adults with malignancies were 5 times more likely to be referred than those with respiratory illnesses. Similarly, in the Netherlands, Van Suijlekom-Smit and colleagues³ found more than an 8-fold variation among childhood diagnosis groups in the likelihood of referral. For patients with similar diagnoses, research has found that severe variants are more likely to be referred.^4-7 Specialty referral is also influenced by the array and complexity of comorbid conditions.⁸

The conceptual foundations of primary care provide further insight into how clinical factors may influence referral to specialty care. A defining feature of primary care is the provision of a comprehensive set of services that meets the majority of a population’s health needs.^9,10 Primary care physicians develop greater experience and expertise for health problems with which they are familiar than those that occur less often. It follows that they would seek specialist assistance for uncommon health problems. However, empirical evidence for this effect is currently lacking.

Our goal was to test the hypothesis that the frequency with which a condition is seen by primary care physicians (practice prevalence) influences the likelihood of referral from primary to specialty care. We use the term practice prevalence to mean the frequency of presentation to primary care physicians and to distinguish it from the frequency of occurrence in the community. Also, we examine the impact of other clinical factors on primary care physicians’ referral decisions, including patient age, sex, comorbidities, and the medical versus surgical nature of the target condition’s management.

Methods

Data Source and Study Sample

We used the 1989 to 1994 National Ambulatory Medical Care Surveys (NAMCS) to examine referrals made to specialist physicians during visits with primary care physicians. NAMCS is a nationally representative survey of office-based physician visits in the United States. Each year, a multistage probability sample of nonfederally funded US physicians who are engaged in patient care activities (excluding radiologists, anesthesiologists, and pathologists) is selected from the master files of the American Medical Association and the American Osteopathic Association. For 1 week each selected physician completes a questionnaire for a 20% to 100% systematic sample of patient visits. Details of the survey methodology and the survey instrument are presented elsewhere.¹¹ The distribution of patient age and sex remained consistent over the 6 years of data collection we used.¹² The 1995 to 1998 surveys were not used, because information on referral was not collected. Using the 1994 and 1998 NAMCS, Forrest and Whelan¹³ found that primary care practice patterns did not substantively differ over time. The pooled data set contained 219,830 visits, of which 78,107 (35.5%) were with generalists (self-reported specialty designation was family/general practice, general pediatrics, or general internal medicine).

Clinical Factors

To examine the referral characteristics of different types of conditions, International Classification of Diseases-ninth revision-Clinical Modification (ICD-9-CM) codes were grouped into clinically similar categories that we called expanded diagnosis clusters (EDCs). The EDCs were based on the diagnosis clusters originally developed by Schneeweiss and coworkers.¹⁴ EDCs were assigned to the vast majority of primary care visits in the data set, matching 93.6% of the unique diagnosis codes. EDCs were grouped into clinical domains based on the nature of the problems and the specialty most responsible for the care. We also classified each EDC according to whether the primary treatment approach was medical or surgical.

A practice-prevalence ratio was calculated for each condition. The numerator was the number of visits for the EDC under consideration, and the denominator was the total number of visits in the data set. These ratios were further divided into tertiles (ie, high-, medium-, and low-frequency EDCs.)

To account for the number and severity of comorbid conditions with which patients presented, we assigned a comorbidity index to each visit. The index was based on the aggregated diagnostic groups (ADGs), the building blocks of The Johns Hopkins Adjusted Clinical Groups Case-Mix system.¹⁵ The 32 ADGs represent morbidity groups that contain ICD-9-CM diagnosis codes that are similar with respect to likelihood of persistence, expected need for health care resources, and other clinical criteria. In the NAMCS, up to 3 ICD-9-CM codes were assigned per visit; thus, each visit was assigned 1 to 3 ADGs. A comorbidity index score was obtained by summing ADG-specific resource intensity weights.* Larger weights suggest more complex health problems and greater expected resource intensity. In previous research,¹³ the comorbidity index increased with patient age and distinguished type of primary care delivery site by morbidity burden. The comorbidity index was divided into tertiles representing high, medium, and low levels of comorbidity.

Data Analysis

The unadjusted association of each clinical factor with the probability of referral was evaluated in bivariate analyses using the chi-square statistic. We further analyzed the relationship between a condition’s practice prevalence and its referral rate using scatter plots and Pearson correlation coefficients. Because both the practice prevalence and referral rate variables were right skewed, we used a logarithmic transformation to normalize both. For the scatter plots we excluded conditions with unstable referral rate estimates because of small sample sizes. Specifically, a condition was included if the ratio of the difference between the 95% confidence interval upper and lower limits of the referral rates divided by the condition-specific referral rate was less than 1. For example, multiple sclerosis was excluded because it presented at a rate of just 6 per 10,000 visits, had a referral rate of 7.1%, and the difference between the 95% confidence interval limits divided by the referral rate was 2.7.

We conducted a multiple logistic regression analysis with the patient visit as the unit of analysis for examining the independent effect of practice prevalence of the presenting problem on the chances of referral. Controlling variables in this analysis included age, sex, comorbidity index tertile, and the medical versus surgical nature of the principal diagnosis. To determine if the clinical complexity of a visit as assessed by the comorbidity index modified the effect of practice prevalence, we also added interaction terms for these variables to the final model. The goodness of fit of the final model was tested using the Hosmer-Lemeshow statistic, which suggested adequate model fit.¹⁶ The regression analyses used the generalized estimating equation¹⁷ to control for the intraphysician correlation of clustered visits.

Results

The distribution of clinical characteristics is shown in Table 1. The population was somewhat more represented by women (57.5%) than men, and by children/adolescents (33.3%) and seniors (39.4%) than middle-aged persons. The average comorbidity index value was 0.63, but there was considerable variability and the distribution was right skewed (median=0.31; interquartile range=0.05-0.83). The mean EDC practice-prevalence estimate for patient visits was 51 per 1000 primary care visits, but this distribution was also right skewed (median=25; interquartile range=10-106).

In the bivariate analyses, the probability of referral was significantly related to age, sex, practice prevalence of the condition, patient comorbidity, and whether the condition was a surgical problem Table 2. For the 65 condition categories with stable referral rate estimates Figure 1, the correlation between the logarithms of the practice-prevalence ratios and the referral rates was -0.87 (P <.001). Also, this relationship remained after stratifying for the 35 medical (r=-0.82; P <.001) and 24 surgical conditions (r=-0.88; P=.001).

Table 3 shows the multiple logistic regression results modeling the probability of referral during an office visit as a function of the clinical factors. There were no significant differences in the chances of referring patients aged 11 to 64 years once measures of morbidity, such as patient comorbidity and practice prevalence of the principal diagnosis, were controlled. Men were 19% more likely to be referred than women, and medical conditions were 39% less likely to be referred than surgical ones.

Only one of the practice prevalence/comorbidity interaction terms was significantly different from 0: commonly occurring conditions presenting among patients with high levels of comorbidity. This finding implies that the comorbidity has a stronger influence on the chances of referral for patients presenting with common problems than those presenting with less common problems.

Table 4 shows the estimated probabilities of referral based on differences in practice prevalence and comorbidity. These probabilities were obtained from the b coefficients in Table 3. The reference group for the probability estimates is women aged 18 to 44 years with health problems categorized as medical conditions. The chances of referral varied as much as 8-fold based on only the practice prevalence of the principal diagnosis and level of comorbidity.

Discussion

Our results support the hypothesis that the frequency with which patients’ health problems present to primary care physicians (practice prevalence) has a strong inverse relationship with the chances of referral to specialty care. Primary care physicians were more likely to send patients with uncommon problems to specialists and retain those with the most common conditions. This finding highlights the responsible judgment primary care physicians employ in recognizing the boundaries of their scope of practice. Practice prevalence is a defining feature of the primary care–specialty care interface.

Referring patients with uncommon problems to specialists is a rational way to organize medical care. Outcomes are related to the volume of patients managed with a specific condition.¹⁸ Specialists need to care for an adequate number of patients with uncommon problems to maintain clinical competence. Patient self-referral, however, which dilutes the prevalence of health problems presenting to specialists, may result in potentially invasive and expensive diagnostic approaches to patients more appropriately evaluated by primary care physicians.¹⁹

In addition to a condition’s practice prevalence, the number and severity of comorbidities managed during the visit influenced primary care physicians’ decisions to make specialty referrals. Also, we found an interaction effect between high practice prevalence and high levels of comorbidity. In other words, patients with uncommon conditions were commonly referred, regardless of the complexity of other conditions. The chances of referral markedly increased for patients with common conditions when they also presented with co-existing medically complex health problems. Thus, the rare presentations for which specialist assistance is sought may be a result of either the practice prevalence of the presenting problem or the overall complexity of a patient.

Men were more commonly referred than were women, after accounting for differences in the nature of their problems. A possible explanation for this finding is that because women make more office visits over a year than men,²⁰ their probability of referral during any given visit will be lower given roughly equal chances of referral between the 2 groups during the course of a year.

Further Research

We demonstrated that the potential need for surgical interventions was an important predictor of referral. Even after other clinical factors were controlled, medical conditions were 39% less likely to be referred than surgical ones. This is not surprising given that primary care physicians generally perform only minor office-based surgical procedures. But which surgical procedures should be in the scope of practice of primary care physicians? This question deserves further research and could be addressed in part by an analysis that is similar to the one presented here. Common outpatient procedures are candidates for inclusion as primary care services. Secondary considerations include the requirements and expense of necessary equipment, technical personnel, and training. Research that builds epidemiologic profiles of office-based procedures would be helpful in determining how responsibilities should be divided between generalists and specialists for these technical services.

Limitations

Several limitations in our study’s data source warrant consideration. First, the data set of visits provided information on primary care physicians’ referral decisions and did not elucidate whether patients actually received specialty care. Second, the sample was restricted to visits made to generalist physicians, excluding both obstetrician-gynecologists and medical subspecialists who may act as primary care physicians. Third, the NAMCS data set did not include hospital-based physicians, who are known to have higher referral rates than their office-based counterparts.¹³ Fourth, the unit of analysis was the visit rather than the patient. Patients with certain chronic conditions may have higher referral rates than suggested by our data if the measure used is the percentage of persons obtaining specialty care over a year. The advantage of focusing on the visit is that physician referral decisions can be examined rather than specialist use. Fifth, some conditions had lower than expected referral rates (eg, appendicitis had a referral rate of 46%), because the denominator for the referral rates was all visits made to generalists for the condition, which included both new presentations and follow-up visits. Finally, because of data limitations we did not assess the extent to which condition prevalence within an individual physician’s own practice affects his or her referral behavior.

Specialist visits can be initiated by primary care physician referral, patient self-referral, or specialist-to-specialist cross-referral. Although our database did not permit us to examine each of these pathways, other research suggests that primary care physician referral is the predominant route, particularly in health maintenance organizations.¹²

Conclusions

Our findings provide evidence that the boundaries between primary care physicians and specialists are defined in part by prevalence of health problems and the overall complexity of patients. Future research should focus on identifying modifiable characteristics of the physician-patient interaction, physicians, their practices, and the health system that influence referral decisions, after accounting for clinical factors. The appreciation of relevant clinical factors is critical to the fair application of administrative and financial constraints on physicians’ abilities to refer. Managed care plans that penalize physicians for high referral behavior, without adjusting for practice prevalence and comorbidity work, are contrary to the goal of providing quality patient care in the most appropriate settings. With more precise definitions of the clinical determinants of referral for populations, health systems can better gauge generalist and specialist workforce requirements.

Acknowledgments

This work was supported by the Agency for Healthcare Research and Quality grants #R01 and #HS09377. Barbara Starfield inspired this work and provided comments on the manuscript. We also thank Barbara Bartman, Norm Smith, MD, MPH, and Jonathan Weiner MD, MPH, for their review and comments on the manuscript. Mia Kang and Sarah von Schrader provided excellent technical assistance.

Related Resources

• Agency for Healthcare Research and Quality, Primary Care Subdirectory Page—includes research articles on primary care referral patterns and coordination of care among referring physicians and specialists. http://www.ahrq.gov/research/primarix.htm

Primary care physicians make specialty referrals to obtain advice for clinically uncertain diagnostic evaluations or treatment plans, to obtain a specialized service that falls outside their scope of practice, because of patient or third-party requests, or because of a combination of these reasons.¹ The clinical reasons for these referral decisions include characteristics of the presenting health problem, the burden and severity of comorbidities, and patient preferences for various treatments and outcomes.

Previous research has shown that certain features ascribed to morbidities influence the likelihood of specialty referral. The type of diagnosis is the most obvious determinant. In one study,² adults with malignancies were 5 times more likely to be referred than those with respiratory illnesses. Similarly, in the Netherlands, Van Suijlekom-Smit and colleagues³ found more than an 8-fold variation among childhood diagnosis groups in the likelihood of referral. For patients with similar diagnoses, research has found that severe variants are more likely to be referred.^4-7 Specialty referral is also influenced by the array and complexity of comorbid conditions.⁸

The conceptual foundations of primary care provide further insight into how clinical factors may influence referral to specialty care. A defining feature of primary care is the provision of a comprehensive set of services that meets the majority of a population’s health needs.^9,10 Primary care physicians develop greater experience and expertise for health problems with which they are familiar than those that occur less often. It follows that they would seek specialist assistance for uncommon health problems. However, empirical evidence for this effect is currently lacking.

Our goal was to test the hypothesis that the frequency with which a condition is seen by primary care physicians (practice prevalence) influences the likelihood of referral from primary to specialty care. We use the term practice prevalence to mean the frequency of presentation to primary care physicians and to distinguish it from the frequency of occurrence in the community. Also, we examine the impact of other clinical factors on primary care physicians’ referral decisions, including patient age, sex, comorbidities, and the medical versus surgical nature of the target condition’s management.

Methods

Data Source and Study Sample

We used the 1989 to 1994 National Ambulatory Medical Care Surveys (NAMCS) to examine referrals made to specialist physicians during visits with primary care physicians. NAMCS is a nationally representative survey of office-based physician visits in the United States. Each year, a multistage probability sample of nonfederally funded US physicians who are engaged in patient care activities (excluding radiologists, anesthesiologists, and pathologists) is selected from the master files of the American Medical Association and the American Osteopathic Association. For 1 week each selected physician completes a questionnaire for a 20% to 100% systematic sample of patient visits. Details of the survey methodology and the survey instrument are presented elsewhere.¹¹ The distribution of patient age and sex remained consistent over the 6 years of data collection we used.¹² The 1995 to 1998 surveys were not used, because information on referral was not collected. Using the 1994 and 1998 NAMCS, Forrest and Whelan¹³ found that primary care practice patterns did not substantively differ over time. The pooled data set contained 219,830 visits, of which 78,107 (35.5%) were with generalists (self-reported specialty designation was family/general practice, general pediatrics, or general internal medicine).

Clinical Factors

To examine the referral characteristics of different types of conditions, International Classification of Diseases-ninth revision-Clinical Modification (ICD-9-CM) codes were grouped into clinically similar categories that we called expanded diagnosis clusters (EDCs). The EDCs were based on the diagnosis clusters originally developed by Schneeweiss and coworkers.¹⁴ EDCs were assigned to the vast majority of primary care visits in the data set, matching 93.6% of the unique diagnosis codes. EDCs were grouped into clinical domains based on the nature of the problems and the specialty most responsible for the care. We also classified each EDC according to whether the primary treatment approach was medical or surgical.

A practice-prevalence ratio was calculated for each condition. The numerator was the number of visits for the EDC under consideration, and the denominator was the total number of visits in the data set. These ratios were further divided into tertiles (ie, high-, medium-, and low-frequency EDCs.)

To account for the number and severity of comorbid conditions with which patients presented, we assigned a comorbidity index to each visit. The index was based on the aggregated diagnostic groups (ADGs), the building blocks of The Johns Hopkins Adjusted Clinical Groups Case-Mix system.¹⁵ The 32 ADGs represent morbidity groups that contain ICD-9-CM diagnosis codes that are similar with respect to likelihood of persistence, expected need for health care resources, and other clinical criteria. In the NAMCS, up to 3 ICD-9-CM codes were assigned per visit; thus, each visit was assigned 1 to 3 ADGs. A comorbidity index score was obtained by summing ADG-specific resource intensity weights.* Larger weights suggest more complex health problems and greater expected resource intensity. In previous research,¹³ the comorbidity index increased with patient age and distinguished type of primary care delivery site by morbidity burden. The comorbidity index was divided into tertiles representing high, medium, and low levels of comorbidity.

