Original Research

We convened a multidisciplinary, multispecialty panel to develop comprehensive evidence and consensus-based guidelines for managing nursing home-acquired pneumonia. The panel began with explicit criteria for process of care quality measures, performed a comprehensive review of the English-language literature, evaluated the quality of the evidence, and drafted a set of proposed guidelines. The panel reviewed the draft, an annotated bibliography, and data from a study of 30-day survival from nursing home-acquired pneumonia, and then participated in an all-day meeting in January 2001. Using a modified Delphi process, the panel refined the guidelines and developed a care pathway. The guidelines recommend a comprehensive approach, including immunization of staff and residents, and communication between nursing staff and the attending physician within 2 hours of symptom onset. Probable pneumonia was defined. An algorithm was delineated for assessing the patient’s wishes for hospitalization and aggressive care, and deciding on hospitalization based on the severity of the illness as well as the capacity of the nursing home to provide acute care. The timing and extent of evaluation in a nursing home relative to the rapid initiation of antibiotics should depend on whether the patient has any unstable vital signs. An antibiotic covering Streptococcus pneumoniae, Haemophilus influenzae, common gram-negative rods, and Staphylococcus aureus should be given for 10 to 14 days, orally if the patient is able to take medications by mouth.

Mortality from nursing home-acquired pneumonia is as high as 44%,¹and nearly a third of those who survive suffer significant functional decline.² Nursing home-acquired pneumonia is an entity distinct from community-acquired pneumonia in the elderly and nosocomial pneumonia. Older adults in nursing homes are more likely than community-dwelling older persons, but less likely than hospitalized elderly patients, to be colonized with gram-negative rods and pathogens with multiple antibiotic resistance.^3-8Inappropriate use of antibiotics in long-term care facilities contributes significantly to the growing problem of antibiotic resistance generally.⁹ In addition, most nursing home residents are cognitively impaired, immunocompromised, have multiple functional deficits, or have dysphagia, which further contributes to their vulnerability.¹⁰

In recent work¹¹ with a national nursing home sample, we showed that high quality care, including appropriate antibiotic use, hospitalization when indicated, and rapid identification of and response to respiratory symptoms, is associated with improved survival of residents who acquire pneumonia. Unfortunately, we also found that many of the 58 nursing homes in our study provided less than adequate care; for example, only 31% of residents received antibiotics within one 8-hour shift of symptom onset.¹¹

Convincing evidence indicates that treatment guidelines improve pneumonia outcomes in acute care settings,^12,13 and preliminary indications suggest that they may have a positive impact on processes of care in nursing facilities.¹⁴ Three guidelines on diagnosis and treatment of infections in nursing homes have been published in the last 2 years,^15-17but these guidelines are not specific to pneumonia. The only guideline specific to nursing home-acquired pneumonia used current community practice in 1 metropolitan area to define guidelines for antibiotic use.¹⁸ The outline specifically does not address issues of diagnostic work-up, decision to hospitalize, or timing of initial antibiotic treatment. We therefore convened a national panel of experts in infectious disease, pharmacology, pulmonology, geriatrics, and nursing to develop comprehensive evidence-based guidelines for management of nursing home-acquired pneumonia. The guidelines address immunization, infection control, timing and thoroughness of nurse and physician evaluation of lower respiratory tract infections, criteria for hospitalization, and criteria for antibiotic spectrum, timing, route, and duration.

Guideline development

Literature review

A comprehensive literature review was undertaken using Ovid (http://www.ovid.com/) and the “explode” version of each of the following key words: nursing homes, long-term care facilities, skilled nursing facilities, and pneumonia, to search MEDLINE from 1975 through 2000. Only studies published in English and germane to nursing home residents in the United States and Canada were examined. Results of the search were augmented by a local mulidisciplinary team, consisting of faculty in geriatric studies, infectious disease, pulmonology, and pharmacology at the University of Colorado Health Sciences Center, an internist who practices exclusively in nursing homes, and a nursing home nurse consultant. This team developed explicit criteria for nursing home-acquired pneumonia processes of care, as described previously.¹¹

Construction of draft guidelines

Working with the explicit criteria developed by the local team, the data from our retrospective study of process of care and survival, and the published literature on nursing home-acquired pneumonia, we drafted an initial set of comprehensive guidelines. We then graded the recommendations according to a standard system for defining quality, with 3 categories for recommendation strength (A, good evidence; B, moderate evidence; C, poor evidence) and 3 grades for quality of evidence (I, at least 1 properly randomized, controlled trial; II, at least 1 well-designed clinical trial without randomization from cohort or case-controlled analytic studies, multiple time series, or dramatic results in uncontrolled experiments; III, opinions of respected authorities based on clinical experience, descriptive studies, or reports of expert committees).¹⁹ Published Canadian and American guidelines for treatment of community-acquired and nosocomial pneumonia formed a template for the antibiotic recommendations.^20,21

Modified delphi process

A multidisciplinary, multispecialty panel was recruited from nationally recognized experts in nursing home-acquired pneumonia, geriatric and infectious disease pharmacology, pneumonia treatment guideline development, and nursing home nurses (see Appendix). Panel members received an annotated version of the draft guidelines with a bibliography and the questionnaire described below. A table of the activity spectrum of currently available antibiotics, a table summarizing published reports on the microbiology of nursing home-acquired pneumonia (see Table 1), and tables summarizing our retrospective study of care processes in nursing homes and 30-day survival¹¹were also included in that mailing.

The questionnaire asked whether each proposed guideline was clear, specific, feasible, measurable, and commensurate with his or her usual practice. Panelists were also asked to score each proposed guideline on its importance in determining the outcome of a nursing home-acquired pneumonia episode on a 5-point Likert scale, with 1 being not important and 5 being extremely important. Suggestions for substantive changes were elicited. The questionnaires were returned 2 weeks before an all-day, face-to-face meeting. For each guideline we calculated the percentage of panelists who agreed that it met each of the 5 criteria outlined. The means and standard deviations of the guideline weights were analyzed. Suggestions for substantive changes in the guidelines were collated and presented in tabular form, along with results of the survey, at the beginning of the meeting.

The most controversial guidelines were found to be classification of symptoms into a diagnosis of probable pneumonia, criteria for hospitalization, evaluation and treatment of residents not being hospitalized, and antibiotic choice. Less controversial were recommendations for immunization and treatment duration. The panel meeting devoted 2 hours to each of the 4 most controversial guidelines. The draft guideline was read and results of the pre-meeting ratings and suggestions were discussed. The guidelines were revised substantially and voted on. A care pathway was also developed at the meeting.

Because revisions were extensive and 1 of the panelists was participating by telephone, a copy of the revised guidelines and care pathway was sent to the panelists for further review and comment after the meeting. At that time, they were asked to rate each guideline on how confident they were that the recommendation should be included as part of the proposed guidelines using a 5-point Likert scale, with 1 denoting not confident and 5 denoting very confident. Because the lowest mean confidence rating was 3.4 after this iterative process, no guidelines were dropped from the set agreed upon at the meeting. Final revisions were made to the guidelines and their respective strength-and-quality-of-evidence grades, and approved by the panel.

TABLE 1
Nursing home pneumonia etiology according to studies using verified sputum* or blood culture

Recommendations

Each of the 25 guidelines is presented below. The strength and quality of evidence rating and panelists’ mean confidence score are shown in parentheses.

Prevention

1. Residents should be vaccinated against Streptococcus pneumoniae at admission unless there is documentation of vaccination within 5 years preceding admission or they were allergic to previous pneumococcal vaccine. (A/I/4.7)^22-27
2. Residents should be vaccinated against influenza by December of each year if they are not allergic to eggs or previous influenza vaccine. Residents admitted between December and March should be vaccinated if not already immunized for the current influenza season, and they are not allergic as described above. (A/I/5.0)^27,28
3. The nursing facility should provide and strongly recommend immunization against influenza for all employees by December of each year if the employee is not allergic to eggs or previous vaccine. (A/I/4.9)^29,30

Initial evaluation of residents with respiratory impairment

Once a resident has been noted to have a significant change in respiratory status, the clinician should use the care pathway outlined in (Figure 1). The panel believed that nurse practitioners and physician assistants with appropriate supervision could substitute for physician care in all pathway activities. Guideline recommendations 4 through 8 address the rapid recognition and physician notification of serious respiratory symptoms:

1. Physicians with nursing home residents should be available or have cross-coverage by pager 24 hours/day, 7 days/week. (Absent/Absent/4.8)
2. Nursing home staff should page the physician within 1 hour when a resident is noted to have any 2 of the following signs or symptoms: new or worsening cough; increased or newly purulent sputum; decline in cognitive, physical, or functional status; fever; hypothermia; dyspnea; tachypnea; chest pain; or new or worsening hypoxemia. (Absent/Absent/4.3)
3. Nurse evaluation at symptom onset should include, at least, vital signs (temperature, pulse rate, respiratory rate, and blood pressure) and oxygen saturation if a pulse oximeter is available in the facility. (Absent/Absent/4.5)
4. When notified as in guideline #5, the physician should call back within 1 hour. (Absent/Absent/4.3)
5. If the nurse does not hear back from the physician within 1 hour, he or she should notify the director of nurses or designee. The nurse and the director of nurses should agree on a plan to notify the medical director or designee and ask him or her to assume care of that episode until the medical director can contact the attending physician. (Absent/Absent/3.5)

Once the physician has been notified, he or she must decide whether pneumonia is a leading consideration in the diagnosis of the reported change.

1. The physician and nursing home staff should concur that pneumonia is a leading consideration in the diagnosis of the change noted in guideline #5 above, if the patient has 2 or more of the following signs or symptoms: new or worsening cough; newly purulent sputum; fever of 100.5°F or 2°F more than baseline; hypothermia < 96°F; dyspnea; respiratory rate > 25 breaths per minute; tachycardia; new or worse hypoxemia; pleuritic chest pain; a decline in cognitive or functional status; or new rales or rhonchi on chest examination. Such patients will be referred to in the remainder of the guidelines as having “probable pneumonia.” (C/III/4.1)^5,31

FIGURE
Care pathway for nursing home-acquired pneumonia

Venue of care

Once pneumonia is considered likely, the physician must decide whether to hospitalize the resident. Because nursing home-acquired pneumonia is frequently the terminal event of a long and debilitating illness, this decision is especially important.

1. The patient’s desire for hospitalization and aggressive care should be assessed directly if possible, or by chart review or discussion with the patient’s health care proxy. Patients with prior orders for no hospitalization or who refuse hospitalization (personally or by proxy) should not be hospitalized. (Absent/Absent/4.8)³²

For patients willing to be hospitalized, the panel suggested categorization into 1 of 3 groups: patients whose severity of illness mandates hospitalization; patients whose severity of illness is such that the nursing home’s capacity to deliver acute care should be considered; and patients so stable that treatment in the nursing home is preferable. This set of recommendations is based on 2 retrospective series which suggest that only patients with elevated respiratory rates benefit from hospitalization² and that patients hospitalized according to criteria similar to these tended to have better survival.¹¹ These recommendations are modeled after Fine’s prediction rule to identify low-risk patients with community-acquired pneumonia,³³ and take into account the mortality prediction model of Naughton and coworkers³⁴ for nursing home-acquired pneumonia.

1. Patients with 2 or more of the following symptoms should be hospitalized (C/III/4.1)^2,11:
2. If the nursing home cannot provide vital sign assessment every 4 hours, laboratory access, parenteral hydration, and 2 licensed nurses per shift in the facility, serious consideration should be given to hospitalizing patients with any one of the above. (C/III/4.1)^11,35,36
3. Patients with none of the above should be treated in the nursing home unless the patient or proxy insists on hospitalization. (B/II/4.5)^2,11,33,34

Evaluating and monitoring patients who remain in nursing homes

1. Nurse evaluation each shift should include, at least, vital signs with measured respiratory rate and oxygen saturation until symptoms resolve. (C/III/4.5)^16,17,37

Although a study published by Mehr and colleagues³⁸ after the panel proceedings suggested that patients with multiple signs and symptoms of nursing home-acquired pneumonia are so likely to have an infiltrate on chest x-ray that treating without first obtaining an x-ray may be acceptable, the panel recommended that:

1. Patients with probable pneumonia should have a chest x-ray. (C/III/4.2)^15,39
2. Patients with probable pneumonia should be evaluated in person by the physician, ideally within 24 hours and certainly within 72 hours. (Absent/Absent/4.5)
3. Because of the issue of antibiotic resistance, and not because of anticipated direct patient benefit, patients for whom antibiotics are ordered should have 1 blood culture drawn if this can be accomplished without delaying initiation of antibiotics longer than 1 hour. (C/III/3.4)^37,40

The panel did not recommend sputum Gram’s stain and culture even though 2 recent guidelines on evaluating infection in long-term care facilities did.^15,16 Obtaining high-quality sputa in this population is notoriously difficult.¹

TABLE 2
Key recommendations of evidence

Antibiotic use

Extrapolating from studies of hospital- and community-acquired pneumonia indicating a survival benefit for patients who receive antibiotics within 8 hours of arrival in hospital or symptom onset,⁴⁰ the panel made the following recommendations, which do not conflict with 3 recently published guidelines on use of antibiotics in long-term care facilities.¹⁶

1. Patients being sent immediately to the hospital should receive 1 dose of parenteral antibiotics prior to leaving the nursing home unless this treatment would delay transport for more than 1 hour. The hospital should be notified of the antibiotic given. (B/II/4.4)⁴⁰
2. A single dose of antibiotics should be ordered immediately (simultaneously with the chest x-ray order) if the patient has any of the following: systolic blood pressure < 90 mm Hg or 20 mm Hg less than baseline; oxygen saturation < 90% on room air at sea level; pulse > 130 beats per minute; respiratory rate > 30 breaths per minute; or temperature > 101.5°F. (B/II/4.5)^40,41
3. For patients with probable pneumonia who are stable (oxygen saturation > 90%, systolic blood pressure > 90 mm Hg, pulse < 120 beats per minute, and respiratory rate < 30 beats per minute), a decision to use antibiotics may await results of the chest x-ray if they will be available within 24 hours. (C/III/3.8)^40,41 If the chest x-ray does not confirm the suspicion of pneumonia, consideration should be given to stopping antibiotic treatment.
4. Antibiotics, if ordered, should be delivered to the patient within 4 hours of the orders being given by the physician. (B/II/4.5)^40,41
5. If the patient is able to take oral medication, that route is preferred, except that patients being sent to the hospital should have 1 parenteral dose. (A/I/4.7)^4,42,43

Because it is difficult to obtain a bacteriologic diagnosis in most nursing home residents, the initial choice of antibiotic must be empiric. Based on the published studies of nursing home-acquired pneumonia bacteriology, half of which used high-quality sputum samples (> 25 white blood cells per low-power field and < 10 squamous epithelial cells per low-power field, (see Table 1), the panel recommended a broad-spectrum antibiotic covering S pneumoniae, Haemophilus influenzae, gram-negative rods, and Staphylococcus aureus. In areas or facilities where resistance of S pneumoniae to penicillin is known to be high, treatment should be modified accordingly. These recommendations differ from those proposed by the Society of Health Care Epidemiology of America in their 2000 position paper,¹⁶ but are similar to those proposed by Naughton and Mylotte¹⁸ in their guideline derived from current community practice in Buffalo, New York, except that the current panel recommends the addition of a macrolide. These differences are driven by controversy over the relative importance of gram-negative rods as etiologic vs colonizing organisms in nursing home-acquired pneumonia; the relative importance of atypical organisms; and by the current panel’s concern that an important percentage of bacteria in long-term care facilities may already be resistant to many of the antibiotics endorsed by the Society of Health Care Epidemiology of America. Moreover, 3 studies have suggested a survival benefit of broader-spectrum empiric treatment in nursing home-acquired pneumonia.^11,44,45

1. The antibiotic chosen as empiric therapy needs to cover S pneumoniae, H influenzae, gram-neg-ative rods, and S aureus. (B/II/4.5)^{37,31,39,44-51}
2. Currently acceptable choices for empiric treatment include: antipneumococcal quinolones or an extended-spectrum beta-lactam plus a macrolide other than erythromycin. (C/III/4.5)^45,52,53
3. Treat for 10 to 14 days. (B/II/4.4)^18,25

Conclusions

A multidisciplinary, multispecialty panel of nationally recognized experts in nursing home-acquired pneumonia achieved consensus in proposing guidelines and a care pathway to prevent, recognize, evaluate, and treat nursing home-acquired pneumonia. For immunization and use of oral antibiotics when possible, evidence was sufficient to make a strong (A/I) recommendation. Even in areas in which evidence was lacking, the panel agreed readily. Thus for most of the 25 proposed guidelines, strong evidence and/or ready consensus existed among national experts from many disciplines.

