Original Research

As US medical care has evolved, physicians have been expected to recognize and treat mental health problems in primary care,¹ “the hidden mental health network.”^2,3 Primary care clinicians are expected to observe signs of possible mental health problems, incorporate those observations into differential diagnoses, and decide which problems to treat or monitor and which to send for consultation or referral.⁴ These decisions can have important financial and health consequences, especially in dealing with depression.

Depression is common in the community⁵ and among primary care patients, 6% to 9% of whom report symptoms of major depression.^6-8 An additional 10% to 15% of primary care patients show signs of less severe but important depressive problems.^8,9 “Subclinical depression” is marked by symptoms that might indicate physical disease, signs of depression, or both; recognition may affect costs of care.^10-13

Research has begun to define the impact of depression on processes¹⁴ and costs of care.^15-17 For example, elderly patients reporting symptoms of depression have more laboratory tests performed at higher cost.¹⁵ Primary care patients diagnosed with depression had total yearly health care costs almost double those of patients without depression, with increased costs secondary to higher medical utilization and not mental health specialty treatment.¹⁶ There is evidence that depressive symptoms and the diagnosis of depression may predict increases in costs of care.¹⁷

Costs of care might be influenced by the model used by primary care physicians to identify depression.¹⁸ For example, a biomedical model might use more laboratory testing to reach a diagnosis of depression by exclusion, whereas a psychosocial model would use fewer laboratory tests while the physician pursues psychosocial issues. To identify optimal strategies for practice, it is important to determine how symptoms of depression and physician diagnosis of depression might interrelate and affect medical care costs.

We explored the following hypotheses: (1) that there are significant differences in each type of charge determined by the presence or absence of symptoms and diagnosis of depression; (2) that depressive symptoms and physician diagnosis of depression predict the occurrence of charges for specialty care, emergency services, laboratory services, and hospitalization; and (3) that depressive symptoms and physician diagnosis of depression predict the magnitude of medical charges for primary care, specialty care, emergency services, laboratory services, hospitalization, and total charges.

Methods

Study design

Five hundred eight adult nonpregnant new patients were assigned randomly to primary care providers in either family practice or general internal medicine clinics in a teaching hospital. Children younger than 18 years and pregnant women were excluded because they are not followed in general internal medicine. At enrollment and follow-up, self-reported depression was determined with the abbreviated Beck Depression Inventory (BDI)¹⁹ and health status was measured with the Medical Outcomes Studies Short Form-36 (MOS SF-36).²⁰ To avoid altering clinician practice, physicians were not provided with either score. Physicians included 105 senior residents (second and third year) in family practice and general internal medicine.

Measures

Beck Depression Inventory. The BDI is a reliable and valid instrument used to measure depressive symptoms.^19,21 The abbreviated version includes 13 items weighted and summed to produce a total score.¹⁹ A score between 9 and 15 indicates moderate depression, and a score of at least 16 indicates severe depression. The BDI is used widely for screening and to assess treatment efficacy.²² In this study, a BDI score between 0 and 8 was considered “low” or normal, and a score of at least 9 was considered “high” or indicative of symptoms of depression.

At study entrance or exit, 130 patients were identified with significant symptoms of depression (BDI > 8) by meeting criteria for moderate or severe depression¹⁹ and thus identifying roughly the top quartile of BDI scores among participants. This proportion approximates that of primary care patients estimated to experience significant depression.^6,7

Medical Outcomes Studies Short Form-36. Health status was measured with the MOS SF-36,²⁰ a 36-item self-report questionnaire. Reliability has been verified for difficult populations.²³ Summary measures can describe a physical component score and a mental component score.^24,25 The physical component score was used in this study to measure physical health status.

Medical chart review. Two physicians (K.D.B. and J.A.R.) reviewed the charts to identify notations of depression on problem lists and in visit notes to signify physician diagnosis of depression.

Charges. Charges were used as a proxy for costs. Electronic data for all health system charges were monitored from the initial visit through 1 full year of care. Six categories were monitored: primary care, specialty care, laboratory testing, emergency department, hospitalization, and total charges. Pharmacy charges were excluded because some patients purchased prescriptions outside the hospital system.

Statistical procedures

Mean log values for each area of medical charges were determined and contrasted with the Duncan multiple range test²⁶ to explore the first hypothesis that charges are associated with symptoms and diagnoses of depression. Next, a double hurdle model was used to test the hypotheses that depressive symptoms and physician diagnosis of depression predict the occurrence and magnitude of charges for a variety of services.^27,28 In a double hurdle model, the first “hurdle,” or step, involves exploring whether there are variables that can significantly predict the occurrence of an event (such as a medical charge). The second step involves exploring whether there are variables that can predict the magnitude of the event (eg, a medical charge).

Log-transformation of charges was performed to eliminate undue influence from outliers. No logistic regression models were developed for the occurrence of primary care charges or total charges (the first hurdle) because all study patients had charges in both categories. Results are presented by hypothesis.

Results

Seventy-seven of 508 study patients (15.1%) were identified as depressed by their primary care providers in chart notes. BDI scores showed considerable spread (range, 0–31) and were significantly associated with the diagnosis of depression (P < .001). Whereas 140 patients reported BDI scores of at least 9, only 36 of these patients were diagnosed as depressed by their physicians. Similarly, 41 patients were diagnosed as depressed despite reporting low (normal) BDI scores. Patients were assigned to 1 of 4 groups: those diagnosed as depressed and having high (abnormal) BDI scores (n = 36); those diagnosed as depressed despite low BDI scores (n = 41); those not diagnosed as depressed despite high BDI scores (n = 94); and those not diagnosed as depressed and not having high BDI scores (n = 337).

Hypothesis 1: overall impact of symptoms and diagnosis on charges

Groups diagnosed with depression had significantly higher log primary care charges than did those not diagnosed (Table 1). Both groups diagnosed with depression showed the highest primary care and total medical charges. Patients diagnosed with depression and reporting high BDI scores had higher specialty charges than those not depressed. Highest laboratory costs were found for those diagnosed as depressed despite low BDI scores and those with elevated BDI scores who were not diagnosed as depressed. There were no significant differences among groups for log charges for emergency care and hospital charges.

TABLE 1
Log charges of care by diagnosis and symptoms of depression

Hypotheses 2 and 3: factors predicting occurrence and magnitude of charges

Cost models are presented as regressions in Table 2. The left side of the table presents logistic regressions exploring which variables predict whether or not a patient accrues charges in all areas except primary care and total charges. Because all patients had at least 1 primary care visit charge and, hence, a total charge, it was not possible to develop a model to predict the occurrence of those charges.

Physical health status (measured by the physical component score of the MOS SF-36) predicted the occurrence of all charges measured with the exception of laboratory tests. Advanced patient age predicted increased likelihood of charges in each area; female sex showed a trend toward predicting occurrence of emergency care charges; and education showed a trend toward predicting occurrence of laboratory charges. BDI scores (measure of symptoms of depression) and physician diagnosis of depression failed to contribute significantly to the prediction of specialty care, emergency care, laboratory testing, or hospital charges. However, there was a trend for depressive symptoms to predict the occurrence of laboratory charges.

The right side of Table 2 presents regression models that predicted the magnitude of the different categories of charges. Physical health status was a significant predictor of the magnitude of all types of charges except emergency care. Patient age contributed to prediction of size of all types of charges except emergency visits and laboratory tests. Female sex was a significant predictor of magnitude of charges in primary care, laboratory tests, and total medical charges. The diagnosis of depression was a significant predictor of magnitude of primary care (P = .0029) and total medical (P = .0158) charges. Neither depressive symptoms nor the diagnosis of depression contributed significantly to the prediction of magnitude of charges for specialty care, emergency care, laboratory testing, or hospital use, although there was a trend for depressive symptoms to predict the magnitude of laboratory costs. Although an interaction term was entered into both kinds of regression equations, there was no evidence of a significant contribution from the interaction of symptoms of depression and diagnosis of depression in any of the predictor models developed.

TABLE 2
Regression analyses predicting charges

Discussion

Medical charges were related to symptoms of depression and physician diagnosis of depression in this study. Although the patient sample was small, it was representative of the primary care population in displaying a wide range of depressive symptoms as measured by the BDI.^6,7 In this study, physician diagnosis of depression was related to self-reported depression ratings: those diagnosed as depressed had significantly higher BDI scores than did those not diagnosed as depressed. However, the relationship between self-reported symptoms and diagnosis was not perfect: 72% of patients with high BDI scores were not recognized as depressed, as often occurs in primary care.^6,7 In fact, more patients diagnosed with depression had low BDI scores (< 9, n = 41) than high BDI scores (> 8, n = 36). Clearly, other factors enter the process by which primary care physicians reach the diagnosis of depression.

Symptoms of depression and the diagnosis of depression probably influence the process of care in different ways. Differences in process of care likely would be reflected in different relationships to medical charges. Physician diagnosis of depression was associated with higher primary care and total costs and contributed to models predicting magnitude of primary care and total charges. However, neither symptoms of depression nor diagnosis of depression predicted which patients were more likely to incur charges for specialty care, emergency care, laboratory tests, or hospitalization. There was a trend only for the symptoms of depression to predict who would incur laboratory charges. These findings suggest that the relationship between depression and primary care charges and total charges is clear but less apparent when looking at less frequently occurring charges.

Other demographic factors showed fairly robust associations with the occurrence of charges. Patient age predicted who would get specialty care, emergency care, laboratory costs, and hospitalization, and there was a trend for female sex to predict occurrence of emergency department charges. Health status proved to be a significant predictor of the magnitude of all charges except those for emergency care. These powerful influences must be considered to accurately assess the impact of depression on charges.

Age also predicted the total amount of charges for primary and all medical care for the year and showed a trend toward prediction of magnitude of specialty charges. Female sex was a significant predictor of magnitude of primary care charges, laboratory charges, and total charges, and less education was a significant predictor of magnitude of emergency department and hospital charges. Some of these demographic predictors are readily explained. For example, as patients age, the number and costs of medical problems often increase. More education may enhance socioeconomic status and self-care, each of which may buffer against the need for emergency care and hospitalization. The reasons that charges are often higher for women are probably more complex. Higher utilization of primary and specialty care for women was associated with lower self-report-ed health status, less education, and lower socioeconomic status in our previous study.²⁹

These results also suggest that physician diagnosis of depression in the absence of elevated BDI scores may flag a different kind of patient presentation. Diagnosis of depression without elevated BDI scores could result from effective treatment controlling the symptoms of previously diagnosed depression, but this does not adequately explain the occurrence. Perhaps other aspects of physician–patient interaction trigger a depression diagnosis without symptoms. This group ranked highest for log-transformed charges for 5 of the 6 areas explored: only for specialty care did those with high BDI scores and diagnosis of depression rank higher in total cost. This strong association with charges implies that these patients represent diagnostic dilemmas, thereby generating more primary care visits and laboratory tests. They may be diagnosed as depressed despite their low BDI scores simply because no organic explanation can be readily identified.

BDI scores showed a trend toward predicting higher laboratory charges in our models. This finding supports the importance of depressive symptoms in influencing the process of primary care, especially laboratory testing.^15,30 Perhaps the diagnosis of depression actually slowed the ordering of laboratory tests.¹⁸ Because our data did not allow a separation of charges for laboratory tests before and after the diagnosis of depression, we did not test this possibility.

The size of this sample (N = 508) and the length of time patients were followed (1 year) might not have provided adequate power to fully test the contributions of symptoms and diagnosis of depression to the 6 sets of charges. This was likely true for hospitalization charges because hospitalization was an infrequent event in this study. Previous, larger studies found indications of increased hospitalization charges for those diagnosed as depressed¹⁷ and those with symptoms of depression.^15,30 Alternatively, the recent emphasis on decreasing hospitalizations to reduce medical costs may mean that hospitalization for depressive symptoms rather than for physical illness is less likely to occur.³¹ In addition, these observations were made by resident physicians and not by community clinicians. It is not clear whether these results would generalize to another setting, although they are consistent with community observations in previous research.

These data do suggest an intriguing interplay of the impact of physician diagnosis of depression and presence of symptoms of depression in a number of indicators of charges and utilization in primary care. Even though each element was associated with increased utilization and charges, their differential impact is unclear. Both may prove important for efforts to enhance recognition of depression; recognition of a mental health problem appeared to shift the process of care in this and previous studies.^14,32 To date, there are no data indicating that the diagnosis of depression reduces utilization or costs of primary care delivery. What is known is that physicians working in primary care are more apt to accurately diagnose those with more severe symptoms of depression than those with more transient or less severe symptoms.^16,33 Although introducing a screening device such as the BDI or the PRIME-MD⁹ likely would increase the number of patients diagnosed with depression, it is unclear what impact that would have on the process, costs, and outcomes of care. Simpler interventions such as training in communication skills such as empathy³⁴ might provide the primary care physician with all the tools needed for identification of emotional distress and mental health problems^14,30 and appropriate treatment or referral.

As US medical care has evolved, physicians have been expected to recognize and treat mental health problems in primary care,¹ “the hidden mental health network.”^2,3 Primary care clinicians are expected to observe signs of possible mental health problems, incorporate those observations into differential diagnoses, and decide which problems to treat or monitor and which to send for consultation or referral.⁴ These decisions can have important financial and health consequences, especially in dealing with depression.

Depression is common in the community⁵ and among primary care patients, 6% to 9% of whom report symptoms of major depression.^6-8 An additional 10% to 15% of primary care patients show signs of less severe but important depressive problems.^8,9 “Subclinical depression” is marked by symptoms that might indicate physical disease, signs of depression, or both; recognition may affect costs of care.^10-13

Research has begun to define the impact of depression on processes¹⁴ and costs of care.^15-17 For example, elderly patients reporting symptoms of depression have more laboratory tests performed at higher cost.¹⁵ Primary care patients diagnosed with depression had total yearly health care costs almost double those of patients without depression, with increased costs secondary to higher medical utilization and not mental health specialty treatment.¹⁶ There is evidence that depressive symptoms and the diagnosis of depression may predict increases in costs of care.¹⁷

Costs of care might be influenced by the model used by primary care physicians to identify depression.¹⁸ For example, a biomedical model might use more laboratory testing to reach a diagnosis of depression by exclusion, whereas a psychosocial model would use fewer laboratory tests while the physician pursues psychosocial issues. To identify optimal strategies for practice, it is important to determine how symptoms of depression and physician diagnosis of depression might interrelate and affect medical care costs.

We explored the following hypotheses: (1) that there are significant differences in each type of charge determined by the presence or absence of symptoms and diagnosis of depression; (2) that depressive symptoms and physician diagnosis of depression predict the occurrence of charges for specialty care, emergency services, laboratory services, and hospitalization; and (3) that depressive symptoms and physician diagnosis of depression predict the magnitude of medical charges for primary care, specialty care, emergency services, laboratory services, hospitalization, and total charges.

Methods

Study design

Five hundred eight adult nonpregnant new patients were assigned randomly to primary care providers in either family practice or general internal medicine clinics in a teaching hospital. Children younger than 18 years and pregnant women were excluded because they are not followed in general internal medicine. At enrollment and follow-up, self-reported depression was determined with the abbreviated Beck Depression Inventory (BDI)¹⁹ and health status was measured with the Medical Outcomes Studies Short Form-36 (MOS SF-36).²⁰ To avoid altering clinician practice, physicians were not provided with either score. Physicians included 105 senior residents (second and third year) in family practice and general internal medicine.

Measures

Beck Depression Inventory. The BDI is a reliable and valid instrument used to measure depressive symptoms.^19,21 The abbreviated version includes 13 items weighted and summed to produce a total score.¹⁹ A score between 9 and 15 indicates moderate depression, and a score of at least 16 indicates severe depression. The BDI is used widely for screening and to assess treatment efficacy.²² In this study, a BDI score between 0 and 8 was considered “low” or normal, and a score of at least 9 was considered “high” or indicative of symptoms of depression.

At study entrance or exit, 130 patients were identified with significant symptoms of depression (BDI > 8) by meeting criteria for moderate or severe depression¹⁹ and thus identifying roughly the top quartile of BDI scores among participants. This proportion approximates that of primary care patients estimated to experience significant depression.^6,7

Medical Outcomes Studies Short Form-36. Health status was measured with the MOS SF-36,²⁰ a 36-item self-report questionnaire. Reliability has been verified for difficult populations.²³ Summary measures can describe a physical component score and a mental component score.^24,25 The physical component score was used in this study to measure physical health status.

Medical chart review. Two physicians (K.D.B. and J.A.R.) reviewed the charts to identify notations of depression on problem lists and in visit notes to signify physician diagnosis of depression.

Charges. Charges were used as a proxy for costs. Electronic data for all health system charges were monitored from the initial visit through 1 full year of care. Six categories were monitored: primary care, specialty care, laboratory testing, emergency department, hospitalization, and total charges. Pharmacy charges were excluded because some patients purchased prescriptions outside the hospital system.

Statistical procedures

Mean log values for each area of medical charges were determined and contrasted with the Duncan multiple range test²⁶ to explore the first hypothesis that charges are associated with symptoms and diagnoses of depression. Next, a double hurdle model was used to test the hypotheses that depressive symptoms and physician diagnosis of depression predict the occurrence and magnitude of charges for a variety of services.^27,28 In a double hurdle model, the first “hurdle,” or step, involves exploring whether there are variables that can significantly predict the occurrence of an event (such as a medical charge). The second step involves exploring whether there are variables that can predict the magnitude of the event (eg, a medical charge).

Log-transformation of charges was performed to eliminate undue influence from outliers. No logistic regression models were developed for the occurrence of primary care charges or total charges (the first hurdle) because all study patients had charges in both categories. Results are presented by hypothesis.

Results

Seventy-seven of 508 study patients (15.1%) were identified as depressed by their primary care providers in chart notes. BDI scores showed considerable spread (range, 0–31) and were significantly associated with the diagnosis of depression (P < .001). Whereas 140 patients reported BDI scores of at least 9, only 36 of these patients were diagnosed as depressed by their physicians. Similarly, 41 patients were diagnosed as depressed despite reporting low (normal) BDI scores. Patients were assigned to 1 of 4 groups: those diagnosed as depressed and having high (abnormal) BDI scores (n = 36); those diagnosed as depressed despite low BDI scores (n = 41); those not diagnosed as depressed despite high BDI scores (n = 94); and those not diagnosed as depressed and not having high BDI scores (n = 337).

Hypothesis 1: overall impact of symptoms and diagnosis on charges

Groups diagnosed with depression had significantly higher log primary care charges than did those not diagnosed (Table 1). Both groups diagnosed with depression showed the highest primary care and total medical charges. Patients diagnosed with depression and reporting high BDI scores had higher specialty charges than those not depressed. Highest laboratory costs were found for those diagnosed as depressed despite low BDI scores and those with elevated BDI scores who were not diagnosed as depressed. There were no significant differences among groups for log charges for emergency care and hospital charges.