Data Analysis

The unadjusted association of each clinical factor with the probability of referral was evaluated in bivariate analyses using the chi-square statistic. We further analyzed the relationship between a condition’s practice prevalence and its referral rate using scatter plots and Pearson correlation coefficients. Because both the practice prevalence and referral rate variables were right skewed, we used a logarithmic transformation to normalize both. For the scatter plots we excluded conditions with unstable referral rate estimates because of small sample sizes. Specifically, a condition was included if the ratio of the difference between the 95% confidence interval upper and lower limits of the referral rates divided by the condition-specific referral rate was less than 1. For example, multiple sclerosis was excluded because it presented at a rate of just 6 per 10,000 visits, had a referral rate of 7.1%, and the difference between the 95% confidence interval limits divided by the referral rate was 2.7.

We conducted a multiple logistic regression analysis with the patient visit as the unit of analysis for examining the independent effect of practice prevalence of the presenting problem on the chances of referral. Controlling variables in this analysis included age, sex, comorbidity index tertile, and the medical versus surgical nature of the principal diagnosis. To determine if the clinical complexity of a visit as assessed by the comorbidity index modified the effect of practice prevalence, we also added interaction terms for these variables to the final model. The goodness of fit of the final model was tested using the Hosmer-Lemeshow statistic, which suggested adequate model fit.¹⁶ The regression analyses used the generalized estimating equation¹⁷ to control for the intraphysician correlation of clustered visits.

Results

The distribution of clinical characteristics is shown in Table 1. The population was somewhat more represented by women (57.5%) than men, and by children/adolescents (33.3%) and seniors (39.4%) than middle-aged persons. The average comorbidity index value was 0.63, but there was considerable variability and the distribution was right skewed (median=0.31; interquartile range=0.05-0.83). The mean EDC practice-prevalence estimate for patient visits was 51 per 1000 primary care visits, but this distribution was also right skewed (median=25; interquartile range=10-106).

In the bivariate analyses, the probability of referral was significantly related to age, sex, practice prevalence of the condition, patient comorbidity, and whether the condition was a surgical problem Table 2. For the 65 condition categories with stable referral rate estimates Figure 1, the correlation between the logarithms of the practice-prevalence ratios and the referral rates was -0.87 (P <.001). Also, this relationship remained after stratifying for the 35 medical (r=-0.82; P <.001) and 24 surgical conditions (r=-0.88; P=.001).

Table 3 shows the multiple logistic regression results modeling the probability of referral during an office visit as a function of the clinical factors. There were no significant differences in the chances of referring patients aged 11 to 64 years once measures of morbidity, such as patient comorbidity and practice prevalence of the principal diagnosis, were controlled. Men were 19% more likely to be referred than women, and medical conditions were 39% less likely to be referred than surgical ones.

Only one of the practice prevalence/comorbidity interaction terms was significantly different from 0: commonly occurring conditions presenting among patients with high levels of comorbidity. This finding implies that the comorbidity has a stronger influence on the chances of referral for patients presenting with common problems than those presenting with less common problems.

Table 4 shows the estimated probabilities of referral based on differences in practice prevalence and comorbidity. These probabilities were obtained from the b coefficients in Table 3. The reference group for the probability estimates is women aged 18 to 44 years with health problems categorized as medical conditions. The chances of referral varied as much as 8-fold based on only the practice prevalence of the principal diagnosis and level of comorbidity.

Discussion

Our results support the hypothesis that the frequency with which patients’ health problems present to primary care physicians (practice prevalence) has a strong inverse relationship with the chances of referral to specialty care. Primary care physicians were more likely to send patients with uncommon problems to specialists and retain those with the most common conditions. This finding highlights the responsible judgment primary care physicians employ in recognizing the boundaries of their scope of practice. Practice prevalence is a defining feature of the primary care–specialty care interface.

Referring patients with uncommon problems to specialists is a rational way to organize medical care. Outcomes are related to the volume of patients managed with a specific condition.¹⁸ Specialists need to care for an adequate number of patients with uncommon problems to maintain clinical competence. Patient self-referral, however, which dilutes the prevalence of health problems presenting to specialists, may result in potentially invasive and expensive diagnostic approaches to patients more appropriately evaluated by primary care physicians.¹⁹

In addition to a condition’s practice prevalence, the number and severity of comorbidities managed during the visit influenced primary care physicians’ decisions to make specialty referrals. Also, we found an interaction effect between high practice prevalence and high levels of comorbidity. In other words, patients with uncommon conditions were commonly referred, regardless of the complexity of other conditions. The chances of referral markedly increased for patients with common conditions when they also presented with co-existing medically complex health problems. Thus, the rare presentations for which specialist assistance is sought may be a result of either the practice prevalence of the presenting problem or the overall complexity of a patient.

Men were more commonly referred than were women, after accounting for differences in the nature of their problems. A possible explanation for this finding is that because women make more office visits over a year than men,²⁰ their probability of referral during any given visit will be lower given roughly equal chances of referral between the 2 groups during the course of a year.

Further Research

We demonstrated that the potential need for surgical interventions was an important predictor of referral. Even after other clinical factors were controlled, medical conditions were 39% less likely to be referred than surgical ones. This is not surprising given that primary care physicians generally perform only minor office-based surgical procedures. But which surgical procedures should be in the scope of practice of primary care physicians? This question deserves further research and could be addressed in part by an analysis that is similar to the one presented here. Common outpatient procedures are candidates for inclusion as primary care services. Secondary considerations include the requirements and expense of necessary equipment, technical personnel, and training. Research that builds epidemiologic profiles of office-based procedures would be helpful in determining how responsibilities should be divided between generalists and specialists for these technical services.

Limitations

Several limitations in our study’s data source warrant consideration. First, the data set of visits provided information on primary care physicians’ referral decisions and did not elucidate whether patients actually received specialty care. Second, the sample was restricted to visits made to generalist physicians, excluding both obstetrician-gynecologists and medical subspecialists who may act as primary care physicians. Third, the NAMCS data set did not include hospital-based physicians, who are known to have higher referral rates than their office-based counterparts.¹³ Fourth, the unit of analysis was the visit rather than the patient. Patients with certain chronic conditions may have higher referral rates than suggested by our data if the measure used is the percentage of persons obtaining specialty care over a year. The advantage of focusing on the visit is that physician referral decisions can be examined rather than specialist use. Fifth, some conditions had lower than expected referral rates (eg, appendicitis had a referral rate of 46%), because the denominator for the referral rates was all visits made to generalists for the condition, which included both new presentations and follow-up visits. Finally, because of data limitations we did not assess the extent to which condition prevalence within an individual physician’s own practice affects his or her referral behavior.

Specialist visits can be initiated by primary care physician referral, patient self-referral, or specialist-to-specialist cross-referral. Although our database did not permit us to examine each of these pathways, other research suggests that primary care physician referral is the predominant route, particularly in health maintenance organizations.¹²

Conclusions

Our findings provide evidence that the boundaries between primary care physicians and specialists are defined in part by prevalence of health problems and the overall complexity of patients. Future research should focus on identifying modifiable characteristics of the physician-patient interaction, physicians, their practices, and the health system that influence referral decisions, after accounting for clinical factors. The appreciation of relevant clinical factors is critical to the fair application of administrative and financial constraints on physicians’ abilities to refer. Managed care plans that penalize physicians for high referral behavior, without adjusting for practice prevalence and comorbidity work, are contrary to the goal of providing quality patient care in the most appropriate settings. With more precise definitions of the clinical determinants of referral for populations, health systems can better gauge generalist and specialist workforce requirements.

Acknowledgments

This work was supported by the Agency for Healthcare Research and Quality grants #R01 and #HS09377. Barbara Starfield inspired this work and provided comments on the manuscript. We also thank Barbara Bartman, Norm Smith, MD, MPH, and Jonathan Weiner MD, MPH, for their review and comments on the manuscript. Mia Kang and Sarah von Schrader provided excellent technical assistance.

Related Resources

• Agency for Healthcare Research and Quality, Primary Care Subdirectory Page—includes research articles on primary care referral patterns and coordination of care among referring physicians and specialists. http://www.ahrq.gov/research/primarix.htm

Primary care physicians make specialty referrals to obtain advice for clinically uncertain diagnostic evaluations or treatment plans, to obtain a specialized service that falls outside their scope of practice, because of patient or third-party requests, or because of a combination of these reasons.¹ The clinical reasons for these referral decisions include characteristics of the presenting health problem, the burden and severity of comorbidities, and patient preferences for various treatments and outcomes.

Previous research has shown that certain features ascribed to morbidities influence the likelihood of specialty referral. The type of diagnosis is the most obvious determinant. In one study,² adults with malignancies were 5 times more likely to be referred than those with respiratory illnesses. Similarly, in the Netherlands, Van Suijlekom-Smit and colleagues³ found more than an 8-fold variation among childhood diagnosis groups in the likelihood of referral. For patients with similar diagnoses, research has found that severe variants are more likely to be referred.^4-7 Specialty referral is also influenced by the array and complexity of comorbid conditions.⁸

The conceptual foundations of primary care provide further insight into how clinical factors may influence referral to specialty care. A defining feature of primary care is the provision of a comprehensive set of services that meets the majority of a population’s health needs.^9,10 Primary care physicians develop greater experience and expertise for health problems with which they are familiar than those that occur less often. It follows that they would seek specialist assistance for uncommon health problems. However, empirical evidence for this effect is currently lacking.

Our goal was to test the hypothesis that the frequency with which a condition is seen by primary care physicians (practice prevalence) influences the likelihood of referral from primary to specialty care. We use the term practice prevalence to mean the frequency of presentation to primary care physicians and to distinguish it from the frequency of occurrence in the community. Also, we examine the impact of other clinical factors on primary care physicians’ referral decisions, including patient age, sex, comorbidities, and the medical versus surgical nature of the target condition’s management.

Methods

Data Source and Study Sample

We used the 1989 to 1994 National Ambulatory Medical Care Surveys (NAMCS) to examine referrals made to specialist physicians during visits with primary care physicians. NAMCS is a nationally representative survey of office-based physician visits in the United States. Each year, a multistage probability sample of nonfederally funded US physicians who are engaged in patient care activities (excluding radiologists, anesthesiologists, and pathologists) is selected from the master files of the American Medical Association and the American Osteopathic Association. For 1 week each selected physician completes a questionnaire for a 20% to 100% systematic sample of patient visits. Details of the survey methodology and the survey instrument are presented elsewhere.¹¹ The distribution of patient age and sex remained consistent over the 6 years of data collection we used.¹² The 1995 to 1998 surveys were not used, because information on referral was not collected. Using the 1994 and 1998 NAMCS, Forrest and Whelan¹³ found that primary care practice patterns did not substantively differ over time. The pooled data set contained 219,830 visits, of which 78,107 (35.5%) were with generalists (self-reported specialty designation was family/general practice, general pediatrics, or general internal medicine).

Clinical Factors

To examine the referral characteristics of different types of conditions, International Classification of Diseases-ninth revision-Clinical Modification (ICD-9-CM) codes were grouped into clinically similar categories that we called expanded diagnosis clusters (EDCs). The EDCs were based on the diagnosis clusters originally developed by Schneeweiss and coworkers.¹⁴ EDCs were assigned to the vast majority of primary care visits in the data set, matching 93.6% of the unique diagnosis codes. EDCs were grouped into clinical domains based on the nature of the problems and the specialty most responsible for the care. We also classified each EDC according to whether the primary treatment approach was medical or surgical.

A practice-prevalence ratio was calculated for each condition. The numerator was the number of visits for the EDC under consideration, and the denominator was the total number of visits in the data set. These ratios were further divided into tertiles (ie, high-, medium-, and low-frequency EDCs.)

To account for the number and severity of comorbid conditions with which patients presented, we assigned a comorbidity index to each visit. The index was based on the aggregated diagnostic groups (ADGs), the building blocks of The Johns Hopkins Adjusted Clinical Groups Case-Mix system.¹⁵ The 32 ADGs represent morbidity groups that contain ICD-9-CM diagnosis codes that are similar with respect to likelihood of persistence, expected need for health care resources, and other clinical criteria. In the NAMCS, up to 3 ICD-9-CM codes were assigned per visit; thus, each visit was assigned 1 to 3 ADGs. A comorbidity index score was obtained by summing ADG-specific resource intensity weights.* Larger weights suggest more complex health problems and greater expected resource intensity. In previous research,¹³ the comorbidity index increased with patient age and distinguished type of primary care delivery site by morbidity burden. The comorbidity index was divided into tertiles representing high, medium, and low levels of comorbidity.

Data Analysis

The unadjusted association of each clinical factor with the probability of referral was evaluated in bivariate analyses using the chi-square statistic. We further analyzed the relationship between a condition’s practice prevalence and its referral rate using scatter plots and Pearson correlation coefficients. Because both the practice prevalence and referral rate variables were right skewed, we used a logarithmic transformation to normalize both. For the scatter plots we excluded conditions with unstable referral rate estimates because of small sample sizes. Specifically, a condition was included if the ratio of the difference between the 95% confidence interval upper and lower limits of the referral rates divided by the condition-specific referral rate was less than 1. For example, multiple sclerosis was excluded because it presented at a rate of just 6 per 10,000 visits, had a referral rate of 7.1%, and the difference between the 95% confidence interval limits divided by the referral rate was 2.7.

We conducted a multiple logistic regression analysis with the patient visit as the unit of analysis for examining the independent effect of practice prevalence of the presenting problem on the chances of referral. Controlling variables in this analysis included age, sex, comorbidity index tertile, and the medical versus surgical nature of the principal diagnosis. To determine if the clinical complexity of a visit as assessed by the comorbidity index modified the effect of practice prevalence, we also added interaction terms for these variables to the final model. The goodness of fit of the final model was tested using the Hosmer-Lemeshow statistic, which suggested adequate model fit.¹⁶ The regression analyses used the generalized estimating equation¹⁷ to control for the intraphysician correlation of clustered visits.

Results

The distribution of clinical characteristics is shown in Table 1. The population was somewhat more represented by women (57.5%) than men, and by children/adolescents (33.3%) and seniors (39.4%) than middle-aged persons. The average comorbidity index value was 0.63, but there was considerable variability and the distribution was right skewed (median=0.31; interquartile range=0.05-0.83). The mean EDC practice-prevalence estimate for patient visits was 51 per 1000 primary care visits, but this distribution was also right skewed (median=25; interquartile range=10-106).

In the bivariate analyses, the probability of referral was significantly related to age, sex, practice prevalence of the condition, patient comorbidity, and whether the condition was a surgical problem Table 2. For the 65 condition categories with stable referral rate estimates Figure 1, the correlation between the logarithms of the practice-prevalence ratios and the referral rates was -0.87 (P <.001). Also, this relationship remained after stratifying for the 35 medical (r=-0.82; P <.001) and 24 surgical conditions (r=-0.88; P=.001).

Table 3 shows the multiple logistic regression results modeling the probability of referral during an office visit as a function of the clinical factors. There were no significant differences in the chances of referring patients aged 11 to 64 years once measures of morbidity, such as patient comorbidity and practice prevalence of the principal diagnosis, were controlled. Men were 19% more likely to be referred than women, and medical conditions were 39% less likely to be referred than surgical ones.

Only one of the practice prevalence/comorbidity interaction terms was significantly different from 0: commonly occurring conditions presenting among patients with high levels of comorbidity. This finding implies that the comorbidity has a stronger influence on the chances of referral for patients presenting with common problems than those presenting with less common problems.

Table 4 shows the estimated probabilities of referral based on differences in practice prevalence and comorbidity. These probabilities were obtained from the b coefficients in Table 3. The reference group for the probability estimates is women aged 18 to 44 years with health problems categorized as medical conditions. The chances of referral varied as much as 8-fold based on only the practice prevalence of the principal diagnosis and level of comorbidity.

Discussion

Our results support the hypothesis that the frequency with which patients’ health problems present to primary care physicians (practice prevalence) has a strong inverse relationship with the chances of referral to specialty care. Primary care physicians were more likely to send patients with uncommon problems to specialists and retain those with the most common conditions. This finding highlights the responsible judgment primary care physicians employ in recognizing the boundaries of their scope of practice. Practice prevalence is a defining feature of the primary care–specialty care interface.

Referring patients with uncommon problems to specialists is a rational way to organize medical care. Outcomes are related to the volume of patients managed with a specific condition.¹⁸ Specialists need to care for an adequate number of patients with uncommon problems to maintain clinical competence. Patient self-referral, however, which dilutes the prevalence of health problems presenting to specialists, may result in potentially invasive and expensive diagnostic approaches to patients more appropriately evaluated by primary care physicians.¹⁹

In addition to a condition’s practice prevalence, the number and severity of comorbidities managed during the visit influenced primary care physicians’ decisions to make specialty referrals. Also, we found an interaction effect between high practice prevalence and high levels of comorbidity. In other words, patients with uncommon conditions were commonly referred, regardless of the complexity of other conditions. The chances of referral markedly increased for patients with common conditions when they also presented with co-existing medically complex health problems. Thus, the rare presentations for which specialist assistance is sought may be a result of either the practice prevalence of the presenting problem or the overall complexity of a patient.