The process of guideline development that we followed was systematic and meticulous. The panel was diverse, including nurses and pharmacologists, nationally recognized experts in infectious disease, nursing home care, geriatrics, and pulmonology. The breadth of the panel’s collective experience, and the rigor of the Delphi process we followed, lend face validity to the proposed guidelines despite some differences with other guidelines, as noted above.^45,53

These guidelines are uniquely comprehensive. They address management of pneumonia from prevention through diagnosis, the decision to hospitalize, and treatment. Such a comprehensive approach is essential given the many people and systems that interact in the care of nursing home residents, a uniquely frail and vulnerable population. Evidence from our work and others^11,13 suggests that only a comprehensive approach will, over time, improve the outcome of this important illness.

The guidelines have 2 important limitations. They have not yet been approved by any official professional society, nor have they been tested prospectively. Prospective testing is required to validate whether, in fact, the proposed guidelines can be implemented and, if implemented, they will improve mortality, function, cost of care, rehospitalization rates, and community discharge rates. We chose not to seek official approval of the guidelines until they can be proven effective.

We believe the proposed guidelines can and should be used to the fullest extent possible by nursing homes and physicians who practice there, because the combination of evidence and consensus is strong. Nursing home-acquired pneumonia causes excess mortality and functional loss. The proposed guidelines, based as they are in empiric evidence, common sense, and expert consensus, offer some hope of decreasing rehospitalization and cost, mitigating functional decline, and improving survival.

APPENDIX: PANEL MEMBERS . Diana Blake, RN, Western Hills Health Care Center, Lakewood, CO; JoAnn G. Congdon, RN, PhD, University of Colorado HSC, Denver, CO; Douglas Fish, PharmD, University of Colorado, Denver, CO; Evelyn Hutt, MD, University of Colorado HSC, Denver, CO; Denise Iskra, RN, Life Care of Westminster, Westminster, CO; Andrew M. Kramer, MD, University of Colorado, Denver, CO; Thomas J. Marrie, MD, University of Alberta, Edmonton, Alberta, Canada; Mark Loeb, MD, McMaster University, Hamilton, Ontario, Canada; Wayne C. McCormick, MD, MPH, Harborview Medical Center, Seattle, WA; Annette M. Medina-Walpole, MD, University of Rochester, Rochester, NY; David R. Mehr, MD, MS, University of Missouri–Columbia, Columbia, MO; Michael S. Niederman, MD, State University of New York at Stony Brook, Mineola, NY; Phillip K. Peterson, MD, University of Minnesota, Minneapolis, MN; J. Mark Ruscin, PharmD, University of Colorado, Denver, CO; Thomas T. Yoshikawa, MD, Charles Drew University, Los Angeles, CA.

Acknowledgments

The authors acknowledge the members of the panel for their thoughtful participation and willingness to review multiple drafts of the guidelines and manuscript; Steven Teutsch, MD, Senior Director, Outcomes Research and Management, Merck & Co., Inc., for his support and advice; Lisa Lampinen for manuscript preparation; and Merck & Co., Inc., for funding.

We convened a multidisciplinary, multispecialty panel to develop comprehensive evidence and consensus-based guidelines for managing nursing home-acquired pneumonia. The panel began with explicit criteria for process of care quality measures, performed a comprehensive review of the English-language literature, evaluated the quality of the evidence, and drafted a set of proposed guidelines. The panel reviewed the draft, an annotated bibliography, and data from a study of 30-day survival from nursing home-acquired pneumonia, and then participated in an all-day meeting in January 2001. Using a modified Delphi process, the panel refined the guidelines and developed a care pathway. The guidelines recommend a comprehensive approach, including immunization of staff and residents, and communication between nursing staff and the attending physician within 2 hours of symptom onset. Probable pneumonia was defined. An algorithm was delineated for assessing the patient’s wishes for hospitalization and aggressive care, and deciding on hospitalization based on the severity of the illness as well as the capacity of the nursing home to provide acute care. The timing and extent of evaluation in a nursing home relative to the rapid initiation of antibiotics should depend on whether the patient has any unstable vital signs. An antibiotic covering Streptococcus pneumoniae, Haemophilus influenzae, common gram-negative rods, and Staphylococcus aureus should be given for 10 to 14 days, orally if the patient is able to take medications by mouth.

Mortality from nursing home-acquired pneumonia is as high as 44%,¹and nearly a third of those who survive suffer significant functional decline.² Nursing home-acquired pneumonia is an entity distinct from community-acquired pneumonia in the elderly and nosocomial pneumonia. Older adults in nursing homes are more likely than community-dwelling older persons, but less likely than hospitalized elderly patients, to be colonized with gram-negative rods and pathogens with multiple antibiotic resistance.^3-8Inappropriate use of antibiotics in long-term care facilities contributes significantly to the growing problem of antibiotic resistance generally.⁹ In addition, most nursing home residents are cognitively impaired, immunocompromised, have multiple functional deficits, or have dysphagia, which further contributes to their vulnerability.¹⁰

In recent work¹¹ with a national nursing home sample, we showed that high quality care, including appropriate antibiotic use, hospitalization when indicated, and rapid identification of and response to respiratory symptoms, is associated with improved survival of residents who acquire pneumonia. Unfortunately, we also found that many of the 58 nursing homes in our study provided less than adequate care; for example, only 31% of residents received antibiotics within one 8-hour shift of symptom onset.¹¹

Convincing evidence indicates that treatment guidelines improve pneumonia outcomes in acute care settings,^12,13 and preliminary indications suggest that they may have a positive impact on processes of care in nursing facilities.¹⁴ Three guidelines on diagnosis and treatment of infections in nursing homes have been published in the last 2 years,^15-17but these guidelines are not specific to pneumonia. The only guideline specific to nursing home-acquired pneumonia used current community practice in 1 metropolitan area to define guidelines for antibiotic use.¹⁸ The outline specifically does not address issues of diagnostic work-up, decision to hospitalize, or timing of initial antibiotic treatment. We therefore convened a national panel of experts in infectious disease, pharmacology, pulmonology, geriatrics, and nursing to develop comprehensive evidence-based guidelines for management of nursing home-acquired pneumonia. The guidelines address immunization, infection control, timing and thoroughness of nurse and physician evaluation of lower respiratory tract infections, criteria for hospitalization, and criteria for antibiotic spectrum, timing, route, and duration.

Guideline development

Literature review

A comprehensive literature review was undertaken using Ovid (http://www.ovid.com/) and the “explode” version of each of the following key words: nursing homes, long-term care facilities, skilled nursing facilities, and pneumonia, to search MEDLINE from 1975 through 2000. Only studies published in English and germane to nursing home residents in the United States and Canada were examined. Results of the search were augmented by a local mulidisciplinary team, consisting of faculty in geriatric studies, infectious disease, pulmonology, and pharmacology at the University of Colorado Health Sciences Center, an internist who practices exclusively in nursing homes, and a nursing home nurse consultant. This team developed explicit criteria for nursing home-acquired pneumonia processes of care, as described previously.¹¹

Construction of draft guidelines

Working with the explicit criteria developed by the local team, the data from our retrospective study of process of care and survival, and the published literature on nursing home-acquired pneumonia, we drafted an initial set of comprehensive guidelines. We then graded the recommendations according to a standard system for defining quality, with 3 categories for recommendation strength (A, good evidence; B, moderate evidence; C, poor evidence) and 3 grades for quality of evidence (I, at least 1 properly randomized, controlled trial; II, at least 1 well-designed clinical trial without randomization from cohort or case-controlled analytic studies, multiple time series, or dramatic results in uncontrolled experiments; III, opinions of respected authorities based on clinical experience, descriptive studies, or reports of expert committees).¹⁹ Published Canadian and American guidelines for treatment of community-acquired and nosocomial pneumonia formed a template for the antibiotic recommendations.^20,21

Modified delphi process

A multidisciplinary, multispecialty panel was recruited from nationally recognized experts in nursing home-acquired pneumonia, geriatric and infectious disease pharmacology, pneumonia treatment guideline development, and nursing home nurses (see Appendix). Panel members received an annotated version of the draft guidelines with a bibliography and the questionnaire described below. A table of the activity spectrum of currently available antibiotics, a table summarizing published reports on the microbiology of nursing home-acquired pneumonia (see Table 1), and tables summarizing our retrospective study of care processes in nursing homes and 30-day survival¹¹were also included in that mailing.

The questionnaire asked whether each proposed guideline was clear, specific, feasible, measurable, and commensurate with his or her usual practice. Panelists were also asked to score each proposed guideline on its importance in determining the outcome of a nursing home-acquired pneumonia episode on a 5-point Likert scale, with 1 being not important and 5 being extremely important. Suggestions for substantive changes were elicited. The questionnaires were returned 2 weeks before an all-day, face-to-face meeting. For each guideline we calculated the percentage of panelists who agreed that it met each of the 5 criteria outlined. The means and standard deviations of the guideline weights were analyzed. Suggestions for substantive changes in the guidelines were collated and presented in tabular form, along with results of the survey, at the beginning of the meeting.

The most controversial guidelines were found to be classification of symptoms into a diagnosis of probable pneumonia, criteria for hospitalization, evaluation and treatment of residents not being hospitalized, and antibiotic choice. Less controversial were recommendations for immunization and treatment duration. The panel meeting devoted 2 hours to each of the 4 most controversial guidelines. The draft guideline was read and results of the pre-meeting ratings and suggestions were discussed. The guidelines were revised substantially and voted on. A care pathway was also developed at the meeting.

Because revisions were extensive and 1 of the panelists was participating by telephone, a copy of the revised guidelines and care pathway was sent to the panelists for further review and comment after the meeting. At that time, they were asked to rate each guideline on how confident they were that the recommendation should be included as part of the proposed guidelines using a 5-point Likert scale, with 1 denoting not confident and 5 denoting very confident. Because the lowest mean confidence rating was 3.4 after this iterative process, no guidelines were dropped from the set agreed upon at the meeting. Final revisions were made to the guidelines and their respective strength-and-quality-of-evidence grades, and approved by the panel.

TABLE 1
Nursing home pneumonia etiology according to studies using verified sputum* or blood culture

Recommendations

Each of the 25 guidelines is presented below. The strength and quality of evidence rating and panelists’ mean confidence score are shown in parentheses.

Prevention

1. Residents should be vaccinated against Streptococcus pneumoniae at admission unless there is documentation of vaccination within 5 years preceding admission or they were allergic to previous pneumococcal vaccine. (A/I/4.7)^22-27
2. Residents should be vaccinated against influenza by December of each year if they are not allergic to eggs or previous influenza vaccine. Residents admitted between December and March should be vaccinated if not already immunized for the current influenza season, and they are not allergic as described above. (A/I/5.0)^27,28
3. The nursing facility should provide and strongly recommend immunization against influenza for all employees by December of each year if the employee is not allergic to eggs or previous vaccine. (A/I/4.9)^29,30

Initial evaluation of residents with respiratory impairment

Once a resident has been noted to have a significant change in respiratory status, the clinician should use the care pathway outlined in (Figure 1). The panel believed that nurse practitioners and physician assistants with appropriate supervision could substitute for physician care in all pathway activities. Guideline recommendations 4 through 8 address the rapid recognition and physician notification of serious respiratory symptoms:

1. Physicians with nursing home residents should be available or have cross-coverage by pager 24 hours/day, 7 days/week. (Absent/Absent/4.8)
2. Nursing home staff should page the physician within 1 hour when a resident is noted to have any 2 of the following signs or symptoms: new or worsening cough; increased or newly purulent sputum; decline in cognitive, physical, or functional status; fever; hypothermia; dyspnea; tachypnea; chest pain; or new or worsening hypoxemia. (Absent/Absent/4.3)
3. Nurse evaluation at symptom onset should include, at least, vital signs (temperature, pulse rate, respiratory rate, and blood pressure) and oxygen saturation if a pulse oximeter is available in the facility. (Absent/Absent/4.5)
4. When notified as in guideline #5, the physician should call back within 1 hour. (Absent/Absent/4.3)
5. If the nurse does not hear back from the physician within 1 hour, he or she should notify the director of nurses or designee. The nurse and the director of nurses should agree on a plan to notify the medical director or designee and ask him or her to assume care of that episode until the medical director can contact the attending physician. (Absent/Absent/3.5)

Once the physician has been notified, he or she must decide whether pneumonia is a leading consideration in the diagnosis of the reported change.

1. The physician and nursing home staff should concur that pneumonia is a leading consideration in the diagnosis of the change noted in guideline #5 above, if the patient has 2 or more of the following signs or symptoms: new or worsening cough; newly purulent sputum; fever of 100.5°F or 2°F more than baseline; hypothermia < 96°F; dyspnea; respiratory rate > 25 breaths per minute; tachycardia; new or worse hypoxemia; pleuritic chest pain; a decline in cognitive or functional status; or new rales or rhonchi on chest examination. Such patients will be referred to in the remainder of the guidelines as having “probable pneumonia.” (C/III/4.1)^5,31

FIGURE
Care pathway for nursing home-acquired pneumonia

Venue of care

Once pneumonia is considered likely, the physician must decide whether to hospitalize the resident. Because nursing home-acquired pneumonia is frequently the terminal event of a long and debilitating illness, this decision is especially important.

1. The patient’s desire for hospitalization and aggressive care should be assessed directly if possible, or by chart review or discussion with the patient’s health care proxy. Patients with prior orders for no hospitalization or who refuse hospitalization (personally or by proxy) should not be hospitalized. (Absent/Absent/4.8)³²

For patients willing to be hospitalized, the panel suggested categorization into 1 of 3 groups: patients whose severity of illness mandates hospitalization; patients whose severity of illness is such that the nursing home’s capacity to deliver acute care should be considered; and patients so stable that treatment in the nursing home is preferable. This set of recommendations is based on 2 retrospective series which suggest that only patients with elevated respiratory rates benefit from hospitalization² and that patients hospitalized according to criteria similar to these tended to have better survival.¹¹ These recommendations are modeled after Fine’s prediction rule to identify low-risk patients with community-acquired pneumonia,³³ and take into account the mortality prediction model of Naughton and coworkers³⁴ for nursing home-acquired pneumonia.