TABLE 1
Log charges of care by diagnosis and symptoms of depression

Hypotheses 2 and 3: factors predicting occurrence and magnitude of charges

Cost models are presented as regressions in Table 2. The left side of the table presents logistic regressions exploring which variables predict whether or not a patient accrues charges in all areas except primary care and total charges. Because all patients had at least 1 primary care visit charge and, hence, a total charge, it was not possible to develop a model to predict the occurrence of those charges.

Physical health status (measured by the physical component score of the MOS SF-36) predicted the occurrence of all charges measured with the exception of laboratory tests. Advanced patient age predicted increased likelihood of charges in each area; female sex showed a trend toward predicting occurrence of emergency care charges; and education showed a trend toward predicting occurrence of laboratory charges. BDI scores (measure of symptoms of depression) and physician diagnosis of depression failed to contribute significantly to the prediction of specialty care, emergency care, laboratory testing, or hospital charges. However, there was a trend for depressive symptoms to predict the occurrence of laboratory charges.

The right side of Table 2 presents regression models that predicted the magnitude of the different categories of charges. Physical health status was a significant predictor of the magnitude of all types of charges except emergency care. Patient age contributed to prediction of size of all types of charges except emergency visits and laboratory tests. Female sex was a significant predictor of magnitude of charges in primary care, laboratory tests, and total medical charges. The diagnosis of depression was a significant predictor of magnitude of primary care (P = .0029) and total medical (P = .0158) charges. Neither depressive symptoms nor the diagnosis of depression contributed significantly to the prediction of magnitude of charges for specialty care, emergency care, laboratory testing, or hospital use, although there was a trend for depressive symptoms to predict the magnitude of laboratory costs. Although an interaction term was entered into both kinds of regression equations, there was no evidence of a significant contribution from the interaction of symptoms of depression and diagnosis of depression in any of the predictor models developed.

TABLE 2
Regression analyses predicting charges

Discussion

Medical charges were related to symptoms of depression and physician diagnosis of depression in this study. Although the patient sample was small, it was representative of the primary care population in displaying a wide range of depressive symptoms as measured by the BDI.^6,7 In this study, physician diagnosis of depression was related to self-reported depression ratings: those diagnosed as depressed had significantly higher BDI scores than did those not diagnosed as depressed. However, the relationship between self-reported symptoms and diagnosis was not perfect: 72% of patients with high BDI scores were not recognized as depressed, as often occurs in primary care.^6,7 In fact, more patients diagnosed with depression had low BDI scores (< 9, n = 41) than high BDI scores (> 8, n = 36). Clearly, other factors enter the process by which primary care physicians reach the diagnosis of depression.

Symptoms of depression and the diagnosis of depression probably influence the process of care in different ways. Differences in process of care likely would be reflected in different relationships to medical charges. Physician diagnosis of depression was associated with higher primary care and total costs and contributed to models predicting magnitude of primary care and total charges. However, neither symptoms of depression nor diagnosis of depression predicted which patients were more likely to incur charges for specialty care, emergency care, laboratory tests, or hospitalization. There was a trend only for the symptoms of depression to predict who would incur laboratory charges. These findings suggest that the relationship between depression and primary care charges and total charges is clear but less apparent when looking at less frequently occurring charges.

Other demographic factors showed fairly robust associations with the occurrence of charges. Patient age predicted who would get specialty care, emergency care, laboratory costs, and hospitalization, and there was a trend for female sex to predict occurrence of emergency department charges. Health status proved to be a significant predictor of the magnitude of all charges except those for emergency care. These powerful influences must be considered to accurately assess the impact of depression on charges.

Age also predicted the total amount of charges for primary and all medical care for the year and showed a trend toward prediction of magnitude of specialty charges. Female sex was a significant predictor of magnitude of primary care charges, laboratory charges, and total charges, and less education was a significant predictor of magnitude of emergency department and hospital charges. Some of these demographic predictors are readily explained. For example, as patients age, the number and costs of medical problems often increase. More education may enhance socioeconomic status and self-care, each of which may buffer against the need for emergency care and hospitalization. The reasons that charges are often higher for women are probably more complex. Higher utilization of primary and specialty care for women was associated with lower self-report-ed health status, less education, and lower socioeconomic status in our previous study.²⁹

These results also suggest that physician diagnosis of depression in the absence of elevated BDI scores may flag a different kind of patient presentation. Diagnosis of depression without elevated BDI scores could result from effective treatment controlling the symptoms of previously diagnosed depression, but this does not adequately explain the occurrence. Perhaps other aspects of physician–patient interaction trigger a depression diagnosis without symptoms. This group ranked highest for log-transformed charges for 5 of the 6 areas explored: only for specialty care did those with high BDI scores and diagnosis of depression rank higher in total cost. This strong association with charges implies that these patients represent diagnostic dilemmas, thereby generating more primary care visits and laboratory tests. They may be diagnosed as depressed despite their low BDI scores simply because no organic explanation can be readily identified.

BDI scores showed a trend toward predicting higher laboratory charges in our models. This finding supports the importance of depressive symptoms in influencing the process of primary care, especially laboratory testing.^15,30 Perhaps the diagnosis of depression actually slowed the ordering of laboratory tests.¹⁸ Because our data did not allow a separation of charges for laboratory tests before and after the diagnosis of depression, we did not test this possibility.

The size of this sample (N = 508) and the length of time patients were followed (1 year) might not have provided adequate power to fully test the contributions of symptoms and diagnosis of depression to the 6 sets of charges. This was likely true for hospitalization charges because hospitalization was an infrequent event in this study. Previous, larger studies found indications of increased hospitalization charges for those diagnosed as depressed¹⁷ and those with symptoms of depression.^15,30 Alternatively, the recent emphasis on decreasing hospitalizations to reduce medical costs may mean that hospitalization for depressive symptoms rather than for physical illness is less likely to occur.³¹ In addition, these observations were made by resident physicians and not by community clinicians. It is not clear whether these results would generalize to another setting, although they are consistent with community observations in previous research.

These data do suggest an intriguing interplay of the impact of physician diagnosis of depression and presence of symptoms of depression in a number of indicators of charges and utilization in primary care. Even though each element was associated with increased utilization and charges, their differential impact is unclear. Both may prove important for efforts to enhance recognition of depression; recognition of a mental health problem appeared to shift the process of care in this and previous studies.^14,32 To date, there are no data indicating that the diagnosis of depression reduces utilization or costs of primary care delivery. What is known is that physicians working in primary care are more apt to accurately diagnose those with more severe symptoms of depression than those with more transient or less severe symptoms.^16,33 Although introducing a screening device such as the BDI or the PRIME-MD⁹ likely would increase the number of patients diagnosed with depression, it is unclear what impact that would have on the process, costs, and outcomes of care. Simpler interventions such as training in communication skills such as empathy³⁴ might provide the primary care physician with all the tools needed for identification of emotional distress and mental health problems^14,30 and appropriate treatment or referral.

As US medical care has evolved, physicians have been expected to recognize and treat mental health problems in primary care,¹ “the hidden mental health network.”^2,3 Primary care clinicians are expected to observe signs of possible mental health problems, incorporate those observations into differential diagnoses, and decide which problems to treat or monitor and which to send for consultation or referral.⁴ These decisions can have important financial and health consequences, especially in dealing with depression.

Depression is common in the community⁵ and among primary care patients, 6% to 9% of whom report symptoms of major depression.^6-8 An additional 10% to 15% of primary care patients show signs of less severe but important depressive problems.^8,9 “Subclinical depression” is marked by symptoms that might indicate physical disease, signs of depression, or both; recognition may affect costs of care.^10-13

Research has begun to define the impact of depression on processes¹⁴ and costs of care.^15-17 For example, elderly patients reporting symptoms of depression have more laboratory tests performed at higher cost.¹⁵ Primary care patients diagnosed with depression had total yearly health care costs almost double those of patients without depression, with increased costs secondary to higher medical utilization and not mental health specialty treatment.¹⁶ There is evidence that depressive symptoms and the diagnosis of depression may predict increases in costs of care.¹⁷

Costs of care might be influenced by the model used by primary care physicians to identify depression.¹⁸ For example, a biomedical model might use more laboratory testing to reach a diagnosis of depression by exclusion, whereas a psychosocial model would use fewer laboratory tests while the physician pursues psychosocial issues. To identify optimal strategies for practice, it is important to determine how symptoms of depression and physician diagnosis of depression might interrelate and affect medical care costs.

We explored the following hypotheses: (1) that there are significant differences in each type of charge determined by the presence or absence of symptoms and diagnosis of depression; (2) that depressive symptoms and physician diagnosis of depression predict the occurrence of charges for specialty care, emergency services, laboratory services, and hospitalization; and (3) that depressive symptoms and physician diagnosis of depression predict the magnitude of medical charges for primary care, specialty care, emergency services, laboratory services, hospitalization, and total charges.

Methods

Study design

Five hundred eight adult nonpregnant new patients were assigned randomly to primary care providers in either family practice or general internal medicine clinics in a teaching hospital. Children younger than 18 years and pregnant women were excluded because they are not followed in general internal medicine. At enrollment and follow-up, self-reported depression was determined with the abbreviated Beck Depression Inventory (BDI)¹⁹ and health status was measured with the Medical Outcomes Studies Short Form-36 (MOS SF-36).²⁰ To avoid altering clinician practice, physicians were not provided with either score. Physicians included 105 senior residents (second and third year) in family practice and general internal medicine.

Measures

Beck Depression Inventory. The BDI is a reliable and valid instrument used to measure depressive symptoms.^19,21 The abbreviated version includes 13 items weighted and summed to produce a total score.¹⁹ A score between 9 and 15 indicates moderate depression, and a score of at least 16 indicates severe depression. The BDI is used widely for screening and to assess treatment efficacy.²² In this study, a BDI score between 0 and 8 was considered “low” or normal, and a score of at least 9 was considered “high” or indicative of symptoms of depression.

At study entrance or exit, 130 patients were identified with significant symptoms of depression (BDI > 8) by meeting criteria for moderate or severe depression¹⁹ and thus identifying roughly the top quartile of BDI scores among participants. This proportion approximates that of primary care patients estimated to experience significant depression.^6,7

Medical Outcomes Studies Short Form-36. Health status was measured with the MOS SF-36,²⁰ a 36-item self-report questionnaire. Reliability has been verified for difficult populations.²³ Summary measures can describe a physical component score and a mental component score.^24,25 The physical component score was used in this study to measure physical health status.

Medical chart review. Two physicians (K.D.B. and J.A.R.) reviewed the charts to identify notations of depression on problem lists and in visit notes to signify physician diagnosis of depression.

Charges. Charges were used as a proxy for costs. Electronic data for all health system charges were monitored from the initial visit through 1 full year of care. Six categories were monitored: primary care, specialty care, laboratory testing, emergency department, hospitalization, and total charges. Pharmacy charges were excluded because some patients purchased prescriptions outside the hospital system.

Statistical procedures

Mean log values for each area of medical charges were determined and contrasted with the Duncan multiple range test²⁶ to explore the first hypothesis that charges are associated with symptoms and diagnoses of depression. Next, a double hurdle model was used to test the hypotheses that depressive symptoms and physician diagnosis of depression predict the occurrence and magnitude of charges for a variety of services.^27,28 In a double hurdle model, the first “hurdle,” or step, involves exploring whether there are variables that can significantly predict the occurrence of an event (such as a medical charge). The second step involves exploring whether there are variables that can predict the magnitude of the event (eg, a medical charge).

Log-transformation of charges was performed to eliminate undue influence from outliers. No logistic regression models were developed for the occurrence of primary care charges or total charges (the first hurdle) because all study patients had charges in both categories. Results are presented by hypothesis.

Results

Seventy-seven of 508 study patients (15.1%) were identified as depressed by their primary care providers in chart notes. BDI scores showed considerable spread (range, 0–31) and were significantly associated with the diagnosis of depression (P < .001). Whereas 140 patients reported BDI scores of at least 9, only 36 of these patients were diagnosed as depressed by their physicians. Similarly, 41 patients were diagnosed as depressed despite reporting low (normal) BDI scores. Patients were assigned to 1 of 4 groups: those diagnosed as depressed and having high (abnormal) BDI scores (n = 36); those diagnosed as depressed despite low BDI scores (n = 41); those not diagnosed as depressed despite high BDI scores (n = 94); and those not diagnosed as depressed and not having high BDI scores (n = 337).

Hypothesis 1: overall impact of symptoms and diagnosis on charges

Groups diagnosed with depression had significantly higher log primary care charges than did those not diagnosed (Table 1). Both groups diagnosed with depression showed the highest primary care and total medical charges. Patients diagnosed with depression and reporting high BDI scores had higher specialty charges than those not depressed. Highest laboratory costs were found for those diagnosed as depressed despite low BDI scores and those with elevated BDI scores who were not diagnosed as depressed. There were no significant differences among groups for log charges for emergency care and hospital charges.

TABLE 1
Log charges of care by diagnosis and symptoms of depression

Hypotheses 2 and 3: factors predicting occurrence and magnitude of charges

Cost models are presented as regressions in Table 2. The left side of the table presents logistic regressions exploring which variables predict whether or not a patient accrues charges in all areas except primary care and total charges. Because all patients had at least 1 primary care visit charge and, hence, a total charge, it was not possible to develop a model to predict the occurrence of those charges.

Physical health status (measured by the physical component score of the MOS SF-36) predicted the occurrence of all charges measured with the exception of laboratory tests. Advanced patient age predicted increased likelihood of charges in each area; female sex showed a trend toward predicting occurrence of emergency care charges; and education showed a trend toward predicting occurrence of laboratory charges. BDI scores (measure of symptoms of depression) and physician diagnosis of depression failed to contribute significantly to the prediction of specialty care, emergency care, laboratory testing, or hospital charges. However, there was a trend for depressive symptoms to predict the occurrence of laboratory charges.

The right side of Table 2 presents regression models that predicted the magnitude of the different categories of charges. Physical health status was a significant predictor of the magnitude of all types of charges except emergency care. Patient age contributed to prediction of size of all types of charges except emergency visits and laboratory tests. Female sex was a significant predictor of magnitude of charges in primary care, laboratory tests, and total medical charges. The diagnosis of depression was a significant predictor of magnitude of primary care (P = .0029) and total medical (P = .0158) charges. Neither depressive symptoms nor the diagnosis of depression contributed significantly to the prediction of magnitude of charges for specialty care, emergency care, laboratory testing, or hospital use, although there was a trend for depressive symptoms to predict the magnitude of laboratory costs. Although an interaction term was entered into both kinds of regression equations, there was no evidence of a significant contribution from the interaction of symptoms of depression and diagnosis of depression in any of the predictor models developed.

TABLE 2
Regression analyses predicting charges

Discussion

Medical charges were related to symptoms of depression and physician diagnosis of depression in this study. Although the patient sample was small, it was representative of the primary care population in displaying a wide range of depressive symptoms as measured by the BDI.^6,7 In this study, physician diagnosis of depression was related to self-reported depression ratings: those diagnosed as depressed had significantly higher BDI scores than did those not diagnosed as depressed. However, the relationship between self-reported symptoms and diagnosis was not perfect: 72% of patients with high BDI scores were not recognized as depressed, as often occurs in primary care.^6,7 In fact, more patients diagnosed with depression had low BDI scores (< 9, n = 41) than high BDI scores (> 8, n = 36). Clearly, other factors enter the process by which primary care physicians reach the diagnosis of depression.

Symptoms of depression and the diagnosis of depression probably influence the process of care in different ways. Differences in process of care likely would be reflected in different relationships to medical charges. Physician diagnosis of depression was associated with higher primary care and total costs and contributed to models predicting magnitude of primary care and total charges. However, neither symptoms of depression nor diagnosis of depression predicted which patients were more likely to incur charges for specialty care, emergency care, laboratory tests, or hospitalization. There was a trend only for the symptoms of depression to predict who would incur laboratory charges. These findings suggest that the relationship between depression and primary care charges and total charges is clear but less apparent when looking at less frequently occurring charges.

Other demographic factors showed fairly robust associations with the occurrence of charges. Patient age predicted who would get specialty care, emergency care, laboratory costs, and hospitalization, and there was a trend for female sex to predict occurrence of emergency department charges. Health status proved to be a significant predictor of the magnitude of all charges except those for emergency care. These powerful influences must be considered to accurately assess the impact of depression on charges.

Age also predicted the total amount of charges for primary and all medical care for the year and showed a trend toward prediction of magnitude of specialty charges. Female sex was a significant predictor of magnitude of primary care charges, laboratory charges, and total charges, and less education was a significant predictor of magnitude of emergency department and hospital charges. Some of these demographic predictors are readily explained. For example, as patients age, the number and costs of medical problems often increase. More education may enhance socioeconomic status and self-care, each of which may buffer against the need for emergency care and hospitalization. The reasons that charges are often higher for women are probably more complex. Higher utilization of primary and specialty care for women was associated with lower self-report-ed health status, less education, and lower socioeconomic status in our previous study.²⁹

These results also suggest that physician diagnosis of depression in the absence of elevated BDI scores may flag a different kind of patient presentation. Diagnosis of depression without elevated BDI scores could result from effective treatment controlling the symptoms of previously diagnosed depression, but this does not adequately explain the occurrence. Perhaps other aspects of physician–patient interaction trigger a depression diagnosis without symptoms. This group ranked highest for log-transformed charges for 5 of the 6 areas explored: only for specialty care did those with high BDI scores and diagnosis of depression rank higher in total cost. This strong association with charges implies that these patients represent diagnostic dilemmas, thereby generating more primary care visits and laboratory tests. They may be diagnosed as depressed despite their low BDI scores simply because no organic explanation can be readily identified.

BDI scores showed a trend toward predicting higher laboratory charges in our models. This finding supports the importance of depressive symptoms in influencing the process of primary care, especially laboratory testing.^15,30 Perhaps the diagnosis of depression actually slowed the ordering of laboratory tests.¹⁸ Because our data did not allow a separation of charges for laboratory tests before and after the diagnosis of depression, we did not test this possibility.

The size of this sample (N = 508) and the length of time patients were followed (1 year) might not have provided adequate power to fully test the contributions of symptoms and diagnosis of depression to the 6 sets of charges. This was likely true for hospitalization charges because hospitalization was an infrequent event in this study. Previous, larger studies found indications of increased hospitalization charges for those diagnosed as depressed¹⁷ and those with symptoms of depression.^15,30 Alternatively, the recent emphasis on decreasing hospitalizations to reduce medical costs may mean that hospitalization for depressive symptoms rather than for physical illness is less likely to occur.³¹ In addition, these observations were made by resident physicians and not by community clinicians. It is not clear whether these results would generalize to another setting, although they are consistent with community observations in previous research.

These data do suggest an intriguing interplay of the impact of physician diagnosis of depression and presence of symptoms of depression in a number of indicators of charges and utilization in primary care. Even though each element was associated with increased utilization and charges, their differential impact is unclear. Both may prove important for efforts to enhance recognition of depression; recognition of a mental health problem appeared to shift the process of care in this and previous studies.^14,32 To date, there are no data indicating that the diagnosis of depression reduces utilization or costs of primary care delivery. What is known is that physicians working in primary care are more apt to accurately diagnose those with more severe symptoms of depression than those with more transient or less severe symptoms.^16,33 Although introducing a screening device such as the BDI or the PRIME-MD⁹ likely would increase the number of patients diagnosed with depression, it is unclear what impact that would have on the process, costs, and outcomes of care. Simpler interventions such as training in communication skills such as empathy³⁴ might provide the primary care physician with all the tools needed for identification of emotional distress and mental health problems^14,30 and appropriate treatment or referral.