Men were more commonly referred than were women, after accounting for differences in the nature of their problems. A possible explanation for this finding is that because women make more office visits over a year than men,²⁰ their probability of referral during any given visit will be lower given roughly equal chances of referral between the 2 groups during the course of a year.

Further Research

We demonstrated that the potential need for surgical interventions was an important predictor of referral. Even after other clinical factors were controlled, medical conditions were 39% less likely to be referred than surgical ones. This is not surprising given that primary care physicians generally perform only minor office-based surgical procedures. But which surgical procedures should be in the scope of practice of primary care physicians? This question deserves further research and could be addressed in part by an analysis that is similar to the one presented here. Common outpatient procedures are candidates for inclusion as primary care services. Secondary considerations include the requirements and expense of necessary equipment, technical personnel, and training. Research that builds epidemiologic profiles of office-based procedures would be helpful in determining how responsibilities should be divided between generalists and specialists for these technical services.

Limitations

Several limitations in our study’s data source warrant consideration. First, the data set of visits provided information on primary care physicians’ referral decisions and did not elucidate whether patients actually received specialty care. Second, the sample was restricted to visits made to generalist physicians, excluding both obstetrician-gynecologists and medical subspecialists who may act as primary care physicians. Third, the NAMCS data set did not include hospital-based physicians, who are known to have higher referral rates than their office-based counterparts.¹³ Fourth, the unit of analysis was the visit rather than the patient. Patients with certain chronic conditions may have higher referral rates than suggested by our data if the measure used is the percentage of persons obtaining specialty care over a year. The advantage of focusing on the visit is that physician referral decisions can be examined rather than specialist use. Fifth, some conditions had lower than expected referral rates (eg, appendicitis had a referral rate of 46%), because the denominator for the referral rates was all visits made to generalists for the condition, which included both new presentations and follow-up visits. Finally, because of data limitations we did not assess the extent to which condition prevalence within an individual physician’s own practice affects his or her referral behavior.

Specialist visits can be initiated by primary care physician referral, patient self-referral, or specialist-to-specialist cross-referral. Although our database did not permit us to examine each of these pathways, other research suggests that primary care physician referral is the predominant route, particularly in health maintenance organizations.¹²

Conclusions

Our findings provide evidence that the boundaries between primary care physicians and specialists are defined in part by prevalence of health problems and the overall complexity of patients. Future research should focus on identifying modifiable characteristics of the physician-patient interaction, physicians, their practices, and the health system that influence referral decisions, after accounting for clinical factors. The appreciation of relevant clinical factors is critical to the fair application of administrative and financial constraints on physicians’ abilities to refer. Managed care plans that penalize physicians for high referral behavior, without adjusting for practice prevalence and comorbidity work, are contrary to the goal of providing quality patient care in the most appropriate settings. With more precise definitions of the clinical determinants of referral for populations, health systems can better gauge generalist and specialist workforce requirements.

Acknowledgments

This work was supported by the Agency for Healthcare Research and Quality grants #R01 and #HS09377. Barbara Starfield inspired this work and provided comments on the manuscript. We also thank Barbara Bartman, Norm Smith, MD, MPH, and Jonathan Weiner MD, MPH, for their review and comments on the manuscript. Mia Kang and Sarah von Schrader provided excellent technical assistance.

Related Resources

• Agency for Healthcare Research and Quality, Primary Care Subdirectory Page—includes research articles on primary care referral patterns and coordination of care among referring physicians and specialists. http://www.ahrq.gov/research/primarix.htm

Telemedicine can be broadly defined as the use of telecommunications to provide medical information and services.¹ It includes a computer connected to the Internet and videoconferencing. The Internet, for example, could be used to improve patient care and enhance biomedical research by connecting practitioners to up-to-date information.² With nearly 110,000 American physicians routinely using the Internet in 1995,³ some believe that it will change the patterns of physician-patient relationships.⁴ A few physicians claim that communicating by E-mail with patients about nonemergent care and test results has saved time and money.^3,5,6

Videoconferencing can help physicians manage the medical and financial risks of providing care to rural and underserved patients.¹ It has been used successfully throughout the United States in such specialties as dermatology,^7-9 psychiatry,^10-15 pulmonary medicine,¹⁶ and cardiology.^17-19 Efforts to expand the use of telemedicine have contributed to making it a cheaper method of providing medical information and education.^20-22

Rural health providers face unique challenges in delivering care: isolation, lack of communication, and lack of access to current medical information and continuing medical education.^23-28 Although telemedicine promises to address these problems with computers and videoconferencing, rural physicians have been slow to accept it.^29-36

The Missouri Telemedicine Network (MTN) consists of 21 videoconferencing sites in 16 Missouri counties. We evaluated a demonstration project in 3 of the counties where a high-speed computer data infrastructure was installed in 10 outpatient practices in 4 communities with populations ranging from 3000 to 8000. The infrastructure included a computer workstation with E-mail, access to the World Wide Web, medical databases including MEDLINE, community-specific demographic information, a calendar, and access to a medical librarian. Important goals of the workstation included fostering networking and access to educational opportunities and current medical information. The videoconferencing facilities were located in the hospitals in the 3 demonstration counties, plus one large group practice clinic. Participation in the project was voluntary.

Because changing physician behavior has proved difficult,^31,36-40 we investigated how rural health care providers perceive the introduction of telemedicine (videoconferencing and a computer workstation) to their practices. We also wanted to create a framework for assessing the readiness of rural providers to adopt telemedicine and to develop a guide for fostering the adoption of this technology.

Methods

We collected qualitative data during semistructured interviews using questions developed from pilot interviews with information specialists and MTN participants. Data were gathered at 10 outpatient practices in the 4 communities with both a computer infrastructure and videoconferencing. Three of the out-patient practices were affiliated with a public tertiary care center; 7 were private practices; and 3 were group practices. Our sampling matrix included physicians, nurses, and administrative staff from all the clinics. Between March and August 1998 we individually interviewed all physicians at the site and at least 2 nurses and administrative staff from each clinic. All interviews were conducted by the second author.

After giving their consent, all participants responded to the following open-ended questions regarding both the video and computer components of telemedicine: (1) What do you perceive are the advantages and disadvantages of the telemedicine technology? (2) What do you perceive are the barriers and facilitators to using the telemedicine technology? (3) How do you use the telemedicine technology? (4) Can you describe the ways in which the telemedicine technology has changed your role? (5) How has the telemedicine technology affected the quality of care you deliver? (6) Do you have any suggestions for improving the telemedicine technology? In addition to these 6 questions, we collected demographic information on age, sex, length of practice, and provider status at the end of the interview.

To guard against any bias toward advocating telemedicine, we stated to respondents at the beginning of the interview that we wanted their honest observations about telemedicine and that their responses would be confidential. We confirmed their observations throughout the interview. Also, before analyzing the data we noted our own bias and preconceptions toward telemedicine, so we could consciously avoid them while reviewing the data.⁴¹

Study staff transcribed the interviews verbatim and entered them into a computer database program, Ethnograph, which was designed to help organize textual material.⁴² We divided interviews by technology type—videoconferencing versus the computer component—and made an initial template analysis to organize and code the data.⁴³ The investigators’ multiple readings of the interviews led to further revisions of the codes until consensus was reached on the identification of salient issues or themes.^44,45 The coding scheme and the salient themes were then reviewed independently by a panel of information specialists and health care providers from nursing and medicine who were familiar with the demonstration project. The panel represented individuals with expertise in informatics and qualitative methods.

Quantitative outcome data were also obtained for each participant. Between March 1998 and February 1999, file servers in each county automatically collected data on use of the Web (number of pages accessed) and E-mail (number sent and received) through the workstation. The content of E-mails remained confidential.

Results

We completed 57 interviews. Thirteen were with physicians (9 men, 4 women) averaging 52 years of age and 19 years in rural practice. Eight were family practice physicians; 4 were in internal medicine; and one physician was in general surgery. Twenty interviews were with nurses or nurse practitioners (17 women, 3 men) averaging 43 years of age and 15 years in rural practice. Twenty-four interviews were with the administrative staff (18 women, 6 men) averaging 45 years of age and 14 years in administration. Before the implementation of telemedicine, all of the participants had minimal experience with information technology.

Those practices that were affiliated with a public tertiary care center had higher telemedicine use than those in private practices, although the overall use level would be considered low. For example, the monthly average number of E-mail messages sent from practices that were affiliated with a public tertiary care center was 25.6, while for those in private practice the average was 11.3. For E-mail messages received, the monthly average was 48.8 and 20.2, respectively. Nine of the 13 physicians used the Web, and those affiliated with the tertiary care center used it far more than those in private practice. A yearly total of 8140 visits to a single Web page was recorded for those affiliated with the tertiary care center (mean = 22.3 per day) compared with a yearly total of 734 visits to a single Web page for those in private practice (mean = 2.01 per day; P=.111). Computer use was also higher for the 4 practice sites that had a nurse practitioner.

Data were systematically gathered on the use of the videoconferencing system. However, the majority of the data represent regular dermatology or psychiatric clinics that were conducted between university physicians and patients from the rural site. The rural physicians rarely participated.

Interviews and other data yielded 6 themes related to the care providers’ receptivity to technologic change: turf, efficacy, practice context, apprehension, time to learn, and ownership. Each of these themes applies to the computer and video components of telemedicine, and each may operate as a perceived barrier or facilitator of change, depending on the provider in question. Some providers saw telemedicine as a welcome opportunity to learn, and others were resistant. The themes inevitably overlap at times, because we were qualitatively assessing the social context in which technologic changes take place.

Turf

This theme summarizes our findings from care providers who perceived telemedicine as a threat to their livelihood or professional autonomy or both. Health care practices are enmeshed in networks of social relationships. Satellite practices with direct ties to larger health care systems employ patterns of referral and consultation as part of the larger system. Private practices are autonomous units that have relationships with other providers and systems based on patterns of referral and consultation initiated by the physician.

Purveyors of telemedicine may assume that simply making this technology available will somehow persuade providers to automatically accept it and use it successfully.⁴⁶ However, some rural physicians see telemedicine as an intrusion on their territory by the urban tertiary care center.⁴⁷

Although some participants affiliated with the tertiary center saw the technology as a “good thing…it was nice to be connected to a big university,” others, particularly those in private practice, saw it as a potential threat to their sense of competency, autonomy, and livelihood.³⁷ One office staff participant in a private practice remarked on the perception among the rural providers that they “are not seen as practicing their craft correctly, that they’re not up to speed, and that’s why this [telemedicine] has come out here.”

A nurse practitioner in a private practice alluded to telemedicine as a threat to professional autonomy when she said, “I have experienced times when, although the intentions were good, the community has rejected it hands down because they didn’t need help from the outside.”

Efficacy

This theme refers to the participants’ desire to know that telemedicine will fill a functional need in their practice before they invest time and money in making such a big change. Telemedicine has no track record of directly improving patient care outcomes. Unlike drug therapies or medical procedures, telemedicine exerts indirect effects on outcomes with its abilities to enhance, streamline, or improve the process of health care delivery.

Some physicians we interviewed saw no compelling reason to integrate telemedicine into their practices. One physician in private practice who rarely used the computer said, “It doesn’t really help a lot. I think computers are good for specialists, and in primary care you know basically most of the stuff…then the other 20% of it that’s more difficult, you look it up in routine journals.”

Although some physicians saw no reason to integrate the new technology, others simply “don’t think about it.” Still other physicians—mostly those affiliated with a tertiary care center where computer technology figured prominently in patient care—welcomed telemedicine and quickly saw capabilities that would enhance their practices.

Practice Context

This theme refers to barriers to adopting telemedicine that clinics may face because they practice in rural areas where technologic change moves at a slower pace than in urban communities. One nurse practitioner in a private practice said, “We got 911 [emergency] 3 years ago. Three years before that, we finally had a 7-digit phone number. So, I think that the expectations for the rate of change and the learning curve should be pretty generous.” However, several participants, particularly those affiliated with a tertiary center, were positive that telemedicine would eventually catch on.

Apprehension

In contrast to the practice context, this theme refers to the apprehension of individual providers. When it comes to adopting new technology, some participants were philosophical about what they described as a human aversion to change. “People are scared of technology,” said one physician. Another physician in private practice said, “We don’t want to change. Everybody’s just fine the way it is…. I’m not prepared for this.”

Some providers had little confidence in their ability to operate the technology, and one nurse feared that her ignorance would get her into legal trouble: “I’m always afraid I’ll push the wrong button and…something will come up and it will say ‘illegal action.’ It scares me. I think ‘Oh my gosh, I’ve done something against the law.’“

Participants were also concerned about whether the information they would get though the videoconferencing channel would be reliable. A similar concern applied to information on the Internet. A physician in a private practice who was reluctant to use the computer workstation said: “The biggest problem I have with it (the Internet) is you don’t know [what] you’re getting…. There’s a lot of stuff on the Internet that’s no good.”

Although several physicians, particularly those in private practice, were apprehensive about telemedicine, they were willing to let others in their practice learn and use the technology. Some physicians in private practice, however, reflected on the seeming inevitability of change and were resigned to having to learn the technology.

Time to learn

This theme refers to hesitancy among providers to take the time to learn a new technology and to persuade patients of its worth.

One nurse/office manager said, “If I’m looking up something in a book, maybe the book is old, but at least I could have it done in 5 minutes…until I get good at this [computer], it’s taking me much longer.”

One physician in private practice bemoaned spending his time persuading patients that this new technology could benefit them. In contrast, a physician affiliated with a tertiary center noted several advantages of videoconferencing.

Ownership

This theme refers to participants who were professionally and emotionally invested in the technology—stakeholders who acknowledged its benefits, adapted it to their needs, and tried to help others learn. Predictably, this higher level of investment was most common in administrators, because of their familiarity with computerized procedures and records. One administrator affiliated with a tertiary center offered an example of this keen interest: “Yeah, we developed our own policies. We took some of the training modules and modified them to match what we thought. And we really had…everybody buy into using the same policies.”

The stakeholders often encourage others to “buy in” to the new technology, as described by this administrator affiliated with a tertiary center: “I don’t worry about the members of this group using it in a negative way. I want them to use it more…. The more exposure that they have to it, the more accustomed they’re going to be to using it.”

Discussion

These 6 themes (turf, efficacy, practice context, apprehension, time to learn, and ownership) provide a framework for understanding some consequences of introducing telemedicine into a rural setting. Although these themes have been noted to varying degrees by others,^29-37,46,48 we grouped all of them as key contextual elements of the rural health environment. Aside from technical issues, such as the user-friendliness of the technology, the elemental themes that emerged from our data helped us explore this broader context.

Introducing telemedicine into a rural setting is analogous in many ways to introducing managed care into such areas. Some rural providers perceive managed care as an opportunity, while others see it as a threat to their practices, taking the local health care dollar away.⁴⁹ Similarly, providers’ perceptions of telemedicine range from seeing it as a chance to improve health care delivery, as nonessential technology, or, at worst, as a threat.

Those introducing telemedicine to these areas appear to be most likely to succeed if they begin with an understanding of how the new technology is perceived by rural providers. The 6 themes we identified provide some essentials for understanding the initial process of technological change in a rural health care practice. Based on our results, rural providers’ acceptance of telemedicine is most likely to occur when there is a greater organizational integration of the new technology, a perceived increase in time efficiency, greater affiliation with a tertiary care center, a perceived increase in ownership, an enhanced ability to accommodate the changes, a reduction in apprehension, and the realization of the slower pace of change in a rural community.

These themes can be considered core issues for developing a plan that can be used when introducing telemedicine. Specific questions can then be formulated to aid in this process, including: Is there a perceived need for the technology? (turf); Who is initiating the technological change, and how it that perceived? (efficacy); How is the rate of technologic change perceived in the community? (practice context); How flexible are the users toward technologic change? What is the level of anxiety about using the technology? (apprehension); How is the time expended to learn and use the technology perceived? (time to learn); and Who manages and supports the technology? (ownership). Answers to these questions can help those introducing telemedicine to structure specific strategies for implementation that are tailored to fit the needs and concerns of each practice.