1. Patients with 2 or more of the following symptoms should be hospitalized (C/III/4.1)^2,11:
2. If the nursing home cannot provide vital sign assessment every 4 hours, laboratory access, parenteral hydration, and 2 licensed nurses per shift in the facility, serious consideration should be given to hospitalizing patients with any one of the above. (C/III/4.1)^11,35,36
3. Patients with none of the above should be treated in the nursing home unless the patient or proxy insists on hospitalization. (B/II/4.5)^2,11,33,34

Evaluating and monitoring patients who remain in nursing homes

1. Nurse evaluation each shift should include, at least, vital signs with measured respiratory rate and oxygen saturation until symptoms resolve. (C/III/4.5)^16,17,37

Although a study published by Mehr and colleagues³⁸ after the panel proceedings suggested that patients with multiple signs and symptoms of nursing home-acquired pneumonia are so likely to have an infiltrate on chest x-ray that treating without first obtaining an x-ray may be acceptable, the panel recommended that:

1. Patients with probable pneumonia should have a chest x-ray. (C/III/4.2)^15,39
2. Patients with probable pneumonia should be evaluated in person by the physician, ideally within 24 hours and certainly within 72 hours. (Absent/Absent/4.5)
3. Because of the issue of antibiotic resistance, and not because of anticipated direct patient benefit, patients for whom antibiotics are ordered should have 1 blood culture drawn if this can be accomplished without delaying initiation of antibiotics longer than 1 hour. (C/III/3.4)^37,40

The panel did not recommend sputum Gram’s stain and culture even though 2 recent guidelines on evaluating infection in long-term care facilities did.^15,16 Obtaining high-quality sputa in this population is notoriously difficult.¹

TABLE 2
Key recommendations of evidence

Antibiotic use

Extrapolating from studies of hospital- and community-acquired pneumonia indicating a survival benefit for patients who receive antibiotics within 8 hours of arrival in hospital or symptom onset,⁴⁰ the panel made the following recommendations, which do not conflict with 3 recently published guidelines on use of antibiotics in long-term care facilities.¹⁶

1. Patients being sent immediately to the hospital should receive 1 dose of parenteral antibiotics prior to leaving the nursing home unless this treatment would delay transport for more than 1 hour. The hospital should be notified of the antibiotic given. (B/II/4.4)⁴⁰
2. A single dose of antibiotics should be ordered immediately (simultaneously with the chest x-ray order) if the patient has any of the following: systolic blood pressure < 90 mm Hg or 20 mm Hg less than baseline; oxygen saturation < 90% on room air at sea level; pulse > 130 beats per minute; respiratory rate > 30 breaths per minute; or temperature > 101.5°F. (B/II/4.5)^40,41
3. For patients with probable pneumonia who are stable (oxygen saturation > 90%, systolic blood pressure > 90 mm Hg, pulse < 120 beats per minute, and respiratory rate < 30 beats per minute), a decision to use antibiotics may await results of the chest x-ray if they will be available within 24 hours. (C/III/3.8)^40,41 If the chest x-ray does not confirm the suspicion of pneumonia, consideration should be given to stopping antibiotic treatment.
4. Antibiotics, if ordered, should be delivered to the patient within 4 hours of the orders being given by the physician. (B/II/4.5)^40,41
5. If the patient is able to take oral medication, that route is preferred, except that patients being sent to the hospital should have 1 parenteral dose. (A/I/4.7)^4,42,43

Because it is difficult to obtain a bacteriologic diagnosis in most nursing home residents, the initial choice of antibiotic must be empiric. Based on the published studies of nursing home-acquired pneumonia bacteriology, half of which used high-quality sputum samples (> 25 white blood cells per low-power field and < 10 squamous epithelial cells per low-power field, (see Table 1), the panel recommended a broad-spectrum antibiotic covering S pneumoniae, Haemophilus influenzae, gram-negative rods, and Staphylococcus aureus. In areas or facilities where resistance of S pneumoniae to penicillin is known to be high, treatment should be modified accordingly. These recommendations differ from those proposed by the Society of Health Care Epidemiology of America in their 2000 position paper,¹⁶ but are similar to those proposed by Naughton and Mylotte¹⁸ in their guideline derived from current community practice in Buffalo, New York, except that the current panel recommends the addition of a macrolide. These differences are driven by controversy over the relative importance of gram-negative rods as etiologic vs colonizing organisms in nursing home-acquired pneumonia; the relative importance of atypical organisms; and by the current panel’s concern that an important percentage of bacteria in long-term care facilities may already be resistant to many of the antibiotics endorsed by the Society of Health Care Epidemiology of America. Moreover, 3 studies have suggested a survival benefit of broader-spectrum empiric treatment in nursing home-acquired pneumonia.^11,44,45

1. The antibiotic chosen as empiric therapy needs to cover S pneumoniae, H influenzae, gram-neg-ative rods, and S aureus. (B/II/4.5)^{37,31,39,44-51}
2. Currently acceptable choices for empiric treatment include: antipneumococcal quinolones or an extended-spectrum beta-lactam plus a macrolide other than erythromycin. (C/III/4.5)^45,52,53
3. Treat for 10 to 14 days. (B/II/4.4)^18,25

Conclusions

A multidisciplinary, multispecialty panel of nationally recognized experts in nursing home-acquired pneumonia achieved consensus in proposing guidelines and a care pathway to prevent, recognize, evaluate, and treat nursing home-acquired pneumonia. For immunization and use of oral antibiotics when possible, evidence was sufficient to make a strong (A/I) recommendation. Even in areas in which evidence was lacking, the panel agreed readily. Thus for most of the 25 proposed guidelines, strong evidence and/or ready consensus existed among national experts from many disciplines.

The process of guideline development that we followed was systematic and meticulous. The panel was diverse, including nurses and pharmacologists, nationally recognized experts in infectious disease, nursing home care, geriatrics, and pulmonology. The breadth of the panel’s collective experience, and the rigor of the Delphi process we followed, lend face validity to the proposed guidelines despite some differences with other guidelines, as noted above.^45,53

These guidelines are uniquely comprehensive. They address management of pneumonia from prevention through diagnosis, the decision to hospitalize, and treatment. Such a comprehensive approach is essential given the many people and systems that interact in the care of nursing home residents, a uniquely frail and vulnerable population. Evidence from our work and others^11,13 suggests that only a comprehensive approach will, over time, improve the outcome of this important illness.

The guidelines have 2 important limitations. They have not yet been approved by any official professional society, nor have they been tested prospectively. Prospective testing is required to validate whether, in fact, the proposed guidelines can be implemented and, if implemented, they will improve mortality, function, cost of care, rehospitalization rates, and community discharge rates. We chose not to seek official approval of the guidelines until they can be proven effective.

We believe the proposed guidelines can and should be used to the fullest extent possible by nursing homes and physicians who practice there, because the combination of evidence and consensus is strong. Nursing home-acquired pneumonia causes excess mortality and functional loss. The proposed guidelines, based as they are in empiric evidence, common sense, and expert consensus, offer some hope of decreasing rehospitalization and cost, mitigating functional decline, and improving survival.

APPENDIX: PANEL MEMBERS . Diana Blake, RN, Western Hills Health Care Center, Lakewood, CO; JoAnn G. Congdon, RN, PhD, University of Colorado HSC, Denver, CO; Douglas Fish, PharmD, University of Colorado, Denver, CO; Evelyn Hutt, MD, University of Colorado HSC, Denver, CO; Denise Iskra, RN, Life Care of Westminster, Westminster, CO; Andrew M. Kramer, MD, University of Colorado, Denver, CO; Thomas J. Marrie, MD, University of Alberta, Edmonton, Alberta, Canada; Mark Loeb, MD, McMaster University, Hamilton, Ontario, Canada; Wayne C. McCormick, MD, MPH, Harborview Medical Center, Seattle, WA; Annette M. Medina-Walpole, MD, University of Rochester, Rochester, NY; David R. Mehr, MD, MS, University of Missouri–Columbia, Columbia, MO; Michael S. Niederman, MD, State University of New York at Stony Brook, Mineola, NY; Phillip K. Peterson, MD, University of Minnesota, Minneapolis, MN; J. Mark Ruscin, PharmD, University of Colorado, Denver, CO; Thomas T. Yoshikawa, MD, Charles Drew University, Los Angeles, CA.

Acknowledgments

The authors acknowledge the members of the panel for their thoughtful participation and willingness to review multiple drafts of the guidelines and manuscript; Steven Teutsch, MD, Senior Director, Outcomes Research and Management, Merck & Co., Inc., for his support and advice; Lisa Lampinen for manuscript preparation; and Merck & Co., Inc., for funding.

We convened a multidisciplinary, multispecialty panel to develop comprehensive evidence and consensus-based guidelines for managing nursing home-acquired pneumonia. The panel began with explicit criteria for process of care quality measures, performed a comprehensive review of the English-language literature, evaluated the quality of the evidence, and drafted a set of proposed guidelines. The panel reviewed the draft, an annotated bibliography, and data from a study of 30-day survival from nursing home-acquired pneumonia, and then participated in an all-day meeting in January 2001. Using a modified Delphi process, the panel refined the guidelines and developed a care pathway. The guidelines recommend a comprehensive approach, including immunization of staff and residents, and communication between nursing staff and the attending physician within 2 hours of symptom onset. Probable pneumonia was defined. An algorithm was delineated for assessing the patient’s wishes for hospitalization and aggressive care, and deciding on hospitalization based on the severity of the illness as well as the capacity of the nursing home to provide acute care. The timing and extent of evaluation in a nursing home relative to the rapid initiation of antibiotics should depend on whether the patient has any unstable vital signs. An antibiotic covering Streptococcus pneumoniae, Haemophilus influenzae, common gram-negative rods, and Staphylococcus aureus should be given for 10 to 14 days, orally if the patient is able to take medications by mouth.

Mortality from nursing home-acquired pneumonia is as high as 44%,¹and nearly a third of those who survive suffer significant functional decline.² Nursing home-acquired pneumonia is an entity distinct from community-acquired pneumonia in the elderly and nosocomial pneumonia. Older adults in nursing homes are more likely than community-dwelling older persons, but less likely than hospitalized elderly patients, to be colonized with gram-negative rods and pathogens with multiple antibiotic resistance.^3-8Inappropriate use of antibiotics in long-term care facilities contributes significantly to the growing problem of antibiotic resistance generally.⁹ In addition, most nursing home residents are cognitively impaired, immunocompromised, have multiple functional deficits, or have dysphagia, which further contributes to their vulnerability.¹⁰

In recent work¹¹ with a national nursing home sample, we showed that high quality care, including appropriate antibiotic use, hospitalization when indicated, and rapid identification of and response to respiratory symptoms, is associated with improved survival of residents who acquire pneumonia. Unfortunately, we also found that many of the 58 nursing homes in our study provided less than adequate care; for example, only 31% of residents received antibiotics within one 8-hour shift of symptom onset.¹¹

Convincing evidence indicates that treatment guidelines improve pneumonia outcomes in acute care settings,^12,13 and preliminary indications suggest that they may have a positive impact on processes of care in nursing facilities.¹⁴ Three guidelines on diagnosis and treatment of infections in nursing homes have been published in the last 2 years,^15-17but these guidelines are not specific to pneumonia. The only guideline specific to nursing home-acquired pneumonia used current community practice in 1 metropolitan area to define guidelines for antibiotic use.¹⁸ The outline specifically does not address issues of diagnostic work-up, decision to hospitalize, or timing of initial antibiotic treatment. We therefore convened a national panel of experts in infectious disease, pharmacology, pulmonology, geriatrics, and nursing to develop comprehensive evidence-based guidelines for management of nursing home-acquired pneumonia. The guidelines address immunization, infection control, timing and thoroughness of nurse and physician evaluation of lower respiratory tract infections, criteria for hospitalization, and criteria for antibiotic spectrum, timing, route, and duration.

Guideline development

Literature review

A comprehensive literature review was undertaken using Ovid (http://www.ovid.com/) and the “explode” version of each of the following key words: nursing homes, long-term care facilities, skilled nursing facilities, and pneumonia, to search MEDLINE from 1975 through 2000. Only studies published in English and germane to nursing home residents in the United States and Canada were examined. Results of the search were augmented by a local mulidisciplinary team, consisting of faculty in geriatric studies, infectious disease, pulmonology, and pharmacology at the University of Colorado Health Sciences Center, an internist who practices exclusively in nursing homes, and a nursing home nurse consultant. This team developed explicit criteria for nursing home-acquired pneumonia processes of care, as described previously.¹¹

Construction of draft guidelines

Working with the explicit criteria developed by the local team, the data from our retrospective study of process of care and survival, and the published literature on nursing home-acquired pneumonia, we drafted an initial set of comprehensive guidelines. We then graded the recommendations according to a standard system for defining quality, with 3 categories for recommendation strength (A, good evidence; B, moderate evidence; C, poor evidence) and 3 grades for quality of evidence (I, at least 1 properly randomized, controlled trial; II, at least 1 well-designed clinical trial without randomization from cohort or case-controlled analytic studies, multiple time series, or dramatic results in uncontrolled experiments; III, opinions of respected authorities based on clinical experience, descriptive studies, or reports of expert committees).¹⁹ Published Canadian and American guidelines for treatment of community-acquired and nosocomial pneumonia formed a template for the antibiotic recommendations.^20,21

Modified delphi process

A multidisciplinary, multispecialty panel was recruited from nationally recognized experts in nursing home-acquired pneumonia, geriatric and infectious disease pharmacology, pneumonia treatment guideline development, and nursing home nurses (see Appendix). Panel members received an annotated version of the draft guidelines with a bibliography and the questionnaire described below. A table of the activity spectrum of currently available antibiotics, a table summarizing published reports on the microbiology of nursing home-acquired pneumonia (see Table 1), and tables summarizing our retrospective study of care processes in nursing homes and 30-day survival¹¹were also included in that mailing.

The questionnaire asked whether each proposed guideline was clear, specific, feasible, measurable, and commensurate with his or her usual practice. Panelists were also asked to score each proposed guideline on its importance in determining the outcome of a nursing home-acquired pneumonia episode on a 5-point Likert scale, with 1 being not important and 5 being extremely important. Suggestions for substantive changes were elicited. The questionnaires were returned 2 weeks before an all-day, face-to-face meeting. For each guideline we calculated the percentage of panelists who agreed that it met each of the 5 criteria outlined. The means and standard deviations of the guideline weights were analyzed. Suggestions for substantive changes in the guidelines were collated and presented in tabular form, along with results of the survey, at the beginning of the meeting.

The most controversial guidelines were found to be classification of symptoms into a diagnosis of probable pneumonia, criteria for hospitalization, evaluation and treatment of residents not being hospitalized, and antibiotic choice. Less controversial were recommendations for immunization and treatment duration. The panel meeting devoted 2 hours to each of the 4 most controversial guidelines. The draft guideline was read and results of the pre-meeting ratings and suggestions were discussed. The guidelines were revised substantially and voted on. A care pathway was also developed at the meeting.

Because revisions were extensive and 1 of the panelists was participating by telephone, a copy of the revised guidelines and care pathway was sent to the panelists for further review and comment after the meeting. At that time, they were asked to rate each guideline on how confident they were that the recommendation should be included as part of the proposed guidelines using a 5-point Likert scale, with 1 denoting not confident and 5 denoting very confident. Because the lowest mean confidence rating was 3.4 after this iterative process, no guidelines were dropped from the set agreed upon at the meeting. Final revisions were made to the guidelines and their respective strength-and-quality-of-evidence grades, and approved by the panel.

TABLE 1
Nursing home pneumonia etiology according to studies using verified sputum* or blood culture

Recommendations

Each of the 25 guidelines is presented below. The strength and quality of evidence rating and panelists’ mean confidence score are shown in parentheses.

Prevention

1. Residents should be vaccinated against Streptococcus pneumoniae at admission unless there is documentation of vaccination within 5 years preceding admission or they were allergic to previous pneumococcal vaccine. (A/I/4.7)^22-27
2. Residents should be vaccinated against influenza by December of each year if they are not allergic to eggs or previous influenza vaccine. Residents admitted between December and March should be vaccinated if not already immunized for the current influenza season, and they are not allergic as described above. (A/I/5.0)^27,28
3. The nursing facility should provide and strongly recommend immunization against influenza for all employees by December of each year if the employee is not allergic to eggs or previous vaccine. (A/I/4.9)^29,30

Initial evaluation of residents with respiratory impairment

Once a resident has been noted to have a significant change in respiratory status, the clinician should use the care pathway outlined in (Figure 1). The panel believed that nurse practitioners and physician assistants with appropriate supervision could substitute for physician care in all pathway activities. Guideline recommendations 4 through 8 address the rapid recognition and physician notification of serious respiratory symptoms:

1. Physicians with nursing home residents should be available or have cross-coverage by pager 24 hours/day, 7 days/week. (Absent/Absent/4.8)
2. Nursing home staff should page the physician within 1 hour when a resident is noted to have any 2 of the following signs or symptoms: new or worsening cough; increased or newly purulent sputum; decline in cognitive, physical, or functional status; fever; hypothermia; dyspnea; tachypnea; chest pain; or new or worsening hypoxemia. (Absent/Absent/4.3)
3. Nurse evaluation at symptom onset should include, at least, vital signs (temperature, pulse rate, respiratory rate, and blood pressure) and oxygen saturation if a pulse oximeter is available in the facility. (Absent/Absent/4.5)
4. When notified as in guideline #5, the physician should call back within 1 hour. (Absent/Absent/4.3)
5. If the nurse does not hear back from the physician within 1 hour, he or she should notify the director of nurses or designee. The nurse and the director of nurses should agree on a plan to notify the medical director or designee and ask him or her to assume care of that episode until the medical director can contact the attending physician. (Absent/Absent/3.5)

Once the physician has been notified, he or she must decide whether pneumonia is a leading consideration in the diagnosis of the reported change.