Current medical therapies for osteoarthritis include conventional nonsteroidal anti-inflammatory drugs (NSAIDs), acetaminophen, glucosamine sulfate, and intra-articular injections of corticosteroids and hyaluronic acid. However, long-term use of corticosteroid injections can exacerbate damage to the affected joints.1,2 Conventional NSAIDs are associated with upper gastrointestinal tract ulceration and inhibition of platelet function.3

Cyclooxygenase-2 (COX-2)–specific inhibitors have demonstrated equivalent efficacy to conventional NSAIDs in treating pain and inflammation associated with osteoarthritis and rheumatoid arthritis. Further, COX-2–specific inhibitors significantly reduce the incidence of gastrointestinal ulceration and bleeding side effects caused by conventional NSAIDs.^4,5 Valdecoxib (Bextra; Pharmacia Corporation and Pfizer Corporation) is a novel COX-2–specific inhibitor that is approximately 28,000-fold more selective against COX-2 than against COX-1. As a potent COX-2–specific inhibitor, valdecoxib is expected to provide efficacy equivalent to conventional NSAIDs for treatment of arthritis and spare the COX-1–related side effects. This randomized, placebo-controlled, double-blind, 12-week study was designed to test this hypothesis by comparing the efficacy and upper gastrointestinal tract safety of valdecoxib with that of naproxen, a leading conventional NSAID comparator.

Methods

Study population

Ambulatory adults who had been diagnosed with moderate to severe osteoarthritis of the knee according to the modified criteria of the American College of Rheumatology^6,7 were eligible to participate in the trial. Patients were recruited from primary care and rheumatology specialty settings. Patients who had baseline scores of at least 40 mm on the Patient’s Assessment of Arthritis Pain–Visual Analog Scale (PAAP-VAS) and baseline categorical scores of poor to very poor on the Patient’s (PaGAA) and Physician’s (PhGAA) Global Assessments of Arthritis were included.^8,9 Any patient suffering from inflammatory arthritis, gout, pseudogout, Paget disease, or any chronic pain syndrome that might interfere with assessment of the Index Knee was excluded from the trial. Patients diagnosed with osteoarthritis of the hip ipsilateral to the Index Knee, severe anserine bursitis, acute joint trauma, or complete loss of articular cartilage on the Index Knee also were excluded. Patients were not eligible if they had active gastrointestinal disease, gastrointestinal tract ulceration 30 days before the trial, a significant bleeding disorder, or a history of gastric or duodenal surgery. Patients with an esophageal, gastric, pyloric channel, or duodenal ulcer or a score of at least 10 for esophageal, gastric, or duodenal erosions at the pretreatment endoscopy examination also were excluded.

FIGURE 1
Patient’s global assessment of arthritis

Study design

This multicenter, randomized, double-blind, placebo-controlled study compared the efficacy and upper gastrointestinal tract safety of valdecoxib at dosages of 5, 10, and 20 mg once daily with placebo and naproxen at a dosage of 500 mg twice daily in relieving moderate to severe osteoarthritis of the knee. The trial was conducted in 85 centers in the United States and Canada, in accordance with the principles of good clinical practice and the Declaration of Helsinki. Eligible patients were randomized to treatment groups and self-administered oral study medication. Patients were randomized to study treatment in the order in which they were enrolled into the study by using a treatment sequence that was determined by a Searle-prepared computer-generated randomization schedule. Patients received their allocated study medications in bottles labeled A and B according to the randomization schedule. Personnel at the study centers carried out the assessments and remained blinded throughout the study. Eligible patients were enrolled and discontinued regular pain medication. Patients discontinued their normal medications at the following specified times before the baseline endoscopy: NSAIDs (including full-dose aspirin at a dosage of ≥325 mg/day) at 48 hours, corticosteroid injections at 4 weeks, and intra-articular injections of corticosteroid or hyaluronic acid preparations at 3 and 6 months, respectively. The use of antiulcer drugs, including H2 blockers, proton pump inhibitors, misoprostol, and sucralfate, was discontinued at least 24 hours before the baseline endoscopy.

Efficacy assessments

The following arthritis assessments were made at baseline and at 2, 6, and 12 weeks or at early termination after study drug administration. PaGAA or PhGAA was measured on a 5-point categorical scale, where 1 = very good, 2 = good, 3 = fair, 4 = poor, and 5 = very poor. The PAAP-VAS was measured on a scale of 0 to 100 mm, where 0 = no pain and 100 = most severe pain. The Western Ontario and McMaster’s Universities (WOMAC) Osteoarthritis indices including Pain, Stiffness, Physical Function, and Composite were measured as described previously.¹⁰

Upper gastrointestinal assessments

Upper gastrointestinal tract endoscopy was performed within 7 days before the first study dose and at the 12-week assessment or at early termination if the patient withdrew. An endoscopy could be performed at any time if the patient experienced symptoms suggestive of an ulcer. The endoscopists performing baseline and 12-week (early termination) assessments remained blinded throughout the study.

General safety assessments

Clinical laboratory tests were performed at screening, baseline, weeks 2, 6, and 12, or at early termination, and a complete physical examination was performed at screening and final visits. The incidence of adverse events occurring in each treatment arm was monitored throughout the study. Adverse events occurring within 7 days and serious adverse events occurring within 30 days of the last study dosage of medication were included in the safety analyses.

Statistical analyses

A sample size of 200 patients per treatment group was deemed sufficient to detect a difference in ulcer rates of 5% for valdecoxib vs 16% for naproxen, with 80% power and type 1 error at .017 (adjusted for 3 primary comparisons against placebo). Homogeneity of treatment groups at baseline with respect to age, height, weight, duration of osteoarthritis, PAAP-VAS, and WOMAC Osteoarthritis Index scores was assessed with 2-way analysis of variance, with treatment group and center as factors. All other demographics and baseline characteristics were compared with the Cochran-Mantel-Haenszel (CMH) test, stratified by center.

All efficacy assessments were performed on the modified intent-to-treat (ITT) cohort by using the last observation carried forward approach. The ITT cohort comprised all patients who were randomized and had taken at least 1 dose of study medication. Analyses of mean change from baseline for PaGAA, PhGAA, PAAP-VAS, and WOMAC Osteoarthritis indices were performed by using analysis of covariance, with treatment and center as factors and the corresponding baseline score as the covariate. Pairwise comparisons of valdecoxib at dosages of 10 and 20 mg once daily vs placebo were interpreted with the Hochberg procedure.¹¹ Primary pairwise comparisons were amended in the statistical analysis plan before data unblinding to compare placebo with 10 and 20 mg valdecoxib, but not with the 5-mg dose. For all other comparisons, including 5 mg valdecoxib and naproxen vs placebo, differences were considered significant if the pairwise P values were less than .05. The incidence of withdrawal due to treatment failure was analyzed by the Fisher exact test, and the time to withdrawal in each treatment group was analyzed by log-rank test and plotted with the Kaplan-Meier product limit.^12,13

Upper gastrointestinal tract endoscopic analyses were performed on the upper gastrointestinal tract ITT population. Randomized patients were included in this cohort if they received at least 1 dose of study medication and had undergone pretreatment and posttreatment endoscopies. Overall and pairwise comparisons of gastroduodenal, gastric, and duodenal ulcers and erosions were assessed with the CMH test stratified by center. The incidence of adverse events was compared between treatment groups with the Fisher exact test. Changes in vital signs were compared between treatment groups with an analysis of covariance using pairwise treatment comparisons, with treatment group as a factor and baseline value as a covariate.

Results

Patient baseline characteristics

Of the 1019 eligible randomized patients, 1 patient randomized to 10 mg/day valdecoxib, 1 to 20 mg/day valdecoxib, and 1 to 500 mg naproxen twice daily did not take the study medication and were excluded from efficacy and safety analyses. The remaining 1016 randomized patients received study medication and were included in the ITT cohort on which analyses of all efficacy end points were based. A total of 269 patients withdrew before the end of the study due to treatment failure, preexisting protocol violations, noncompliance, or adverse signs and symptoms, or were lost to follow-up: 74 patients in the placebo group, 39 in the 5-mg valdecoxib group, 56 in the 10-mg valdecoxib group, 44 in the 20-mg valdecoxib group, and 56 in the naproxen group. The upper gastrointestinal tract ITT cohort comprised 908 patients who were included in the upper gastrointestinal tract safety analyses. More than 90% of patients included in the study evaluated their osteoarthritis as poor to very poor as assessed by baseline PaGAA scores. Treatment groups were homogeneous with respect to demographics, vital signs, medical history, and all baseline arthritis assessments (Table 1).

TABLE 1
Patient baseline characteristics

Efficacy

The least square mean change in the PaGAA was significantly improved at most assessments in response to valdecoxib (10 and 20 mg/day) and 500 mg naproxen twice daily compared with placebo (Table 2). However, the improvement in response to valdecoxib 5 mg qd did not reach statistical significance (Table 2). Significant improvements in the PhGAA were observed in response to valdecoxib and naproxen at all assessments (Table 2).

The dosages of 20 mg/day valdecoxib and 500 mg naproxen twice daily were associated with a reduction in pain, as assessed by the PAAP-VAS scores. Pain reduction associated with 5 and 10 mg/day valdecoxib was significantly better than that with placebo at all assessments except for week 12 (Table 2).

Valdecoxib and naproxen treatments improved the WOMAC Pain, Stiffness, Physical Function, and Composite indices compared with placebo at 2, 6, and 12 weeks. Valdecoxib 20 mg/day and naproxen 500 mg twice daily produced statistically significant changes in all WOMAC Osteoarthritis scores throughout the 12-week study period compared with placebo (P < .05). WOMAC Pain scores for 10 mg valdecoxib were significantly different from those for placebo at 2 weeks (P < .001) but not at 6 or 12 weeks. No significant differences were noted between any of the valdecoxib treatment doses and naproxen in terms of improvement in WOMAC indices.

The incidences of withdrawal due to treatment failure were 20% (95% confidence interval [CI], 15.3–26.8) in the placebo group; 8% (95% CI, 4.8–12.8), 12% (95% CI, 7.8–17.1), and 10% (95% CI, 6.3–15.2) in the 5-, 10-, and 20-mg/day valdecoxib groups; and 6% (95% CI, 3.6–10.9) in the 500-mg naproxen group (P < .05; Table 3). Patients in the placebo group withdrew at a significantly faster rate than those in the 4 active treatment groups (P < .05), but there were no significant differences in withdrawal rates across the 4 active treatment groups.

TABLE 2
Baseline arthritis assessments and mean changes from baseline scores

TABLE 3
Incidence of gastroduodenal, gastric, and duodenal ulcers (>5 mm) at final endoscopic evaluation

Safety

Valdecoxib and placebo had comparable upper gastrointestinal tract ulceration rates, whereas naproxen produced a significantly higher incidence of upper gastrointestinal tract ulcers than did 5 and 10 mg valdecoxib and placebo (P < .05). There were 14 adjudicated symptomatic ulcers during the study: 1 in the 5-mg valdecoxib group, 2 in the 10-mg valdecoxib group, 3 in the 20-mg valdecoxib group, and 7 in the 500-mg naproxen group.

Adverse events with an incidence of at least 5% in any treatment group and adverse events leading to withdrawal from the study are summarized by body system in Table 4. There were no significant differences in the incidence of adverse events between the valdecoxib and placebo groups. In contrast, 500 mg naproxen twice daily was associated with significantly more adverse events than 5 or 10 mg/day valdecoxib (P < .05). The incidence of adverse events was similar in the 20-mg valdecoxib and naproxen groups. Most adverse events were reported in the gastrointestinal system and consisted of abdominal pain, constipation, diarrhea, dyspepsia, flatulence, and nausea. The incidences of constipation, diarrhea, and flatulence were significantly higher in the naproxen group than in the 5-, 10-, and 20-mg valdecoxib groups, respectively. Other adverse events included accidental injury, headache, myalgia, and upper respiratory tract infections. Valdecoxib at 5 mg/day produced a significantly higher incidence of myalgia than did placebo, and valdecoxib at 20 mg/day produced a significantly lower incidence of upper respiratory tract infections than did placebo. Adverse events causing withdrawal with an incidence of at least 1% were accidental injury, abdominal pain, diarrhea, dyspepsia, nausea, abnormal hepatic function, rash, and blurred vision. The proportion of patients in the naproxen group (12.7%) who withdrew from the study was significantly greater than those for the 5-and 20-mg valdecoxib (6.0% and 5.5%) groups (P < .05), although the incidence of withdrawal due to adverse events in the 10-mg valdecoxib and naproxen groups were similar. In addition, gastrointestinal adverse events commonly related to NSAID treatment, such as dyspepsia and constipation, were more frequent in the naproxen group than in the valdecoxib and placebo groups.

TABLE 4
Adverse events

FIGURE 2
Western Ontario and McMaster’s Universities Osteoarthritis Pain Index

Discussion

This study confirmed that the novel COX-2–specific inhibitor valdecoxib at a dosage of 10 or 20 mg/day is as effective as naproxen at a dosage of 500 mg twice daily in relieving moderate to severe osteoarthritis of the knee over 12 weeks. In addition, treatment with 10 mg/day valdecoxib orally, the recommended dosage for treatment of osteoarthritis, is associated with a significantly lower gastroduodenal ulceration rate than occurs with the conventional NSAID, naproxen.

Patients receiving 10 and 20 mg/day valdecoxib experienced significant improvements in the signs and symptoms of osteoarthritis, and in all assessments the efficacies of valdecoxib 10 and 20 mg/day were numerically similar to that of naproxen. This finding is consistent with the inhibition of prostaglandin production in inflamed synovial tissue and in the central pain pathway. Increased COX-2 activity in the spinal cord in response to tissue damage and in the synovial membrane of osteoarthritis patients is at least partly responsible for joint inflammation and sensitization to inflammatory pain.^14–16 The efficacy of valdecoxib in treating moderate to severe osteoarthritis of the knee was consistent with reports of other COX-2–specific inhibitors that are comparable to conventional NSAIDs in relieving chronic pain and inflammation.^17,18 These data confirmed that 10 mg/day valdecoxib is as effective as 500 mg naproxen twice daily in treating the pain and inflammation associated with osteoarthritis. The efficacy of 10 mg/day valdecoxib makes it one of the most potent COX-2–specific inhibitors for treating moderate to severe osteoarthritis.

Conventional NSAIDs were associated with a significant risk of serious gastrointestinal complications such as ulceration and perforation and low gastrointestinal tolerability.^20–22 Naproxen treatment of osteoarthritis and rheumatoid arthritis demonstrated a higher rate of endoscopically proven gastrointestinal ulceration than did COX-2–specific inhibitors,^17,23 and that finding was confirmed in this study for 10 mg valdecoxib. Naproxen treatment was associated with significantly more gastroduodenal ulcers than 5 or 10 mg valdecoxib. We found no significant difference between 20 mg valdecoxib and naproxen, which might be explained by a lower incidence of ulcers with naproxen than reported in previous studies.²⁴ In terms of numbers needed to treat, 14 patients would be needed to observe a difference in endoscopic ulcer rates between valdecoxib (5 or 10 mg) and naproxen compared with 20 patients to observe a difference between 20 mg valdecoxib and naproxen and 16 to observe a difference in ulcer rates between naproxen and placebo.

Valdecoxib at a dosage of 10 mg/day also demonstrated overall improved gastrointestinal tolerability, with significantly fewer adverse events and withdrawals due to adverse events, in particular gastrointestinal-related events such as constipation and dyspepsia, than did naproxen. The improved upper gastrointestinal tract safety of valdecoxib was as expected because the COX-1–sparing nature of this agent allows effective inhibition of COX-2 without inhibiting COX-1 in the gastric mucosa and platelets. An improved gastrointestinal safety profile is an important consideration in the treatment of osteoarthritis because the moderate to severe gastrointestinal complications associated with conventional NSAID therapy frequently lead to poor patient compliance or discontinuation of the medication.^25,26

Overall, this study suggests clinical benefits of single daily doses of 10 and 20 mg valdecoxib and improved upper gastrointestinal tract safety for the 10-mg dose, compared with 500 mg/day naproxen. No additional efficacy benefit was obtained from a 20-mg dose as opposed to a 10-mg dose. Valdecoxib (10 mg) is a potent and effective once-daily alternative to conventional NSAIDs, with a gastrointestinal safety advantage that will be of value to rheumatologists and primary care physicians alike.

FIGURE 3
Western Ontario and Western Universities Osteoarthritis Physical Function Index

Current medical therapies for osteoarthritis include conventional nonsteroidal anti-inflammatory drugs (NSAIDs), acetaminophen, glucosamine sulfate, and intra-articular injections of corticosteroids and hyaluronic acid. However, long-term use of corticosteroid injections can exacerbate damage to the affected joints.1,2 Conventional NSAIDs are associated with upper gastrointestinal tract ulceration and inhibition of platelet function.3

Cyclooxygenase-2 (COX-2)–specific inhibitors have demonstrated equivalent efficacy to conventional NSAIDs in treating pain and inflammation associated with osteoarthritis and rheumatoid arthritis. Further, COX-2–specific inhibitors significantly reduce the incidence of gastrointestinal ulceration and bleeding side effects caused by conventional NSAIDs.^4,5 Valdecoxib (Bextra; Pharmacia Corporation and Pfizer Corporation) is a novel COX-2–specific inhibitor that is approximately 28,000-fold more selective against COX-2 than against COX-1. As a potent COX-2–specific inhibitor, valdecoxib is expected to provide efficacy equivalent to conventional NSAIDs for treatment of arthritis and spare the COX-1–related side effects. This randomized, placebo-controlled, double-blind, 12-week study was designed to test this hypothesis by comparing the efficacy and upper gastrointestinal tract safety of valdecoxib with that of naproxen, a leading conventional NSAID comparator.

Methods

Study population

Ambulatory adults who had been diagnosed with moderate to severe osteoarthritis of the knee according to the modified criteria of the American College of Rheumatology^6,7 were eligible to participate in the trial. Patients were recruited from primary care and rheumatology specialty settings. Patients who had baseline scores of at least 40 mm on the Patient’s Assessment of Arthritis Pain–Visual Analog Scale (PAAP-VAS) and baseline categorical scores of poor to very poor on the Patient’s (PaGAA) and Physician’s (PhGAA) Global Assessments of Arthritis were included.^8,9 Any patient suffering from inflammatory arthritis, gout, pseudogout, Paget disease, or any chronic pain syndrome that might interfere with assessment of the Index Knee was excluded from the trial. Patients diagnosed with osteoarthritis of the hip ipsilateral to the Index Knee, severe anserine bursitis, acute joint trauma, or complete loss of articular cartilage on the Index Knee also were excluded. Patients were not eligible if they had active gastrointestinal disease, gastrointestinal tract ulceration 30 days before the trial, a significant bleeding disorder, or a history of gastric or duodenal surgery. Patients with an esophageal, gastric, pyloric channel, or duodenal ulcer or a score of at least 10 for esophageal, gastric, or duodenal erosions at the pretreatment endoscopy examination also were excluded.