Strategies for change

After acknowledging the variability of rural practices and practice behaviors and the environmental conditions of the 10 practices in our study, we grouped them into 3 categories according to their readiness for implanting telemedicine: fertile soil, somewhat fertile soil, and barren soil. For each of these conditions, we propose strategies for change that enhance the potential for the growth of telemedicine as illustrated in the Table 1.^50-52 For those practices that have been identified as fertile soil, it is important to include the physicians and administrators in the entire planning and implementation process.⁵³ They are more likely to use physician extenders, so it is important to facilitate team building with regard to new innovations, while at the same time building various coalitions with other affiliated physicians. Empowerment is also a key to making sure the innovation is successfully implemented. Appropriate resources need to be provided, such as space for the innovation (or technology) and adequate personnel, access, and training.⁵²

For practices identified as having partly fertile soil for change, it is important to establish a sense of urgency for the implementation of the new technology and to engage in coalition building within the community and with other specialty physicians. Help is needed to create a new vision for the practice, and this should be communicated to all employees. It also helps to provide for short-term incentives regarding the new technology.⁵²

Engaging barren soil types of practices in implementing new technologies is difficult. It is possible, however, to facilitate change in the practice by developing a perceived need for the technology through presenting the physician(s) with current evidence-based medical information, for example. All the physicians need to be included in the planning and implementing process,⁵³ and steps should be taken to facilitate coalition building within the community.⁵²

Implementation strategies need to be tailored to the environmental conditions of practice sites that are carefully chosen for their potential to cultivate telemedicine. Successful sites can become exemplars to others. Establishing relationships with a practice site, however, begins with diplomatic negotiation that is sensitive to local conditions. A commitment must be made to nurture the relationship.

Limitations

The strength of our study lies in the initial investigation of rural health care providers’ perceptions of telemedicine, and we are not aware of any similar qualitative studies in the literature. The results of our study are limited, however, to the recent introduction of telemedicine technologies into rural settings. We presented perceptions of providers who were just beginning to adjust to new technologies. Future research is needed to determine the extent of these perceptions among rural health care providers in general and in particular whether some of the negative perceptions of telemedicine of the providers in our study will change over time.

Conclusions

Rural health care providers and administrators consider a range of factors, including economic ramifications, efficacy, social pressure, and apprehension, in deciding whether and how fast to adopt telemedicine technology. Since adopting this technology can be a major change, agencies trying to introduce it into rural areas should take all these factors into account in their approach to rural health care providers, staff, and communities.

Acknowledgments

Our study was funded by the National Library of Medicine, contract number: NO1-LM-6-3538.

Related resources

• Telemedicine Information Exchange (TIE) A National Library of Medicine-funded web page which offers comprehensive information on telemedicine and telehealth. http://tie.telemed.org
• Telemedicine And Health Care Informatics Legal Issues Web site A resource for providers, lawyers, professionals or anyone interested in learning more about health care law and, more specifically, the regulatory and transactional aspects of health care. http://www.netreach.net

Telemedicine can be broadly defined as the use of telecommunications to provide medical information and services.¹ It includes a computer connected to the Internet and videoconferencing. The Internet, for example, could be used to improve patient care and enhance biomedical research by connecting practitioners to up-to-date information.² With nearly 110,000 American physicians routinely using the Internet in 1995,³ some believe that it will change the patterns of physician-patient relationships.⁴ A few physicians claim that communicating by E-mail with patients about nonemergent care and test results has saved time and money.^3,5,6

Videoconferencing can help physicians manage the medical and financial risks of providing care to rural and underserved patients.¹ It has been used successfully throughout the United States in such specialties as dermatology,^7-9 psychiatry,^10-15 pulmonary medicine,¹⁶ and cardiology.^17-19 Efforts to expand the use of telemedicine have contributed to making it a cheaper method of providing medical information and education.^20-22

Rural health providers face unique challenges in delivering care: isolation, lack of communication, and lack of access to current medical information and continuing medical education.^23-28 Although telemedicine promises to address these problems with computers and videoconferencing, rural physicians have been slow to accept it.^29-36

The Missouri Telemedicine Network (MTN) consists of 21 videoconferencing sites in 16 Missouri counties. We evaluated a demonstration project in 3 of the counties where a high-speed computer data infrastructure was installed in 10 outpatient practices in 4 communities with populations ranging from 3000 to 8000. The infrastructure included a computer workstation with E-mail, access to the World Wide Web, medical databases including MEDLINE, community-specific demographic information, a calendar, and access to a medical librarian. Important goals of the workstation included fostering networking and access to educational opportunities and current medical information. The videoconferencing facilities were located in the hospitals in the 3 demonstration counties, plus one large group practice clinic. Participation in the project was voluntary.

Because changing physician behavior has proved difficult,^31,36-40 we investigated how rural health care providers perceive the introduction of telemedicine (videoconferencing and a computer workstation) to their practices. We also wanted to create a framework for assessing the readiness of rural providers to adopt telemedicine and to develop a guide for fostering the adoption of this technology.

Methods

We collected qualitative data during semistructured interviews using questions developed from pilot interviews with information specialists and MTN participants. Data were gathered at 10 outpatient practices in the 4 communities with both a computer infrastructure and videoconferencing. Three of the out-patient practices were affiliated with a public tertiary care center; 7 were private practices; and 3 were group practices. Our sampling matrix included physicians, nurses, and administrative staff from all the clinics. Between March and August 1998 we individually interviewed all physicians at the site and at least 2 nurses and administrative staff from each clinic. All interviews were conducted by the second author.

After giving their consent, all participants responded to the following open-ended questions regarding both the video and computer components of telemedicine: (1) What do you perceive are the advantages and disadvantages of the telemedicine technology? (2) What do you perceive are the barriers and facilitators to using the telemedicine technology? (3) How do you use the telemedicine technology? (4) Can you describe the ways in which the telemedicine technology has changed your role? (5) How has the telemedicine technology affected the quality of care you deliver? (6) Do you have any suggestions for improving the telemedicine technology? In addition to these 6 questions, we collected demographic information on age, sex, length of practice, and provider status at the end of the interview.

To guard against any bias toward advocating telemedicine, we stated to respondents at the beginning of the interview that we wanted their honest observations about telemedicine and that their responses would be confidential. We confirmed their observations throughout the interview. Also, before analyzing the data we noted our own bias and preconceptions toward telemedicine, so we could consciously avoid them while reviewing the data.⁴¹

Study staff transcribed the interviews verbatim and entered them into a computer database program, Ethnograph, which was designed to help organize textual material.⁴² We divided interviews by technology type—videoconferencing versus the computer component—and made an initial template analysis to organize and code the data.⁴³ The investigators’ multiple readings of the interviews led to further revisions of the codes until consensus was reached on the identification of salient issues or themes.^44,45 The coding scheme and the salient themes were then reviewed independently by a panel of information specialists and health care providers from nursing and medicine who were familiar with the demonstration project. The panel represented individuals with expertise in informatics and qualitative methods.

Quantitative outcome data were also obtained for each participant. Between March 1998 and February 1999, file servers in each county automatically collected data on use of the Web (number of pages accessed) and E-mail (number sent and received) through the workstation. The content of E-mails remained confidential.

Results

We completed 57 interviews. Thirteen were with physicians (9 men, 4 women) averaging 52 years of age and 19 years in rural practice. Eight were family practice physicians; 4 were in internal medicine; and one physician was in general surgery. Twenty interviews were with nurses or nurse practitioners (17 women, 3 men) averaging 43 years of age and 15 years in rural practice. Twenty-four interviews were with the administrative staff (18 women, 6 men) averaging 45 years of age and 14 years in administration. Before the implementation of telemedicine, all of the participants had minimal experience with information technology.

Those practices that were affiliated with a public tertiary care center had higher telemedicine use than those in private practices, although the overall use level would be considered low. For example, the monthly average number of E-mail messages sent from practices that were affiliated with a public tertiary care center was 25.6, while for those in private practice the average was 11.3. For E-mail messages received, the monthly average was 48.8 and 20.2, respectively. Nine of the 13 physicians used the Web, and those affiliated with the tertiary care center used it far more than those in private practice. A yearly total of 8140 visits to a single Web page was recorded for those affiliated with the tertiary care center (mean = 22.3 per day) compared with a yearly total of 734 visits to a single Web page for those in private practice (mean = 2.01 per day; P=.111). Computer use was also higher for the 4 practice sites that had a nurse practitioner.

Data were systematically gathered on the use of the videoconferencing system. However, the majority of the data represent regular dermatology or psychiatric clinics that were conducted between university physicians and patients from the rural site. The rural physicians rarely participated.

Interviews and other data yielded 6 themes related to the care providers’ receptivity to technologic change: turf, efficacy, practice context, apprehension, time to learn, and ownership. Each of these themes applies to the computer and video components of telemedicine, and each may operate as a perceived barrier or facilitator of change, depending on the provider in question. Some providers saw telemedicine as a welcome opportunity to learn, and others were resistant. The themes inevitably overlap at times, because we were qualitatively assessing the social context in which technologic changes take place.

Turf

This theme summarizes our findings from care providers who perceived telemedicine as a threat to their livelihood or professional autonomy or both. Health care practices are enmeshed in networks of social relationships. Satellite practices with direct ties to larger health care systems employ patterns of referral and consultation as part of the larger system. Private practices are autonomous units that have relationships with other providers and systems based on patterns of referral and consultation initiated by the physician.

Purveyors of telemedicine may assume that simply making this technology available will somehow persuade providers to automatically accept it and use it successfully.⁴⁶ However, some rural physicians see telemedicine as an intrusion on their territory by the urban tertiary care center.⁴⁷

Although some participants affiliated with the tertiary center saw the technology as a “good thing…it was nice to be connected to a big university,” others, particularly those in private practice, saw it as a potential threat to their sense of competency, autonomy, and livelihood.³⁷ One office staff participant in a private practice remarked on the perception among the rural providers that they “are not seen as practicing their craft correctly, that they’re not up to speed, and that’s why this [telemedicine] has come out here.”

A nurse practitioner in a private practice alluded to telemedicine as a threat to professional autonomy when she said, “I have experienced times when, although the intentions were good, the community has rejected it hands down because they didn’t need help from the outside.”

Efficacy

This theme refers to the participants’ desire to know that telemedicine will fill a functional need in their practice before they invest time and money in making such a big change. Telemedicine has no track record of directly improving patient care outcomes. Unlike drug therapies or medical procedures, telemedicine exerts indirect effects on outcomes with its abilities to enhance, streamline, or improve the process of health care delivery.

Some physicians we interviewed saw no compelling reason to integrate telemedicine into their practices. One physician in private practice who rarely used the computer said, “It doesn’t really help a lot. I think computers are good for specialists, and in primary care you know basically most of the stuff…then the other 20% of it that’s more difficult, you look it up in routine journals.”

Although some physicians saw no reason to integrate the new technology, others simply “don’t think about it.” Still other physicians—mostly those affiliated with a tertiary care center where computer technology figured prominently in patient care—welcomed telemedicine and quickly saw capabilities that would enhance their practices.

Practice Context

This theme refers to barriers to adopting telemedicine that clinics may face because they practice in rural areas where technologic change moves at a slower pace than in urban communities. One nurse practitioner in a private practice said, “We got 911 [emergency] 3 years ago. Three years before that, we finally had a 7-digit phone number. So, I think that the expectations for the rate of change and the learning curve should be pretty generous.” However, several participants, particularly those affiliated with a tertiary center, were positive that telemedicine would eventually catch on.

Apprehension

In contrast to the practice context, this theme refers to the apprehension of individual providers. When it comes to adopting new technology, some participants were philosophical about what they described as a human aversion to change. “People are scared of technology,” said one physician. Another physician in private practice said, “We don’t want to change. Everybody’s just fine the way it is…. I’m not prepared for this.”

Some providers had little confidence in their ability to operate the technology, and one nurse feared that her ignorance would get her into legal trouble: “I’m always afraid I’ll push the wrong button and…something will come up and it will say ‘illegal action.’ It scares me. I think ‘Oh my gosh, I’ve done something against the law.’“

Participants were also concerned about whether the information they would get though the videoconferencing channel would be reliable. A similar concern applied to information on the Internet. A physician in a private practice who was reluctant to use the computer workstation said: “The biggest problem I have with it (the Internet) is you don’t know [what] you’re getting…. There’s a lot of stuff on the Internet that’s no good.”

Although several physicians, particularly those in private practice, were apprehensive about telemedicine, they were willing to let others in their practice learn and use the technology. Some physicians in private practice, however, reflected on the seeming inevitability of change and were resigned to having to learn the technology.

Time to learn

This theme refers to hesitancy among providers to take the time to learn a new technology and to persuade patients of its worth.

One nurse/office manager said, “If I’m looking up something in a book, maybe the book is old, but at least I could have it done in 5 minutes…until I get good at this [computer], it’s taking me much longer.”

One physician in private practice bemoaned spending his time persuading patients that this new technology could benefit them. In contrast, a physician affiliated with a tertiary center noted several advantages of videoconferencing.

Ownership

This theme refers to participants who were professionally and emotionally invested in the technology—stakeholders who acknowledged its benefits, adapted it to their needs, and tried to help others learn. Predictably, this higher level of investment was most common in administrators, because of their familiarity with computerized procedures and records. One administrator affiliated with a tertiary center offered an example of this keen interest: “Yeah, we developed our own policies. We took some of the training modules and modified them to match what we thought. And we really had…everybody buy into using the same policies.”

The stakeholders often encourage others to “buy in” to the new technology, as described by this administrator affiliated with a tertiary center: “I don’t worry about the members of this group using it in a negative way. I want them to use it more…. The more exposure that they have to it, the more accustomed they’re going to be to using it.”

Discussion

These 6 themes (turf, efficacy, practice context, apprehension, time to learn, and ownership) provide a framework for understanding some consequences of introducing telemedicine into a rural setting. Although these themes have been noted to varying degrees by others,^29-37,46,48 we grouped all of them as key contextual elements of the rural health environment. Aside from technical issues, such as the user-friendliness of the technology, the elemental themes that emerged from our data helped us explore this broader context.

Introducing telemedicine into a rural setting is analogous in many ways to introducing managed care into such areas. Some rural providers perceive managed care as an opportunity, while others see it as a threat to their practices, taking the local health care dollar away.⁴⁹ Similarly, providers’ perceptions of telemedicine range from seeing it as a chance to improve health care delivery, as nonessential technology, or, at worst, as a threat.

Those introducing telemedicine to these areas appear to be most likely to succeed if they begin with an understanding of how the new technology is perceived by rural providers. The 6 themes we identified provide some essentials for understanding the initial process of technological change in a rural health care practice. Based on our results, rural providers’ acceptance of telemedicine is most likely to occur when there is a greater organizational integration of the new technology, a perceived increase in time efficiency, greater affiliation with a tertiary care center, a perceived increase in ownership, an enhanced ability to accommodate the changes, a reduction in apprehension, and the realization of the slower pace of change in a rural community.

These themes can be considered core issues for developing a plan that can be used when introducing telemedicine. Specific questions can then be formulated to aid in this process, including: Is there a perceived need for the technology? (turf); Who is initiating the technological change, and how it that perceived? (efficacy); How is the rate of technologic change perceived in the community? (practice context); How flexible are the users toward technologic change? What is the level of anxiety about using the technology? (apprehension); How is the time expended to learn and use the technology perceived? (time to learn); and Who manages and supports the technology? (ownership). Answers to these questions can help those introducing telemedicine to structure specific strategies for implementation that are tailored to fit the needs and concerns of each practice.

Strategies for change

After acknowledging the variability of rural practices and practice behaviors and the environmental conditions of the 10 practices in our study, we grouped them into 3 categories according to their readiness for implanting telemedicine: fertile soil, somewhat fertile soil, and barren soil. For each of these conditions, we propose strategies for change that enhance the potential for the growth of telemedicine as illustrated in the Table 1.^50-52 For those practices that have been identified as fertile soil, it is important to include the physicians and administrators in the entire planning and implementation process.⁵³ They are more likely to use physician extenders, so it is important to facilitate team building with regard to new innovations, while at the same time building various coalitions with other affiliated physicians. Empowerment is also a key to making sure the innovation is successfully implemented. Appropriate resources need to be provided, such as space for the innovation (or technology) and adequate personnel, access, and training.⁵²

For practices identified as having partly fertile soil for change, it is important to establish a sense of urgency for the implementation of the new technology and to engage in coalition building within the community and with other specialty physicians. Help is needed to create a new vision for the practice, and this should be communicated to all employees. It also helps to provide for short-term incentives regarding the new technology.⁵²

Engaging barren soil types of practices in implementing new technologies is difficult. It is possible, however, to facilitate change in the practice by developing a perceived need for the technology through presenting the physician(s) with current evidence-based medical information, for example. All the physicians need to be included in the planning and implementing process,⁵³ and steps should be taken to facilitate coalition building within the community.⁵²

Implementation strategies need to be tailored to the environmental conditions of practice sites that are carefully chosen for their potential to cultivate telemedicine. Successful sites can become exemplars to others. Establishing relationships with a practice site, however, begins with diplomatic negotiation that is sensitive to local conditions. A commitment must be made to nurture the relationship.

Limitations

The strength of our study lies in the initial investigation of rural health care providers’ perceptions of telemedicine, and we are not aware of any similar qualitative studies in the literature. The results of our study are limited, however, to the recent introduction of telemedicine technologies into rural settings. We presented perceptions of providers who were just beginning to adjust to new technologies. Future research is needed to determine the extent of these perceptions among rural health care providers in general and in particular whether some of the negative perceptions of telemedicine of the providers in our study will change over time.