1. The physician and nursing home staff should concur that pneumonia is a leading consideration in the diagnosis of the change noted in guideline #5 above, if the patient has 2 or more of the following signs or symptoms: new or worsening cough; newly purulent sputum; fever of 100.5°F or 2°F more than baseline; hypothermia < 96°F; dyspnea; respiratory rate > 25 breaths per minute; tachycardia; new or worse hypoxemia; pleuritic chest pain; a decline in cognitive or functional status; or new rales or rhonchi on chest examination. Such patients will be referred to in the remainder of the guidelines as having “probable pneumonia.” (C/III/4.1)^5,31

FIGURE
Care pathway for nursing home-acquired pneumonia

Venue of care

Once pneumonia is considered likely, the physician must decide whether to hospitalize the resident. Because nursing home-acquired pneumonia is frequently the terminal event of a long and debilitating illness, this decision is especially important.

1. The patient’s desire for hospitalization and aggressive care should be assessed directly if possible, or by chart review or discussion with the patient’s health care proxy. Patients with prior orders for no hospitalization or who refuse hospitalization (personally or by proxy) should not be hospitalized. (Absent/Absent/4.8)³²

For patients willing to be hospitalized, the panel suggested categorization into 1 of 3 groups: patients whose severity of illness mandates hospitalization; patients whose severity of illness is such that the nursing home’s capacity to deliver acute care should be considered; and patients so stable that treatment in the nursing home is preferable. This set of recommendations is based on 2 retrospective series which suggest that only patients with elevated respiratory rates benefit from hospitalization² and that patients hospitalized according to criteria similar to these tended to have better survival.¹¹ These recommendations are modeled after Fine’s prediction rule to identify low-risk patients with community-acquired pneumonia,³³ and take into account the mortality prediction model of Naughton and coworkers³⁴ for nursing home-acquired pneumonia.

1. Patients with 2 or more of the following symptoms should be hospitalized (C/III/4.1)^2,11:
2. If the nursing home cannot provide vital sign assessment every 4 hours, laboratory access, parenteral hydration, and 2 licensed nurses per shift in the facility, serious consideration should be given to hospitalizing patients with any one of the above. (C/III/4.1)^11,35,36
3. Patients with none of the above should be treated in the nursing home unless the patient or proxy insists on hospitalization. (B/II/4.5)^2,11,33,34

Evaluating and monitoring patients who remain in nursing homes

1. Nurse evaluation each shift should include, at least, vital signs with measured respiratory rate and oxygen saturation until symptoms resolve. (C/III/4.5)^16,17,37

Although a study published by Mehr and colleagues³⁸ after the panel proceedings suggested that patients with multiple signs and symptoms of nursing home-acquired pneumonia are so likely to have an infiltrate on chest x-ray that treating without first obtaining an x-ray may be acceptable, the panel recommended that:

1. Patients with probable pneumonia should have a chest x-ray. (C/III/4.2)^15,39
2. Patients with probable pneumonia should be evaluated in person by the physician, ideally within 24 hours and certainly within 72 hours. (Absent/Absent/4.5)
3. Because of the issue of antibiotic resistance, and not because of anticipated direct patient benefit, patients for whom antibiotics are ordered should have 1 blood culture drawn if this can be accomplished without delaying initiation of antibiotics longer than 1 hour. (C/III/3.4)^37,40

The panel did not recommend sputum Gram’s stain and culture even though 2 recent guidelines on evaluating infection in long-term care facilities did.^15,16 Obtaining high-quality sputa in this population is notoriously difficult.¹

TABLE 2
Key recommendations of evidence

Antibiotic use

Extrapolating from studies of hospital- and community-acquired pneumonia indicating a survival benefit for patients who receive antibiotics within 8 hours of arrival in hospital or symptom onset,⁴⁰ the panel made the following recommendations, which do not conflict with 3 recently published guidelines on use of antibiotics in long-term care facilities.¹⁶

1. Patients being sent immediately to the hospital should receive 1 dose of parenteral antibiotics prior to leaving the nursing home unless this treatment would delay transport for more than 1 hour. The hospital should be notified of the antibiotic given. (B/II/4.4)⁴⁰
2. A single dose of antibiotics should be ordered immediately (simultaneously with the chest x-ray order) if the patient has any of the following: systolic blood pressure < 90 mm Hg or 20 mm Hg less than baseline; oxygen saturation < 90% on room air at sea level; pulse > 130 beats per minute; respiratory rate > 30 breaths per minute; or temperature > 101.5°F. (B/II/4.5)^40,41
3. For patients with probable pneumonia who are stable (oxygen saturation > 90%, systolic blood pressure > 90 mm Hg, pulse < 120 beats per minute, and respiratory rate < 30 beats per minute), a decision to use antibiotics may await results of the chest x-ray if they will be available within 24 hours. (C/III/3.8)^40,41 If the chest x-ray does not confirm the suspicion of pneumonia, consideration should be given to stopping antibiotic treatment.
4. Antibiotics, if ordered, should be delivered to the patient within 4 hours of the orders being given by the physician. (B/II/4.5)^40,41
5. If the patient is able to take oral medication, that route is preferred, except that patients being sent to the hospital should have 1 parenteral dose. (A/I/4.7)^4,42,43

Because it is difficult to obtain a bacteriologic diagnosis in most nursing home residents, the initial choice of antibiotic must be empiric. Based on the published studies of nursing home-acquired pneumonia bacteriology, half of which used high-quality sputum samples (> 25 white blood cells per low-power field and < 10 squamous epithelial cells per low-power field, (see Table 1), the panel recommended a broad-spectrum antibiotic covering S pneumoniae, Haemophilus influenzae, gram-negative rods, and Staphylococcus aureus. In areas or facilities where resistance of S pneumoniae to penicillin is known to be high, treatment should be modified accordingly. These recommendations differ from those proposed by the Society of Health Care Epidemiology of America in their 2000 position paper,¹⁶ but are similar to those proposed by Naughton and Mylotte¹⁸ in their guideline derived from current community practice in Buffalo, New York, except that the current panel recommends the addition of a macrolide. These differences are driven by controversy over the relative importance of gram-negative rods as etiologic vs colonizing organisms in nursing home-acquired pneumonia; the relative importance of atypical organisms; and by the current panel’s concern that an important percentage of bacteria in long-term care facilities may already be resistant to many of the antibiotics endorsed by the Society of Health Care Epidemiology of America. Moreover, 3 studies have suggested a survival benefit of broader-spectrum empiric treatment in nursing home-acquired pneumonia.^11,44,45

1. The antibiotic chosen as empiric therapy needs to cover S pneumoniae, H influenzae, gram-neg-ative rods, and S aureus. (B/II/4.5)^{37,31,39,44-51}
2. Currently acceptable choices for empiric treatment include: antipneumococcal quinolones or an extended-spectrum beta-lactam plus a macrolide other than erythromycin. (C/III/4.5)^45,52,53
3. Treat for 10 to 14 days. (B/II/4.4)^18,25

Conclusions

A multidisciplinary, multispecialty panel of nationally recognized experts in nursing home-acquired pneumonia achieved consensus in proposing guidelines and a care pathway to prevent, recognize, evaluate, and treat nursing home-acquired pneumonia. For immunization and use of oral antibiotics when possible, evidence was sufficient to make a strong (A/I) recommendation. Even in areas in which evidence was lacking, the panel agreed readily. Thus for most of the 25 proposed guidelines, strong evidence and/or ready consensus existed among national experts from many disciplines.

The process of guideline development that we followed was systematic and meticulous. The panel was diverse, including nurses and pharmacologists, nationally recognized experts in infectious disease, nursing home care, geriatrics, and pulmonology. The breadth of the panel’s collective experience, and the rigor of the Delphi process we followed, lend face validity to the proposed guidelines despite some differences with other guidelines, as noted above.^45,53

These guidelines are uniquely comprehensive. They address management of pneumonia from prevention through diagnosis, the decision to hospitalize, and treatment. Such a comprehensive approach is essential given the many people and systems that interact in the care of nursing home residents, a uniquely frail and vulnerable population. Evidence from our work and others^11,13 suggests that only a comprehensive approach will, over time, improve the outcome of this important illness.

The guidelines have 2 important limitations. They have not yet been approved by any official professional society, nor have they been tested prospectively. Prospective testing is required to validate whether, in fact, the proposed guidelines can be implemented and, if implemented, they will improve mortality, function, cost of care, rehospitalization rates, and community discharge rates. We chose not to seek official approval of the guidelines until they can be proven effective.

We believe the proposed guidelines can and should be used to the fullest extent possible by nursing homes and physicians who practice there, because the combination of evidence and consensus is strong. Nursing home-acquired pneumonia causes excess mortality and functional loss. The proposed guidelines, based as they are in empiric evidence, common sense, and expert consensus, offer some hope of decreasing rehospitalization and cost, mitigating functional decline, and improving survival.

APPENDIX: PANEL MEMBERS . Diana Blake, RN, Western Hills Health Care Center, Lakewood, CO; JoAnn G. Congdon, RN, PhD, University of Colorado HSC, Denver, CO; Douglas Fish, PharmD, University of Colorado, Denver, CO; Evelyn Hutt, MD, University of Colorado HSC, Denver, CO; Denise Iskra, RN, Life Care of Westminster, Westminster, CO; Andrew M. Kramer, MD, University of Colorado, Denver, CO; Thomas J. Marrie, MD, University of Alberta, Edmonton, Alberta, Canada; Mark Loeb, MD, McMaster University, Hamilton, Ontario, Canada; Wayne C. McCormick, MD, MPH, Harborview Medical Center, Seattle, WA; Annette M. Medina-Walpole, MD, University of Rochester, Rochester, NY; David R. Mehr, MD, MS, University of Missouri–Columbia, Columbia, MO; Michael S. Niederman, MD, State University of New York at Stony Brook, Mineola, NY; Phillip K. Peterson, MD, University of Minnesota, Minneapolis, MN; J. Mark Ruscin, PharmD, University of Colorado, Denver, CO; Thomas T. Yoshikawa, MD, Charles Drew University, Los Angeles, CA.

Acknowledgments

The authors acknowledge the members of the panel for their thoughtful participation and willingness to review multiple drafts of the guidelines and manuscript; Steven Teutsch, MD, Senior Director, Outcomes Research and Management, Merck & Co., Inc., for his support and advice; Lisa Lampinen for manuscript preparation; and Merck & Co., Inc., for funding.

Cardiovascular disease is the leading cause of death in the United States, and aspirin, a platelet aggregate inhibitor, is often recommended as prophylaxis for cardiovascular disease.^1-3Clinical studies have demonstrated the benefit of aspirin use for secondary prevention of cardiovascular disease and stroke.^1,4-10 In high-risk subjects, aspirin has been proven effective in primary prevention of major cardiovascular events and nonfatal ischemic heart disease.^11-13 Sanmuganathan and colleagues recently reported a meta-analysis of 4 randomized trials of aspirin for primary prevention. Although they determined that aspirin treatment is safe if the coronary event rate is at least 1.5% each year and unsafe if the rate is no higher than 0.5% each year, they did not address all-cause mortality, and 2 of the 4 trials did not include low-risk subjects.⁹

Many physicians and patients are prescribing aspirin with the expectation of reduced mortality in high-risk and low-risk individuals. Media advertisements and health programs may not clearly delineate the population for whom aspirin has clear benefits. A recent review suggested that aspirin is likely to be effective for primary prevention in yet to be defined groups.¹⁴ This review seeks to answer 2 questions. First, are there any primary prevention studies using aspirin that included only low-risk subjects? Second, should aspirin be prescribed routinely to persons at low risk for cardiovascular disease to decrease total mortality?

Methods

Search strategy

The MEDLINE database and the Cochrane Library were systematically searched using the terms aspirin or antiplatelet therapy and primary prevention or prevention and primary and mortality. An additional search was made with primary prevention and myocardial infarction or stroke. The Internet was searched (http://www.google.com) by using the same search terms. The studies were limited to human populations. Search results consisted of abstracts, complete reviews, and reference lists from articles. Morbidity associated with aspirin use also was reviewed.

Selection criteria: end points

Only those studies that investigated primary prevention of cardiovascular disease using aspirin, had low-risk subjects, and included a measure of total mortality were part of our analysis. We used the 2001 Adult Treatment Panel III Guidelines and the recent British Medical Journal clinical evidence guidelines on primary prevention of cardiovascular disorders to define the low-risk patient.^15,16 Those guidelines classified major risk factors for ischemic vascular disease as hypertension, low high-density lipoprotein cholesterol, high lowdensity lipoprotein cholesterol, family history of premature coronary heart disease, smoking, diabetes, and advancing age (men ≥ 45 years, women ≥ 55 years). We also classified those patients with past cerebrovascular events, myocardial infarction, and angina as high risk. We defined low risk as having no more than 1 of these risk factors.

Every trial was evaluated independently by each author according to the Jadad scale.¹⁷ Based on information in the original articles, we recalculated the odds ratios (ORs) for each study. The results of the 2 randomized trails were combined by means of the Mantel-Haenszel method for combining ORs, and StatXact 4 for Windows was used for the analysis.^18,19 The data were used to create a forest plot of mortal-ity.¹⁹ The decision to combine studies of like type was made a priority.

Results

MEDLINE search results for aspirin and primary prevention yielded 291 articles. Antiplatelet therapy and primary prevention yielded 64 articles. Myocardial infarction or stroke and primary prevention yielded 514 articles. Cross-referencing aspirin, prevention, and mortality yielded 690 articles. The Cochrane Library search of antiplatelet therapy and prevention and primary yielded 17 complete reviews and 6 abstracts of systematic reviews. No additional studies published or unpublished were identified through the Internet.

Five clinical trials and 1 cohort study that evaluated aspirin for primary prevention were identi-fied.^11,12,20-23 One of those, a pilot study, was excluded because it did not provide mortality data for the aspirin and placebo groups.²³ Two clinical trials, the Hypertension Optimal Treatment Trial and the Thrombosis Prevention Trial, did not include low-risk subjects.^11,12 Although no studies were identified that included only low-risk subjects, 3 studies met our inclusion criteria. Characteristics of those 3 studies are reported in Table 1.

The US Physicians Health Study (USPHS) randomized physicians into 4 treatment groups: aspirin plus beta-carotene, aspirin plus placebo, betacarotene plus placebo, and placebo plus placebo.²⁰ Both aspirin groups took 325 mg every other day. The mean age was 53.2 years.²⁴ Fifty percent of the participants were current or past smokers, and 9% had hypertension. Although the rate of myocardial infarction was significantly lower in the aspirin group, there was no reduction in total cardiovascular mortality. The results are reported in Table 2. More side effects were noted in the aspirin group, including gastric ulcers, gastrointestinal bleeding, hemorrhagic stroke, and other bleeding disorders.²⁰ No separate analysis of low-risk subjects’ risk was performed.

In the British Doctors Study (BDS), 66% of patients were randomized to take aspirin once daily and 33% were to avoid aspirin.²¹ More than half of the subjects were at least 60 years old. Physicians with stroke, myocardial infarction, ulcer disease, or currently taking any aspirin products were excluded. Six percent of the subjects had a history of heart disease other than myocardial infarction, 10% had hypertension, and 75% of participants were currently smoking or had a history of smoking. No significant differences were noted between groups for myocardial infarction or total mortality. By the end of the study, 44% of the aspirin group had discontinued aspirin secondary to side effects, the most common being dyspepsia. Of the control group, 2% per year started using aspirin because they developed risk factors such as vascular disease or for primary prevention. Low-risk individuals were not evaluated separately.

The Nurses Health Study (NHS) was a cohort study of women who were free of diagnosed coronary heart disease, stroke, and cancer at the start of the study. However, 29% of the women smoked and 15% had hypertension.²² The mean age was 46.0 years, and the follow-up was 96.7% of total potential person years. The study respondents were asked how many aspirin tablets they took per week: 0, 1 to 3, 4 to 6, 7 to 14, or 15+. Those who smoked or were overweight were more likely to take aspirin. No separate analysis of low-risk subjects was performed. Mortality from aspirin use was clearly dose dependent. For study participants taking 1 to 6 aspirin tablets each week, mortality was 0.84% (OR, 1.51; 95% confidence interval [CI], 1.26–1.82); for those taking 7 to 14 aspirin tablets, mortality was 0.99% (OR, 1.80; CI, 1.39–2.33); and for those taking 15+ aspirin tablets, mortality was 1.82% (OR, 3.32; CI, 2.62–4.21). Rates of myocardial infarction and stroke also were higher for all groups taking aspirin (see Table 2). When combined, the USPHS and the BDS demonstrated no significant difference in mortality between aspirin and placebo groups, whereas the NHS found increased mortality from aspirin (Figure).