FIGURE 1
Patient’s global assessment of arthritis

Study design

This multicenter, randomized, double-blind, placebo-controlled study compared the efficacy and upper gastrointestinal tract safety of valdecoxib at dosages of 5, 10, and 20 mg once daily with placebo and naproxen at a dosage of 500 mg twice daily in relieving moderate to severe osteoarthritis of the knee. The trial was conducted in 85 centers in the United States and Canada, in accordance with the principles of good clinical practice and the Declaration of Helsinki. Eligible patients were randomized to treatment groups and self-administered oral study medication. Patients were randomized to study treatment in the order in which they were enrolled into the study by using a treatment sequence that was determined by a Searle-prepared computer-generated randomization schedule. Patients received their allocated study medications in bottles labeled A and B according to the randomization schedule. Personnel at the study centers carried out the assessments and remained blinded throughout the study. Eligible patients were enrolled and discontinued regular pain medication. Patients discontinued their normal medications at the following specified times before the baseline endoscopy: NSAIDs (including full-dose aspirin at a dosage of ≥325 mg/day) at 48 hours, corticosteroid injections at 4 weeks, and intra-articular injections of corticosteroid or hyaluronic acid preparations at 3 and 6 months, respectively. The use of antiulcer drugs, including H2 blockers, proton pump inhibitors, misoprostol, and sucralfate, was discontinued at least 24 hours before the baseline endoscopy.

Efficacy assessments

The following arthritis assessments were made at baseline and at 2, 6, and 12 weeks or at early termination after study drug administration. PaGAA or PhGAA was measured on a 5-point categorical scale, where 1 = very good, 2 = good, 3 = fair, 4 = poor, and 5 = very poor. The PAAP-VAS was measured on a scale of 0 to 100 mm, where 0 = no pain and 100 = most severe pain. The Western Ontario and McMaster’s Universities (WOMAC) Osteoarthritis indices including Pain, Stiffness, Physical Function, and Composite were measured as described previously.¹⁰

Upper gastrointestinal assessments

Upper gastrointestinal tract endoscopy was performed within 7 days before the first study dose and at the 12-week assessment or at early termination if the patient withdrew. An endoscopy could be performed at any time if the patient experienced symptoms suggestive of an ulcer. The endoscopists performing baseline and 12-week (early termination) assessments remained blinded throughout the study.

General safety assessments

Clinical laboratory tests were performed at screening, baseline, weeks 2, 6, and 12, or at early termination, and a complete physical examination was performed at screening and final visits. The incidence of adverse events occurring in each treatment arm was monitored throughout the study. Adverse events occurring within 7 days and serious adverse events occurring within 30 days of the last study dosage of medication were included in the safety analyses.

Statistical analyses

A sample size of 200 patients per treatment group was deemed sufficient to detect a difference in ulcer rates of 5% for valdecoxib vs 16% for naproxen, with 80% power and type 1 error at .017 (adjusted for 3 primary comparisons against placebo). Homogeneity of treatment groups at baseline with respect to age, height, weight, duration of osteoarthritis, PAAP-VAS, and WOMAC Osteoarthritis Index scores was assessed with 2-way analysis of variance, with treatment group and center as factors. All other demographics and baseline characteristics were compared with the Cochran-Mantel-Haenszel (CMH) test, stratified by center.

All efficacy assessments were performed on the modified intent-to-treat (ITT) cohort by using the last observation carried forward approach. The ITT cohort comprised all patients who were randomized and had taken at least 1 dose of study medication. Analyses of mean change from baseline for PaGAA, PhGAA, PAAP-VAS, and WOMAC Osteoarthritis indices were performed by using analysis of covariance, with treatment and center as factors and the corresponding baseline score as the covariate. Pairwise comparisons of valdecoxib at dosages of 10 and 20 mg once daily vs placebo were interpreted with the Hochberg procedure.¹¹ Primary pairwise comparisons were amended in the statistical analysis plan before data unblinding to compare placebo with 10 and 20 mg valdecoxib, but not with the 5-mg dose. For all other comparisons, including 5 mg valdecoxib and naproxen vs placebo, differences were considered significant if the pairwise P values were less than .05. The incidence of withdrawal due to treatment failure was analyzed by the Fisher exact test, and the time to withdrawal in each treatment group was analyzed by log-rank test and plotted with the Kaplan-Meier product limit.^12,13

Upper gastrointestinal tract endoscopic analyses were performed on the upper gastrointestinal tract ITT population. Randomized patients were included in this cohort if they received at least 1 dose of study medication and had undergone pretreatment and posttreatment endoscopies. Overall and pairwise comparisons of gastroduodenal, gastric, and duodenal ulcers and erosions were assessed with the CMH test stratified by center. The incidence of adverse events was compared between treatment groups with the Fisher exact test. Changes in vital signs were compared between treatment groups with an analysis of covariance using pairwise treatment comparisons, with treatment group as a factor and baseline value as a covariate.

Results

Patient baseline characteristics

Of the 1019 eligible randomized patients, 1 patient randomized to 10 mg/day valdecoxib, 1 to 20 mg/day valdecoxib, and 1 to 500 mg naproxen twice daily did not take the study medication and were excluded from efficacy and safety analyses. The remaining 1016 randomized patients received study medication and were included in the ITT cohort on which analyses of all efficacy end points were based. A total of 269 patients withdrew before the end of the study due to treatment failure, preexisting protocol violations, noncompliance, or adverse signs and symptoms, or were lost to follow-up: 74 patients in the placebo group, 39 in the 5-mg valdecoxib group, 56 in the 10-mg valdecoxib group, 44 in the 20-mg valdecoxib group, and 56 in the naproxen group. The upper gastrointestinal tract ITT cohort comprised 908 patients who were included in the upper gastrointestinal tract safety analyses. More than 90% of patients included in the study evaluated their osteoarthritis as poor to very poor as assessed by baseline PaGAA scores. Treatment groups were homogeneous with respect to demographics, vital signs, medical history, and all baseline arthritis assessments (Table 1).

TABLE 1
Patient baseline characteristics

Efficacy

The least square mean change in the PaGAA was significantly improved at most assessments in response to valdecoxib (10 and 20 mg/day) and 500 mg naproxen twice daily compared with placebo (Table 2). However, the improvement in response to valdecoxib 5 mg qd did not reach statistical significance (Table 2). Significant improvements in the PhGAA were observed in response to valdecoxib and naproxen at all assessments (Table 2).

The dosages of 20 mg/day valdecoxib and 500 mg naproxen twice daily were associated with a reduction in pain, as assessed by the PAAP-VAS scores. Pain reduction associated with 5 and 10 mg/day valdecoxib was significantly better than that with placebo at all assessments except for week 12 (Table 2).

Valdecoxib and naproxen treatments improved the WOMAC Pain, Stiffness, Physical Function, and Composite indices compared with placebo at 2, 6, and 12 weeks. Valdecoxib 20 mg/day and naproxen 500 mg twice daily produced statistically significant changes in all WOMAC Osteoarthritis scores throughout the 12-week study period compared with placebo (P < .05). WOMAC Pain scores for 10 mg valdecoxib were significantly different from those for placebo at 2 weeks (P < .001) but not at 6 or 12 weeks. No significant differences were noted between any of the valdecoxib treatment doses and naproxen in terms of improvement in WOMAC indices.

The incidences of withdrawal due to treatment failure were 20% (95% confidence interval [CI], 15.3–26.8) in the placebo group; 8% (95% CI, 4.8–12.8), 12% (95% CI, 7.8–17.1), and 10% (95% CI, 6.3–15.2) in the 5-, 10-, and 20-mg/day valdecoxib groups; and 6% (95% CI, 3.6–10.9) in the 500-mg naproxen group (P < .05; Table 3). Patients in the placebo group withdrew at a significantly faster rate than those in the 4 active treatment groups (P < .05), but there were no significant differences in withdrawal rates across the 4 active treatment groups.

TABLE 2
Baseline arthritis assessments and mean changes from baseline scores

TABLE 3
Incidence of gastroduodenal, gastric, and duodenal ulcers (>5 mm) at final endoscopic evaluation

Safety

Valdecoxib and placebo had comparable upper gastrointestinal tract ulceration rates, whereas naproxen produced a significantly higher incidence of upper gastrointestinal tract ulcers than did 5 and 10 mg valdecoxib and placebo (P < .05). There were 14 adjudicated symptomatic ulcers during the study: 1 in the 5-mg valdecoxib group, 2 in the 10-mg valdecoxib group, 3 in the 20-mg valdecoxib group, and 7 in the 500-mg naproxen group.

Adverse events with an incidence of at least 5% in any treatment group and adverse events leading to withdrawal from the study are summarized by body system in Table 4. There were no significant differences in the incidence of adverse events between the valdecoxib and placebo groups. In contrast, 500 mg naproxen twice daily was associated with significantly more adverse events than 5 or 10 mg/day valdecoxib (P < .05). The incidence of adverse events was similar in the 20-mg valdecoxib and naproxen groups. Most adverse events were reported in the gastrointestinal system and consisted of abdominal pain, constipation, diarrhea, dyspepsia, flatulence, and nausea. The incidences of constipation, diarrhea, and flatulence were significantly higher in the naproxen group than in the 5-, 10-, and 20-mg valdecoxib groups, respectively. Other adverse events included accidental injury, headache, myalgia, and upper respiratory tract infections. Valdecoxib at 5 mg/day produced a significantly higher incidence of myalgia than did placebo, and valdecoxib at 20 mg/day produced a significantly lower incidence of upper respiratory tract infections than did placebo. Adverse events causing withdrawal with an incidence of at least 1% were accidental injury, abdominal pain, diarrhea, dyspepsia, nausea, abnormal hepatic function, rash, and blurred vision. The proportion of patients in the naproxen group (12.7%) who withdrew from the study was significantly greater than those for the 5-and 20-mg valdecoxib (6.0% and 5.5%) groups (P < .05), although the incidence of withdrawal due to adverse events in the 10-mg valdecoxib and naproxen groups were similar. In addition, gastrointestinal adverse events commonly related to NSAID treatment, such as dyspepsia and constipation, were more frequent in the naproxen group than in the valdecoxib and placebo groups.

TABLE 4
Adverse events

FIGURE 2
Western Ontario and McMaster’s Universities Osteoarthritis Pain Index

Discussion

This study confirmed that the novel COX-2–specific inhibitor valdecoxib at a dosage of 10 or 20 mg/day is as effective as naproxen at a dosage of 500 mg twice daily in relieving moderate to severe osteoarthritis of the knee over 12 weeks. In addition, treatment with 10 mg/day valdecoxib orally, the recommended dosage for treatment of osteoarthritis, is associated with a significantly lower gastroduodenal ulceration rate than occurs with the conventional NSAID, naproxen.

Patients receiving 10 and 20 mg/day valdecoxib experienced significant improvements in the signs and symptoms of osteoarthritis, and in all assessments the efficacies of valdecoxib 10 and 20 mg/day were numerically similar to that of naproxen. This finding is consistent with the inhibition of prostaglandin production in inflamed synovial tissue and in the central pain pathway. Increased COX-2 activity in the spinal cord in response to tissue damage and in the synovial membrane of osteoarthritis patients is at least partly responsible for joint inflammation and sensitization to inflammatory pain.^14–16 The efficacy of valdecoxib in treating moderate to severe osteoarthritis of the knee was consistent with reports of other COX-2–specific inhibitors that are comparable to conventional NSAIDs in relieving chronic pain and inflammation.^17,18 These data confirmed that 10 mg/day valdecoxib is as effective as 500 mg naproxen twice daily in treating the pain and inflammation associated with osteoarthritis. The efficacy of 10 mg/day valdecoxib makes it one of the most potent COX-2–specific inhibitors for treating moderate to severe osteoarthritis.

Conventional NSAIDs were associated with a significant risk of serious gastrointestinal complications such as ulceration and perforation and low gastrointestinal tolerability.^20–22 Naproxen treatment of osteoarthritis and rheumatoid arthritis demonstrated a higher rate of endoscopically proven gastrointestinal ulceration than did COX-2–specific inhibitors,^17,23 and that finding was confirmed in this study for 10 mg valdecoxib. Naproxen treatment was associated with significantly more gastroduodenal ulcers than 5 or 10 mg valdecoxib. We found no significant difference between 20 mg valdecoxib and naproxen, which might be explained by a lower incidence of ulcers with naproxen than reported in previous studies.²⁴ In terms of numbers needed to treat, 14 patients would be needed to observe a difference in endoscopic ulcer rates between valdecoxib (5 or 10 mg) and naproxen compared with 20 patients to observe a difference between 20 mg valdecoxib and naproxen and 16 to observe a difference in ulcer rates between naproxen and placebo.

Valdecoxib at a dosage of 10 mg/day also demonstrated overall improved gastrointestinal tolerability, with significantly fewer adverse events and withdrawals due to adverse events, in particular gastrointestinal-related events such as constipation and dyspepsia, than did naproxen. The improved upper gastrointestinal tract safety of valdecoxib was as expected because the COX-1–sparing nature of this agent allows effective inhibition of COX-2 without inhibiting COX-1 in the gastric mucosa and platelets. An improved gastrointestinal safety profile is an important consideration in the treatment of osteoarthritis because the moderate to severe gastrointestinal complications associated with conventional NSAID therapy frequently lead to poor patient compliance or discontinuation of the medication.^25,26

Overall, this study suggests clinical benefits of single daily doses of 10 and 20 mg valdecoxib and improved upper gastrointestinal tract safety for the 10-mg dose, compared with 500 mg/day naproxen. No additional efficacy benefit was obtained from a 20-mg dose as opposed to a 10-mg dose. Valdecoxib (10 mg) is a potent and effective once-daily alternative to conventional NSAIDs, with a gastrointestinal safety advantage that will be of value to rheumatologists and primary care physicians alike.

FIGURE 3
Western Ontario and Western Universities Osteoarthritis Physical Function Index

Current medical therapies for osteoarthritis include conventional nonsteroidal anti-inflammatory drugs (NSAIDs), acetaminophen, glucosamine sulfate, and intra-articular injections of corticosteroids and hyaluronic acid. However, long-term use of corticosteroid injections can exacerbate damage to the affected joints.1,2 Conventional NSAIDs are associated with upper gastrointestinal tract ulceration and inhibition of platelet function.3

Cyclooxygenase-2 (COX-2)–specific inhibitors have demonstrated equivalent efficacy to conventional NSAIDs in treating pain and inflammation associated with osteoarthritis and rheumatoid arthritis. Further, COX-2–specific inhibitors significantly reduce the incidence of gastrointestinal ulceration and bleeding side effects caused by conventional NSAIDs.^4,5 Valdecoxib (Bextra; Pharmacia Corporation and Pfizer Corporation) is a novel COX-2–specific inhibitor that is approximately 28,000-fold more selective against COX-2 than against COX-1. As a potent COX-2–specific inhibitor, valdecoxib is expected to provide efficacy equivalent to conventional NSAIDs for treatment of arthritis and spare the COX-1–related side effects. This randomized, placebo-controlled, double-blind, 12-week study was designed to test this hypothesis by comparing the efficacy and upper gastrointestinal tract safety of valdecoxib with that of naproxen, a leading conventional NSAID comparator.

Methods

Study population

Ambulatory adults who had been diagnosed with moderate to severe osteoarthritis of the knee according to the modified criteria of the American College of Rheumatology^6,7 were eligible to participate in the trial. Patients were recruited from primary care and rheumatology specialty settings. Patients who had baseline scores of at least 40 mm on the Patient’s Assessment of Arthritis Pain–Visual Analog Scale (PAAP-VAS) and baseline categorical scores of poor to very poor on the Patient’s (PaGAA) and Physician’s (PhGAA) Global Assessments of Arthritis were included.^8,9 Any patient suffering from inflammatory arthritis, gout, pseudogout, Paget disease, or any chronic pain syndrome that might interfere with assessment of the Index Knee was excluded from the trial. Patients diagnosed with osteoarthritis of the hip ipsilateral to the Index Knee, severe anserine bursitis, acute joint trauma, or complete loss of articular cartilage on the Index Knee also were excluded. Patients were not eligible if they had active gastrointestinal disease, gastrointestinal tract ulceration 30 days before the trial, a significant bleeding disorder, or a history of gastric or duodenal surgery. Patients with an esophageal, gastric, pyloric channel, or duodenal ulcer or a score of at least 10 for esophageal, gastric, or duodenal erosions at the pretreatment endoscopy examination also were excluded.

FIGURE 1
Patient’s global assessment of arthritis

Study design

This multicenter, randomized, double-blind, placebo-controlled study compared the efficacy and upper gastrointestinal tract safety of valdecoxib at dosages of 5, 10, and 20 mg once daily with placebo and naproxen at a dosage of 500 mg twice daily in relieving moderate to severe osteoarthritis of the knee. The trial was conducted in 85 centers in the United States and Canada, in accordance with the principles of good clinical practice and the Declaration of Helsinki. Eligible patients were randomized to treatment groups and self-administered oral study medication. Patients were randomized to study treatment in the order in which they were enrolled into the study by using a treatment sequence that was determined by a Searle-prepared computer-generated randomization schedule. Patients received their allocated study medications in bottles labeled A and B according to the randomization schedule. Personnel at the study centers carried out the assessments and remained blinded throughout the study. Eligible patients were enrolled and discontinued regular pain medication. Patients discontinued their normal medications at the following specified times before the baseline endoscopy: NSAIDs (including full-dose aspirin at a dosage of ≥325 mg/day) at 48 hours, corticosteroid injections at 4 weeks, and intra-articular injections of corticosteroid or hyaluronic acid preparations at 3 and 6 months, respectively. The use of antiulcer drugs, including H2 blockers, proton pump inhibitors, misoprostol, and sucralfate, was discontinued at least 24 hours before the baseline endoscopy.

Efficacy assessments

The following arthritis assessments were made at baseline and at 2, 6, and 12 weeks or at early termination after study drug administration. PaGAA or PhGAA was measured on a 5-point categorical scale, where 1 = very good, 2 = good, 3 = fair, 4 = poor, and 5 = very poor. The PAAP-VAS was measured on a scale of 0 to 100 mm, where 0 = no pain and 100 = most severe pain. The Western Ontario and McMaster’s Universities (WOMAC) Osteoarthritis indices including Pain, Stiffness, Physical Function, and Composite were measured as described previously.¹⁰

Upper gastrointestinal assessments

Upper gastrointestinal tract endoscopy was performed within 7 days before the first study dose and at the 12-week assessment or at early termination if the patient withdrew. An endoscopy could be performed at any time if the patient experienced symptoms suggestive of an ulcer. The endoscopists performing baseline and 12-week (early termination) assessments remained blinded throughout the study.