Conclusions

Rural health care providers and administrators consider a range of factors, including economic ramifications, efficacy, social pressure, and apprehension, in deciding whether and how fast to adopt telemedicine technology. Since adopting this technology can be a major change, agencies trying to introduce it into rural areas should take all these factors into account in their approach to rural health care providers, staff, and communities.

Acknowledgments

Our study was funded by the National Library of Medicine, contract number: NO1-LM-6-3538.

Related resources

• Telemedicine Information Exchange (TIE) A National Library of Medicine-funded web page which offers comprehensive information on telemedicine and telehealth. http://tie.telemed.org
• Telemedicine And Health Care Informatics Legal Issues Web site A resource for providers, lawyers, professionals or anyone interested in learning more about health care law and, more specifically, the regulatory and transactional aspects of health care. http://www.netreach.net

Telemedicine can be broadly defined as the use of telecommunications to provide medical information and services.¹ It includes a computer connected to the Internet and videoconferencing. The Internet, for example, could be used to improve patient care and enhance biomedical research by connecting practitioners to up-to-date information.² With nearly 110,000 American physicians routinely using the Internet in 1995,³ some believe that it will change the patterns of physician-patient relationships.⁴ A few physicians claim that communicating by E-mail with patients about nonemergent care and test results has saved time and money.^3,5,6

Videoconferencing can help physicians manage the medical and financial risks of providing care to rural and underserved patients.¹ It has been used successfully throughout the United States in such specialties as dermatology,^7-9 psychiatry,^10-15 pulmonary medicine,¹⁶ and cardiology.^17-19 Efforts to expand the use of telemedicine have contributed to making it a cheaper method of providing medical information and education.^20-22

Rural health providers face unique challenges in delivering care: isolation, lack of communication, and lack of access to current medical information and continuing medical education.^23-28 Although telemedicine promises to address these problems with computers and videoconferencing, rural physicians have been slow to accept it.^29-36

The Missouri Telemedicine Network (MTN) consists of 21 videoconferencing sites in 16 Missouri counties. We evaluated a demonstration project in 3 of the counties where a high-speed computer data infrastructure was installed in 10 outpatient practices in 4 communities with populations ranging from 3000 to 8000. The infrastructure included a computer workstation with E-mail, access to the World Wide Web, medical databases including MEDLINE, community-specific demographic information, a calendar, and access to a medical librarian. Important goals of the workstation included fostering networking and access to educational opportunities and current medical information. The videoconferencing facilities were located in the hospitals in the 3 demonstration counties, plus one large group practice clinic. Participation in the project was voluntary.

Because changing physician behavior has proved difficult,^31,36-40 we investigated how rural health care providers perceive the introduction of telemedicine (videoconferencing and a computer workstation) to their practices. We also wanted to create a framework for assessing the readiness of rural providers to adopt telemedicine and to develop a guide for fostering the adoption of this technology.

Methods

We collected qualitative data during semistructured interviews using questions developed from pilot interviews with information specialists and MTN participants. Data were gathered at 10 outpatient practices in the 4 communities with both a computer infrastructure and videoconferencing. Three of the out-patient practices were affiliated with a public tertiary care center; 7 were private practices; and 3 were group practices. Our sampling matrix included physicians, nurses, and administrative staff from all the clinics. Between March and August 1998 we individually interviewed all physicians at the site and at least 2 nurses and administrative staff from each clinic. All interviews were conducted by the second author.

After giving their consent, all participants responded to the following open-ended questions regarding both the video and computer components of telemedicine: (1) What do you perceive are the advantages and disadvantages of the telemedicine technology? (2) What do you perceive are the barriers and facilitators to using the telemedicine technology? (3) How do you use the telemedicine technology? (4) Can you describe the ways in which the telemedicine technology has changed your role? (5) How has the telemedicine technology affected the quality of care you deliver? (6) Do you have any suggestions for improving the telemedicine technology? In addition to these 6 questions, we collected demographic information on age, sex, length of practice, and provider status at the end of the interview.

To guard against any bias toward advocating telemedicine, we stated to respondents at the beginning of the interview that we wanted their honest observations about telemedicine and that their responses would be confidential. We confirmed their observations throughout the interview. Also, before analyzing the data we noted our own bias and preconceptions toward telemedicine, so we could consciously avoid them while reviewing the data.⁴¹

Study staff transcribed the interviews verbatim and entered them into a computer database program, Ethnograph, which was designed to help organize textual material.⁴² We divided interviews by technology type—videoconferencing versus the computer component—and made an initial template analysis to organize and code the data.⁴³ The investigators’ multiple readings of the interviews led to further revisions of the codes until consensus was reached on the identification of salient issues or themes.^44,45 The coding scheme and the salient themes were then reviewed independently by a panel of information specialists and health care providers from nursing and medicine who were familiar with the demonstration project. The panel represented individuals with expertise in informatics and qualitative methods.

Quantitative outcome data were also obtained for each participant. Between March 1998 and February 1999, file servers in each county automatically collected data on use of the Web (number of pages accessed) and E-mail (number sent and received) through the workstation. The content of E-mails remained confidential.

Results

We completed 57 interviews. Thirteen were with physicians (9 men, 4 women) averaging 52 years of age and 19 years in rural practice. Eight were family practice physicians; 4 were in internal medicine; and one physician was in general surgery. Twenty interviews were with nurses or nurse practitioners (17 women, 3 men) averaging 43 years of age and 15 years in rural practice. Twenty-four interviews were with the administrative staff (18 women, 6 men) averaging 45 years of age and 14 years in administration. Before the implementation of telemedicine, all of the participants had minimal experience with information technology.

Those practices that were affiliated with a public tertiary care center had higher telemedicine use than those in private practices, although the overall use level would be considered low. For example, the monthly average number of E-mail messages sent from practices that were affiliated with a public tertiary care center was 25.6, while for those in private practice the average was 11.3. For E-mail messages received, the monthly average was 48.8 and 20.2, respectively. Nine of the 13 physicians used the Web, and those affiliated with the tertiary care center used it far more than those in private practice. A yearly total of 8140 visits to a single Web page was recorded for those affiliated with the tertiary care center (mean = 22.3 per day) compared with a yearly total of 734 visits to a single Web page for those in private practice (mean = 2.01 per day; P=.111). Computer use was also higher for the 4 practice sites that had a nurse practitioner.

Data were systematically gathered on the use of the videoconferencing system. However, the majority of the data represent regular dermatology or psychiatric clinics that were conducted between university physicians and patients from the rural site. The rural physicians rarely participated.

Interviews and other data yielded 6 themes related to the care providers’ receptivity to technologic change: turf, efficacy, practice context, apprehension, time to learn, and ownership. Each of these themes applies to the computer and video components of telemedicine, and each may operate as a perceived barrier or facilitator of change, depending on the provider in question. Some providers saw telemedicine as a welcome opportunity to learn, and others were resistant. The themes inevitably overlap at times, because we were qualitatively assessing the social context in which technologic changes take place.

Turf

This theme summarizes our findings from care providers who perceived telemedicine as a threat to their livelihood or professional autonomy or both. Health care practices are enmeshed in networks of social relationships. Satellite practices with direct ties to larger health care systems employ patterns of referral and consultation as part of the larger system. Private practices are autonomous units that have relationships with other providers and systems based on patterns of referral and consultation initiated by the physician.

Purveyors of telemedicine may assume that simply making this technology available will somehow persuade providers to automatically accept it and use it successfully.⁴⁶ However, some rural physicians see telemedicine as an intrusion on their territory by the urban tertiary care center.⁴⁷

Although some participants affiliated with the tertiary center saw the technology as a “good thing…it was nice to be connected to a big university,” others, particularly those in private practice, saw it as a potential threat to their sense of competency, autonomy, and livelihood.³⁷ One office staff participant in a private practice remarked on the perception among the rural providers that they “are not seen as practicing their craft correctly, that they’re not up to speed, and that’s why this [telemedicine] has come out here.”

A nurse practitioner in a private practice alluded to telemedicine as a threat to professional autonomy when she said, “I have experienced times when, although the intentions were good, the community has rejected it hands down because they didn’t need help from the outside.”

Efficacy

This theme refers to the participants’ desire to know that telemedicine will fill a functional need in their practice before they invest time and money in making such a big change. Telemedicine has no track record of directly improving patient care outcomes. Unlike drug therapies or medical procedures, telemedicine exerts indirect effects on outcomes with its abilities to enhance, streamline, or improve the process of health care delivery.

Some physicians we interviewed saw no compelling reason to integrate telemedicine into their practices. One physician in private practice who rarely used the computer said, “It doesn’t really help a lot. I think computers are good for specialists, and in primary care you know basically most of the stuff…then the other 20% of it that’s more difficult, you look it up in routine journals.”

Although some physicians saw no reason to integrate the new technology, others simply “don’t think about it.” Still other physicians—mostly those affiliated with a tertiary care center where computer technology figured prominently in patient care—welcomed telemedicine and quickly saw capabilities that would enhance their practices.

Practice Context

This theme refers to barriers to adopting telemedicine that clinics may face because they practice in rural areas where technologic change moves at a slower pace than in urban communities. One nurse practitioner in a private practice said, “We got 911 [emergency] 3 years ago. Three years before that, we finally had a 7-digit phone number. So, I think that the expectations for the rate of change and the learning curve should be pretty generous.” However, several participants, particularly those affiliated with a tertiary center, were positive that telemedicine would eventually catch on.

Apprehension

In contrast to the practice context, this theme refers to the apprehension of individual providers. When it comes to adopting new technology, some participants were philosophical about what they described as a human aversion to change. “People are scared of technology,” said one physician. Another physician in private practice said, “We don’t want to change. Everybody’s just fine the way it is…. I’m not prepared for this.”

Some providers had little confidence in their ability to operate the technology, and one nurse feared that her ignorance would get her into legal trouble: “I’m always afraid I’ll push the wrong button and…something will come up and it will say ‘illegal action.’ It scares me. I think ‘Oh my gosh, I’ve done something against the law.’“

Participants were also concerned about whether the information they would get though the videoconferencing channel would be reliable. A similar concern applied to information on the Internet. A physician in a private practice who was reluctant to use the computer workstation said: “The biggest problem I have with it (the Internet) is you don’t know [what] you’re getting…. There’s a lot of stuff on the Internet that’s no good.”

Although several physicians, particularly those in private practice, were apprehensive about telemedicine, they were willing to let others in their practice learn and use the technology. Some physicians in private practice, however, reflected on the seeming inevitability of change and were resigned to having to learn the technology.

Time to learn

This theme refers to hesitancy among providers to take the time to learn a new technology and to persuade patients of its worth.

One nurse/office manager said, “If I’m looking up something in a book, maybe the book is old, but at least I could have it done in 5 minutes…until I get good at this [computer], it’s taking me much longer.”

One physician in private practice bemoaned spending his time persuading patients that this new technology could benefit them. In contrast, a physician affiliated with a tertiary center noted several advantages of videoconferencing.

Ownership

This theme refers to participants who were professionally and emotionally invested in the technology—stakeholders who acknowledged its benefits, adapted it to their needs, and tried to help others learn. Predictably, this higher level of investment was most common in administrators, because of their familiarity with computerized procedures and records. One administrator affiliated with a tertiary center offered an example of this keen interest: “Yeah, we developed our own policies. We took some of the training modules and modified them to match what we thought. And we really had…everybody buy into using the same policies.”

The stakeholders often encourage others to “buy in” to the new technology, as described by this administrator affiliated with a tertiary center: “I don’t worry about the members of this group using it in a negative way. I want them to use it more…. The more exposure that they have to it, the more accustomed they’re going to be to using it.”

Discussion

These 6 themes (turf, efficacy, practice context, apprehension, time to learn, and ownership) provide a framework for understanding some consequences of introducing telemedicine into a rural setting. Although these themes have been noted to varying degrees by others,^29-37,46,48 we grouped all of them as key contextual elements of the rural health environment. Aside from technical issues, such as the user-friendliness of the technology, the elemental themes that emerged from our data helped us explore this broader context.

Introducing telemedicine into a rural setting is analogous in many ways to introducing managed care into such areas. Some rural providers perceive managed care as an opportunity, while others see it as a threat to their practices, taking the local health care dollar away.⁴⁹ Similarly, providers’ perceptions of telemedicine range from seeing it as a chance to improve health care delivery, as nonessential technology, or, at worst, as a threat.

Those introducing telemedicine to these areas appear to be most likely to succeed if they begin with an understanding of how the new technology is perceived by rural providers. The 6 themes we identified provide some essentials for understanding the initial process of technological change in a rural health care practice. Based on our results, rural providers’ acceptance of telemedicine is most likely to occur when there is a greater organizational integration of the new technology, a perceived increase in time efficiency, greater affiliation with a tertiary care center, a perceived increase in ownership, an enhanced ability to accommodate the changes, a reduction in apprehension, and the realization of the slower pace of change in a rural community.

These themes can be considered core issues for developing a plan that can be used when introducing telemedicine. Specific questions can then be formulated to aid in this process, including: Is there a perceived need for the technology? (turf); Who is initiating the technological change, and how it that perceived? (efficacy); How is the rate of technologic change perceived in the community? (practice context); How flexible are the users toward technologic change? What is the level of anxiety about using the technology? (apprehension); How is the time expended to learn and use the technology perceived? (time to learn); and Who manages and supports the technology? (ownership). Answers to these questions can help those introducing telemedicine to structure specific strategies for implementation that are tailored to fit the needs and concerns of each practice.

Strategies for change

After acknowledging the variability of rural practices and practice behaviors and the environmental conditions of the 10 practices in our study, we grouped them into 3 categories according to their readiness for implanting telemedicine: fertile soil, somewhat fertile soil, and barren soil. For each of these conditions, we propose strategies for change that enhance the potential for the growth of telemedicine as illustrated in the Table 1.^50-52 For those practices that have been identified as fertile soil, it is important to include the physicians and administrators in the entire planning and implementation process.⁵³ They are more likely to use physician extenders, so it is important to facilitate team building with regard to new innovations, while at the same time building various coalitions with other affiliated physicians. Empowerment is also a key to making sure the innovation is successfully implemented. Appropriate resources need to be provided, such as space for the innovation (or technology) and adequate personnel, access, and training.⁵²

For practices identified as having partly fertile soil for change, it is important to establish a sense of urgency for the implementation of the new technology and to engage in coalition building within the community and with other specialty physicians. Help is needed to create a new vision for the practice, and this should be communicated to all employees. It also helps to provide for short-term incentives regarding the new technology.⁵²

Engaging barren soil types of practices in implementing new technologies is difficult. It is possible, however, to facilitate change in the practice by developing a perceived need for the technology through presenting the physician(s) with current evidence-based medical information, for example. All the physicians need to be included in the planning and implementing process,⁵³ and steps should be taken to facilitate coalition building within the community.⁵²

Implementation strategies need to be tailored to the environmental conditions of practice sites that are carefully chosen for their potential to cultivate telemedicine. Successful sites can become exemplars to others. Establishing relationships with a practice site, however, begins with diplomatic negotiation that is sensitive to local conditions. A commitment must be made to nurture the relationship.

Limitations

The strength of our study lies in the initial investigation of rural health care providers’ perceptions of telemedicine, and we are not aware of any similar qualitative studies in the literature. The results of our study are limited, however, to the recent introduction of telemedicine technologies into rural settings. We presented perceptions of providers who were just beginning to adjust to new technologies. Future research is needed to determine the extent of these perceptions among rural health care providers in general and in particular whether some of the negative perceptions of telemedicine of the providers in our study will change over time.

Conclusions

Rural health care providers and administrators consider a range of factors, including economic ramifications, efficacy, social pressure, and apprehension, in deciding whether and how fast to adopt telemedicine technology. Since adopting this technology can be a major change, agencies trying to introduce it into rural areas should take all these factors into account in their approach to rural health care providers, staff, and communities.

Acknowledgments

Our study was funded by the National Library of Medicine, contract number: NO1-LM-6-3538.

Related resources

• Telemedicine Information Exchange (TIE) A National Library of Medicine-funded web page which offers comprehensive information on telemedicine and telehealth. http://tie.telemed.org
• Telemedicine And Health Care Informatics Legal Issues Web site A resource for providers, lawyers, professionals or anyone interested in learning more about health care law and, more specifically, the regulatory and transactional aspects of health care. http://www.netreach.net

E-mail use has been reported in a variety of broad areas, including biomedical communication, general patient surveys^2,3 and medical practice^4-8; it is also used by several institutions.⁹ Approximately half of all US adults report that they currently use E-mail at home or at work, and as many as 40% of patients would use E-mail to communicate with their physicians.¹⁰ Experts estimate that 5% to 10% of physicians are already communicating with their patients by E-mail.¹¹

There are many potential benefits to E-mail in medical practice.^10,12 It allows for efficient asynchronous communication. It eliminates phone tag, and the caller does not incur long-distance phone charges. E-mail is also a good marketing tool and lends itself well to linkage with patient education Web sites. Another advantage is improved documentation. By simply printing and including or copying all E-mail communications in the medical record, excellent documentation of the provider-patient discourse is obtained. This form of provider-patient communication may be very beneficial financially in capitated environments, where simple medical problems can be addressed without an office visit.