TABLE 1

Characteristics of aspirin studies that included low-risk subjects

TABLE 2

Rates of myocardial infarction, stroke, total cardiovascular mortality, and total mortality in studies of aspirin vs no aspirin that included low-risk patients

FIGURE Forest plot of mortality in healthy patients on aspirin

Discussion

To date, there has been no study of aspirin for primary prevention that included a separate analysis of patients who were free of cardiovascular risk factors. Each of the 3 studies that included low-risk subjects grouped them with subjects at higher risk, those known to benefit from aspirin.^9,11,12 Even so, none of those studies demonstrated a statistically significant decrease in all-cause mortality. Even when combined, the BDS and the USPHS demonstrated no significant improvement in mortality. Mortality in the BDS was nearly 4 times greater than that in the USPHS. This finding is likely due to the higher baseline rate of smoking and other risk factors in the British doctors. In contrast to the BDS, the USPHS demonstrated significantly decreased rates for fatal and nonfatal myocardial infarction. Our analysis of the NHS associated aspirin with increased mortality, fatal myocardial infarction, and nonfatal myocardial infarction at any dose. The nurses on average had lower risk than the doctors, fewer smoked, and they were younger.

Many studies have clearly demonstrated the benefits of aspirin for primary prevention in high-risk subjects.^10-12,25 There may be other benefits to taking prophylactic aspirin. In the Cancer Prevention Study II, aspirin use was associated with decreased death rates from colon cancer.²⁶ Unfortunately, that study did not measure all-cause mortality.

There are a number of limitations to this study. There were no strictly low-risk studies of aspirin for primary prevention of cardiovascular mortality, and there was a paucity of studies that included low-risk subjects. Because the studies analyzed did not include only low-risk subjects, the results may not apply to all low-risk patients. The BDS did not include a placebo and was not blinded. Although not statistically significant, the ORs tended toward a protective effect for aspirin in the 2 randomized trials. The large difference in mortality between those 2 trials remains unexplained. The NHS was the only study to include women, and it was a cohort study, which is subject to selection and reporting biases. Therefore, aspirin users may have been at higher mortality risk due to smoking, obesity, or other illness, thus rendering the association between aspirin and higher mortality meaningless.

Many studies have shown significant side effects of aspirin, including epistaxis, peptic ulcer disease, gastrointestinal bleeds, and hemorrhagic stroke.^{15,20-22,27-32} In the BDS, 17% more subjects in the aspirin group developed peptic ulcer disease, and 19% stopped treatment during the first year secondary to gastrointestinal complaints.²¹

In conclusion, there is currently no evidence to recommend for or against the use of aspirin in low-risk individuals to decrease mortality. There may be other reasons to take aspirin prophylactically such as to reduce myocardial infarction or colon cancer. However, these benefits have not been established in a low-risk population. Health care providers should ask all patients whether they are taking aspirin and evaluate the risk-benefit ratio before recommending it.

Cardiovascular disease is the leading cause of death in the United States, and aspirin, a platelet aggregate inhibitor, is often recommended as prophylaxis for cardiovascular disease.^1-3Clinical studies have demonstrated the benefit of aspirin use for secondary prevention of cardiovascular disease and stroke.^1,4-10 In high-risk subjects, aspirin has been proven effective in primary prevention of major cardiovascular events and nonfatal ischemic heart disease.^11-13 Sanmuganathan and colleagues recently reported a meta-analysis of 4 randomized trials of aspirin for primary prevention. Although they determined that aspirin treatment is safe if the coronary event rate is at least 1.5% each year and unsafe if the rate is no higher than 0.5% each year, they did not address all-cause mortality, and 2 of the 4 trials did not include low-risk subjects.⁹

Many physicians and patients are prescribing aspirin with the expectation of reduced mortality in high-risk and low-risk individuals. Media advertisements and health programs may not clearly delineate the population for whom aspirin has clear benefits. A recent review suggested that aspirin is likely to be effective for primary prevention in yet to be defined groups.¹⁴ This review seeks to answer 2 questions. First, are there any primary prevention studies using aspirin that included only low-risk subjects? Second, should aspirin be prescribed routinely to persons at low risk for cardiovascular disease to decrease total mortality?

Methods

Search strategy

The MEDLINE database and the Cochrane Library were systematically searched using the terms aspirin or antiplatelet therapy and primary prevention or prevention and primary and mortality. An additional search was made with primary prevention and myocardial infarction or stroke. The Internet was searched (http://www.google.com) by using the same search terms. The studies were limited to human populations. Search results consisted of abstracts, complete reviews, and reference lists from articles. Morbidity associated with aspirin use also was reviewed.

Selection criteria: end points

Only those studies that investigated primary prevention of cardiovascular disease using aspirin, had low-risk subjects, and included a measure of total mortality were part of our analysis. We used the 2001 Adult Treatment Panel III Guidelines and the recent British Medical Journal clinical evidence guidelines on primary prevention of cardiovascular disorders to define the low-risk patient.^15,16 Those guidelines classified major risk factors for ischemic vascular disease as hypertension, low high-density lipoprotein cholesterol, high lowdensity lipoprotein cholesterol, family history of premature coronary heart disease, smoking, diabetes, and advancing age (men ≥ 45 years, women ≥ 55 years). We also classified those patients with past cerebrovascular events, myocardial infarction, and angina as high risk. We defined low risk as having no more than 1 of these risk factors.

Every trial was evaluated independently by each author according to the Jadad scale.¹⁷ Based on information in the original articles, we recalculated the odds ratios (ORs) for each study. The results of the 2 randomized trails were combined by means of the Mantel-Haenszel method for combining ORs, and StatXact 4 for Windows was used for the analysis.^18,19 The data were used to create a forest plot of mortal-ity.¹⁹ The decision to combine studies of like type was made a priority.

Results

MEDLINE search results for aspirin and primary prevention yielded 291 articles. Antiplatelet therapy and primary prevention yielded 64 articles. Myocardial infarction or stroke and primary prevention yielded 514 articles. Cross-referencing aspirin, prevention, and mortality yielded 690 articles. The Cochrane Library search of antiplatelet therapy and prevention and primary yielded 17 complete reviews and 6 abstracts of systematic reviews. No additional studies published or unpublished were identified through the Internet.

Five clinical trials and 1 cohort study that evaluated aspirin for primary prevention were identi-fied.^11,12,20-23 One of those, a pilot study, was excluded because it did not provide mortality data for the aspirin and placebo groups.²³ Two clinical trials, the Hypertension Optimal Treatment Trial and the Thrombosis Prevention Trial, did not include low-risk subjects.^11,12 Although no studies were identified that included only low-risk subjects, 3 studies met our inclusion criteria. Characteristics of those 3 studies are reported in Table 1.

The US Physicians Health Study (USPHS) randomized physicians into 4 treatment groups: aspirin plus beta-carotene, aspirin plus placebo, betacarotene plus placebo, and placebo plus placebo.²⁰ Both aspirin groups took 325 mg every other day. The mean age was 53.2 years.²⁴ Fifty percent of the participants were current or past smokers, and 9% had hypertension. Although the rate of myocardial infarction was significantly lower in the aspirin group, there was no reduction in total cardiovascular mortality. The results are reported in Table 2. More side effects were noted in the aspirin group, including gastric ulcers, gastrointestinal bleeding, hemorrhagic stroke, and other bleeding disorders.²⁰ No separate analysis of low-risk subjects’ risk was performed.

In the British Doctors Study (BDS), 66% of patients were randomized to take aspirin once daily and 33% were to avoid aspirin.²¹ More than half of the subjects were at least 60 years old. Physicians with stroke, myocardial infarction, ulcer disease, or currently taking any aspirin products were excluded. Six percent of the subjects had a history of heart disease other than myocardial infarction, 10% had hypertension, and 75% of participants were currently smoking or had a history of smoking. No significant differences were noted between groups for myocardial infarction or total mortality. By the end of the study, 44% of the aspirin group had discontinued aspirin secondary to side effects, the most common being dyspepsia. Of the control group, 2% per year started using aspirin because they developed risk factors such as vascular disease or for primary prevention. Low-risk individuals were not evaluated separately.

The Nurses Health Study (NHS) was a cohort study of women who were free of diagnosed coronary heart disease, stroke, and cancer at the start of the study. However, 29% of the women smoked and 15% had hypertension.²² The mean age was 46.0 years, and the follow-up was 96.7% of total potential person years. The study respondents were asked how many aspirin tablets they took per week: 0, 1 to 3, 4 to 6, 7 to 14, or 15+. Those who smoked or were overweight were more likely to take aspirin. No separate analysis of low-risk subjects was performed. Mortality from aspirin use was clearly dose dependent. For study participants taking 1 to 6 aspirin tablets each week, mortality was 0.84% (OR, 1.51; 95% confidence interval [CI], 1.26–1.82); for those taking 7 to 14 aspirin tablets, mortality was 0.99% (OR, 1.80; CI, 1.39–2.33); and for those taking 15+ aspirin tablets, mortality was 1.82% (OR, 3.32; CI, 2.62–4.21). Rates of myocardial infarction and stroke also were higher for all groups taking aspirin (see Table 2). When combined, the USPHS and the BDS demonstrated no significant difference in mortality between aspirin and placebo groups, whereas the NHS found increased mortality from aspirin (Figure).

TABLE 1

Characteristics of aspirin studies that included low-risk subjects

TABLE 2

Rates of myocardial infarction, stroke, total cardiovascular mortality, and total mortality in studies of aspirin vs no aspirin that included low-risk patients

FIGURE Forest plot of mortality in healthy patients on aspirin

Discussion

To date, there has been no study of aspirin for primary prevention that included a separate analysis of patients who were free of cardiovascular risk factors. Each of the 3 studies that included low-risk subjects grouped them with subjects at higher risk, those known to benefit from aspirin.^9,11,12 Even so, none of those studies demonstrated a statistically significant decrease in all-cause mortality. Even when combined, the BDS and the USPHS demonstrated no significant improvement in mortality. Mortality in the BDS was nearly 4 times greater than that in the USPHS. This finding is likely due to the higher baseline rate of smoking and other risk factors in the British doctors. In contrast to the BDS, the USPHS demonstrated significantly decreased rates for fatal and nonfatal myocardial infarction. Our analysis of the NHS associated aspirin with increased mortality, fatal myocardial infarction, and nonfatal myocardial infarction at any dose. The nurses on average had lower risk than the doctors, fewer smoked, and they were younger.

Many studies have clearly demonstrated the benefits of aspirin for primary prevention in high-risk subjects.^10-12,25 There may be other benefits to taking prophylactic aspirin. In the Cancer Prevention Study II, aspirin use was associated with decreased death rates from colon cancer.²⁶ Unfortunately, that study did not measure all-cause mortality.

There are a number of limitations to this study. There were no strictly low-risk studies of aspirin for primary prevention of cardiovascular mortality, and there was a paucity of studies that included low-risk subjects. Because the studies analyzed did not include only low-risk subjects, the results may not apply to all low-risk patients. The BDS did not include a placebo and was not blinded. Although not statistically significant, the ORs tended toward a protective effect for aspirin in the 2 randomized trials. The large difference in mortality between those 2 trials remains unexplained. The NHS was the only study to include women, and it was a cohort study, which is subject to selection and reporting biases. Therefore, aspirin users may have been at higher mortality risk due to smoking, obesity, or other illness, thus rendering the association between aspirin and higher mortality meaningless.

Many studies have shown significant side effects of aspirin, including epistaxis, peptic ulcer disease, gastrointestinal bleeds, and hemorrhagic stroke.^{15,20-22,27-32} In the BDS, 17% more subjects in the aspirin group developed peptic ulcer disease, and 19% stopped treatment during the first year secondary to gastrointestinal complaints.²¹

In conclusion, there is currently no evidence to recommend for or against the use of aspirin in low-risk individuals to decrease mortality. There may be other reasons to take aspirin prophylactically such as to reduce myocardial infarction or colon cancer. However, these benefits have not been established in a low-risk population. Health care providers should ask all patients whether they are taking aspirin and evaluate the risk-benefit ratio before recommending it.

Cardiovascular disease is the leading cause of death in the United States, and aspirin, a platelet aggregate inhibitor, is often recommended as prophylaxis for cardiovascular disease.^1-3Clinical studies have demonstrated the benefit of aspirin use for secondary prevention of cardiovascular disease and stroke.^1,4-10 In high-risk subjects, aspirin has been proven effective in primary prevention of major cardiovascular events and nonfatal ischemic heart disease.^11-13 Sanmuganathan and colleagues recently reported a meta-analysis of 4 randomized trials of aspirin for primary prevention. Although they determined that aspirin treatment is safe if the coronary event rate is at least 1.5% each year and unsafe if the rate is no higher than 0.5% each year, they did not address all-cause mortality, and 2 of the 4 trials did not include low-risk subjects.⁹

Many physicians and patients are prescribing aspirin with the expectation of reduced mortality in high-risk and low-risk individuals. Media advertisements and health programs may not clearly delineate the population for whom aspirin has clear benefits. A recent review suggested that aspirin is likely to be effective for primary prevention in yet to be defined groups.¹⁴ This review seeks to answer 2 questions. First, are there any primary prevention studies using aspirin that included only low-risk subjects? Second, should aspirin be prescribed routinely to persons at low risk for cardiovascular disease to decrease total mortality?

Methods

Search strategy

The MEDLINE database and the Cochrane Library were systematically searched using the terms aspirin or antiplatelet therapy and primary prevention or prevention and primary and mortality. An additional search was made with primary prevention and myocardial infarction or stroke. The Internet was searched (http://www.google.com) by using the same search terms. The studies were limited to human populations. Search results consisted of abstracts, complete reviews, and reference lists from articles. Morbidity associated with aspirin use also was reviewed.

Selection criteria: end points

Only those studies that investigated primary prevention of cardiovascular disease using aspirin, had low-risk subjects, and included a measure of total mortality were part of our analysis. We used the 2001 Adult Treatment Panel III Guidelines and the recent British Medical Journal clinical evidence guidelines on primary prevention of cardiovascular disorders to define the low-risk patient.^15,16 Those guidelines classified major risk factors for ischemic vascular disease as hypertension, low high-density lipoprotein cholesterol, high lowdensity lipoprotein cholesterol, family history of premature coronary heart disease, smoking, diabetes, and advancing age (men ≥ 45 years, women ≥ 55 years). We also classified those patients with past cerebrovascular events, myocardial infarction, and angina as high risk. We defined low risk as having no more than 1 of these risk factors.

Every trial was evaluated independently by each author according to the Jadad scale.¹⁷ Based on information in the original articles, we recalculated the odds ratios (ORs) for each study. The results of the 2 randomized trails were combined by means of the Mantel-Haenszel method for combining ORs, and StatXact 4 for Windows was used for the analysis.^18,19 The data were used to create a forest plot of mortal-ity.¹⁹ The decision to combine studies of like type was made a priority.

Results

MEDLINE search results for aspirin and primary prevention yielded 291 articles. Antiplatelet therapy and primary prevention yielded 64 articles. Myocardial infarction or stroke and primary prevention yielded 514 articles. Cross-referencing aspirin, prevention, and mortality yielded 690 articles. The Cochrane Library search of antiplatelet therapy and prevention and primary yielded 17 complete reviews and 6 abstracts of systematic reviews. No additional studies published or unpublished were identified through the Internet.

Five clinical trials and 1 cohort study that evaluated aspirin for primary prevention were identi-fied.^11,12,20-23 One of those, a pilot study, was excluded because it did not provide mortality data for the aspirin and placebo groups.²³ Two clinical trials, the Hypertension Optimal Treatment Trial and the Thrombosis Prevention Trial, did not include low-risk subjects.^11,12 Although no studies were identified that included only low-risk subjects, 3 studies met our inclusion criteria. Characteristics of those 3 studies are reported in Table 1.