General safety assessments

Clinical laboratory tests were performed at screening, baseline, weeks 2, 6, and 12, or at early termination, and a complete physical examination was performed at screening and final visits. The incidence of adverse events occurring in each treatment arm was monitored throughout the study. Adverse events occurring within 7 days and serious adverse events occurring within 30 days of the last study dosage of medication were included in the safety analyses.

Statistical analyses

A sample size of 200 patients per treatment group was deemed sufficient to detect a difference in ulcer rates of 5% for valdecoxib vs 16% for naproxen, with 80% power and type 1 error at .017 (adjusted for 3 primary comparisons against placebo). Homogeneity of treatment groups at baseline with respect to age, height, weight, duration of osteoarthritis, PAAP-VAS, and WOMAC Osteoarthritis Index scores was assessed with 2-way analysis of variance, with treatment group and center as factors. All other demographics and baseline characteristics were compared with the Cochran-Mantel-Haenszel (CMH) test, stratified by center.

All efficacy assessments were performed on the modified intent-to-treat (ITT) cohort by using the last observation carried forward approach. The ITT cohort comprised all patients who were randomized and had taken at least 1 dose of study medication. Analyses of mean change from baseline for PaGAA, PhGAA, PAAP-VAS, and WOMAC Osteoarthritis indices were performed by using analysis of covariance, with treatment and center as factors and the corresponding baseline score as the covariate. Pairwise comparisons of valdecoxib at dosages of 10 and 20 mg once daily vs placebo were interpreted with the Hochberg procedure.¹¹ Primary pairwise comparisons were amended in the statistical analysis plan before data unblinding to compare placebo with 10 and 20 mg valdecoxib, but not with the 5-mg dose. For all other comparisons, including 5 mg valdecoxib and naproxen vs placebo, differences were considered significant if the pairwise P values were less than .05. The incidence of withdrawal due to treatment failure was analyzed by the Fisher exact test, and the time to withdrawal in each treatment group was analyzed by log-rank test and plotted with the Kaplan-Meier product limit.^12,13

Upper gastrointestinal tract endoscopic analyses were performed on the upper gastrointestinal tract ITT population. Randomized patients were included in this cohort if they received at least 1 dose of study medication and had undergone pretreatment and posttreatment endoscopies. Overall and pairwise comparisons of gastroduodenal, gastric, and duodenal ulcers and erosions were assessed with the CMH test stratified by center. The incidence of adverse events was compared between treatment groups with the Fisher exact test. Changes in vital signs were compared between treatment groups with an analysis of covariance using pairwise treatment comparisons, with treatment group as a factor and baseline value as a covariate.

Results

Patient baseline characteristics

Of the 1019 eligible randomized patients, 1 patient randomized to 10 mg/day valdecoxib, 1 to 20 mg/day valdecoxib, and 1 to 500 mg naproxen twice daily did not take the study medication and were excluded from efficacy and safety analyses. The remaining 1016 randomized patients received study medication and were included in the ITT cohort on which analyses of all efficacy end points were based. A total of 269 patients withdrew before the end of the study due to treatment failure, preexisting protocol violations, noncompliance, or adverse signs and symptoms, or were lost to follow-up: 74 patients in the placebo group, 39 in the 5-mg valdecoxib group, 56 in the 10-mg valdecoxib group, 44 in the 20-mg valdecoxib group, and 56 in the naproxen group. The upper gastrointestinal tract ITT cohort comprised 908 patients who were included in the upper gastrointestinal tract safety analyses. More than 90% of patients included in the study evaluated their osteoarthritis as poor to very poor as assessed by baseline PaGAA scores. Treatment groups were homogeneous with respect to demographics, vital signs, medical history, and all baseline arthritis assessments (Table 1).

TABLE 1
Patient baseline characteristics

Efficacy

The least square mean change in the PaGAA was significantly improved at most assessments in response to valdecoxib (10 and 20 mg/day) and 500 mg naproxen twice daily compared with placebo (Table 2). However, the improvement in response to valdecoxib 5 mg qd did not reach statistical significance (Table 2). Significant improvements in the PhGAA were observed in response to valdecoxib and naproxen at all assessments (Table 2).

The dosages of 20 mg/day valdecoxib and 500 mg naproxen twice daily were associated with a reduction in pain, as assessed by the PAAP-VAS scores. Pain reduction associated with 5 and 10 mg/day valdecoxib was significantly better than that with placebo at all assessments except for week 12 (Table 2).

Valdecoxib and naproxen treatments improved the WOMAC Pain, Stiffness, Physical Function, and Composite indices compared with placebo at 2, 6, and 12 weeks. Valdecoxib 20 mg/day and naproxen 500 mg twice daily produced statistically significant changes in all WOMAC Osteoarthritis scores throughout the 12-week study period compared with placebo (P < .05). WOMAC Pain scores for 10 mg valdecoxib were significantly different from those for placebo at 2 weeks (P < .001) but not at 6 or 12 weeks. No significant differences were noted between any of the valdecoxib treatment doses and naproxen in terms of improvement in WOMAC indices.

The incidences of withdrawal due to treatment failure were 20% (95% confidence interval [CI], 15.3–26.8) in the placebo group; 8% (95% CI, 4.8–12.8), 12% (95% CI, 7.8–17.1), and 10% (95% CI, 6.3–15.2) in the 5-, 10-, and 20-mg/day valdecoxib groups; and 6% (95% CI, 3.6–10.9) in the 500-mg naproxen group (P < .05; Table 3). Patients in the placebo group withdrew at a significantly faster rate than those in the 4 active treatment groups (P < .05), but there were no significant differences in withdrawal rates across the 4 active treatment groups.

TABLE 2
Baseline arthritis assessments and mean changes from baseline scores

TABLE 3
Incidence of gastroduodenal, gastric, and duodenal ulcers (>5 mm) at final endoscopic evaluation

Safety

Valdecoxib and placebo had comparable upper gastrointestinal tract ulceration rates, whereas naproxen produced a significantly higher incidence of upper gastrointestinal tract ulcers than did 5 and 10 mg valdecoxib and placebo (P < .05). There were 14 adjudicated symptomatic ulcers during the study: 1 in the 5-mg valdecoxib group, 2 in the 10-mg valdecoxib group, 3 in the 20-mg valdecoxib group, and 7 in the 500-mg naproxen group.

Adverse events with an incidence of at least 5% in any treatment group and adverse events leading to withdrawal from the study are summarized by body system in Table 4. There were no significant differences in the incidence of adverse events between the valdecoxib and placebo groups. In contrast, 500 mg naproxen twice daily was associated with significantly more adverse events than 5 or 10 mg/day valdecoxib (P < .05). The incidence of adverse events was similar in the 20-mg valdecoxib and naproxen groups. Most adverse events were reported in the gastrointestinal system and consisted of abdominal pain, constipation, diarrhea, dyspepsia, flatulence, and nausea. The incidences of constipation, diarrhea, and flatulence were significantly higher in the naproxen group than in the 5-, 10-, and 20-mg valdecoxib groups, respectively. Other adverse events included accidental injury, headache, myalgia, and upper respiratory tract infections. Valdecoxib at 5 mg/day produced a significantly higher incidence of myalgia than did placebo, and valdecoxib at 20 mg/day produced a significantly lower incidence of upper respiratory tract infections than did placebo. Adverse events causing withdrawal with an incidence of at least 1% were accidental injury, abdominal pain, diarrhea, dyspepsia, nausea, abnormal hepatic function, rash, and blurred vision. The proportion of patients in the naproxen group (12.7%) who withdrew from the study was significantly greater than those for the 5-and 20-mg valdecoxib (6.0% and 5.5%) groups (P < .05), although the incidence of withdrawal due to adverse events in the 10-mg valdecoxib and naproxen groups were similar. In addition, gastrointestinal adverse events commonly related to NSAID treatment, such as dyspepsia and constipation, were more frequent in the naproxen group than in the valdecoxib and placebo groups.

TABLE 4
Adverse events

FIGURE 2
Western Ontario and McMaster’s Universities Osteoarthritis Pain Index

Discussion

This study confirmed that the novel COX-2–specific inhibitor valdecoxib at a dosage of 10 or 20 mg/day is as effective as naproxen at a dosage of 500 mg twice daily in relieving moderate to severe osteoarthritis of the knee over 12 weeks. In addition, treatment with 10 mg/day valdecoxib orally, the recommended dosage for treatment of osteoarthritis, is associated with a significantly lower gastroduodenal ulceration rate than occurs with the conventional NSAID, naproxen.

Patients receiving 10 and 20 mg/day valdecoxib experienced significant improvements in the signs and symptoms of osteoarthritis, and in all assessments the efficacies of valdecoxib 10 and 20 mg/day were numerically similar to that of naproxen. This finding is consistent with the inhibition of prostaglandin production in inflamed synovial tissue and in the central pain pathway. Increased COX-2 activity in the spinal cord in response to tissue damage and in the synovial membrane of osteoarthritis patients is at least partly responsible for joint inflammation and sensitization to inflammatory pain.^14–16 The efficacy of valdecoxib in treating moderate to severe osteoarthritis of the knee was consistent with reports of other COX-2–specific inhibitors that are comparable to conventional NSAIDs in relieving chronic pain and inflammation.^17,18 These data confirmed that 10 mg/day valdecoxib is as effective as 500 mg naproxen twice daily in treating the pain and inflammation associated with osteoarthritis. The efficacy of 10 mg/day valdecoxib makes it one of the most potent COX-2–specific inhibitors for treating moderate to severe osteoarthritis.

Conventional NSAIDs were associated with a significant risk of serious gastrointestinal complications such as ulceration and perforation and low gastrointestinal tolerability.^20–22 Naproxen treatment of osteoarthritis and rheumatoid arthritis demonstrated a higher rate of endoscopically proven gastrointestinal ulceration than did COX-2–specific inhibitors,^17,23 and that finding was confirmed in this study for 10 mg valdecoxib. Naproxen treatment was associated with significantly more gastroduodenal ulcers than 5 or 10 mg valdecoxib. We found no significant difference between 20 mg valdecoxib and naproxen, which might be explained by a lower incidence of ulcers with naproxen than reported in previous studies.²⁴ In terms of numbers needed to treat, 14 patients would be needed to observe a difference in endoscopic ulcer rates between valdecoxib (5 or 10 mg) and naproxen compared with 20 patients to observe a difference between 20 mg valdecoxib and naproxen and 16 to observe a difference in ulcer rates between naproxen and placebo.

Valdecoxib at a dosage of 10 mg/day also demonstrated overall improved gastrointestinal tolerability, with significantly fewer adverse events and withdrawals due to adverse events, in particular gastrointestinal-related events such as constipation and dyspepsia, than did naproxen. The improved upper gastrointestinal tract safety of valdecoxib was as expected because the COX-1–sparing nature of this agent allows effective inhibition of COX-2 without inhibiting COX-1 in the gastric mucosa and platelets. An improved gastrointestinal safety profile is an important consideration in the treatment of osteoarthritis because the moderate to severe gastrointestinal complications associated with conventional NSAID therapy frequently lead to poor patient compliance or discontinuation of the medication.^25,26

Overall, this study suggests clinical benefits of single daily doses of 10 and 20 mg valdecoxib and improved upper gastrointestinal tract safety for the 10-mg dose, compared with 500 mg/day naproxen. No additional efficacy benefit was obtained from a 20-mg dose as opposed to a 10-mg dose. Valdecoxib (10 mg) is a potent and effective once-daily alternative to conventional NSAIDs, with a gastrointestinal safety advantage that will be of value to rheumatologists and primary care physicians alike.

FIGURE 3
Western Ontario and Western Universities Osteoarthritis Physical Function Index

Cryotherapy is an accepted procedure for treating low-grade cervical dysplasia.^1,2 Only minor modifications of the precise technique of cryotherapy application have occurred since its inception. Currently the double-freeze technique of cryotherapy is an accepted treatment for mild and focal moderate dysplasia of the uterine cervix.³ Cervical cryotherapy is used widely not only because of its proven efficacy but also because of its ease of use in the outpatient setting and lack of known significant side effects. The procedure can be performed in the office setting without the use of local or general anesthesia, making it superior to the more invasive procedures performed before the availability of cryotherapy (eg, cervical conization and hysterectomy).

There has been limited investigation of the effects of cryotherapy on the anatomy of the uterine cervix. Whereas one study showed that cryotherapy has no effect on subsequent fertility or pregnancy outcome,⁴ another in adolescents reported cervical stenosis and pelvic inflammatory disease as possible treatment side effects.⁵ In addition, a study published in 1984 by Jobson and Homesley reported higher rates of retraction of the proximal squamocolumnar junction into the endocervical canal in patients undergoing cryotherapy compared with patients undergoing carbon dioxide laser ablation⁶; 47% of the follow-up colposcopic examinations were inadequate in that study population. Adequacy of colposcopic examination is defined as complete visualization of the transformation zone, visualization of the entire lesion, if present, and correlation between cytologic and histologic findings and the colposcopist’s impression.⁷ Failure to meet any one of these criteria leads to an inadequate colposcopic examination requiring further, more invasive evaluation. This study compared the rate of adequate and inadequate colposcopic examinations in women with and without a history of cryotherapy. Other factors found to influence the adequacy of follow-up colposcopy also are described.

Methods

We performed a retrospective cohort study using data collected from 7 of 14 state-run public health clinics. These 7 sites included rural and urban clinics. All women undergoing at least 2 colposcopic examinations in these clinics between January 1, 1991, and December 1, 1995, were included. Women underwent initial colposcopic examination after an abnormality was noted on a screening Pap test. Only women who had both colposcopic examinations in the same clinic were included. Care provided in these clinics included Pap test screening, colposcopic examinations, and treatment of identified cervical dysplasia with cervical cryotherapy, conization, and loop excision of the ectocervical portion (LEEP). State-contracted physicians trained in obstetrics and gynecology followed women who attended these clinics.

Chart review was used to determine the adequacy of the initial examination, whether an intervening procedure was done, and the adequacy of follow-up colposcopic examination. Adequacy was documented by the physician performing the colposcopic examination with the use of a standard form consistent among clinics. The accepted criteria for adequacy were used, and each colposcopic examination was documented as adequate or inadequate based on the colposcopist’s findings. Charts were reviewed and data were abstracted by 3 reviewers. Cervical biopsy results, presence of human papilloma virus (HPV) noted on routine cytology, endocervical curettage (ECC) results, and routine demographic data also were recorded. The management and therapeutic protocols were consistent across the 7 clinics.

Women were excluded from the analysis if (1) they had cryotherapy performed before their initial colposcopic examination (n = 1), (2) the date of the initial colposcopic examination was not available (n = 36), (3) information confirming the type of treatment used between colposcopic examinations was unknown (n = 32), (4) initial colposcopic examination was inadequate (n = 16), or (5) the adequacy of the follow-up colposcopic examination was not documented (n = 6). The total number of women excluded was 65 because some women met multiple exclusion criteria.

The management after initial colposcopic examination was done according to whether the women had cryotherapy, cone, LEEP, or no procedure between initial and follow-up colposcopic examinations. Univariate analysis of the association between the management group with clinic of treatment, performance of ECC, biopsy results, presence of HPV, and cytologic presence of glandular atypia was performed. Mean age and duration (interval between initial and follow-up colposcopic examinations or between the procedure and follow-up examination) were calculated for all groups.

The odds ratio of an inadequate follow-up colposcopic examination was estimated for type of treatment (cryotherapy, cone/LEEP) compared with no treatment, age, clinic where treatment was provided, performance of ECC, biopsy results, presence of HPV, and presence of glandular atypia. The 95% confidence intervals about the relative odds estimates were calculated. Mean age and duration between initial colposcopy and follow-up colposcopy were calculated for the groups with adequate and inadequate follow-up colposcopic examinations.

Multivariable logistic regression analysis was used to evaluate the association of adequacy of follow-up colposcopic examination with age (years), clinic where colposcopic examination was performed, duration (days), whether or not ECC was performed, biopsy results from the initial colposcopic examination, presence of HPV, and presence of glandular atypia noted on initial colposcopy. Biopsy results were categorized as normal or abnormal in the model that is reported. The stepwise backward elimination technique was used to evaluate the best model. The 95% confidence intervals about the adjusted odds ratio were calculated.

The Pearson chi-square test was used to test the significance of the association between binary variables. The significance of the difference between means was tested with the one-way analysis of variance. Data were analyzed with the personal computer version of the Statistical Package for the Social Sciences (SPSS/PC+ version 8.0).

Results

Between January 1, 1991, and December 31, 1995, 3225 women underwent colposcopic evaluation or treatment at 7 county colposcopy clinics in Oklahoma. Two hundred sixty-eight of these women underwent 2 examinations during the study period. There were 203 of 268 subjects available for analysis after exclusions for missing data. Eighty-three patients (41.1%) had cryotherapy, 24 (11.9%) underwent a cone biopsy or a LEEP procedure, and 96 (47.5%) underwent no procedure between initial and follow-up colposcopic examinations.

Table 1 shows characteristics of women who had cryotherapy, cone/LEEP, and no procedure. The groups were similar with respect to age, whether ECC was performed, presence of HPV, and whether glandular involvement was noted. There was an association between the degree of cervical dysplasia and the three treatment groups, which was expected because degree of dysplasia determines treatment modality. Women who had cryotherapy had follow-up colposcopy (mean = 565 days) later than women who had cone or LEEP (mean = 319 days) or no procedure (mean = 339 days; P < .0001).

Thirty-three percent (n = 67) had inadequate follow-up colposcopic examinations. These included a large proportion of women, 61.4%, who had cryotherapy (51/83) compared with 20.8% (5/24) of women who had cone or LEEP and 11.5% (11/96) of women who had no procedure.

Table 2 shows the relationship between inadequate second colposcopy and previous cryotherapy, cone/LEEP, abnormal cervical biopsy, ECC, presence of HPV, and presence of glandular atypia. Patients who had cryotherapy had an increased likelihood of inadequate follow-up compared with patients who had no procedure (adjusted odds ratio =18.67, 95% confidence interval = 6.99–49.81). Cone/LEEP increased the likelihood of inadequate follow-up but was not statistically significant. Age, duration, ECC, presence of HPV, or presence of glandular atypia did not increase the likelihood of subsequent inadequate colposcopic examination. Odds ratio estimates for different clinics are not reported but were imprecise due to small numbers.

TABLE 1
Characteristics of patients with and without cryotherapy between initial and follow-up colposcopy

TABLE 2
Likelihood of inadequate follow-up colposcopic examination*

Discussion

Undergoing cryotherapy of the uterine cervix increases the risk that a follow-up colposcopic examination will be inadequate. This agrees with the findings of Jobson and Homesley’s 1984 study,⁶ which looked at the efficacy of cryotherapy vs carbon dioxide laser ablation in the treatment of cervical dysplasia. Although it was not the focus of their study, a high rate of inadequacy was noted on follow-up colposcopic examinations after cryotherapy.

Because of the retrospective design of this study, we could not randomly assign women to a treatment group. However, the study groups were similar with respect to other variables potentially associated with the outcome measure. In addition, we attempted to control confounding variables by using multivariable analysis. By including the clinic where the examination was performed, we attempted to limit the effect of the subjective assignment of adequacy by the physician. This is a limitation of this study.