Along with these benefits, however, come significant potential disadvantages.^3,12,13 Many physicians are afraid that E-mail would allow patients too much access, and consequently they are reluctant to embrace this innovative communication tool. There are related concerns that patients will barrage their physicians with excessive E-mails on trivial matters. It could become another physician hassle factor of practice and create one more thing to do at the end of the day. Also, there are genuine fears that patients will think of E-mail as a hot line to the physician’s office and inappropriately use it for emergent situations, creating additional liabilities for the physician and staff. There are also very real concerns about privacy and security: How can this exchange of information between providers and patients be protected and kept confidential if it is on the Internet?

Although it is generally agreed that some guidelines are required to manage and regulate E-mail communication between patients and their health care providers,^14,15 it is equally important to assess the actual desire for this technology in specific practices. We conducted a needs assessment for E-mail communication between family physicians, other health care providers, and their patients attending 6 family practice clinics in central Texas.

Methods

Study Design and Setting

We performed a cross-sectional, in-person prevalence survey using patients with scheduled appointments to see a primary care physician in 6 of the 18 clinics of the Scott & White Healthcare System in central Texas: Northside Clinic and Santa Fe Clinic in Temple, Belton Clinic, Killeen Clinic, Bryan/College Station Clinic, and Waco Clinic. Temple, Belton, and Killeen are all located in Bell County, while Bryan/College Station is located in Brazos County, the site of the main campus of Texas A&M University. The Scott & White Institutional Review Board reviewed the study protocol.

Study Participants and Data Collection

A concerted effort was made to enroll all patients who presented to each of the 6 clinics on preselected days for the surveys. The days differed by clinic and were selected to enroll a specified number of patients according to clinic size for a total of approximately 1000 subjects. The newly opened Northside Clinic was an exception.

The survey included questions about: (1) current Internet and E-mail access; (2) how likely it was that patients would use E-mail for selected clinical services, if available, scored on a 5-point Likert scale; (3) what in their opinion was a reasonable response time to their E-mail communication about routine laboratory results, prescription refills, and medical questions; and (4) demographic information including sex, age, race/ethnicity, education, and annual family income.

Statistical Analysis

Data management and analysis were performed using SPSS software¹⁶ on a personal computer. We determined the proportion of patients with access to the Internet and E-mail by clinic. Overall, reported desired areas for using E-mail for selected clinical services were computed as the combined responses of 3 to 5 on the 5-point Likert scale. Reported desired areas for using E-mail were also computed as mean responses on the scale. We determined patient expectations regarding the timeliness of their E-mail queries. Group differences were assessed for significance using the c² statistic or Fisher exact test for categorical data and the nonparametric Kruskal-Wallis analysis of variance test for ordinal (Likert-style) data. Finally, multivariate logistic regression modeling was used to control for measured covariates on the 5 main outcome variables. All tests were 2-tailed and considered significant at P less than .05.

Results

E-mail Access

Overall, 54.3% of the patients reported having E-mail access, with a significant wide variation (33%-75%) among the 6 clinics Table 1. Internet access rates mirrored those of E-mail access rates.

Desired Areas for Using E-mail

On the basis of the combined responses of 3 to 5 on the 5-point Likert scale, we found that patients most wanted to use E-mail to request prescription refills (90%), for nonurgent consultations (87%), and to obtain routine laboratory results or test reports (84%). Using E-mail to make or cancel appointments (78%) was the area of least interest reported by all patients Figure 1.

The reported desire to use E-mail for selected clinical services varied by patient demographic characteristics, using mean responses on the 5-point Likert scale. However, after multivariate adjustment for other measured variables, we only found 2 significant associations. African Americans were somewhat less likely than other groups to want to use E-mail to get laboratory results or test reports, and older patients were significantly less interested in using E-mail to consult a nurse on nonurgent simple medical questions.

Timeliness of Responses to E-mail Queries

Patients’ expectations of the timeliness of responses to their E-mail queries varied significantly by selected clinical services but not by clinic. For routine laboratory results, for example, their expectations were: less than 9 hours, 21%; 9 to 24 hours, 53%; and more than 24 hours, 26% Table 2.

No sex or racial/ethnic differences were found regarding the timeliness of responses to E-mail queries for the 3 selected clinical services. Additionally, there were no demographic differences for the timeliness of responses to E-mail queries about prescription refills. However, there were significant age group and income differences in the timeliness of responses to patients’ E-mail queries on laboratory results or test reports. Although the majority of patients in each age group expected a response to their E-mail queries on laboratory results or test reports within 24 hours, only 6% of patients aged 65 years and older expected a response later than 24 hours compared with 20% to 29% of patients in other age groups. Surprisingly, patients with annual family incomes at both extremes had significantly higher expectations for the timeliness of responses to E-mail queries on laboratory results or test reports than their counterparts in the middle income brackets. Also, patients with educational levels at both extremes, particularly those with less than a high school education, had significantly higher expectations regarding the timeliness of responses to queries on medical questions Table 3.

Discussion

Our findings confirmed some of our suppositions and brought new information to light. We were not surprised to find that slightly more than 50% of our patients had E-mail and Internet access. This statistic is similar to those reported in other published literature.^2,3 However, we could not completely explain the large variation from site to site within our own health care system. One plausible explanation for the relatively high rate of E-mail access observed at the Bryan/College Station Clinic is its location in a university town, where the clinic clientele is more likely to have an overall higher level of education than that in some of our other sites.

Recently it was reported in the Dallas Business Journal¹⁷ that nearly half of 1000 adult patients interviewed during a Laurus Health.com telephone survey said they would like to have E-mail access to their physician’s appointment scheduling system. In our study, we found a very strong desire for this service (78%) among our patient population. The Laurus Health.com survey also reported that 37% of all patients wanted electronic access to their test results. Our study found that 84% of our patients with E-mail access desired this capability. In fact, our study found a degree of interest in electronic communication with their health care provider that was very similar to that of a University of Michigan study of patients in a general medicine clinic. In the University of Michigan study, 70% of patients surveyed indicated their willingness to communicate with their health care provider using E-mail.¹⁸

Although it has been reported that consumers are 35% more likely to choose a physician who offers to communicate with patients using E-mail,¹⁷ this is not a marketing strategy that physicians should take lightly. From our study, one could assume that patients have very high expectations regarding response times for this form of provider-patient communication. One of the more disconcerting findings of our study was the exceptionally rapid turnaround time patients expected for obtaining laboratory results or test reports, prescription refills, and answers to their medical questions. Knowing that patients would expect these results within 24 hours at least 70% of the time may be unsettling to many physicians who would feel that this time frame is not attainable with the current system of laboratory processing and handling patient requests. Meeting those expectations may require major changes for physician practices.

The Health Institution Portability and Accountability Act of 1996 places comprehensive new security requirements on the US health care industry.^19,20 The standards for privacy and protection of all health information that can be linked directly to an individual mandate that all patient E-mail communication be as secure as possible. Physicians using E-mail with their patients must be familiar and be compliant with these federal regulations.* The Journal of the American Medical Informatics Association also recently published “guidelines for the clinical use of electronic mail with patients.”¹⁴ This is an excellent reference for any physician considering E-mail communication and is available at their Web site (www.amia.org).

Limitations

Our study has several limitations. The surveys involved patients who were being seen in 6 clinics in central Texas; therefore, this sample may not truly represent the population at large. Also, all patients were scheduled to see family physicians, limiting the ability to generalize our findings to other disciplines. Future studies should expand beyond one discipline to include other primary care and specialty care departments. Also, the survey results reflect patients’ self-reported anticipated behavior if services were available and do not reflect actual usage. Another limitation is the small number of racial/ethnic minority groups.

The survey instrument we used lacked specificity on some questions. For example, we do not know whether patient expectations vary by test (ie, Do patients expect a faster response to a blood test than to an x-ray or a Papanicolaou test?) Future studies should use more specific survey instruments. Our study did not include an assessment of urban versus rural differences in E-mail communication, although it would seem that factors such as access, time, and lack of knowledge about this new technology may make a difference.²¹ Additionally, many previous studies have unveiled disparities in health care access between urban and rural populations, defined as places with fewer than 2500 residents. For example, the 25% of Americans who live in rural areas are less likely to use preventive screening services and wear seat belts. Also, in 1996, 20% of the rural population was uninsured, compared with 16% of the urban population.²² Future studies should incorporate this variable in the data collection process.

Conclusions

In central Texas the majority of patients attending⁶ family practice clinics reported having access to E-mail and indicated they would use it to request prescription refills, obtain routine laboratory results or test reports, and for nonurgent consultations independent of their age group, sex, education, or income. Also, there was a wide variability of E-mail access from practice to practice. Independent of sex or race, patients have high expectations that these tasks can be completed in a relatively short time.

Acknowledgments

We wish to acknowledge the contributions made by all the family physicians, operations managers, and supervisors at the 6 participating clinics during the data collection. We are also grateful to Pat Kirkpatrick for her initial ideas, Saundra Mason for data management, and Marcine Chambers, Linda Teer, and Virginia Gray for secretarial support.

Related resources

• American Medical Informatics Association—nonprofit organization of individuals, institutions and corporations dedicated to developing and using information technologies to improve health care. http://www.amia.org
• California Academy Of Family Physicians—offers monograph on “Making the Most of Physician-Patient E-mail.” http://www.familydocs.org

E-mail use has been reported in a variety of broad areas, including biomedical communication, general patient surveys^2,3 and medical practice^4-8; it is also used by several institutions.⁹ Approximately half of all US adults report that they currently use E-mail at home or at work, and as many as 40% of patients would use E-mail to communicate with their physicians.¹⁰ Experts estimate that 5% to 10% of physicians are already communicating with their patients by E-mail.¹¹

There are many potential benefits to E-mail in medical practice.^10,12 It allows for efficient asynchronous communication. It eliminates phone tag, and the caller does not incur long-distance phone charges. E-mail is also a good marketing tool and lends itself well to linkage with patient education Web sites. Another advantage is improved documentation. By simply printing and including or copying all E-mail communications in the medical record, excellent documentation of the provider-patient discourse is obtained. This form of provider-patient communication may be very beneficial financially in capitated environments, where simple medical problems can be addressed without an office visit.

Along with these benefits, however, come significant potential disadvantages.^3,12,13 Many physicians are afraid that E-mail would allow patients too much access, and consequently they are reluctant to embrace this innovative communication tool. There are related concerns that patients will barrage their physicians with excessive E-mails on trivial matters. It could become another physician hassle factor of practice and create one more thing to do at the end of the day. Also, there are genuine fears that patients will think of E-mail as a hot line to the physician’s office and inappropriately use it for emergent situations, creating additional liabilities for the physician and staff. There are also very real concerns about privacy and security: How can this exchange of information between providers and patients be protected and kept confidential if it is on the Internet?

Although it is generally agreed that some guidelines are required to manage and regulate E-mail communication between patients and their health care providers,^14,15 it is equally important to assess the actual desire for this technology in specific practices. We conducted a needs assessment for E-mail communication between family physicians, other health care providers, and their patients attending 6 family practice clinics in central Texas.

Methods

Study Design and Setting

We performed a cross-sectional, in-person prevalence survey using patients with scheduled appointments to see a primary care physician in 6 of the 18 clinics of the Scott & White Healthcare System in central Texas: Northside Clinic and Santa Fe Clinic in Temple, Belton Clinic, Killeen Clinic, Bryan/College Station Clinic, and Waco Clinic. Temple, Belton, and Killeen are all located in Bell County, while Bryan/College Station is located in Brazos County, the site of the main campus of Texas A&M University. The Scott & White Institutional Review Board reviewed the study protocol.

Study Participants and Data Collection

A concerted effort was made to enroll all patients who presented to each of the 6 clinics on preselected days for the surveys. The days differed by clinic and were selected to enroll a specified number of patients according to clinic size for a total of approximately 1000 subjects. The newly opened Northside Clinic was an exception.

The survey included questions about: (1) current Internet and E-mail access; (2) how likely it was that patients would use E-mail for selected clinical services, if available, scored on a 5-point Likert scale; (3) what in their opinion was a reasonable response time to their E-mail communication about routine laboratory results, prescription refills, and medical questions; and (4) demographic information including sex, age, race/ethnicity, education, and annual family income.

Statistical Analysis

Data management and analysis were performed using SPSS software¹⁶ on a personal computer. We determined the proportion of patients with access to the Internet and E-mail by clinic. Overall, reported desired areas for using E-mail for selected clinical services were computed as the combined responses of 3 to 5 on the 5-point Likert scale. Reported desired areas for using E-mail were also computed as mean responses on the scale. We determined patient expectations regarding the timeliness of their E-mail queries. Group differences were assessed for significance using the c² statistic or Fisher exact test for categorical data and the nonparametric Kruskal-Wallis analysis of variance test for ordinal (Likert-style) data. Finally, multivariate logistic regression modeling was used to control for measured covariates on the 5 main outcome variables. All tests were 2-tailed and considered significant at P less than .05.

Results

E-mail Access

Overall, 54.3% of the patients reported having E-mail access, with a significant wide variation (33%-75%) among the 6 clinics Table 1. Internet access rates mirrored those of E-mail access rates.

Desired Areas for Using E-mail

On the basis of the combined responses of 3 to 5 on the 5-point Likert scale, we found that patients most wanted to use E-mail to request prescription refills (90%), for nonurgent consultations (87%), and to obtain routine laboratory results or test reports (84%). Using E-mail to make or cancel appointments (78%) was the area of least interest reported by all patients Figure 1.

The reported desire to use E-mail for selected clinical services varied by patient demographic characteristics, using mean responses on the 5-point Likert scale. However, after multivariate adjustment for other measured variables, we only found 2 significant associations. African Americans were somewhat less likely than other groups to want to use E-mail to get laboratory results or test reports, and older patients were significantly less interested in using E-mail to consult a nurse on nonurgent simple medical questions.

Timeliness of Responses to E-mail Queries

Patients’ expectations of the timeliness of responses to their E-mail queries varied significantly by selected clinical services but not by clinic. For routine laboratory results, for example, their expectations were: less than 9 hours, 21%; 9 to 24 hours, 53%; and more than 24 hours, 26% Table 2.

No sex or racial/ethnic differences were found regarding the timeliness of responses to E-mail queries for the 3 selected clinical services. Additionally, there were no demographic differences for the timeliness of responses to E-mail queries about prescription refills. However, there were significant age group and income differences in the timeliness of responses to patients’ E-mail queries on laboratory results or test reports. Although the majority of patients in each age group expected a response to their E-mail queries on laboratory results or test reports within 24 hours, only 6% of patients aged 65 years and older expected a response later than 24 hours compared with 20% to 29% of patients in other age groups. Surprisingly, patients with annual family incomes at both extremes had significantly higher expectations for the timeliness of responses to E-mail queries on laboratory results or test reports than their counterparts in the middle income brackets. Also, patients with educational levels at both extremes, particularly those with less than a high school education, had significantly higher expectations regarding the timeliness of responses to queries on medical questions Table 3.

Discussion

Our findings confirmed some of our suppositions and brought new information to light. We were not surprised to find that slightly more than 50% of our patients had E-mail and Internet access. This statistic is similar to those reported in other published literature.^2,3 However, we could not completely explain the large variation from site to site within our own health care system. One plausible explanation for the relatively high rate of E-mail access observed at the Bryan/College Station Clinic is its location in a university town, where the clinic clientele is more likely to have an overall higher level of education than that in some of our other sites.

Recently it was reported in the Dallas Business Journal¹⁷ that nearly half of 1000 adult patients interviewed during a Laurus Health.com telephone survey said they would like to have E-mail access to their physician’s appointment scheduling system. In our study, we found a very strong desire for this service (78%) among our patient population. The Laurus Health.com survey also reported that 37% of all patients wanted electronic access to their test results. Our study found that 84% of our patients with E-mail access desired this capability. In fact, our study found a degree of interest in electronic communication with their health care provider that was very similar to that of a University of Michigan study of patients in a general medicine clinic. In the University of Michigan study, 70% of patients surveyed indicated their willingness to communicate with their health care provider using E-mail.¹⁸

Although it has been reported that consumers are 35% more likely to choose a physician who offers to communicate with patients using E-mail,¹⁷ this is not a marketing strategy that physicians should take lightly. From our study, one could assume that patients have very high expectations regarding response times for this form of provider-patient communication. One of the more disconcerting findings of our study was the exceptionally rapid turnaround time patients expected for obtaining laboratory results or test reports, prescription refills, and answers to their medical questions. Knowing that patients would expect these results within 24 hours at least 70% of the time may be unsettling to many physicians who would feel that this time frame is not attainable with the current system of laboratory processing and handling patient requests. Meeting those expectations may require major changes for physician practices.