The US Physicians Health Study (USPHS) randomized physicians into 4 treatment groups: aspirin plus beta-carotene, aspirin plus placebo, betacarotene plus placebo, and placebo plus placebo.²⁰ Both aspirin groups took 325 mg every other day. The mean age was 53.2 years.²⁴ Fifty percent of the participants were current or past smokers, and 9% had hypertension. Although the rate of myocardial infarction was significantly lower in the aspirin group, there was no reduction in total cardiovascular mortality. The results are reported in Table 2. More side effects were noted in the aspirin group, including gastric ulcers, gastrointestinal bleeding, hemorrhagic stroke, and other bleeding disorders.²⁰ No separate analysis of low-risk subjects’ risk was performed.

In the British Doctors Study (BDS), 66% of patients were randomized to take aspirin once daily and 33% were to avoid aspirin.²¹ More than half of the subjects were at least 60 years old. Physicians with stroke, myocardial infarction, ulcer disease, or currently taking any aspirin products were excluded. Six percent of the subjects had a history of heart disease other than myocardial infarction, 10% had hypertension, and 75% of participants were currently smoking or had a history of smoking. No significant differences were noted between groups for myocardial infarction or total mortality. By the end of the study, 44% of the aspirin group had discontinued aspirin secondary to side effects, the most common being dyspepsia. Of the control group, 2% per year started using aspirin because they developed risk factors such as vascular disease or for primary prevention. Low-risk individuals were not evaluated separately.

The Nurses Health Study (NHS) was a cohort study of women who were free of diagnosed coronary heart disease, stroke, and cancer at the start of the study. However, 29% of the women smoked and 15% had hypertension.²² The mean age was 46.0 years, and the follow-up was 96.7% of total potential person years. The study respondents were asked how many aspirin tablets they took per week: 0, 1 to 3, 4 to 6, 7 to 14, or 15+. Those who smoked or were overweight were more likely to take aspirin. No separate analysis of low-risk subjects was performed. Mortality from aspirin use was clearly dose dependent. For study participants taking 1 to 6 aspirin tablets each week, mortality was 0.84% (OR, 1.51; 95% confidence interval [CI], 1.26–1.82); for those taking 7 to 14 aspirin tablets, mortality was 0.99% (OR, 1.80; CI, 1.39–2.33); and for those taking 15+ aspirin tablets, mortality was 1.82% (OR, 3.32; CI, 2.62–4.21). Rates of myocardial infarction and stroke also were higher for all groups taking aspirin (see Table 2). When combined, the USPHS and the BDS demonstrated no significant difference in mortality between aspirin and placebo groups, whereas the NHS found increased mortality from aspirin (Figure).

TABLE 1

Characteristics of aspirin studies that included low-risk subjects

TABLE 2

Rates of myocardial infarction, stroke, total cardiovascular mortality, and total mortality in studies of aspirin vs no aspirin that included low-risk patients

FIGURE Forest plot of mortality in healthy patients on aspirin

Discussion

To date, there has been no study of aspirin for primary prevention that included a separate analysis of patients who were free of cardiovascular risk factors. Each of the 3 studies that included low-risk subjects grouped them with subjects at higher risk, those known to benefit from aspirin.^9,11,12 Even so, none of those studies demonstrated a statistically significant decrease in all-cause mortality. Even when combined, the BDS and the USPHS demonstrated no significant improvement in mortality. Mortality in the BDS was nearly 4 times greater than that in the USPHS. This finding is likely due to the higher baseline rate of smoking and other risk factors in the British doctors. In contrast to the BDS, the USPHS demonstrated significantly decreased rates for fatal and nonfatal myocardial infarction. Our analysis of the NHS associated aspirin with increased mortality, fatal myocardial infarction, and nonfatal myocardial infarction at any dose. The nurses on average had lower risk than the doctors, fewer smoked, and they were younger.

Many studies have clearly demonstrated the benefits of aspirin for primary prevention in high-risk subjects.^10-12,25 There may be other benefits to taking prophylactic aspirin. In the Cancer Prevention Study II, aspirin use was associated with decreased death rates from colon cancer.²⁶ Unfortunately, that study did not measure all-cause mortality.

There are a number of limitations to this study. There were no strictly low-risk studies of aspirin for primary prevention of cardiovascular mortality, and there was a paucity of studies that included low-risk subjects. Because the studies analyzed did not include only low-risk subjects, the results may not apply to all low-risk patients. The BDS did not include a placebo and was not blinded. Although not statistically significant, the ORs tended toward a protective effect for aspirin in the 2 randomized trials. The large difference in mortality between those 2 trials remains unexplained. The NHS was the only study to include women, and it was a cohort study, which is subject to selection and reporting biases. Therefore, aspirin users may have been at higher mortality risk due to smoking, obesity, or other illness, thus rendering the association between aspirin and higher mortality meaningless.

Many studies have shown significant side effects of aspirin, including epistaxis, peptic ulcer disease, gastrointestinal bleeds, and hemorrhagic stroke.^{15,20-22,27-32} In the BDS, 17% more subjects in the aspirin group developed peptic ulcer disease, and 19% stopped treatment during the first year secondary to gastrointestinal complaints.²¹

In conclusion, there is currently no evidence to recommend for or against the use of aspirin in low-risk individuals to decrease mortality. There may be other reasons to take aspirin prophylactically such as to reduce myocardial infarction or colon cancer. However, these benefits have not been established in a low-risk population. Health care providers should ask all patients whether they are taking aspirin and evaluate the risk-benefit ratio before recommending it.

Although common, third party involvement in adult medical care has not been well studied. This prospective study explored the frequency of companions, the reasons for accompaniment, and the companion’s influence on the medical encounter from the perspective of the patient, the companion, and the physician.

Methods

Setting, physicians, and patients

The study was conducted at the general internal medicine practice of the University of Colorado Health Sciences Center. Fifteen full-time faculty and 42 internal medicine residents participated. A companion was defined as any person older than 18 years who accompanied a patient to a medical visit and was designated as an examination room companion if that person spent any portion of the visit in the examination room; otherwise, that person was designated as a waiting room companion. Persons employed solely to provide transportation services for patients were not considered companions.

Study design

The study consisted of 2 parts: a prospective study to document the frequency of patient accompaniment by a third person to ambulatory medicine visits, and a survey of patients, companions, and physicians to explore the rationales and influence of the companion during the medical encounter. A professional research assistant was present for an average of 8 of 10 half-day clinics per week. To accomplish the first objective, a research assistant directly observed 1294 consecutive patient visits from September 22 to October 29, 1998. To accomplish the second objective, from mid-September to mid-November, a professional research assistant attempted to enroll all consecutive patients accompanied to their appointments. Unaccompanied patients were approached for consent at the convenience of the research assistant. Patients and their companions were approached for consent in the waiting room immediately before their visits.

For inclusion, patients and companions provided consent and were literate in English. Patients, companions, and physicians independently completed self-administered questionnaires immediately after their visits. All were informed that responses were confidential and would not be disclosed to one another. This study was reviewed and approved by the Colorado Multiple Institutional Review Board.

Questionnaire development

The survey instruments were developed after a thorough review of existing research^1,6-10and refined by pilot testing and review with a professional survey consultant. Patients and companions completed demographic questions (Table 1). Patients rated their overall health, stated their relationship to the companion, and indicated the reasons for companion accompaniment (Table 2). Patients and companions indicated from a list of 7 items (Table 2) how the companion influenced the visit and rated the companion’s helpfulness during the encounter (5-point Likert scale: 1 = very unhelpful to 5 = very helpful).

TABLE 1

Patient demographics and characteristics*

Patients’ and companions’ reports of companion’s reasons for accompaniment and influence on the medical encounter*

TABLE 3

Physician report of examination room companion’s influence on the medical encounter*

Physician report of examination room companion’s behaviors

Data analysis

Data were used from each member of a set regardless of survey completion by other set members. One patient had 2 examination room companions and 2 patients had 2 waiting room companions. In each case, both companions were surveyed. In the 2 cases in which a patient had waiting and examination room companions, the examination room companion was considered more influential for the medical encounter and only that person was surveyed.

The data were analyzed with SAS version 6.12 (SAS Inc, Cary, NC) using bivariate and multivariable methods. Comparisons were made between patient categories (patient alone, patient with examination room companion, patients with waiting room companion) using the chi-square statistic for categorical variables.

Multivariable analyses were conducted to explore the effects of various independent variables on the decision to bring a companion into the examination room. The outcome variables for the regression models were defined by patient status (patient alone, patient with examination room companion, patient with waiting room companion). All significant variables (P ≤ .05) in bivariate analyses were entered into the multivariate analyses. Odds ratios (ORs) with 95% confidence intervals (CIs) were obtained for each variable in the model.

Patient and companion agreement on the reasons for accompaniment and influence on the medical encounter were measured with the kappa statistic. Kappas (κ) of 1.0 to .75 denote excellent agreement, .4 to .75 denote good agreement, and 0 to .4 denote marginal agreement.¹¹

RESULTS

Of the 1294 patient visits, 834 (64%) were to faculty physicians and 451 (35%) were to resident physicians. Overall, companions were present for 29% (n = 374) of patient visits and accompanied the patient into the examination room for 16% (n = 212) of visits. Companions accompanied patients to 23% (n = 196) of faculty visits and 39% (n = 178) of resident visits (P P = .98).

Ninety-three percent (121/130) of unaccompanied patients and 92% (200/217) of consecutive patient–companion pairs approached for consent agreed to participate in the study. In 26 cases the patient or the companion refused to participate for 1of the following reasons: language barrier, too ill, lack of time, invasion of privacy, or uncomfortable with process of consent. Patients and physicians completed surveys for 97% of patient encounters, and companions completed surveys for 99% of patient encounters.

Comparisons between accompanied and unaccompanied patients are presented in Table 1. The fact that faculty physicians had a greater proportion of accompanied patients who received and completed surveys is likely due to the necessity of English literacy for inclusion into the survey study. At this practice site, patients cared for by residents are more likely to be non-English speaking and have a companion for the purpose of translation. Examination room companions were often the spouse or partner (55%), parents (17%), or less frequently roommate or friend (7%), whereas waiting room companions were commonly a spouse or partner (46%) or roommate or friend (24%). Overall, family members accounted for 93% of examination room companions and 76% of waiting room companions. Examination room companions were more likely to be female than waiting room companions (65% vs 51%, P = .05). Patients who were older, less well educated, and whose cases had greater medical or social complexity were more likely to have a companion in the examination room.

Patients’ and companions’ assessments of the reasons for accompaniment and the companions’ influence on communication and resource use are shown in Table 2. The patients’ and companions’ stated reasons for companion accompaniment were in good agreement with the kappa statistic ranging from 0.41 for “help with insurance forms” to 0.61 for “help remember the physician’s advice.” Patients’ and companions’ agreement regarding the influence of the companion on the medical visit was less than 0.4, suggesting marginal agreement for tests ordered (κ = .29), prescribed treatment (κ = .36), and length of visit (κ = .33). There was good agreement for number of referrals (κ = .45) and for physician and patient understanding (κ = .62 and .60, respectively).

Table 3 displays the physicians’ reports of the examination room companion’s influence on the medical encounter. Table 4 shows the physicians’ reports of the behavior of the examination room companions.

Patients regarded examination and waiting room companions as “very helpful” for 84% and 71% of visits, respectively, and as “very unhelpful” for 1% of visits. Of the 121 patients who came alone to their medical visits, 7% indicated that they considered bringing a companion to their visits and 16% thought a companion’s presence would have been helpful.

Physicians regarded examination room companions as “somewhat to very helpful” for 66% of visits. When physicians did not have contact with a companion, they indicated that contact would have been helpful for 16% of patient encounters.

Multivariable analyses explored the effects of independent variables on the decision to bring a companion into the examination room. A physician rating of a case of having greater medical and social complexity was the only variable associated with companion accompaniment to the examination room vs not having a companion (OR, 1.7; 95% CI, 1.4–2.1). Patient characteristics and patients’ reported reasons for accompaniment were factors influencing accompaniment into the examination room vs the waiting room. A need for help with communicating concerns to the physician (OR, 7.8; 95% CI, 2.4–25.6), help with remembering the physician’s advice and instructions (OR, 7.1; 95% CI, 2.0–25.3), and greater medical and social complexity of cases (OR, 1.5; 95% CI, 1.1–2.0) were associated with being accompanied to the examination room over just the waiting room. In contrast, needing help with transportation was negatively associated with having a companion in the examination room vs the waiting room (OR, 0.2; 95% CI, 0.1–0.5).

DISCUSSION

Companions frequently accompany adult patients and participate in ambulatory medical encounters. We found that companions accompanied 29% of patients and were present in the examination room for 16% of outpatient medical encounters. Examination room companions often were present to aid communication with the physician and to help the patient remember instructions. Physicians, patients, and companions believed that physician and patient understanding often were favorably affected by the presence of a companion. A companion’s presence in the examination room had beneficial effects on patient and physician understanding and very rarely had a negative effect. These findings reflect the results of previous studies in which approximately 33% of patients were accompanied to family medicine and geriatric encounters and 66% of these companions were present in the examination room.^1,7-10,12,13 The only variable associated with accompaniment into the examination room vs presenting to one’s appointment alone was a physician rating of greater medical and social complexity. This finding also corroborates the findings of other studies.^7,14,15

Some research has suggested a negative effect of a third person during the medical encounter. For example, older patients in triadic encounters raise fewer topics, are less assertive, and participate less in humor and joint decision making.⁶ Other research has failed to find a benefit in health outcomes with the presence of a companion, and unaccompanied patients rated themselves as having greater understanding of their medical problems and greater faith that their physicians were doing everything possible for them than patients accompanied to their visits.¹⁵ Still others have found that physicians provide more information and time but less emotional support to accompanied patients.⁷

Unique to our study was the specific assessment of companion influence on various aspects of the medical encounter from the perspectives of the companion, patient, and physician. As expected, examination room companions had significant influence on aspects of communication. Examination room companions were generally considered helpful by patients and physicians. Physicians may use the companion and patient as barometers of the visit’s accomplished goals. Hence, the time spent listening to a companion provide information about a patient’s medical problems might be balanced by the provision of less emotional support to the patient, especially if the companion is providing that support. The physician may offer an explanation until confident that either party (patient or companion) has a complete understanding. As demonstrated by the results, physicians, patients, and companions thought that patient understanding was increased in approximately 50% of encounters by a companion’s presence, and companions overwhelmingly were considered very helpful by patients.

Our study was limited to 1 urban, academic, general internal medicine practice and may not be generalizable to other settings. The method of assessing medical and social complexities was simple, and the very presence of an examination room companion may have biased physicians to rate these patients as having more complex problems than unaccompanied patients. Also, the longer period of the second part of the study necessary to enroll 200 patient–companion pairs, compared with the companion frequency data of the first part of the study, suggested that we did not enroll “consecutive” patient–companion pairs. This may be explained by the exclusion of patient–companion pairs when either party was not literate in English. Also, the logistics of obtaining consent and administering post visit questionnaires by a single research assistant interfered with the attempt to enroll all patient–companion pairs. Convenience enrollment of unaccompanied patients may have been biased. The effect of the companion on the medical encounter was not verified by objective measures such as timing visit length. Further, patients, companions, and physicians rated the effect of the companion’s presence immediately after the encounter; the full effect of the companion’s presence might require more time to emerge.

Companions frequently accompany patients to their ambulatory general medicine visits. The companion is usually a family member who is present at the request of the patient. Companions assume important roles and are overwhelmingly considered helpful by patients. Nonetheless, some of the behaviors that contribute to an effective physician–patient relationship might be diluted by the presence of a companion. Future directions of study include (1) assessing the effect of the companion on the physician–patient relationship, including specific aspects of communication and behavior, and (2) determining whether companion involvement influences health outcomes or resource use.

· Acknowledgments ·

We thank Marcia Blake, MA, for her statistical expertise; Jean Kutner, MD, MSPH, for her assistance with data analysis and statistical expertise; and Robert Dellavalle, MD, PhD, for his editorial assistance.