We found an association between the clinics where the follow-up colposcopist’s examinations were performed and whether a follow-up examination was adequate or inadequate. The determination of adequacy depends on the physician’s observations during the colposcopic examination. We were unable to measure the intra- or interobservation variation between the examinations. However, we attempted to control for this effect by including the clinic site in the multivariable analysis.

The current standard of care for inadequate colposcopic examination recommends more invasive evaluation with a procedure such as cervical conization or LEEP. This allows clarification of discordance between cytology, histology, and the colposcopist’s impression; sampling of any lesion that may extend past the view of standard colposcopy; and histologic evaluation of the entire transformation zone. Given the reported high rates of spontaneous regression of mild and moderate cervical dysplasias with a watchful waiting approach,^{8 12} we wonder whether we are performing unnecessary procedures (LEEP and conization) after cryotherapy as a result of inadequate follow-up colposcopic examinations. A study evaluating the pathologic findings of cone or LEEP specimens from inadequate colposcopic examinations after cryotherapy would help answer these questions. If there is no persistence or progression of dysplasia, then this would support the hypothesis that cryotherapy leads to unnecessary, invasive procedures. Further controlled trials are required to answer these questions.

ACKNOWLEDGMENTS

The authors acknowledge the assistance of Adeline Yerkes, of the Chronic Disease Division, Oklahoma State Department of Health, in facilitating access to the county clinic records.

Cryotherapy is an accepted procedure for treating low-grade cervical dysplasia.^1,2 Only minor modifications of the precise technique of cryotherapy application have occurred since its inception. Currently the double-freeze technique of cryotherapy is an accepted treatment for mild and focal moderate dysplasia of the uterine cervix.³ Cervical cryotherapy is used widely not only because of its proven efficacy but also because of its ease of use in the outpatient setting and lack of known significant side effects. The procedure can be performed in the office setting without the use of local or general anesthesia, making it superior to the more invasive procedures performed before the availability of cryotherapy (eg, cervical conization and hysterectomy).

There has been limited investigation of the effects of cryotherapy on the anatomy of the uterine cervix. Whereas one study showed that cryotherapy has no effect on subsequent fertility or pregnancy outcome,⁴ another in adolescents reported cervical stenosis and pelvic inflammatory disease as possible treatment side effects.⁵ In addition, a study published in 1984 by Jobson and Homesley reported higher rates of retraction of the proximal squamocolumnar junction into the endocervical canal in patients undergoing cryotherapy compared with patients undergoing carbon dioxide laser ablation⁶; 47% of the follow-up colposcopic examinations were inadequate in that study population. Adequacy of colposcopic examination is defined as complete visualization of the transformation zone, visualization of the entire lesion, if present, and correlation between cytologic and histologic findings and the colposcopist’s impression.⁷ Failure to meet any one of these criteria leads to an inadequate colposcopic examination requiring further, more invasive evaluation. This study compared the rate of adequate and inadequate colposcopic examinations in women with and without a history of cryotherapy. Other factors found to influence the adequacy of follow-up colposcopy also are described.

Methods

We performed a retrospective cohort study using data collected from 7 of 14 state-run public health clinics. These 7 sites included rural and urban clinics. All women undergoing at least 2 colposcopic examinations in these clinics between January 1, 1991, and December 1, 1995, were included. Women underwent initial colposcopic examination after an abnormality was noted on a screening Pap test. Only women who had both colposcopic examinations in the same clinic were included. Care provided in these clinics included Pap test screening, colposcopic examinations, and treatment of identified cervical dysplasia with cervical cryotherapy, conization, and loop excision of the ectocervical portion (LEEP). State-contracted physicians trained in obstetrics and gynecology followed women who attended these clinics.

Chart review was used to determine the adequacy of the initial examination, whether an intervening procedure was done, and the adequacy of follow-up colposcopic examination. Adequacy was documented by the physician performing the colposcopic examination with the use of a standard form consistent among clinics. The accepted criteria for adequacy were used, and each colposcopic examination was documented as adequate or inadequate based on the colposcopist’s findings. Charts were reviewed and data were abstracted by 3 reviewers. Cervical biopsy results, presence of human papilloma virus (HPV) noted on routine cytology, endocervical curettage (ECC) results, and routine demographic data also were recorded. The management and therapeutic protocols were consistent across the 7 clinics.

Women were excluded from the analysis if (1) they had cryotherapy performed before their initial colposcopic examination (n = 1), (2) the date of the initial colposcopic examination was not available (n = 36), (3) information confirming the type of treatment used between colposcopic examinations was unknown (n = 32), (4) initial colposcopic examination was inadequate (n = 16), or (5) the adequacy of the follow-up colposcopic examination was not documented (n = 6). The total number of women excluded was 65 because some women met multiple exclusion criteria.

The management after initial colposcopic examination was done according to whether the women had cryotherapy, cone, LEEP, or no procedure between initial and follow-up colposcopic examinations. Univariate analysis of the association between the management group with clinic of treatment, performance of ECC, biopsy results, presence of HPV, and cytologic presence of glandular atypia was performed. Mean age and duration (interval between initial and follow-up colposcopic examinations or between the procedure and follow-up examination) were calculated for all groups.

The odds ratio of an inadequate follow-up colposcopic examination was estimated for type of treatment (cryotherapy, cone/LEEP) compared with no treatment, age, clinic where treatment was provided, performance of ECC, biopsy results, presence of HPV, and presence of glandular atypia. The 95% confidence intervals about the relative odds estimates were calculated. Mean age and duration between initial colposcopy and follow-up colposcopy were calculated for the groups with adequate and inadequate follow-up colposcopic examinations.

Multivariable logistic regression analysis was used to evaluate the association of adequacy of follow-up colposcopic examination with age (years), clinic where colposcopic examination was performed, duration (days), whether or not ECC was performed, biopsy results from the initial colposcopic examination, presence of HPV, and presence of glandular atypia noted on initial colposcopy. Biopsy results were categorized as normal or abnormal in the model that is reported. The stepwise backward elimination technique was used to evaluate the best model. The 95% confidence intervals about the adjusted odds ratio were calculated.

The Pearson chi-square test was used to test the significance of the association between binary variables. The significance of the difference between means was tested with the one-way analysis of variance. Data were analyzed with the personal computer version of the Statistical Package for the Social Sciences (SPSS/PC+ version 8.0).

Results

Between January 1, 1991, and December 31, 1995, 3225 women underwent colposcopic evaluation or treatment at 7 county colposcopy clinics in Oklahoma. Two hundred sixty-eight of these women underwent 2 examinations during the study period. There were 203 of 268 subjects available for analysis after exclusions for missing data. Eighty-three patients (41.1%) had cryotherapy, 24 (11.9%) underwent a cone biopsy or a LEEP procedure, and 96 (47.5%) underwent no procedure between initial and follow-up colposcopic examinations.

Table 1 shows characteristics of women who had cryotherapy, cone/LEEP, and no procedure. The groups were similar with respect to age, whether ECC was performed, presence of HPV, and whether glandular involvement was noted. There was an association between the degree of cervical dysplasia and the three treatment groups, which was expected because degree of dysplasia determines treatment modality. Women who had cryotherapy had follow-up colposcopy (mean = 565 days) later than women who had cone or LEEP (mean = 319 days) or no procedure (mean = 339 days; P < .0001).

Thirty-three percent (n = 67) had inadequate follow-up colposcopic examinations. These included a large proportion of women, 61.4%, who had cryotherapy (51/83) compared with 20.8% (5/24) of women who had cone or LEEP and 11.5% (11/96) of women who had no procedure.

Table 2 shows the relationship between inadequate second colposcopy and previous cryotherapy, cone/LEEP, abnormal cervical biopsy, ECC, presence of HPV, and presence of glandular atypia. Patients who had cryotherapy had an increased likelihood of inadequate follow-up compared with patients who had no procedure (adjusted odds ratio =18.67, 95% confidence interval = 6.99–49.81). Cone/LEEP increased the likelihood of inadequate follow-up but was not statistically significant. Age, duration, ECC, presence of HPV, or presence of glandular atypia did not increase the likelihood of subsequent inadequate colposcopic examination. Odds ratio estimates for different clinics are not reported but were imprecise due to small numbers.

TABLE 1
Characteristics of patients with and without cryotherapy between initial and follow-up colposcopy

TABLE 2
Likelihood of inadequate follow-up colposcopic examination*

Discussion

Undergoing cryotherapy of the uterine cervix increases the risk that a follow-up colposcopic examination will be inadequate. This agrees with the findings of Jobson and Homesley’s 1984 study,⁶ which looked at the efficacy of cryotherapy vs carbon dioxide laser ablation in the treatment of cervical dysplasia. Although it was not the focus of their study, a high rate of inadequacy was noted on follow-up colposcopic examinations after cryotherapy.

Because of the retrospective design of this study, we could not randomly assign women to a treatment group. However, the study groups were similar with respect to other variables potentially associated with the outcome measure. In addition, we attempted to control confounding variables by using multivariable analysis. By including the clinic where the examination was performed, we attempted to limit the effect of the subjective assignment of adequacy by the physician. This is a limitation of this study.

We found an association between the clinics where the follow-up colposcopist’s examinations were performed and whether a follow-up examination was adequate or inadequate. The determination of adequacy depends on the physician’s observations during the colposcopic examination. We were unable to measure the intra- or interobservation variation between the examinations. However, we attempted to control for this effect by including the clinic site in the multivariable analysis.

The current standard of care for inadequate colposcopic examination recommends more invasive evaluation with a procedure such as cervical conization or LEEP. This allows clarification of discordance between cytology, histology, and the colposcopist’s impression; sampling of any lesion that may extend past the view of standard colposcopy; and histologic evaluation of the entire transformation zone. Given the reported high rates of spontaneous regression of mild and moderate cervical dysplasias with a watchful waiting approach,^{8 12} we wonder whether we are performing unnecessary procedures (LEEP and conization) after cryotherapy as a result of inadequate follow-up colposcopic examinations. A study evaluating the pathologic findings of cone or LEEP specimens from inadequate colposcopic examinations after cryotherapy would help answer these questions. If there is no persistence or progression of dysplasia, then this would support the hypothesis that cryotherapy leads to unnecessary, invasive procedures. Further controlled trials are required to answer these questions.

ACKNOWLEDGMENTS

The authors acknowledge the assistance of Adeline Yerkes, of the Chronic Disease Division, Oklahoma State Department of Health, in facilitating access to the county clinic records.

Cryotherapy is an accepted procedure for treating low-grade cervical dysplasia.^1,2 Only minor modifications of the precise technique of cryotherapy application have occurred since its inception. Currently the double-freeze technique of cryotherapy is an accepted treatment for mild and focal moderate dysplasia of the uterine cervix.³ Cervical cryotherapy is used widely not only because of its proven efficacy but also because of its ease of use in the outpatient setting and lack of known significant side effects. The procedure can be performed in the office setting without the use of local or general anesthesia, making it superior to the more invasive procedures performed before the availability of cryotherapy (eg, cervical conization and hysterectomy).

There has been limited investigation of the effects of cryotherapy on the anatomy of the uterine cervix. Whereas one study showed that cryotherapy has no effect on subsequent fertility or pregnancy outcome,⁴ another in adolescents reported cervical stenosis and pelvic inflammatory disease as possible treatment side effects.⁵ In addition, a study published in 1984 by Jobson and Homesley reported higher rates of retraction of the proximal squamocolumnar junction into the endocervical canal in patients undergoing cryotherapy compared with patients undergoing carbon dioxide laser ablation⁶; 47% of the follow-up colposcopic examinations were inadequate in that study population. Adequacy of colposcopic examination is defined as complete visualization of the transformation zone, visualization of the entire lesion, if present, and correlation between cytologic and histologic findings and the colposcopist’s impression.⁷ Failure to meet any one of these criteria leads to an inadequate colposcopic examination requiring further, more invasive evaluation. This study compared the rate of adequate and inadequate colposcopic examinations in women with and without a history of cryotherapy. Other factors found to influence the adequacy of follow-up colposcopy also are described.

Methods

We performed a retrospective cohort study using data collected from 7 of 14 state-run public health clinics. These 7 sites included rural and urban clinics. All women undergoing at least 2 colposcopic examinations in these clinics between January 1, 1991, and December 1, 1995, were included. Women underwent initial colposcopic examination after an abnormality was noted on a screening Pap test. Only women who had both colposcopic examinations in the same clinic were included. Care provided in these clinics included Pap test screening, colposcopic examinations, and treatment of identified cervical dysplasia with cervical cryotherapy, conization, and loop excision of the ectocervical portion (LEEP). State-contracted physicians trained in obstetrics and gynecology followed women who attended these clinics.

Chart review was used to determine the adequacy of the initial examination, whether an intervening procedure was done, and the adequacy of follow-up colposcopic examination. Adequacy was documented by the physician performing the colposcopic examination with the use of a standard form consistent among clinics. The accepted criteria for adequacy were used, and each colposcopic examination was documented as adequate or inadequate based on the colposcopist’s findings. Charts were reviewed and data were abstracted by 3 reviewers. Cervical biopsy results, presence of human papilloma virus (HPV) noted on routine cytology, endocervical curettage (ECC) results, and routine demographic data also were recorded. The management and therapeutic protocols were consistent across the 7 clinics.

Women were excluded from the analysis if (1) they had cryotherapy performed before their initial colposcopic examination (n = 1), (2) the date of the initial colposcopic examination was not available (n = 36), (3) information confirming the type of treatment used between colposcopic examinations was unknown (n = 32), (4) initial colposcopic examination was inadequate (n = 16), or (5) the adequacy of the follow-up colposcopic examination was not documented (n = 6). The total number of women excluded was 65 because some women met multiple exclusion criteria.

The management after initial colposcopic examination was done according to whether the women had cryotherapy, cone, LEEP, or no procedure between initial and follow-up colposcopic examinations. Univariate analysis of the association between the management group with clinic of treatment, performance of ECC, biopsy results, presence of HPV, and cytologic presence of glandular atypia was performed. Mean age and duration (interval between initial and follow-up colposcopic examinations or between the procedure and follow-up examination) were calculated for all groups.

The odds ratio of an inadequate follow-up colposcopic examination was estimated for type of treatment (cryotherapy, cone/LEEP) compared with no treatment, age, clinic where treatment was provided, performance of ECC, biopsy results, presence of HPV, and presence of glandular atypia. The 95% confidence intervals about the relative odds estimates were calculated. Mean age and duration between initial colposcopy and follow-up colposcopy were calculated for the groups with adequate and inadequate follow-up colposcopic examinations.

Multivariable logistic regression analysis was used to evaluate the association of adequacy of follow-up colposcopic examination with age (years), clinic where colposcopic examination was performed, duration (days), whether or not ECC was performed, biopsy results from the initial colposcopic examination, presence of HPV, and presence of glandular atypia noted on initial colposcopy. Biopsy results were categorized as normal or abnormal in the model that is reported. The stepwise backward elimination technique was used to evaluate the best model. The 95% confidence intervals about the adjusted odds ratio were calculated.

The Pearson chi-square test was used to test the significance of the association between binary variables. The significance of the difference between means was tested with the one-way analysis of variance. Data were analyzed with the personal computer version of the Statistical Package for the Social Sciences (SPSS/PC+ version 8.0).

Results

Between January 1, 1991, and December 31, 1995, 3225 women underwent colposcopic evaluation or treatment at 7 county colposcopy clinics in Oklahoma. Two hundred sixty-eight of these women underwent 2 examinations during the study period. There were 203 of 268 subjects available for analysis after exclusions for missing data. Eighty-three patients (41.1%) had cryotherapy, 24 (11.9%) underwent a cone biopsy or a LEEP procedure, and 96 (47.5%) underwent no procedure between initial and follow-up colposcopic examinations.

Table 1 shows characteristics of women who had cryotherapy, cone/LEEP, and no procedure. The groups were similar with respect to age, whether ECC was performed, presence of HPV, and whether glandular involvement was noted. There was an association between the degree of cervical dysplasia and the three treatment groups, which was expected because degree of dysplasia determines treatment modality. Women who had cryotherapy had follow-up colposcopy (mean = 565 days) later than women who had cone or LEEP (mean = 319 days) or no procedure (mean = 339 days; P < .0001).

Thirty-three percent (n = 67) had inadequate follow-up colposcopic examinations. These included a large proportion of women, 61.4%, who had cryotherapy (51/83) compared with 20.8% (5/24) of women who had cone or LEEP and 11.5% (11/96) of women who had no procedure.

Table 2 shows the relationship between inadequate second colposcopy and previous cryotherapy, cone/LEEP, abnormal cervical biopsy, ECC, presence of HPV, and presence of glandular atypia. Patients who had cryotherapy had an increased likelihood of inadequate follow-up compared with patients who had no procedure (adjusted odds ratio =18.67, 95% confidence interval = 6.99–49.81). Cone/LEEP increased the likelihood of inadequate follow-up but was not statistically significant. Age, duration, ECC, presence of HPV, or presence of glandular atypia did not increase the likelihood of subsequent inadequate colposcopic examination. Odds ratio estimates for different clinics are not reported but were imprecise due to small numbers.

TABLE 1
Characteristics of patients with and without cryotherapy between initial and follow-up colposcopy

TABLE 2
Likelihood of inadequate follow-up colposcopic examination*

Discussion

Undergoing cryotherapy of the uterine cervix increases the risk that a follow-up colposcopic examination will be inadequate. This agrees with the findings of Jobson and Homesley’s 1984 study,⁶ which looked at the efficacy of cryotherapy vs carbon dioxide laser ablation in the treatment of cervical dysplasia. Although it was not the focus of their study, a high rate of inadequacy was noted on follow-up colposcopic examinations after cryotherapy.

Because of the retrospective design of this study, we could not randomly assign women to a treatment group. However, the study groups were similar with respect to other variables potentially associated with the outcome measure. In addition, we attempted to control confounding variables by using multivariable analysis. By including the clinic where the examination was performed, we attempted to limit the effect of the subjective assignment of adequacy by the physician. This is a limitation of this study.

We found an association between the clinics where the follow-up colposcopist’s examinations were performed and whether a follow-up examination was adequate or inadequate. The determination of adequacy depends on the physician’s observations during the colposcopic examination. We were unable to measure the intra- or interobservation variation between the examinations. However, we attempted to control for this effect by including the clinic site in the multivariable analysis.

The current standard of care for inadequate colposcopic examination recommends more invasive evaluation with a procedure such as cervical conization or LEEP. This allows clarification of discordance between cytology, histology, and the colposcopist’s impression; sampling of any lesion that may extend past the view of standard colposcopy; and histologic evaluation of the entire transformation zone. Given the reported high rates of spontaneous regression of mild and moderate cervical dysplasias with a watchful waiting approach,^{8 12} we wonder whether we are performing unnecessary procedures (LEEP and conization) after cryotherapy as a result of inadequate follow-up colposcopic examinations. A study evaluating the pathologic findings of cone or LEEP specimens from inadequate colposcopic examinations after cryotherapy would help answer these questions. If there is no persistence or progression of dysplasia, then this would support the hypothesis that cryotherapy leads to unnecessary, invasive procedures. Further controlled trials are required to answer these questions.