The Health Institution Portability and Accountability Act of 1996 places comprehensive new security requirements on the US health care industry.^19,20 The standards for privacy and protection of all health information that can be linked directly to an individual mandate that all patient E-mail communication be as secure as possible. Physicians using E-mail with their patients must be familiar and be compliant with these federal regulations.* The Journal of the American Medical Informatics Association also recently published “guidelines for the clinical use of electronic mail with patients.”¹⁴ This is an excellent reference for any physician considering E-mail communication and is available at their Web site (www.amia.org).

Limitations

Our study has several limitations. The surveys involved patients who were being seen in 6 clinics in central Texas; therefore, this sample may not truly represent the population at large. Also, all patients were scheduled to see family physicians, limiting the ability to generalize our findings to other disciplines. Future studies should expand beyond one discipline to include other primary care and specialty care departments. Also, the survey results reflect patients’ self-reported anticipated behavior if services were available and do not reflect actual usage. Another limitation is the small number of racial/ethnic minority groups.

The survey instrument we used lacked specificity on some questions. For example, we do not know whether patient expectations vary by test (ie, Do patients expect a faster response to a blood test than to an x-ray or a Papanicolaou test?) Future studies should use more specific survey instruments. Our study did not include an assessment of urban versus rural differences in E-mail communication, although it would seem that factors such as access, time, and lack of knowledge about this new technology may make a difference.²¹ Additionally, many previous studies have unveiled disparities in health care access between urban and rural populations, defined as places with fewer than 2500 residents. For example, the 25% of Americans who live in rural areas are less likely to use preventive screening services and wear seat belts. Also, in 1996, 20% of the rural population was uninsured, compared with 16% of the urban population.²² Future studies should incorporate this variable in the data collection process.

Conclusions

In central Texas the majority of patients attending⁶ family practice clinics reported having access to E-mail and indicated they would use it to request prescription refills, obtain routine laboratory results or test reports, and for nonurgent consultations independent of their age group, sex, education, or income. Also, there was a wide variability of E-mail access from practice to practice. Independent of sex or race, patients have high expectations that these tasks can be completed in a relatively short time.

Acknowledgments

We wish to acknowledge the contributions made by all the family physicians, operations managers, and supervisors at the 6 participating clinics during the data collection. We are also grateful to Pat Kirkpatrick for her initial ideas, Saundra Mason for data management, and Marcine Chambers, Linda Teer, and Virginia Gray for secretarial support.

Related resources

• American Medical Informatics Association—nonprofit organization of individuals, institutions and corporations dedicated to developing and using information technologies to improve health care. http://www.amia.org
• California Academy Of Family Physicians—offers monograph on “Making the Most of Physician-Patient E-mail.” http://www.familydocs.org

E-mail use has been reported in a variety of broad areas, including biomedical communication, general patient surveys^2,3 and medical practice^4-8; it is also used by several institutions.⁹ Approximately half of all US adults report that they currently use E-mail at home or at work, and as many as 40% of patients would use E-mail to communicate with their physicians.¹⁰ Experts estimate that 5% to 10% of physicians are already communicating with their patients by E-mail.¹¹

There are many potential benefits to E-mail in medical practice.^10,12 It allows for efficient asynchronous communication. It eliminates phone tag, and the caller does not incur long-distance phone charges. E-mail is also a good marketing tool and lends itself well to linkage with patient education Web sites. Another advantage is improved documentation. By simply printing and including or copying all E-mail communications in the medical record, excellent documentation of the provider-patient discourse is obtained. This form of provider-patient communication may be very beneficial financially in capitated environments, where simple medical problems can be addressed without an office visit.

Along with these benefits, however, come significant potential disadvantages.^3,12,13 Many physicians are afraid that E-mail would allow patients too much access, and consequently they are reluctant to embrace this innovative communication tool. There are related concerns that patients will barrage their physicians with excessive E-mails on trivial matters. It could become another physician hassle factor of practice and create one more thing to do at the end of the day. Also, there are genuine fears that patients will think of E-mail as a hot line to the physician’s office and inappropriately use it for emergent situations, creating additional liabilities for the physician and staff. There are also very real concerns about privacy and security: How can this exchange of information between providers and patients be protected and kept confidential if it is on the Internet?

Although it is generally agreed that some guidelines are required to manage and regulate E-mail communication between patients and their health care providers,^14,15 it is equally important to assess the actual desire for this technology in specific practices. We conducted a needs assessment for E-mail communication between family physicians, other health care providers, and their patients attending 6 family practice clinics in central Texas.

Methods

Study Design and Setting

We performed a cross-sectional, in-person prevalence survey using patients with scheduled appointments to see a primary care physician in 6 of the 18 clinics of the Scott & White Healthcare System in central Texas: Northside Clinic and Santa Fe Clinic in Temple, Belton Clinic, Killeen Clinic, Bryan/College Station Clinic, and Waco Clinic. Temple, Belton, and Killeen are all located in Bell County, while Bryan/College Station is located in Brazos County, the site of the main campus of Texas A&M University. The Scott & White Institutional Review Board reviewed the study protocol.

Study Participants and Data Collection

A concerted effort was made to enroll all patients who presented to each of the 6 clinics on preselected days for the surveys. The days differed by clinic and were selected to enroll a specified number of patients according to clinic size for a total of approximately 1000 subjects. The newly opened Northside Clinic was an exception.

The survey included questions about: (1) current Internet and E-mail access; (2) how likely it was that patients would use E-mail for selected clinical services, if available, scored on a 5-point Likert scale; (3) what in their opinion was a reasonable response time to their E-mail communication about routine laboratory results, prescription refills, and medical questions; and (4) demographic information including sex, age, race/ethnicity, education, and annual family income.

Statistical Analysis

Data management and analysis were performed using SPSS software¹⁶ on a personal computer. We determined the proportion of patients with access to the Internet and E-mail by clinic. Overall, reported desired areas for using E-mail for selected clinical services were computed as the combined responses of 3 to 5 on the 5-point Likert scale. Reported desired areas for using E-mail were also computed as mean responses on the scale. We determined patient expectations regarding the timeliness of their E-mail queries. Group differences were assessed for significance using the c² statistic or Fisher exact test for categorical data and the nonparametric Kruskal-Wallis analysis of variance test for ordinal (Likert-style) data. Finally, multivariate logistic regression modeling was used to control for measured covariates on the 5 main outcome variables. All tests were 2-tailed and considered significant at P less than .05.

Results

E-mail Access

Overall, 54.3% of the patients reported having E-mail access, with a significant wide variation (33%-75%) among the 6 clinics Table 1. Internet access rates mirrored those of E-mail access rates.

Desired Areas for Using E-mail

On the basis of the combined responses of 3 to 5 on the 5-point Likert scale, we found that patients most wanted to use E-mail to request prescription refills (90%), for nonurgent consultations (87%), and to obtain routine laboratory results or test reports (84%). Using E-mail to make or cancel appointments (78%) was the area of least interest reported by all patients Figure 1.

The reported desire to use E-mail for selected clinical services varied by patient demographic characteristics, using mean responses on the 5-point Likert scale. However, after multivariate adjustment for other measured variables, we only found 2 significant associations. African Americans were somewhat less likely than other groups to want to use E-mail to get laboratory results or test reports, and older patients were significantly less interested in using E-mail to consult a nurse on nonurgent simple medical questions.

Timeliness of Responses to E-mail Queries

Patients’ expectations of the timeliness of responses to their E-mail queries varied significantly by selected clinical services but not by clinic. For routine laboratory results, for example, their expectations were: less than 9 hours, 21%; 9 to 24 hours, 53%; and more than 24 hours, 26% Table 2.

No sex or racial/ethnic differences were found regarding the timeliness of responses to E-mail queries for the 3 selected clinical services. Additionally, there were no demographic differences for the timeliness of responses to E-mail queries about prescription refills. However, there were significant age group and income differences in the timeliness of responses to patients’ E-mail queries on laboratory results or test reports. Although the majority of patients in each age group expected a response to their E-mail queries on laboratory results or test reports within 24 hours, only 6% of patients aged 65 years and older expected a response later than 24 hours compared with 20% to 29% of patients in other age groups. Surprisingly, patients with annual family incomes at both extremes had significantly higher expectations for the timeliness of responses to E-mail queries on laboratory results or test reports than their counterparts in the middle income brackets. Also, patients with educational levels at both extremes, particularly those with less than a high school education, had significantly higher expectations regarding the timeliness of responses to queries on medical questions Table 3.

Discussion

Our findings confirmed some of our suppositions and brought new information to light. We were not surprised to find that slightly more than 50% of our patients had E-mail and Internet access. This statistic is similar to those reported in other published literature.^2,3 However, we could not completely explain the large variation from site to site within our own health care system. One plausible explanation for the relatively high rate of E-mail access observed at the Bryan/College Station Clinic is its location in a university town, where the clinic clientele is more likely to have an overall higher level of education than that in some of our other sites.

Recently it was reported in the Dallas Business Journal¹⁷ that nearly half of 1000 adult patients interviewed during a Laurus Health.com telephone survey said they would like to have E-mail access to their physician’s appointment scheduling system. In our study, we found a very strong desire for this service (78%) among our patient population. The Laurus Health.com survey also reported that 37% of all patients wanted electronic access to their test results. Our study found that 84% of our patients with E-mail access desired this capability. In fact, our study found a degree of interest in electronic communication with their health care provider that was very similar to that of a University of Michigan study of patients in a general medicine clinic. In the University of Michigan study, 70% of patients surveyed indicated their willingness to communicate with their health care provider using E-mail.¹⁸

Although it has been reported that consumers are 35% more likely to choose a physician who offers to communicate with patients using E-mail,¹⁷ this is not a marketing strategy that physicians should take lightly. From our study, one could assume that patients have very high expectations regarding response times for this form of provider-patient communication. One of the more disconcerting findings of our study was the exceptionally rapid turnaround time patients expected for obtaining laboratory results or test reports, prescription refills, and answers to their medical questions. Knowing that patients would expect these results within 24 hours at least 70% of the time may be unsettling to many physicians who would feel that this time frame is not attainable with the current system of laboratory processing and handling patient requests. Meeting those expectations may require major changes for physician practices.

The Health Institution Portability and Accountability Act of 1996 places comprehensive new security requirements on the US health care industry.^19,20 The standards for privacy and protection of all health information that can be linked directly to an individual mandate that all patient E-mail communication be as secure as possible. Physicians using E-mail with their patients must be familiar and be compliant with these federal regulations.* The Journal of the American Medical Informatics Association also recently published “guidelines for the clinical use of electronic mail with patients.”¹⁴ This is an excellent reference for any physician considering E-mail communication and is available at their Web site (www.amia.org).

Limitations

Our study has several limitations. The surveys involved patients who were being seen in 6 clinics in central Texas; therefore, this sample may not truly represent the population at large. Also, all patients were scheduled to see family physicians, limiting the ability to generalize our findings to other disciplines. Future studies should expand beyond one discipline to include other primary care and specialty care departments. Also, the survey results reflect patients’ self-reported anticipated behavior if services were available and do not reflect actual usage. Another limitation is the small number of racial/ethnic minority groups.

The survey instrument we used lacked specificity on some questions. For example, we do not know whether patient expectations vary by test (ie, Do patients expect a faster response to a blood test than to an x-ray or a Papanicolaou test?) Future studies should use more specific survey instruments. Our study did not include an assessment of urban versus rural differences in E-mail communication, although it would seem that factors such as access, time, and lack of knowledge about this new technology may make a difference.²¹ Additionally, many previous studies have unveiled disparities in health care access between urban and rural populations, defined as places with fewer than 2500 residents. For example, the 25% of Americans who live in rural areas are less likely to use preventive screening services and wear seat belts. Also, in 1996, 20% of the rural population was uninsured, compared with 16% of the urban population.²² Future studies should incorporate this variable in the data collection process.

Conclusions

In central Texas the majority of patients attending⁶ family practice clinics reported having access to E-mail and indicated they would use it to request prescription refills, obtain routine laboratory results or test reports, and for nonurgent consultations independent of their age group, sex, education, or income. Also, there was a wide variability of E-mail access from practice to practice. Independent of sex or race, patients have high expectations that these tasks can be completed in a relatively short time.

Acknowledgments

We wish to acknowledge the contributions made by all the family physicians, operations managers, and supervisors at the 6 participating clinics during the data collection. We are also grateful to Pat Kirkpatrick for her initial ideas, Saundra Mason for data management, and Marcine Chambers, Linda Teer, and Virginia Gray for secretarial support.

Related resources

• American Medical Informatics Association—nonprofit organization of individuals, institutions and corporations dedicated to developing and using information technologies to improve health care. http://www.amia.org
• California Academy Of Family Physicians—offers monograph on “Making the Most of Physician-Patient E-mail.” http://www.familydocs.org

Dysthymia and minor depression are common depressive disorders in patients in primary care settings.^1-3 Together with major depression, these 3 disorders account for the vast majority of depressive illness present in primary care. Although the level of depressive symptomatology for these patients is less than that for major depression, these disorders are accompanied by significant morbidity,^4-6 and their impact on the health delivery system is considerable.^4,7-9 However, there are relatively few controlled trials in primary care examining the effectiveness of recommended treatments for these disorders.^10-13 Studies in this area have typically involved small groups of patients, and generalizability was limited because of such factors as stringent entrance criteria that would exclude many primary care patients with these disorders. The need for treatment outcome data for the majority of these patients seen in primary care was a principal reason for our study.

Antidepressant medications, particularly the selective serotonin reuptake inhibitors, are commonly used for treatment of depression in primary care.^14,15 Support and watchful waiting make up another common method of treatment.¹⁴ Psychologic treatments that customarily require referral to mental health providers have also been used, although stigma, fear of loss of confidentiality, increased cost, limited access in some localities, and local cultural preferences have limited their use as a treatment option. In part to address these issues a behaviorally based psychologic treatment—Problem-Solving Treatment for Primary Care (PST-PC)—was developed in the United Kingdom.¹⁶ This treatment was relatively brief and could be applied in the primary care setting. In studies involving patients with major depression in the United Kingdom, the treatment had high patient acceptance and an effectiveness comparable with antidepressants,^17,18 making it an attractive alternative when patients did not want pharmacotherapy or if such treatment was contraindicated for medical reasons. For dysthymia and minor depression there are no studies specifically examining the effectiveness of PST-PC, but this treatment has potential utility for those conditions.

In 1995 the MacArthur Foundation and the Hartford Foundation provided funding for a comparative treatment trial. The project’s development and methodology have been outlined in an earlier report.¹⁹ Four sites recruited patients 60 years and older; the results of that study have been reported elsewhere.²⁰ Two sites recruited patients aged 18 to 59 years. We present outcome data for this younger group.

Methods

Patients aged 18 to 59 years were recruited from primary care settings at 2 participating sites (Lebanon, New Hampshire, and Seattle, Washington). To be eligible, patients had to meet Diagnostic and Statistical Manual of Mental Disorders, third edition, revised (DSM-III-R) criteria for dysthymia,²¹ or specified criteria for minor depression and score 10 or higher on the 17-item Hamilton Depression Rating Scale (HDRS).²² To receive a diagnosis of minor depression, 3 of the 9 DSM-III-R symptoms for major depression (1 of these had to be depressed mood or anhedonia) had to be present for at least 4 weeks. Depression diagnoses were made by a research psychiatrist using the Primary Care Evaluation of Mental Disorders (PRIME-MD), a diagnostic instrument designed for use in primary care.²³

Design

Patients who met the entrance criteria and consented to the study were randomly assigned to paroxetine, placebo, or PST-PC using a computer-generated random allocation table. Randomization was blocked and stratified by site and by diagnosis. Treatment assignments were held by a local pharmacist and were available to study personnel only in the event of a medical emergency.

Treatment

Patients were scheduled for 6 treatment sessions occurring over 11 weeks. The treatment sessions took place in the general medical setting. Medication visits were 10 to 15 minutes each, were conducted by psychiatrists or psychiatric residents, and consisted of medication dose titration, symptom assessment, a review of adverse effects, and general support. Specific psychologic treatments or counseling were prohibited. Paroxetine and placebo were given in a double-blind fashion. Paroxetine was initiated at 10 mg per day and increased at week 2 to the target dose of 20 mg. At week 4 or 6, the dose could be further increased to 30 mg per day and at week 6 or 8 to 40 mg if there had been limited clinical improvement. Placebo was titrated in an identical fashion.