Although common, third party involvement in adult medical care has not been well studied. This prospective study explored the frequency of companions, the reasons for accompaniment, and the companion’s influence on the medical encounter from the perspective of the patient, the companion, and the physician.

Methods

Setting, physicians, and patients

The study was conducted at the general internal medicine practice of the University of Colorado Health Sciences Center. Fifteen full-time faculty and 42 internal medicine residents participated. A companion was defined as any person older than 18 years who accompanied a patient to a medical visit and was designated as an examination room companion if that person spent any portion of the visit in the examination room; otherwise, that person was designated as a waiting room companion. Persons employed solely to provide transportation services for patients were not considered companions.

Study design

The study consisted of 2 parts: a prospective study to document the frequency of patient accompaniment by a third person to ambulatory medicine visits, and a survey of patients, companions, and physicians to explore the rationales and influence of the companion during the medical encounter. A professional research assistant was present for an average of 8 of 10 half-day clinics per week. To accomplish the first objective, a research assistant directly observed 1294 consecutive patient visits from September 22 to October 29, 1998. To accomplish the second objective, from mid-September to mid-November, a professional research assistant attempted to enroll all consecutive patients accompanied to their appointments. Unaccompanied patients were approached for consent at the convenience of the research assistant. Patients and their companions were approached for consent in the waiting room immediately before their visits.

For inclusion, patients and companions provided consent and were literate in English. Patients, companions, and physicians independently completed self-administered questionnaires immediately after their visits. All were informed that responses were confidential and would not be disclosed to one another. This study was reviewed and approved by the Colorado Multiple Institutional Review Board.

Questionnaire development

The survey instruments were developed after a thorough review of existing research^1,6-10and refined by pilot testing and review with a professional survey consultant. Patients and companions completed demographic questions (Table 1). Patients rated their overall health, stated their relationship to the companion, and indicated the reasons for companion accompaniment (Table 2). Patients and companions indicated from a list of 7 items (Table 2) how the companion influenced the visit and rated the companion’s helpfulness during the encounter (5-point Likert scale: 1 = very unhelpful to 5 = very helpful).

TABLE 1

Patient demographics and characteristics*

Patients’ and companions’ reports of companion’s reasons for accompaniment and influence on the medical encounter*

TABLE 3

Physician report of examination room companion’s influence on the medical encounter*

Physician report of examination room companion’s behaviors

Data analysis

Data were used from each member of a set regardless of survey completion by other set members. One patient had 2 examination room companions and 2 patients had 2 waiting room companions. In each case, both companions were surveyed. In the 2 cases in which a patient had waiting and examination room companions, the examination room companion was considered more influential for the medical encounter and only that person was surveyed.

The data were analyzed with SAS version 6.12 (SAS Inc, Cary, NC) using bivariate and multivariable methods. Comparisons were made between patient categories (patient alone, patient with examination room companion, patients with waiting room companion) using the chi-square statistic for categorical variables.

Multivariable analyses were conducted to explore the effects of various independent variables on the decision to bring a companion into the examination room. The outcome variables for the regression models were defined by patient status (patient alone, patient with examination room companion, patient with waiting room companion). All significant variables (P ≤ .05) in bivariate analyses were entered into the multivariate analyses. Odds ratios (ORs) with 95% confidence intervals (CIs) were obtained for each variable in the model.

Patient and companion agreement on the reasons for accompaniment and influence on the medical encounter were measured with the kappa statistic. Kappas (κ) of 1.0 to .75 denote excellent agreement, .4 to .75 denote good agreement, and 0 to .4 denote marginal agreement.¹¹

RESULTS

Of the 1294 patient visits, 834 (64%) were to faculty physicians and 451 (35%) were to resident physicians. Overall, companions were present for 29% (n = 374) of patient visits and accompanied the patient into the examination room for 16% (n = 212) of visits. Companions accompanied patients to 23% (n = 196) of faculty visits and 39% (n = 178) of resident visits (P P = .98).

Ninety-three percent (121/130) of unaccompanied patients and 92% (200/217) of consecutive patient–companion pairs approached for consent agreed to participate in the study. In 26 cases the patient or the companion refused to participate for 1of the following reasons: language barrier, too ill, lack of time, invasion of privacy, or uncomfortable with process of consent. Patients and physicians completed surveys for 97% of patient encounters, and companions completed surveys for 99% of patient encounters.

Comparisons between accompanied and unaccompanied patients are presented in Table 1. The fact that faculty physicians had a greater proportion of accompanied patients who received and completed surveys is likely due to the necessity of English literacy for inclusion into the survey study. At this practice site, patients cared for by residents are more likely to be non-English speaking and have a companion for the purpose of translation. Examination room companions were often the spouse or partner (55%), parents (17%), or less frequently roommate or friend (7%), whereas waiting room companions were commonly a spouse or partner (46%) or roommate or friend (24%). Overall, family members accounted for 93% of examination room companions and 76% of waiting room companions. Examination room companions were more likely to be female than waiting room companions (65% vs 51%, P = .05). Patients who were older, less well educated, and whose cases had greater medical or social complexity were more likely to have a companion in the examination room.

Patients’ and companions’ assessments of the reasons for accompaniment and the companions’ influence on communication and resource use are shown in Table 2. The patients’ and companions’ stated reasons for companion accompaniment were in good agreement with the kappa statistic ranging from 0.41 for “help with insurance forms” to 0.61 for “help remember the physician’s advice.” Patients’ and companions’ agreement regarding the influence of the companion on the medical visit was less than 0.4, suggesting marginal agreement for tests ordered (κ = .29), prescribed treatment (κ = .36), and length of visit (κ = .33). There was good agreement for number of referrals (κ = .45) and for physician and patient understanding (κ = .62 and .60, respectively).

Table 3 displays the physicians’ reports of the examination room companion’s influence on the medical encounter. Table 4 shows the physicians’ reports of the behavior of the examination room companions.

Patients regarded examination and waiting room companions as “very helpful” for 84% and 71% of visits, respectively, and as “very unhelpful” for 1% of visits. Of the 121 patients who came alone to their medical visits, 7% indicated that they considered bringing a companion to their visits and 16% thought a companion’s presence would have been helpful.

Physicians regarded examination room companions as “somewhat to very helpful” for 66% of visits. When physicians did not have contact with a companion, they indicated that contact would have been helpful for 16% of patient encounters.

Multivariable analyses explored the effects of independent variables on the decision to bring a companion into the examination room. A physician rating of a case of having greater medical and social complexity was the only variable associated with companion accompaniment to the examination room vs not having a companion (OR, 1.7; 95% CI, 1.4–2.1). Patient characteristics and patients’ reported reasons for accompaniment were factors influencing accompaniment into the examination room vs the waiting room. A need for help with communicating concerns to the physician (OR, 7.8; 95% CI, 2.4–25.6), help with remembering the physician’s advice and instructions (OR, 7.1; 95% CI, 2.0–25.3), and greater medical and social complexity of cases (OR, 1.5; 95% CI, 1.1–2.0) were associated with being accompanied to the examination room over just the waiting room. In contrast, needing help with transportation was negatively associated with having a companion in the examination room vs the waiting room (OR, 0.2; 95% CI, 0.1–0.5).

DISCUSSION

Companions frequently accompany adult patients and participate in ambulatory medical encounters. We found that companions accompanied 29% of patients and were present in the examination room for 16% of outpatient medical encounters. Examination room companions often were present to aid communication with the physician and to help the patient remember instructions. Physicians, patients, and companions believed that physician and patient understanding often were favorably affected by the presence of a companion. A companion’s presence in the examination room had beneficial effects on patient and physician understanding and very rarely had a negative effect. These findings reflect the results of previous studies in which approximately 33% of patients were accompanied to family medicine and geriatric encounters and 66% of these companions were present in the examination room.^1,7-10,12,13 The only variable associated with accompaniment into the examination room vs presenting to one’s appointment alone was a physician rating of greater medical and social complexity. This finding also corroborates the findings of other studies.^7,14,15

Some research has suggested a negative effect of a third person during the medical encounter. For example, older patients in triadic encounters raise fewer topics, are less assertive, and participate less in humor and joint decision making.⁶ Other research has failed to find a benefit in health outcomes with the presence of a companion, and unaccompanied patients rated themselves as having greater understanding of their medical problems and greater faith that their physicians were doing everything possible for them than patients accompanied to their visits.¹⁵ Still others have found that physicians provide more information and time but less emotional support to accompanied patients.⁷

Unique to our study was the specific assessment of companion influence on various aspects of the medical encounter from the perspectives of the companion, patient, and physician. As expected, examination room companions had significant influence on aspects of communication. Examination room companions were generally considered helpful by patients and physicians. Physicians may use the companion and patient as barometers of the visit’s accomplished goals. Hence, the time spent listening to a companion provide information about a patient’s medical problems might be balanced by the provision of less emotional support to the patient, especially if the companion is providing that support. The physician may offer an explanation until confident that either party (patient or companion) has a complete understanding. As demonstrated by the results, physicians, patients, and companions thought that patient understanding was increased in approximately 50% of encounters by a companion’s presence, and companions overwhelmingly were considered very helpful by patients.

Our study was limited to 1 urban, academic, general internal medicine practice and may not be generalizable to other settings. The method of assessing medical and social complexities was simple, and the very presence of an examination room companion may have biased physicians to rate these patients as having more complex problems than unaccompanied patients. Also, the longer period of the second part of the study necessary to enroll 200 patient–companion pairs, compared with the companion frequency data of the first part of the study, suggested that we did not enroll “consecutive” patient–companion pairs. This may be explained by the exclusion of patient–companion pairs when either party was not literate in English. Also, the logistics of obtaining consent and administering post visit questionnaires by a single research assistant interfered with the attempt to enroll all patient–companion pairs. Convenience enrollment of unaccompanied patients may have been biased. The effect of the companion on the medical encounter was not verified by objective measures such as timing visit length. Further, patients, companions, and physicians rated the effect of the companion’s presence immediately after the encounter; the full effect of the companion’s presence might require more time to emerge.

Companions frequently accompany patients to their ambulatory general medicine visits. The companion is usually a family member who is present at the request of the patient. Companions assume important roles and are overwhelmingly considered helpful by patients. Nonetheless, some of the behaviors that contribute to an effective physician–patient relationship might be diluted by the presence of a companion. Future directions of study include (1) assessing the effect of the companion on the physician–patient relationship, including specific aspects of communication and behavior, and (2) determining whether companion involvement influences health outcomes or resource use.

· Acknowledgments ·

We thank Marcia Blake, MA, for her statistical expertise; Jean Kutner, MD, MSPH, for her assistance with data analysis and statistical expertise; and Robert Dellavalle, MD, PhD, for his editorial assistance.

Although common, third party involvement in adult medical care has not been well studied. This prospective study explored the frequency of companions, the reasons for accompaniment, and the companion’s influence on the medical encounter from the perspective of the patient, the companion, and the physician.

Methods

Setting, physicians, and patients

The study was conducted at the general internal medicine practice of the University of Colorado Health Sciences Center. Fifteen full-time faculty and 42 internal medicine residents participated. A companion was defined as any person older than 18 years who accompanied a patient to a medical visit and was designated as an examination room companion if that person spent any portion of the visit in the examination room; otherwise, that person was designated as a waiting room companion. Persons employed solely to provide transportation services for patients were not considered companions.

Study design

The study consisted of 2 parts: a prospective study to document the frequency of patient accompaniment by a third person to ambulatory medicine visits, and a survey of patients, companions, and physicians to explore the rationales and influence of the companion during the medical encounter. A professional research assistant was present for an average of 8 of 10 half-day clinics per week. To accomplish the first objective, a research assistant directly observed 1294 consecutive patient visits from September 22 to October 29, 1998. To accomplish the second objective, from mid-September to mid-November, a professional research assistant attempted to enroll all consecutive patients accompanied to their appointments. Unaccompanied patients were approached for consent at the convenience of the research assistant. Patients and their companions were approached for consent in the waiting room immediately before their visits.

For inclusion, patients and companions provided consent and were literate in English. Patients, companions, and physicians independently completed self-administered questionnaires immediately after their visits. All were informed that responses were confidential and would not be disclosed to one another. This study was reviewed and approved by the Colorado Multiple Institutional Review Board.

Questionnaire development

The survey instruments were developed after a thorough review of existing research^1,6-10and refined by pilot testing and review with a professional survey consultant. Patients and companions completed demographic questions (Table 1). Patients rated their overall health, stated their relationship to the companion, and indicated the reasons for companion accompaniment (Table 2). Patients and companions indicated from a list of 7 items (Table 2) how the companion influenced the visit and rated the companion’s helpfulness during the encounter (5-point Likert scale: 1 = very unhelpful to 5 = very helpful).

TABLE 1

Patient demographics and characteristics*

Patients’ and companions’ reports of companion’s reasons for accompaniment and influence on the medical encounter*

TABLE 3

Physician report of examination room companion’s influence on the medical encounter*

Physician report of examination room companion’s behaviors

Data analysis

Data were used from each member of a set regardless of survey completion by other set members. One patient had 2 examination room companions and 2 patients had 2 waiting room companions. In each case, both companions were surveyed. In the 2 cases in which a patient had waiting and examination room companions, the examination room companion was considered more influential for the medical encounter and only that person was surveyed.

The data were analyzed with SAS version 6.12 (SAS Inc, Cary, NC) using bivariate and multivariable methods. Comparisons were made between patient categories (patient alone, patient with examination room companion, patients with waiting room companion) using the chi-square statistic for categorical variables.

Multivariable analyses were conducted to explore the effects of various independent variables on the decision to bring a companion into the examination room. The outcome variables for the regression models were defined by patient status (patient alone, patient with examination room companion, patient with waiting room companion). All significant variables (P ≤ .05) in bivariate analyses were entered into the multivariate analyses. Odds ratios (ORs) with 95% confidence intervals (CIs) were obtained for each variable in the model.

Patient and companion agreement on the reasons for accompaniment and influence on the medical encounter were measured with the kappa statistic. Kappas (κ) of 1.0 to .75 denote excellent agreement, .4 to .75 denote good agreement, and 0 to .4 denote marginal agreement.¹¹

RESULTS

Of the 1294 patient visits, 834 (64%) were to faculty physicians and 451 (35%) were to resident physicians. Overall, companions were present for 29% (n = 374) of patient visits and accompanied the patient into the examination room for 16% (n = 212) of visits. Companions accompanied patients to 23% (n = 196) of faculty visits and 39% (n = 178) of resident visits (P P = .98).

Ninety-three percent (121/130) of unaccompanied patients and 92% (200/217) of consecutive patient–companion pairs approached for consent agreed to participate in the study. In 26 cases the patient or the companion refused to participate for 1of the following reasons: language barrier, too ill, lack of time, invasion of privacy, or uncomfortable with process of consent. Patients and physicians completed surveys for 97% of patient encounters, and companions completed surveys for 99% of patient encounters.

Comparisons between accompanied and unaccompanied patients are presented in Table 1. The fact that faculty physicians had a greater proportion of accompanied patients who received and completed surveys is likely due to the necessity of English literacy for inclusion into the survey study. At this practice site, patients cared for by residents are more likely to be non-English speaking and have a companion for the purpose of translation. Examination room companions were often the spouse or partner (55%), parents (17%), or less frequently roommate or friend (7%), whereas waiting room companions were commonly a spouse or partner (46%) or roommate or friend (24%). Overall, family members accounted for 93% of examination room companions and 76% of waiting room companions. Examination room companions were more likely to be female than waiting room companions (65% vs 51%, P = .05). Patients who were older, less well educated, and whose cases had greater medical or social complexity were more likely to have a companion in the examination room.

Patients’ and companions’ assessments of the reasons for accompaniment and the companions’ influence on communication and resource use are shown in Table 2. The patients’ and companions’ stated reasons for companion accompaniment were in good agreement with the kappa statistic ranging from 0.41 for “help with insurance forms” to 0.61 for “help remember the physician’s advice.” Patients’ and companions’ agreement regarding the influence of the companion on the medical visit was less than 0.4, suggesting marginal agreement for tests ordered (κ = .29), prescribed treatment (κ = .36), and length of visit (κ = .33). There was good agreement for number of referrals (κ = .45) and for physician and patient understanding (κ = .62 and .60, respectively).