ACKNOWLEDGMENTS

The authors acknowledge the assistance of Adeline Yerkes, of the Chronic Disease Division, Oklahoma State Department of Health, in facilitating access to the county clinic records.

Patients are increasingly using the Internet to obtain medical information. Few practice Web sites provide services beyond information about the clinic and common medical diseases. We surveyed computer-using patients at 4 family medicine clinics in Denver, Colorado, by assessing their desire for Internet services from their providers. Patients were especially interested in getting e-mail reminders about appointments, online booking of appointments in real time, and receiving updates about new advances in treatment. Patients were also interested in virtual visits for simple and chronic medical problems and for following chronic conditions through virtual means. We concluded that computer-using patients desire Internet services to augment their medical care. As growth and communication via the Internet continue, primary care physicians should move more aggressively toward adding services to their practices’ Internet Web sites beyond the simple provision of information.

Patients are increasingly using the Internet to obtain medical information. A recent Harris poll estimated that 98 million Americans have retrieved health-related information online, an increase of 44 million since 1998.¹ Previous studies examined patients’ subjective ratings² of medical information sites and assessed the quality of medical information available through the World Wide Web.³ However, very little research has been published regarding patients’ interest in “e-health” services.^4,5 The health care industry lags far behind other industries in terms of providing useful Internet services for the consumer.

We hypothesized that computer-using patients were interested in using the current and potential future services of Web-based technology to augment their care through clinic-based Web sites. The purpose of this study was to specifically determine the interests and needs of computer-using patients in clinic Web services beyond informational services alone.

Methods

An anonymous survey was given to a convenience sample of patients from 4 Denver Family Medicine clinics, with each surveying anywhere from 40 to 110 patients. The clinical sites used in this survey were socioeconomically diverse and included 1 community-based residency clinic, 1 university-based residency clinic, and 2 health maintenance organization clinics. A total of 600 surveys were distributed. Patient surveys were placed at the front desk, where the personnel were requested to ask patients to complete this volunteer survey. Computer and noncomputer users were asked to take the survey and their computer-using status was noted on the survey. Surveys were completed during the visit and returned to the front desk for collection. The surveys represented visits in these clinics from July 2000 to November 2000. This anonymous survey assessed patient demographics, Internet use, and patients’ interest in Internet services. Preferences for 22 Internet services were assessed on a Likert scale of 1 (definitely would not use) to 10 (definitely would use).

Data were analyzed using SPSS version 10 for Windows (SPSS Inc., Chicago, IL). Only computer users were included in the final calculations because of the very small percentage of noncomputer users (7.4%) who volunteered to take the survey. Frequencies were used to describe the computer-using survey respondents, their use of computers, and their preferences for Web-based services. Tests were used to evaluate significant variations among the survey respondents.

Results

Of 600 surveys, 227 were returned (37.8%). Most respondents were female (66.3%) with a mean age of 44.7 years. The vast majority of those who responded to this survey owned computers at home (90.0%) and/or had them at work (83.7%); 44.5% were college graduates and 52.1% had chronic medical conditions. Data on patients’ current use of the Internet are shown in Table 1.

Patient’s desires for Web-based services are summarized in Table 2. Patients displayed a strong interest in front desk services such as being able to book appointments in real time (mean Likert score, 8.50) over the Internet and getting e-mail reminders about appointments (mean Likert score, 8.61). Back office services ranking high included requesting medication refills online (mean Likert score, 8.47) to requesting a referral (mean Likert score, 8.26). The ability to send a message to “your doctor” also ranked high (mean Likert score, 8.40). There was relatively little interest in taking a virtual tour of the clinic (mean Likert score, 6.26) or having a page of links to health insurance company Web sites (mean Likert score, 6.73).

Patients displayed moderate interest in virtual visits (a patient-to-physician encounter conducted using the Internet alone), with 66.0% showing interest in a virtual visit for a simple medical problem. A slightly lower percentage (57.7%) was interested in a virtual visit for a chronic medical problem. Approximately a third of patients (32.6%) was more interested in a real-time virtual visit that used a personal computer (PC) videoconference rather than a real-time e-mail conversation (ie, “chat room” or one-on-one “chat”). Not surprisingly, a larger percentage of patients was more willing to make a virtual visit if it offered a lower co-payment (62%). Only 46.7% of patients indicated they would be interested in a virtual visit if it required the usual co-payment.

Interest in virtual visits for simple medical problems was higher among patients who had previously used the Internet to order products online (74.6% vs 45.0%, P < .001). Patients with chronic diseases were more likely to be interested in virtual visits for simple medical problems (70.8% vs 62.2%, P = .213), although this association was not statistically significant. A higher education level was associated with obtaining medical information over the Internet. College graduates were more likely than nongraduates to have used the Internet to obtain medical information (50% vs 33.6%, P < .05).

TABLE 1
Internet use among computer-using patients

TABLE 2
Internet services desired by computer-using patients

Discussion

Patients who used computers and the Internet showed significant interest in using Web-based services from their family physicians. These patients were especially interested in using the Internet for front desk services and common tasks, which are frequently provided over a busy telephone line. Services related to providing information were of less interest, and patients displayed only moderate interest in virtual visits. Using PC videoconferencing instead of e-mail communication would increase patients’ interest in a virtual visit. Poor videoconferencing capability over PCs, lack of access, or perhaps a fear of insufficient security over Web-based communications might limit interest.^6-8

The survey had several limitations. As noted, only 7.4% of noncomputer users took the survey when requested by front desk staff. Therefore, we limited our analysis to computer-using patients. However, given the current statistics of Internet use and growth in access to all sectors of our population, it is likely that most practices will find a sufficient percentage of “connected” patients to apply the study’s findings. Assessment of online use at a specific clinic site will be useful in prioritizing the need and application of Internet services. The low response rate of our survey is likely due to the voluntary nature of the survey and the challenge of the front desk staff in finding time to encourage patients to take the survey. The practices that participated were busy ones that must move patients in a timely fashion from the front desk area to examination rooms.

Businesses with many employees who use e-commerce and banking services may especially benefit from signing up with a practice that offers online services. Patients with chronic diseases usually require more frequent visits with their physicians. We hope that patients with chronic disease will take advantage of “virtual visits” as they become available, thereby freeing them from transportation costs, lost time, and productivity.

Other desired services such as online appointment scheduling, medication refills, and referral requests might improve the efficiency in front and back office functions by reducing the number of lengthy telephone calls. We hope to perform future studies that evaluate the impact of Internet services on efficiency and patient/provider satisfaction.

Physicians should place a high priority on building service components into their practice Web sites. Interfacing these Web-based services with electronic medical records is another important task that needs further programmer development and attention by physicians. We hope that continued research in e-health care will further catalyze technologic developments that improve disease management, increase practice efficiency and patient satisfaction, and reduce medical errors.

Acknowledgments

The authors thank Lu Sandoval and Coline Bublitz for their help in preparing the data. They also thank Richard Drexilius, MD, at the Swedish Family Medicine Center; Manoj Pawar, MD, at the Exempla Family Medicine Center; and Carl Severin, MD, at the Kaiser Centerpointe Clinic for allowing the authors to perform the survey at their facilities. Special thanks to Perry Dickinson, MD, for his editorial assistance.

Patients are increasingly using the Internet to obtain medical information. Few practice Web sites provide services beyond information about the clinic and common medical diseases. We surveyed computer-using patients at 4 family medicine clinics in Denver, Colorado, by assessing their desire for Internet services from their providers. Patients were especially interested in getting e-mail reminders about appointments, online booking of appointments in real time, and receiving updates about new advances in treatment. Patients were also interested in virtual visits for simple and chronic medical problems and for following chronic conditions through virtual means. We concluded that computer-using patients desire Internet services to augment their medical care. As growth and communication via the Internet continue, primary care physicians should move more aggressively toward adding services to their practices’ Internet Web sites beyond the simple provision of information.

Patients are increasingly using the Internet to obtain medical information. A recent Harris poll estimated that 98 million Americans have retrieved health-related information online, an increase of 44 million since 1998.¹ Previous studies examined patients’ subjective ratings² of medical information sites and assessed the quality of medical information available through the World Wide Web.³ However, very little research has been published regarding patients’ interest in “e-health” services.^4,5 The health care industry lags far behind other industries in terms of providing useful Internet services for the consumer.

We hypothesized that computer-using patients were interested in using the current and potential future services of Web-based technology to augment their care through clinic-based Web sites. The purpose of this study was to specifically determine the interests and needs of computer-using patients in clinic Web services beyond informational services alone.

Methods

An anonymous survey was given to a convenience sample of patients from 4 Denver Family Medicine clinics, with each surveying anywhere from 40 to 110 patients. The clinical sites used in this survey were socioeconomically diverse and included 1 community-based residency clinic, 1 university-based residency clinic, and 2 health maintenance organization clinics. A total of 600 surveys were distributed. Patient surveys were placed at the front desk, where the personnel were requested to ask patients to complete this volunteer survey. Computer and noncomputer users were asked to take the survey and their computer-using status was noted on the survey. Surveys were completed during the visit and returned to the front desk for collection. The surveys represented visits in these clinics from July 2000 to November 2000. This anonymous survey assessed patient demographics, Internet use, and patients’ interest in Internet services. Preferences for 22 Internet services were assessed on a Likert scale of 1 (definitely would not use) to 10 (definitely would use).

Data were analyzed using SPSS version 10 for Windows (SPSS Inc., Chicago, IL). Only computer users were included in the final calculations because of the very small percentage of noncomputer users (7.4%) who volunteered to take the survey. Frequencies were used to describe the computer-using survey respondents, their use of computers, and their preferences for Web-based services. Tests were used to evaluate significant variations among the survey respondents.

Results

Of 600 surveys, 227 were returned (37.8%). Most respondents were female (66.3%) with a mean age of 44.7 years. The vast majority of those who responded to this survey owned computers at home (90.0%) and/or had them at work (83.7%); 44.5% were college graduates and 52.1% had chronic medical conditions. Data on patients’ current use of the Internet are shown in Table 1.

Patient’s desires for Web-based services are summarized in Table 2. Patients displayed a strong interest in front desk services such as being able to book appointments in real time (mean Likert score, 8.50) over the Internet and getting e-mail reminders about appointments (mean Likert score, 8.61). Back office services ranking high included requesting medication refills online (mean Likert score, 8.47) to requesting a referral (mean Likert score, 8.26). The ability to send a message to “your doctor” also ranked high (mean Likert score, 8.40). There was relatively little interest in taking a virtual tour of the clinic (mean Likert score, 6.26) or having a page of links to health insurance company Web sites (mean Likert score, 6.73).

Patients displayed moderate interest in virtual visits (a patient-to-physician encounter conducted using the Internet alone), with 66.0% showing interest in a virtual visit for a simple medical problem. A slightly lower percentage (57.7%) was interested in a virtual visit for a chronic medical problem. Approximately a third of patients (32.6%) was more interested in a real-time virtual visit that used a personal computer (PC) videoconference rather than a real-time e-mail conversation (ie, “chat room” or one-on-one “chat”). Not surprisingly, a larger percentage of patients was more willing to make a virtual visit if it offered a lower co-payment (62%). Only 46.7% of patients indicated they would be interested in a virtual visit if it required the usual co-payment.

Interest in virtual visits for simple medical problems was higher among patients who had previously used the Internet to order products online (74.6% vs 45.0%, P < .001). Patients with chronic diseases were more likely to be interested in virtual visits for simple medical problems (70.8% vs 62.2%, P = .213), although this association was not statistically significant. A higher education level was associated with obtaining medical information over the Internet. College graduates were more likely than nongraduates to have used the Internet to obtain medical information (50% vs 33.6%, P < .05).

TABLE 1
Internet use among computer-using patients

TABLE 2
Internet services desired by computer-using patients

Discussion

Patients who used computers and the Internet showed significant interest in using Web-based services from their family physicians. These patients were especially interested in using the Internet for front desk services and common tasks, which are frequently provided over a busy telephone line. Services related to providing information were of less interest, and patients displayed only moderate interest in virtual visits. Using PC videoconferencing instead of e-mail communication would increase patients’ interest in a virtual visit. Poor videoconferencing capability over PCs, lack of access, or perhaps a fear of insufficient security over Web-based communications might limit interest.^6-8

The survey had several limitations. As noted, only 7.4% of noncomputer users took the survey when requested by front desk staff. Therefore, we limited our analysis to computer-using patients. However, given the current statistics of Internet use and growth in access to all sectors of our population, it is likely that most practices will find a sufficient percentage of “connected” patients to apply the study’s findings. Assessment of online use at a specific clinic site will be useful in prioritizing the need and application of Internet services. The low response rate of our survey is likely due to the voluntary nature of the survey and the challenge of the front desk staff in finding time to encourage patients to take the survey. The practices that participated were busy ones that must move patients in a timely fashion from the front desk area to examination rooms.

Businesses with many employees who use e-commerce and banking services may especially benefit from signing up with a practice that offers online services. Patients with chronic diseases usually require more frequent visits with their physicians. We hope that patients with chronic disease will take advantage of “virtual visits” as they become available, thereby freeing them from transportation costs, lost time, and productivity.

Other desired services such as online appointment scheduling, medication refills, and referral requests might improve the efficiency in front and back office functions by reducing the number of lengthy telephone calls. We hope to perform future studies that evaluate the impact of Internet services on efficiency and patient/provider satisfaction.

Physicians should place a high priority on building service components into their practice Web sites. Interfacing these Web-based services with electronic medical records is another important task that needs further programmer development and attention by physicians. We hope that continued research in e-health care will further catalyze technologic developments that improve disease management, increase practice efficiency and patient satisfaction, and reduce medical errors.

Acknowledgments

The authors thank Lu Sandoval and Coline Bublitz for their help in preparing the data. They also thank Richard Drexilius, MD, at the Swedish Family Medicine Center; Manoj Pawar, MD, at the Exempla Family Medicine Center; and Carl Severin, MD, at the Kaiser Centerpointe Clinic for allowing the authors to perform the survey at their facilities. Special thanks to Perry Dickinson, MD, for his editorial assistance.

Patients are increasingly using the Internet to obtain medical information. Few practice Web sites provide services beyond information about the clinic and common medical diseases. We surveyed computer-using patients at 4 family medicine clinics in Denver, Colorado, by assessing their desire for Internet services from their providers. Patients were especially interested in getting e-mail reminders about appointments, online booking of appointments in real time, and receiving updates about new advances in treatment. Patients were also interested in virtual visits for simple and chronic medical problems and for following chronic conditions through virtual means. We concluded that computer-using patients desire Internet services to augment their medical care. As growth and communication via the Internet continue, primary care physicians should move more aggressively toward adding services to their practices’ Internet Web sites beyond the simple provision of information.

Patients are increasingly using the Internet to obtain medical information. A recent Harris poll estimated that 98 million Americans have retrieved health-related information online, an increase of 44 million since 1998.¹ Previous studies examined patients’ subjective ratings² of medical information sites and assessed the quality of medical information available through the World Wide Web.³ However, very little research has been published regarding patients’ interest in “e-health” services.^4,5 The health care industry lags far behind other industries in terms of providing useful Internet services for the consumer.

We hypothesized that computer-using patients were interested in using the current and potential future services of Web-based technology to augment their care through clinic-based Web sites. The purpose of this study was to specifically determine the interests and needs of computer-using patients in clinic Web services beyond informational services alone.

Methods

An anonymous survey was given to a convenience sample of patients from 4 Denver Family Medicine clinics, with each surveying anywhere from 40 to 110 patients. The clinical sites used in this survey were socioeconomically diverse and included 1 community-based residency clinic, 1 university-based residency clinic, and 2 health maintenance organization clinics. A total of 600 surveys were distributed. Patient surveys were placed at the front desk, where the personnel were requested to ask patients to complete this volunteer survey. Computer and noncomputer users were asked to take the survey and their computer-using status was noted on the survey. Surveys were completed during the visit and returned to the front desk for collection. The surveys represented visits in these clinics from July 2000 to November 2000. This anonymous survey assessed patient demographics, Internet use, and patients’ interest in Internet services. Preferences for 22 Internet services were assessed on a Likert scale of 1 (definitely would not use) to 10 (definitely would use).

Data were analyzed using SPSS version 10 for Windows (SPSS Inc., Chicago, IL). Only computer users were included in the final calculations because of the very small percentage of noncomputer users (7.4%) who volunteered to take the survey. Frequencies were used to describe the computer-using survey respondents, their use of computers, and their preferences for Web-based services. Tests were used to evaluate significant variations among the survey respondents.

Results

Of 600 surveys, 227 were returned (37.8%). Most respondents were female (66.3%) with a mean age of 44.7 years. The vast majority of those who responded to this survey owned computers at home (90.0%) and/or had them at work (83.7%); 44.5% were college graduates and 52.1% had chronic medical conditions. Data on patients’ current use of the Internet are shown in Table 1.

Patient’s desires for Web-based services are summarized in Table 2. Patients displayed a strong interest in front desk services such as being able to book appointments in real time (mean Likert score, 8.50) over the Internet and getting e-mail reminders about appointments (mean Likert score, 8.61). Back office services ranking high included requesting medication refills online (mean Likert score, 8.47) to requesting a referral (mean Likert score, 8.26). The ability to send a message to “your doctor” also ranked high (mean Likert score, 8.40). There was relatively little interest in taking a virtual tour of the clinic (mean Likert score, 6.26) or having a page of links to health insurance company Web sites (mean Likert score, 6.73).

Patients displayed moderate interest in virtual visits (a patient-to-physician encounter conducted using the Internet alone), with 66.0% showing interest in a virtual visit for a simple medical problem. A slightly lower percentage (57.7%) was interested in a virtual visit for a chronic medical problem. Approximately a third of patients (32.6%) was more interested in a real-time virtual visit that used a personal computer (PC) videoconference rather than a real-time e-mail conversation (ie, “chat room” or one-on-one “chat”). Not surprisingly, a larger percentage of patients was more willing to make a virtual visit if it offered a lower co-payment (62%). Only 46.7% of patients indicated they would be interested in a virtual visit if it required the usual co-payment.

Interest in virtual visits for simple medical problems was higher among patients who had previously used the Internet to order products online (74.6% vs 45.0%, P < .001). Patients with chronic diseases were more likely to be interested in virtual visits for simple medical problems (70.8% vs 62.2%, P = .213), although this association was not statistically significant. A higher education level was associated with obtaining medical information over the Internet. College graduates were more likely than nongraduates to have used the Internet to obtain medical information (50% vs 33.6%, P < .05).