The PST-PC therapists were PhD psychologists. All therapists received training in PST-PC. The patients received 6 PST-PC sessions, lasting approximately 1 hour for the first visit and 30 minutes for each subsequent visit. Antidepressant medication was prohibited for the PST-PC group.

Assessments

Sociodemographic and clinical information was collected at baseline. Coexisting medical illness was evaluated by chart review using the Duke Severity of Illness Checklist.²⁵ Outcome measurements included self-report and interviewer rated instruments; the latter were completed blind to the patient’s treatment assignment. There were 3 principal outcome measures. One was the 20-item Hopkins Depression self-report scale²⁶ (HSCL-D-20) consisting of the 13-item depression scale and 7 additional depression-related items added to increase responsiveness.²⁷ The HSCL-D-20 score was obtained at baseline and at each treatment visit. The other principal outcome measures were a 17-item Hamilton Depression Rating Scale (HDRS), used to determine remission status, and the 36-item Medical Outcomes Study Short Form (SF-36), that provided 2 functional status measures—a mental health component (MHC) and a physical health component (PHC).^28,29 Both these measures were obtained at baseline and at 6 and 11 weeks.

Data Analysis

For continuous demographic and clinical data, we used parametric and nonparametric analysis of variance to analyze baseline differences across site, diagnostic group, and treatment assignment. Stratified contingency table analyses were used to analyze baseline differences in categorical patient variables. For all analyses, design variables (specified in advance) included diagnosis, treatment provided, and site.

We analyzed the HSCL-D-20 using a nonlinear piece-wise random coefficient model with 2 random intercepts and a random slope fit to the individual patient data.³⁰ Random intercepts were defined at baseline and at week 2. The random intercept at week 2 enabled us to model a nonlinear response to treatment. Treatment effects were evaluated by comparing the slopes of the fitted function from week 2 through week 11. Restricted maximum likelihood estimation was used to fit the random coefficient model to the data.³¹ The Tukey-Kramer multiple comparison procedure³² was used to adjust P values for multiple comparisons. For the HSCL-D-20, analyses were performed both on an intention-to-treat group (full sample) and on an adequate treatment exposure subgroup defined as patients who completed at least 4 treatment sessions.

For the HDRS data, patients were classified as remitted (HDRS≤6) or as nonremitters³³ at week 11 on the basis of previously reported normative data. The analytic method we used was a generalized linear model with binomial response and logit link function; adjustment of P values for multiple comparisons was by the Sidak procedure.³² Six-week assessment scores were carried forward for patients for whom HDRS data were unavailable at the 11-week follow-up. The analysis reported was based on the adequate treatment exposure patient sample. This analysis gives clinicians an estimate of treatment effects for patients who actually received the treatment.

For the SF-36 data, analyses were performed both on the intention-to-treat group and the adequate exposure subgroup. The analytic method used was a mixed model analysis of covariance. Baseline SF-36 MHC and PHC component scores served as covariates in each respective analysis.

Results

Patient Enrollment and Characteristics

Of the 407 patients who received a study assessment, 241 (59%) were eligible and were randomized. Of those patients assessed but not randomized, 22 were eligible but refused participation, and 144 were ineligible Figure 1.The most common reasons for ineligibility were major depression (n=77), depression with an HDRS score of less than 10 (n=26), and no depression diagnosis (n=21).

Patients were randomized to paroxetine (n=80), PST-PC (n=80), and placebo (n=81). Sociodemographic and clinical characteristics were similar for the 3 treatment groups Table 1. Comorbid anxiety disorders assessed by the PRIME-MD at baseline were present in approximately 25% of the patients but with no significant difference in prevalence across the 3 treatment groups. Depression severity was mild to moderate as reflected by a mean HDRS of 14.2 (standard deviation [SD]=3.33) and mean HSCL-D-20 of 1.6 (SD=0.65). On the SF-36, mental health functioning was more impaired (MHC mean=33.7; SD=10.2) than physical health functioning (PHC mean=47.1; SD=12.1). At baseline, there were no significant differences between patients with dysthymia and those with minor depression on any of these 4 outcome measure scales.

Treatment Received and Follow-Up

Of the 241 patients randomized, 197 (81.7%) attended at least 4 treatment sessions Figure 1; 191 (79.3%) completed all scheduled treatment sessions. Twenty patients (8.3%) did not attend any treatment sessions; they dropped out after randomization. Of these 20 patients, 16 (80%) were assigned to paroxetine or placebo; 4 (20%) were assigned to PST-PC. Subsequently, 6 patients (2.4%) discontinued treatment for adverse effects; all of these were in the paroxetine group. One patient also in the paroxetine group discontinued because of medical illness. Twenty-three patients (9.5%) with at least 1 treatment visit discontinued for a variety of other reasons, such as relocation, self-medication, or because they felt they were not getting better.

Adherence to paroxetine and placebo was high. By the second treatment visit, 85% of patients initiating treatment achieved the target dose of 20 mg (2 pills) per day (81% of those receiving paroxetine, 89% receiving placebo). By study end, 94% had achieved the target dose or higher. Of patients who came for at least 1 visit, more patients randomized to placebo were increased to 40 mg per day (21/72, 29.2%) than those randomized to paroxetine (10/73, 13.7%; P=.023). For patients randomized to PST-PC, treatment attendance was high. Of those beginning treatment, 84.2% (64/76) completed all 6 treatment sessions.

Outcomes

HSCL-D-20

One principal outcome measure was change in depression level on the HSCL-D-20 scale. In the intention-to-treat analysis, all treatment groups showed significant improvement over the 11-weeks Figure 2. The average mean change was 0.88 (SE=0.08) for paroxetine, 0.79 (0.09) for PST-PC, and 0.85 (0.09) for placebo. For paroxetine and for placebo, the rate of symptom resolution was similar and rapid during the first 2 weeks of treatment: 0.60 (.06) and 0.56 (.06), respectively; from week 2 to week 11 it slowed and remained similar: paroxetine, 0.28 (.06); placebo, 0.29 (.07). For PST-PC in the first 2 weeks, the rate of symptom resolution was slower 0.36 (.06) compared with paroxetine or placebo, but it was more rapid from week 2 through week 11; 0.43 (.07).

In this overall analysis, from baseline to 2 weeks there were significant differences in outcome by site (P=.006) and by treatment group (P=.007) but not by diagnosis (P=.497). For this time period the site by treatment group interaction was marginal (P=.101). Lebanon accounted for the majority of these treatment differences. At that site, from baseline to week 2 the improvement was significantly more rapid for paroxetine (P=.003) and for placebo (P=.016) compared with PST-PC. When outcome was examined from week 2 to week 11, there were no significant differences at the .05 level, although there was a trend toward the earlier pattern of differences by site (P=.104) and by treatment group (P=.190), with PST-PC marginally better than paroxetine (P=.090) and placebo Figure 2. On this measure diagnostic group again showed no relationship to outcome (P=.718). When the overall outcome (baseline to week 11) was examined, there were no significant differences between the 3 intervention groups.

When these analyses were repeated on the adequate treatment exposure group of patients, the results were essentially similar. There was significant reduction in symptomatology for all 3 treatment groups, but from baseline to week 11 there were essentially no differences in the amount of this reduction between the 3 treatment groups.

Remission over 11 Weeks as Measured by HDRS

The proportion of patients achieving remission status (HDRS score 6) was examined using the 197 patients with adequate treatment exposure (4 or more visits). This group was compared with the 44 patients with less than 4 visits on baseline variables. There were no significant differences except for education: 54.5% of those with fewer than 4 visits had 13 or more years of education compared with 75.6% of those with 4 or more visits (P=.005).

In the generalized linear model used to analyze the HDRS remission data, diagnostic group, treatment group, site, and all the interactions (diagnosis by treatment, site by treatment, site by diagnosis, and site by diagnosis by treatment) were entered into the analysis. There was a significant site by treatment group interaction (P=.001) and a significant diagnostic group by treatment group interaction (P=.005). To understand these interaction terms, results were examined separated by diagnosis and by site. For dysthymia at the Lebanon site, there were 2 significant effects: paroxetine had a better outcome than placebo (P <.001) or PST-PC (P<.001). For dysthymia at the Seattle site, both paroxetine and PST-PC had marginally better outcomes than placebo (P=.093 and P=.073, respectively). To display these findings, bivariate analyses were carried out for each diagnosis by site Table 2. Table 2 also shows the remission rates when patients with each diagnosis were combined across sites. For dysthymia, the remission rates were 80% for paroxetine, 56.8% for PST-PC, and 44.4% for placebo (P=.008). For minor depression, the overall remission rate was high (64.0%), and it was similar for each treatment group: 60.7% for paroxetine, 65.5% for PST-PC, and 65.6% for placebo (P=.906).

SF-36 Mental Health Component and Physical Health Component Scales

For the SF-36 MHC, on the intention-to-treat sample there was a significant baseline level by treatment group by time interaction (P=.006). Baseline MHC was then used as a covariate by dividing patient groups into tertiles on the basis of the baseline scores Table 3. Change from week 6 to week 11 was examined after controlling for baseline MHC within each group. With paroxetine there was significant improvement for the more impaired MHC group, +7.4 (SE=1.5), P <.001; and for the intermediate group, +4.3 (1.1), P <.003. For PST-PC, the absolute change for each MHC group was essentially similar: +3.1 (1.6) for the low group, +3.2 (1.3) for the intermediate group, and +3.2 (1.5) for the high group. These changes were not significant at the .05 level. For placebo, the amount of change was lower than that for the 2 active interventions; none of those changes approached statistical significance.

Results using the adequate exposure sample for the SF-36 MHC were similar overall to those obtained on the intention-to-treat analysis.

For the SF-36 PHC analyses there were no significant differences between any of the treatment groups.

Discussion

The findings from this study provide information about treatment response for these 2 diagnostic conditions, dysthymia and minor depression, in primary care patients. There are few data from other studies with which to compare these results; most treatment outcome results for these disorders come from patients treated in psychiatric settings. One study that does provide such data used a similar design and methodology on older patients (60 years and older) and was done in parallel with this study.²⁰ In that study, the patients showed improvement on all the interventions for the measures examined. However, whether outcome with the active treatments showed a significant difference over placebo plus nonspecific clinical management is clearly of interest. For this question, the results are more complex, with variations in outcome by site, diagnosis, and treatment for both age groups, depending on the measure used. The most easily interpreted results are the remission results obtained using the HDRS. These are also the reported results when all individuals received an adequate exposure to the treatment (4 or more visits). For dysthymia in the patients aged 18 to 59, there was an overall gradient with the highest recovery rate obtained for paroxetine, the next highest for PST-PC, and the lowest for placebo. The same pattern was evident for dysthymia in the patients aged 60 years and older; higher remission rates were obtained for both paroxetine and PST-PC than placebo.

When change was measured by decline over the 11-week trial on the HSCL-D-20 as the outcome variable, in the patients 60 years or older, those taking paroxetine had a significantly greater decline compared with those taking placebo at 11 weeks and a greater rate of decline from week 2 to 11. Patients receiving PST-PC did not show a significantly greater symptom reduction than those on placebo at 11 weeks, but they did show a significantly more rapid symptom reduction in weeks 2 to 11. For patients aged 18 to 59 years, there were no significant differences between the active treatments and placebo on this measure.

Results obtained using the SF-36 MHC are difficult to compare between the age groups because diagnosis showed a significant improvement in the patients 60 years and older but not in the patients aged 18 to 59 years. Dysthymic patients taking paroxetine who were 60 years and older with higher baseline MHC (less impaired) did significantly better at 11 weeks compared with those taking placebo; those receiving PST-PC did better but not significantly so. Patients with minor depression and low baseline MHC (more impaired) improved significantly more on both paroxetine and PST-PC compared with placebo. The patients aged 18 to 59 years showed a different pattern with diagnosis not relating to outcome, but all patients with low or intermediate baseline MHC improved significantly on paroxetine. Improvement amount on PST-PC fell between paroxetine and placebo, but was not significant. These results are difficult to interpret except to note that paroxetine did have a beneficial but modest effect in both age groups for some patients.

Taking an overview of the findings from both studies, it is worth noting that there are some consistent patterns of outcome related to treatment across the 2 age groups. In general, those patients taking paroxetine showed a greater improvement compared with placebo on one or more of the measures used. Similarly for PST-PC, on some measures there was a significant difference compared with placebo, although these results were more variable than those obtained with paroxetine. The greatest PST-PC versus placebo differences were present on the remission analyses; for both age groups, diagnosis was an important predictor with the best remission results obtained for patients with dysthymia. In both age groups, for those with minor depression there was a higher placebo response and almost no significant differences between either active treatment and placebo.

Strengths and Limitations

Our study has several strengths. It is focused on those depressive disorders, dysthymia and minor depression, that are common in primary care and are treated most often in that setting. The inclusion criteria were broad, permitting results to be generalizable to the majority of patients with these disorders presenting in primary care. The treatments were provided in the primary care setting, emphasizing their potential practicality for primary care practice. For the medication intervention, this placebo-controlled trial contributes to the scientific knowledge base concerning treatments in primary care. There are relatively few such controlled trials for dysthymia and even fewer for minor depression.

This is the first treatment trial outside of the United Kingdom in which the behavioral treatment PST-PC was used. As in the United Kingdom, PST-PC had a high patient acceptance rate; 80% of the patients assigned to it completed all 6 visits and 87% of those coming for one visit completed 4. On some outcome measures, it had effectiveness similar to paroxetine and greater than placebo plus clinical management, although it showed greater variability by site than paroxetine. In this trial, PST-PC therapists varied on the level of previous experience with behavioral therapy treatment, overall experience, and number of patients treated with PST-PC, all variables that may have related to their skill in delivering the treatment. Analyses are in progress to examine the effects of these and other variables on PST-PC outcome. The results reported here indicate PST-PC has promise but cannot be considered an established treatment alternative to antidepressants in depressed primary care patients, as it is in the United Kingdom.

Our study also has shortcomings. The placebo-controlled condition involved contact with a clinician for 6 visits over the 11-week trial, considerably more than usually takes place in primary care. Whether this nonspecific clinician contact related to the relatively high improvement (remission rates) for placebo, particularly for those with minor depression, cannot be assessed in our study. In retrospect, including a true “treatment as usual” group making 2 to 3 in-person visits over 11 weeks would have clarified these results. Also, the clinical significance of the amount of symptom reduction observed in the scale analyses (SF-36-MHC, HSCL-D-20) is difficult to establish. The amounts of those reductions were modest, even when statistically significant. For clinical significance, one must rely primarily on the remission analyses that were based on those patients receiving adequate exposure to the treatments, not an intention-to-treat group.

Further Research

Variation in outcome by site was a problem in this data, as it was in the group of patients 60 years and older. Further analyses have taken place, to be reported in separate publications,^34,35 in an attempt to examine the effect of other variables, such as demographics, level of medical comorbidity, or personality variables such as neuroticism. The findings we reported on patients aged 18 to 59 years and those reported elsewhere for the patients 60 years and older are examining only the effects of diagnosis, treatment received, and site. The effect, if any, of various moderator variables was not examined but will be in these later reports.

Conclusions

Evidence-based guidelines are available to direct primary care physicians’ treatment for major depression, and when implemented well, they improve patient outcomes.^27,36,37 For the treatment of minor depression and dysthymia, evidence-based guidelines are unavailable, because the evidence base is insufficient to develop recommendations.

Our results showed that paroxetine andto a lesser degree PST-PC improved remission of dysthymia more than the use of placebo plus nonspecific clinical management. Results varied for the other outcomes measured. For minor depression, the 3 interventions (paroxetine, PST-PC, and placebo) were equally effective, so general clinical management is an appropriate treatment option.

Acknowledgments

Our project was supported by the John D. and Catherine T. MacArthur Foundation, Chicago, Illinois, and by the John A. Hartford Foundation, New York, New York. In the early planning stages of the project design advice was provided by George Alexopoulis, MD; Daniel Blazer, MD; Christopher Callahan, MD; Charles Reynolds, MD; Carl Salzman, PhD; and Herbert Schulberg, PhD. Helena Kraemer, PhD, from Stanford University provided statistical and data analytic advice; Laurence Mynors-Wallis, MD, currently at the University of Southampton provided the initial training in PST-PC.

Related resources

• Agency for Health Care Policy and Research—Consumer Guideline #5: Depression is a Treatable llness http://text.nlm.nih.gov/ftrs/pick?collect=ahcpr&dbName=depp&cd=1&t=96928766
• American Medical Association—Consumer Health Information: Insight on Depression http://www.ama-assn.org/insight/spec_con/depressn/depressn.htm
• American Psychiatric Association—APA On-line Public Information http://www.psych.org/public_info
• American Psychological Association—APA Resources for the Public: What You Need to Know About Depression http://www.apa.org/psychnet/depression.html
• MacArthur Initiative on Depression in Primary care www.depression-primarycare.org