TABLE 1
Internet use among computer-using patients

TABLE 2
Internet services desired by computer-using patients

Discussion

Patients who used computers and the Internet showed significant interest in using Web-based services from their family physicians. These patients were especially interested in using the Internet for front desk services and common tasks, which are frequently provided over a busy telephone line. Services related to providing information were of less interest, and patients displayed only moderate interest in virtual visits. Using PC videoconferencing instead of e-mail communication would increase patients’ interest in a virtual visit. Poor videoconferencing capability over PCs, lack of access, or perhaps a fear of insufficient security over Web-based communications might limit interest.^6-8

The survey had several limitations. As noted, only 7.4% of noncomputer users took the survey when requested by front desk staff. Therefore, we limited our analysis to computer-using patients. However, given the current statistics of Internet use and growth in access to all sectors of our population, it is likely that most practices will find a sufficient percentage of “connected” patients to apply the study’s findings. Assessment of online use at a specific clinic site will be useful in prioritizing the need and application of Internet services. The low response rate of our survey is likely due to the voluntary nature of the survey and the challenge of the front desk staff in finding time to encourage patients to take the survey. The practices that participated were busy ones that must move patients in a timely fashion from the front desk area to examination rooms.

Businesses with many employees who use e-commerce and banking services may especially benefit from signing up with a practice that offers online services. Patients with chronic diseases usually require more frequent visits with their physicians. We hope that patients with chronic disease will take advantage of “virtual visits” as they become available, thereby freeing them from transportation costs, lost time, and productivity.

Other desired services such as online appointment scheduling, medication refills, and referral requests might improve the efficiency in front and back office functions by reducing the number of lengthy telephone calls. We hope to perform future studies that evaluate the impact of Internet services on efficiency and patient/provider satisfaction.

Physicians should place a high priority on building service components into their practice Web sites. Interfacing these Web-based services with electronic medical records is another important task that needs further programmer development and attention by physicians. We hope that continued research in e-health care will further catalyze technologic developments that improve disease management, increase practice efficiency and patient satisfaction, and reduce medical errors.

Acknowledgments

The authors thank Lu Sandoval and Coline Bublitz for their help in preparing the data. They also thank Richard Drexilius, MD, at the Swedish Family Medicine Center; Manoj Pawar, MD, at the Exempla Family Medicine Center; and Carl Severin, MD, at the Kaiser Centerpointe Clinic for allowing the authors to perform the survey at their facilities. Special thanks to Perry Dickinson, MD, for his editorial assistance.

Previous studies of after-hours calls to family physicians focused on caller demographics, medical triage skills, and patient satisfaction, and were usually conducted for a limited time. We examined the frequency and nature of calls to a family practice residency over 1 year. Caller and patient information, date, time, and chief complaint were obtained from answering service logs. The 5 most frequent chief complaints related to medications, pain, obstetric issues, fever, and nausea. Interestingly, 56 “high utilizers” (0.6% of all patients) accounted for 23% of the calls.

Although telephone calls may account for 10% to 25% of all patient contacts,^1,2 few studies have examined the frequency and nature of these calls over an extended time. A month-long study³ found that patients who telephoned after hours were 3 times more likely to rate their problem in the highest severity category compared with the physician’s rating of the problem. This study, done in July, may not reflect the diversity of patient problems, because of seasonal variations; also, it did not appear to include obstetric problems, which are a prominent reason for calls to family practice physicians.^4,5 Many physician groups use answering services to screen calls as a method for decreasing the number of calls. The purpose of this study was to document the frequency and nature of after-hours calls to a family practice office over 1 year.

METHODS

All after-hours telephone calls (5 PM to 8 AM, weekends and holidays) made to a freestanding community-based family practice training program were collected for the 12-month period between April 2000 and March 2001. A recorded message directed the caller to call 911 for a life-threatening emergency or stay on the line for operator assistance. Emergency calls were forwarded to the resident physician on call. Sixteen family medicine residents supported by 8 faculty physicians took primary calls on a rotating basis. The practice had approximately 9000 active patients (at least 1 visit in the last 3 years), and about 1350 patient visits per month. Approximately 30% were covered by Medicaid, 10% by Medicare, 35% by managed care, and 12% by indemnity insurance; 13% were uninsured.

The operator recorded date and time, caller’s and patient’s first and last names, primary care physician, patient’s pregnancy status, date of last office visit, chief complaint(s), and whether the caller felt the situation was an emergency.

Previous studies variously classified patient calls based on diagnostic group, chief complaint, symptom, treatment and medication, injury, and organ system affected.^1,3,6-10 We followed the lead of Benjamin⁸and Perkins and colleagues,¹ who used the patient’s chief complaint to categorize calls. We classified the patient’s chief complaint by searching for key words such as “heart” (eg, “fast heartbeat,” “pains near heart,” or “isn’t feeling well, heart failure a couple of years ago”). This allowed for the broad inclusion of chief complaints while avoiding the risk of premature diagnosis.

A research assistant entered information from the operator’s records into a Microsoft Access Database. Patients who called more than 6 times after hours during the year were arbitrarily defined as “high utilizers.” We also gathered data on these callers’ hospital emergency room visits and admissions to affiliated hospitals. The HealthOne Institutional Review Board approved the study.

RESULTS

A total of 3538 calls were made by 1564 patients; 2465 were clinical calls, and key words or phrases were used to classify them under chief complaint headings. If a caller had a multiple-symptom complaint (ie, fever and headache), it was classified under all appropriate headings and counted twice. The total number of complaints is therefore higher than the total number of calls. Table 1 presents the frequency and percentage of after-hours clinical calls for all subjects, and separately for high utilizers. Table 2 presents the average number of clinical calls organized by season and day of the week. Thirty-three percent of all calls were made by the patient, 31% by a proxy (spouse, parent, friend), and 36% by other parties (nurse, pharmacy, unidentified party).

Although the rankings of calls for all patients and high utilizers in Table 1 were similar, several differences stand out. High utilizers account for only 0.6% of patients, but 23% of all calls. High utilizers called substantially more for complaints relating to medication, pain, asthma/breathing and chest problems; 39% of their calls were for medication or pain concerns. Of the high utilizers, 39% (22/56) made 46 emergency room visits, but only 7% (4/56) were hospitalized during the year.

TABLE 1
Percentage of after-hours calls, by chief complaint

TABLE 2
Average number of clinical calls by season and day of week

DISCUSSION

This study expands on previous work by describing the total variety of after-hours phone calls to a family practice office over an entire year. Our findings on reasons for call, time of call, and demographics are similar to those of previous work.^3,10 However, our study is one of the first to describe the subset of high utilizers. Introducing a patient health handbook, practice Web site, pharmacy help line, or other practice management tools might reduce the number of “information only” calls. Contrary to our expectation, the highest numbers of average daily calls were in the spring and summer and not in the winter. Saturdays and Sundays were the busiest days of the week for such calls.

Patients called for diverse clinical reasons (Table 1) and therefore physicians might focus their attention on the most frequent reasons for calls, in order to improve the effectiveness of their educational efforts. For example, physicians might discuss the patient’s medication concerns, give specific recommendations to talk to the pharmacist, and possibly offer an automated medication “tracking system” to alert patients during the week when their medications were running out, as a way of reducing the number of calls and allaying patient concerns.

Pain symptoms clearly account for a substantial number of calls. Although some of these calls might be serious emergencies (chest pain) and require immediate action, other calls, such as for migraine headaches, may point to a need to educate and set limits with patients during their regular appointments. For example, patients could be told that migraine headaches are not a “life-threatening” emergency and be urged to use self-management strategies until the next day.

Discussing fever management with new parents at well-child visits might decrease future calls. There is some research to suggest that providing new parents with specific guidelines about when to call if their child has a fever can dramatically reduce after-hours visits to the emergency room.¹¹ Obstetric calls represent an important group requiring immediate callback with very specific questions (eg, fetal movement, bleeding), and might be a target area for physician education.

Out of approximately 9000 patients in the practice and 1564 patients who called the practice during the year, we identified 56 high utilizers (0.6% of all patients). They averaged nearly 10 calls per year in contrast to 1.3 calls for all other callers. Future research might be directed at trying to determine why these patients feel a need to call at nearly 10 times the rate of other patients.

These findings should be interpreted in light of several limitations. Because our findings are based on a family practice residency, the patient population may be different from the typical private family practice office and have less continuity. However, the wide range of calls is likely to be typical of the diverse problems managed by family physicians. This study did not collect information on the management and disposition of these after-hours calls. Certainly, understanding the entire episode of after-hours contact (reason for call, management, outcome, satisfaction) is important, and is the next step in our research.

The diversity and seriousness of medical problems addressed by the after-hours physician highlight the need to provide specific training to physicians for dealing with patient calls and educating patients on the many issues leading to after-hours calls.

ACKNOWLEDGMENTS

The authors thank Ellie Jensen for help with data collection, entry, and analysis.

Previous studies of after-hours calls to family physicians focused on caller demographics, medical triage skills, and patient satisfaction, and were usually conducted for a limited time. We examined the frequency and nature of calls to a family practice residency over 1 year. Caller and patient information, date, time, and chief complaint were obtained from answering service logs. The 5 most frequent chief complaints related to medications, pain, obstetric issues, fever, and nausea. Interestingly, 56 “high utilizers” (0.6% of all patients) accounted for 23% of the calls.

Although telephone calls may account for 10% to 25% of all patient contacts,^1,2 few studies have examined the frequency and nature of these calls over an extended time. A month-long study³ found that patients who telephoned after hours were 3 times more likely to rate their problem in the highest severity category compared with the physician’s rating of the problem. This study, done in July, may not reflect the diversity of patient problems, because of seasonal variations; also, it did not appear to include obstetric problems, which are a prominent reason for calls to family practice physicians.^4,5 Many physician groups use answering services to screen calls as a method for decreasing the number of calls. The purpose of this study was to document the frequency and nature of after-hours calls to a family practice office over 1 year.

METHODS

All after-hours telephone calls (5 PM to 8 AM, weekends and holidays) made to a freestanding community-based family practice training program were collected for the 12-month period between April 2000 and March 2001. A recorded message directed the caller to call 911 for a life-threatening emergency or stay on the line for operator assistance. Emergency calls were forwarded to the resident physician on call. Sixteen family medicine residents supported by 8 faculty physicians took primary calls on a rotating basis. The practice had approximately 9000 active patients (at least 1 visit in the last 3 years), and about 1350 patient visits per month. Approximately 30% were covered by Medicaid, 10% by Medicare, 35% by managed care, and 12% by indemnity insurance; 13% were uninsured.

The operator recorded date and time, caller’s and patient’s first and last names, primary care physician, patient’s pregnancy status, date of last office visit, chief complaint(s), and whether the caller felt the situation was an emergency.

Previous studies variously classified patient calls based on diagnostic group, chief complaint, symptom, treatment and medication, injury, and organ system affected.^1,3,6-10 We followed the lead of Benjamin⁸and Perkins and colleagues,¹ who used the patient’s chief complaint to categorize calls. We classified the patient’s chief complaint by searching for key words such as “heart” (eg, “fast heartbeat,” “pains near heart,” or “isn’t feeling well, heart failure a couple of years ago”). This allowed for the broad inclusion of chief complaints while avoiding the risk of premature diagnosis.

A research assistant entered information from the operator’s records into a Microsoft Access Database. Patients who called more than 6 times after hours during the year were arbitrarily defined as “high utilizers.” We also gathered data on these callers’ hospital emergency room visits and admissions to affiliated hospitals. The HealthOne Institutional Review Board approved the study.

RESULTS

A total of 3538 calls were made by 1564 patients; 2465 were clinical calls, and key words or phrases were used to classify them under chief complaint headings. If a caller had a multiple-symptom complaint (ie, fever and headache), it was classified under all appropriate headings and counted twice. The total number of complaints is therefore higher than the total number of calls. Table 1 presents the frequency and percentage of after-hours clinical calls for all subjects, and separately for high utilizers. Table 2 presents the average number of clinical calls organized by season and day of the week. Thirty-three percent of all calls were made by the patient, 31% by a proxy (spouse, parent, friend), and 36% by other parties (nurse, pharmacy, unidentified party).

Although the rankings of calls for all patients and high utilizers in Table 1 were similar, several differences stand out. High utilizers account for only 0.6% of patients, but 23% of all calls. High utilizers called substantially more for complaints relating to medication, pain, asthma/breathing and chest problems; 39% of their calls were for medication or pain concerns. Of the high utilizers, 39% (22/56) made 46 emergency room visits, but only 7% (4/56) were hospitalized during the year.

TABLE 1
Percentage of after-hours calls, by chief complaint

TABLE 2
Average number of clinical calls by season and day of week

DISCUSSION

This study expands on previous work by describing the total variety of after-hours phone calls to a family practice office over an entire year. Our findings on reasons for call, time of call, and demographics are similar to those of previous work.^3,10 However, our study is one of the first to describe the subset of high utilizers. Introducing a patient health handbook, practice Web site, pharmacy help line, or other practice management tools might reduce the number of “information only” calls. Contrary to our expectation, the highest numbers of average daily calls were in the spring and summer and not in the winter. Saturdays and Sundays were the busiest days of the week for such calls.

Patients called for diverse clinical reasons (Table 1) and therefore physicians might focus their attention on the most frequent reasons for calls, in order to improve the effectiveness of their educational efforts. For example, physicians might discuss the patient’s medication concerns, give specific recommendations to talk to the pharmacist, and possibly offer an automated medication “tracking system” to alert patients during the week when their medications were running out, as a way of reducing the number of calls and allaying patient concerns.

Pain symptoms clearly account for a substantial number of calls. Although some of these calls might be serious emergencies (chest pain) and require immediate action, other calls, such as for migraine headaches, may point to a need to educate and set limits with patients during their regular appointments. For example, patients could be told that migraine headaches are not a “life-threatening” emergency and be urged to use self-management strategies until the next day.

Discussing fever management with new parents at well-child visits might decrease future calls. There is some research to suggest that providing new parents with specific guidelines about when to call if their child has a fever can dramatically reduce after-hours visits to the emergency room.¹¹ Obstetric calls represent an important group requiring immediate callback with very specific questions (eg, fetal movement, bleeding), and might be a target area for physician education.

Out of approximately 9000 patients in the practice and 1564 patients who called the practice during the year, we identified 56 high utilizers (0.6% of all patients). They averaged nearly 10 calls per year in contrast to 1.3 calls for all other callers. Future research might be directed at trying to determine why these patients feel a need to call at nearly 10 times the rate of other patients.

These findings should be interpreted in light of several limitations. Because our findings are based on a family practice residency, the patient population may be different from the typical private family practice office and have less continuity. However, the wide range of calls is likely to be typical of the diverse problems managed by family physicians. This study did not collect information on the management and disposition of these after-hours calls. Certainly, understanding the entire episode of after-hours contact (reason for call, management, outcome, satisfaction) is important, and is the next step in our research.

The diversity and seriousness of medical problems addressed by the after-hours physician highlight the need to provide specific training to physicians for dealing with patient calls and educating patients on the many issues leading to after-hours calls.

ACKNOWLEDGMENTS

The authors thank Ellie Jensen for help with data collection, entry, and analysis.

Previous studies of after-hours calls to family physicians focused on caller demographics, medical triage skills, and patient satisfaction, and were usually conducted for a limited time. We examined the frequency and nature of calls to a family practice residency over 1 year. Caller and patient information, date, time, and chief complaint were obtained from answering service logs. The 5 most frequent chief complaints related to medications, pain, obstetric issues, fever, and nausea. Interestingly, 56 “high utilizers” (0.6% of all patients) accounted for 23% of the calls.

Although telephone calls may account for 10% to 25% of all patient contacts,^1,2 few studies have examined the frequency and nature of these calls over an extended time. A month-long study³ found that patients who telephoned after hours were 3 times more likely to rate their problem in the highest severity category compared with the physician’s rating of the problem. This study, done in July, may not reflect the diversity of patient problems, because of seasonal variations; also, it did not appear to include obstetric problems, which are a prominent reason for calls to family practice physicians.^4,5 Many physician groups use answering services to screen calls as a method for decreasing the number of calls. The purpose of this study was to document the frequency and nature of after-hours calls to a family practice office over 1 year.

METHODS

All after-hours telephone calls (5 PM to 8 AM, weekends and holidays) made to a freestanding community-based family practice training program were collected for the 12-month period between April 2000 and March 2001. A recorded message directed the caller to call 911 for a life-threatening emergency or stay on the line for operator assistance. Emergency calls were forwarded to the resident physician on call. Sixteen family medicine residents supported by 8 faculty physicians took primary calls on a rotating basis. The practice had approximately 9000 active patients (at least 1 visit in the last 3 years), and about 1350 patient visits per month. Approximately 30% were covered by Medicaid, 10% by Medicare, 35% by managed care, and 12% by indemnity insurance; 13% were uninsured.

The operator recorded date and time, caller’s and patient’s first and last names, primary care physician, patient’s pregnancy status, date of last office visit, chief complaint(s), and whether the caller felt the situation was an emergency.

Previous studies variously classified patient calls based on diagnostic group, chief complaint, symptom, treatment and medication, injury, and organ system affected.^1,3,6-10 We followed the lead of Benjamin⁸and Perkins and colleagues,¹ who used the patient’s chief complaint to categorize calls. We classified the patient’s chief complaint by searching for key words such as “heart” (eg, “fast heartbeat,” “pains near heart,” or “isn’t feeling well, heart failure a couple of years ago”). This allowed for the broad inclusion of chief complaints while avoiding the risk of premature diagnosis.

A research assistant entered information from the operator’s records into a Microsoft Access Database. Patients who called more than 6 times after hours during the year were arbitrarily defined as “high utilizers.” We also gathered data on these callers’ hospital emergency room visits and admissions to affiliated hospitals. The HealthOne Institutional Review Board approved the study.

RESULTS

A total of 3538 calls were made by 1564 patients; 2465 were clinical calls, and key words or phrases were used to classify them under chief complaint headings. If a caller had a multiple-symptom complaint (ie, fever and headache), it was classified under all appropriate headings and counted twice. The total number of complaints is therefore higher than the total number of calls. Table 1 presents the frequency and percentage of after-hours clinical calls for all subjects, and separately for high utilizers. Table 2 presents the average number of clinical calls organized by season and day of the week. Thirty-three percent of all calls were made by the patient, 31% by a proxy (spouse, parent, friend), and 36% by other parties (nurse, pharmacy, unidentified party).

Although the rankings of calls for all patients and high utilizers in Table 1 were similar, several differences stand out. High utilizers account for only 0.6% of patients, but 23% of all calls. High utilizers called substantially more for complaints relating to medication, pain, asthma/breathing and chest problems; 39% of their calls were for medication or pain concerns. Of the high utilizers, 39% (22/56) made 46 emergency room visits, but only 7% (4/56) were hospitalized during the year.

TABLE 1
Percentage of after-hours calls, by chief complaint

TABLE 2
Average number of clinical calls by season and day of week

DISCUSSION

This study expands on previous work by describing the total variety of after-hours phone calls to a family practice office over an entire year. Our findings on reasons for call, time of call, and demographics are similar to those of previous work.^3,10 However, our study is one of the first to describe the subset of high utilizers. Introducing a patient health handbook, practice Web site, pharmacy help line, or other practice management tools might reduce the number of “information only” calls. Contrary to our expectation, the highest numbers of average daily calls were in the spring and summer and not in the winter. Saturdays and Sundays were the busiest days of the week for such calls.