Table 3 displays the physicians’ reports of the examination room companion’s influence on the medical encounter. Table 4 shows the physicians’ reports of the behavior of the examination room companions.

Patients regarded examination and waiting room companions as “very helpful” for 84% and 71% of visits, respectively, and as “very unhelpful” for 1% of visits. Of the 121 patients who came alone to their medical visits, 7% indicated that they considered bringing a companion to their visits and 16% thought a companion’s presence would have been helpful.

Physicians regarded examination room companions as “somewhat to very helpful” for 66% of visits. When physicians did not have contact with a companion, they indicated that contact would have been helpful for 16% of patient encounters.

Multivariable analyses explored the effects of independent variables on the decision to bring a companion into the examination room. A physician rating of a case of having greater medical and social complexity was the only variable associated with companion accompaniment to the examination room vs not having a companion (OR, 1.7; 95% CI, 1.4–2.1). Patient characteristics and patients’ reported reasons for accompaniment were factors influencing accompaniment into the examination room vs the waiting room. A need for help with communicating concerns to the physician (OR, 7.8; 95% CI, 2.4–25.6), help with remembering the physician’s advice and instructions (OR, 7.1; 95% CI, 2.0–25.3), and greater medical and social complexity of cases (OR, 1.5; 95% CI, 1.1–2.0) were associated with being accompanied to the examination room over just the waiting room. In contrast, needing help with transportation was negatively associated with having a companion in the examination room vs the waiting room (OR, 0.2; 95% CI, 0.1–0.5).

DISCUSSION

Companions frequently accompany adult patients and participate in ambulatory medical encounters. We found that companions accompanied 29% of patients and were present in the examination room for 16% of outpatient medical encounters. Examination room companions often were present to aid communication with the physician and to help the patient remember instructions. Physicians, patients, and companions believed that physician and patient understanding often were favorably affected by the presence of a companion. A companion’s presence in the examination room had beneficial effects on patient and physician understanding and very rarely had a negative effect. These findings reflect the results of previous studies in which approximately 33% of patients were accompanied to family medicine and geriatric encounters and 66% of these companions were present in the examination room.^1,7-10,12,13 The only variable associated with accompaniment into the examination room vs presenting to one’s appointment alone was a physician rating of greater medical and social complexity. This finding also corroborates the findings of other studies.^7,14,15

Some research has suggested a negative effect of a third person during the medical encounter. For example, older patients in triadic encounters raise fewer topics, are less assertive, and participate less in humor and joint decision making.⁶ Other research has failed to find a benefit in health outcomes with the presence of a companion, and unaccompanied patients rated themselves as having greater understanding of their medical problems and greater faith that their physicians were doing everything possible for them than patients accompanied to their visits.¹⁵ Still others have found that physicians provide more information and time but less emotional support to accompanied patients.⁷

Unique to our study was the specific assessment of companion influence on various aspects of the medical encounter from the perspectives of the companion, patient, and physician. As expected, examination room companions had significant influence on aspects of communication. Examination room companions were generally considered helpful by patients and physicians. Physicians may use the companion and patient as barometers of the visit’s accomplished goals. Hence, the time spent listening to a companion provide information about a patient’s medical problems might be balanced by the provision of less emotional support to the patient, especially if the companion is providing that support. The physician may offer an explanation until confident that either party (patient or companion) has a complete understanding. As demonstrated by the results, physicians, patients, and companions thought that patient understanding was increased in approximately 50% of encounters by a companion’s presence, and companions overwhelmingly were considered very helpful by patients.

Our study was limited to 1 urban, academic, general internal medicine practice and may not be generalizable to other settings. The method of assessing medical and social complexities was simple, and the very presence of an examination room companion may have biased physicians to rate these patients as having more complex problems than unaccompanied patients. Also, the longer period of the second part of the study necessary to enroll 200 patient–companion pairs, compared with the companion frequency data of the first part of the study, suggested that we did not enroll “consecutive” patient–companion pairs. This may be explained by the exclusion of patient–companion pairs when either party was not literate in English. Also, the logistics of obtaining consent and administering post visit questionnaires by a single research assistant interfered with the attempt to enroll all patient–companion pairs. Convenience enrollment of unaccompanied patients may have been biased. The effect of the companion on the medical encounter was not verified by objective measures such as timing visit length. Further, patients, companions, and physicians rated the effect of the companion’s presence immediately after the encounter; the full effect of the companion’s presence might require more time to emerge.

Companions frequently accompany patients to their ambulatory general medicine visits. The companion is usually a family member who is present at the request of the patient. Companions assume important roles and are overwhelmingly considered helpful by patients. Nonetheless, some of the behaviors that contribute to an effective physician–patient relationship might be diluted by the presence of a companion. Future directions of study include (1) assessing the effect of the companion on the physician–patient relationship, including specific aspects of communication and behavior, and (2) determining whether companion involvement influences health outcomes or resource use.

· Acknowledgments ·

We thank Marcia Blake, MA, for her statistical expertise; Jean Kutner, MD, MSPH, for her assistance with data analysis and statistical expertise; and Robert Dellavalle, MD, PhD, for his editorial assistance.

Until recently, the satisfaction of physicians was studied only among those working in special settings, such as in rural and urban underserved areas. In these settings, a less than satisfying work environment sometimes leads physicians to leave their jobs, with subsequent costs for needy communities.^1,2 Now, with US physicians often employed by others, physicians’ autonomy constrained, medicine managed as a business, and incomes falling for some specialties, the satisfaction of rank and file physicians in all locations and specialties can no longer be assumed.^3-5 Indeed, unhappiness is increasing among physicians, with growing numbers considering job changes, nonclinical work, and early retirement.^6-8 Some physicians regret ever pursuing a medical career.^8-10

Dissatisfaction among physicians is a concern even if medical work still pays handsomely and medical schools still have an abundance of applicants. Physicians’ satisfaction relates to their job effectiveness, including the quality of care they render, their patients’ compliance with recommended care, patients’ satisfaction with care, and patients’ health outcomes.^6,9,11-15 The most obvious consequence of a dissatisfying job, however, is that physicians will quit to work elsewhere, disrupting patient-physician continuity and creating organizational instability and replacement costs, estimated at nearly $250,000 per physician lost.^16-18 A job change also brings personal and financial costs to physicians and their families.

Although common experience and prior studies show that physicians’ job satisfaction generally influences whether they leave their jobs, little more is known about how satisfaction relates to job changes for physicians. From studies of other professions, we know that satisfaction is a multidimensional concept in which individuals can be satisfied with some aspects of work while dissatisfied with others.^16,19 Further, we know that workers’ satisfaction with only some aspects of work correlates with turnover.^16,20 For example, in 1 recent study rural generalist physicians who were satisfied with their local communities remained longer in their jobs, whereas their satisfaction with most other aspects of work, including with autonomy and doctor-patient relationships, was unrelated to their retention.² The effects of workers’ satisfaction and dissatisfaction may also require separate assessment, because greater dissatisfaction with a given aspect of work may lead a worker to leave a job, but increasingly higher levels of satisfaction with the same issue may not lead to ever-increasing chances of retention, or vice versa.^15,21

The goal of this study was to better understand the relationship between physicians’ satisfaction with various aspects of their work and whether or not they have thoughts or plans to leave their jobs. We assessed whether anticipated job departure related to satisfaction at both the upper and lower ends of its range; that is, whether particularly high satisfaction levels relative to other physicians protected against job change plans and significant relative dissatisfaction promoted such plans. We also assessed how the satisfaction/job-change relationship varied for physician groups with differing needs and values, specifically for generalists compared with specialists, and for younger, midcareer, and older physicians.

Methods

Subjects, survey mailings, and response rates

A sampling frame, previously described,^22,23 was constructed from the American Medical Association’s Physician Masterfile. The target population was the 171,252 civilian physicians listed as participating in direct patient care (rather than in training, administration, teaching, inactive, etc), including generalists (family physicians, general internists, and general pediatricians; general practitioners were not included) and specialists (medical and pediatric nonsurgical subspecialists) in the United States. We stratified the frame into geographic regions of high and low penetration of managed care, non-Hispanic white vs other physician ethnicity, and the 5-targeted specialty groupings.

We drew a random sample of 5704 physicians, oversampling smaller strata. Questionnaires and up to 3 follow-up mailings were sent to the sample from January through September 1997. A total of 2325 physicians responded, 52% when corrected for noncontact and ineligibility.²⁴ Based on information on the Masterfile, response rates did not differ by age, for those in office vs hospital-based practices, or for those in communities of high vs low managed care penetration. However, response rates were higher for subspecialists than for generalists, US medical graduates than for international graduates, and those for whom race-ethnicity was listed as black, “missing,” or “other” than for those listed as Hispanic, non-Hispanic white, or Asian. Statistical weighting in analyses adjusted for group contact and response rate differences.

We dropped the 218 respondents who worked less than 25 hours per week in activities related to patient care and another 169 physicians who primarily practiced outside of their specialty, such as general internists who worked as endocrinologists or oncologists. Analyses were based on the remaining 1939 physicians.

Questionnaire design and analyses

The questionnaire included 36 declarative statements about physicians’ views and regard for various features of their work and practices, to which respondents provided Likert-scaled responses from “strongly agree” to “strongly disagree.”²² Representative statements were, “I feel a strong personal connection with my patients” and “My total compensation package is fair.” These items were drawn from previous studies of physician satisfaction, from issues raised in physician focus groups we assembled, and through the suggestions of experts. We then used exploratory factor analysis with a pilot physician sample, followed by confirmatory factor analysis with a validation physician sample to group these items into 10 scales reflecting physicians’ satisfaction with 10 aspects of their work.^22,23 Scale alpha coefficients of reliability ranged from 0.65 to 0.77.

Physicians’ anticipated departure from their jobs was measured through a questionnaire item asking “What is the likelihood that you will leave your current practice situation within two years?”, with allowable responses of “none,” “slight,” “moderate,” “likely,” and “definitely.” Social psychology research has substantiated the use of people’s expressed intentions as measures of their future behavior.²⁵ Prior studies^26-28 have used anticipated departure and anticipated retention as expeditious proxy indicators for actual job departure and retention, with correlation coefficients between workers’ plans to leave and subsequent departure typically measured at about 0.5. In the only validating physician study, rural physicians in Western Australia much more often left their practices if they had predicted 10 years earlier that they would (odds ratio [OR], 3.3).²⁹ Shorter-term predictions, such as the 2-year time horizon used in this study, should be even more accurate.

This study’s dichotomous outcome indicator-whether physicians anticipated leaving their jobs within 2 years-defined planned “leavers” as those who indicated a moderate, likely, or definite likelihood of leaving their practices; planned “retainees” were those who envisioned only a slight or no chance of leaving. This cut point was selected for both its face validity-falling at the scale point at which leaving became a real possibility for subjects and no longer an unlikely event-and because it yielded an adequately balanced split in outcome values.

In analyses, levels of satisfaction with the 10 aspects of work and anticipated job departure were described for generalists, specialists, and physicians in 3 age groups. We compared groups on satisfaction levels and anticipated departure rates with t tests and 1-way analysis of variance.

We used logistic regression analysis to model separately the anticipated departure of each of the 5 groups. We entered satisfaction levels for each of the 10 areas of work into the models as 2 dichotomous (dummy) variables, yielding 20 satisfaction indicator variables. Ten of these indicators reflected values in the top quartile of satisfaction (ie, top quartile vs lower 3 quartiles) in each area of work and another 10 indicators reflected values in the lowest quartile of satisfaction (bottom quartile vs upper 3 quartiles). Cut points for the middle vs high satisfaction groups for the 10 areas of work ranged from 3.0 to 4.25 on the satisfaction scale (in which 1 indicated dissatisfaction; 3, neutral; and 5, satisfaction); for the middle vs lower satisfaction group, cut points ranged from 2.0 to 3.33. The inclusion of dichotomous indicators of high and low satisfaction for each area of work treated the middle 2 satisfaction quartiles statistically as the omitted, comparison category. This analytic approach modeled the effects on anticipated departure of relatively high and low satisfaction levels for each area of work, compared with more typical, midrange satisfaction levels, simultaneously controlling for relative satisfaction in the other 9 areas of work. We also included control variables for sex, age, and whether physicians owned their practices.

The logistic models for the 5 physician groups were repeated with the more parsimonious forward stepwise variable selection approach to test the robustness of the model findings and as a functional colinearity check.³⁰ With only 1 exception of the 100 satisfaction-departure associations tested across the 5 physician groups, the findings were identical to those of the full logistic models. We reported only the full logistic models.

The statistical significance level was set at .01, to partially adjust for this study’s numerous comparisons.

Results

Unweighted characterizations of the 1939 eligible respondent physicians found that 70% were male and 72% non-Hispanic white, with a median age of 46 years (range, 27-88 years). Most were married (84%) and had children (85%). There were 459 family physicians (24%), 375 general internists (19%), 494 general pediatricians (25%), and 611 pediatric and medical subspecialists (32%). Eighteen percent worked in solo practices, 36% worked in small groups, and the rest worked in larger groups. Most (57%) owned their practices. Physicians had worked in their current practices for a median duration of 9 years when surveyed (range, 0-58 years).

Satisfaction scores

As a whole, the generalist and specialist physicians reported midrange satisfaction levels in the 10 areas of work assessed Table 1). For 2 work facets-administrative responsibilities and having adequate personal time-the average satisfaction scores for both groups indicated overall dissatisfaction. Physicians indicated greatest satisfaction with their intrinsic rewards from patient care, specifically their personal connection with and gratitude from patients.

The mean satisfaction scores of generalists and specialists were comparable for 7 of the 10 aspects of practice. In the remaining 3 areas, the generalist-specialist differences in satisfaction were small. Satisfaction scores varied more often with age, with physicians 55 years and older more satisfied than younger physicians in 8 of the 10 areas queried.

Anticipated job turnover

In total, 27.1% of physicians foresaw leaving their practices within the next 2 years: 13.8% believed leaving was likely or definite and 13.3% believed they had a moderate chance of leaving. Generalist and specialist physicians perceived similar chances of leaving (26.7% vs 30.0%, respectively; P = .31), with and without adjusting for age and sex. In contrast, physicians of various age groups anticipated different likelihoods of leaving Figure 1), with a 29% rate for the youngest group, a nadir of 22% to 23% for physicians aged 45 to 54 years, and again higher rates for older physicians, reaching 63% for those 70 years and older (P < .001).

Those who anticipated leaving their jobs within 2 years frequently also anticipated leaving the practice of clinical medicine entirely within 5 years, including 84% of leavers older than 54 years, 50% of leavers 45 to 54 years, and even 33% of leavers younger than 45 years.

Satisfaction with work and anticipated job turnover

Overall observations. In analyses that controlled for satisfaction with other facets of work and for several physician characteristics, we found that physicians’ regard for only some aspects of work was associated with their anticipated departure from their jobs (Table 2 and Table 3). Further, we found that physicians who were relatively dissatisfied with a variety of aspects of their work more often foresaw leaving their jobs than those with median satisfaction levels. In contrast, intentions to leave were generally just as likely for those of high and median satisfaction levels.Figure 2 shows the typical relationship between satisfaction and anticipated departure. In this example, as in most instances, the transition from lower to higher likelihood of anticipated departure occurs only when physicians’ feelings drop below neutral into dissatisfied levels.

Generalists and specialists. Among generalists and compared with those of average satisfaction (the omitted group in the statistical models), leaving one’s job was anticipated more often by those less satisfied with their pay (OR, 3.65; P < .0001), with their relationships with their communities (OR, 2.26; P < .0001), and with their relationships with the nonphysician staff in their offices (OR, 1.59; P < .01) Table 2). Only for 1 area of work, patient care issues (eg, feeling overwhelmed with patients’ demands and feeling that time constraints prevented good relationships with patients), were generalists in the highest satisfaction quartile less likely to anticipate leaving their practices than generalists with midrange satisfaction scores (OR, 0.60; P < .01).

For specialists, those reporting lowest satisfaction with their pay (OR, 3.24; P = .01) and community relationships (OR, 3.89; P < .01) anticipated leaving more often than those with midrange scores Table 2). For no aspects of work were the most satisfied specialists less likely to anticipate leaving than specialists of average satisfaction.

Younger, midcareer, and older physicians. Among physicians in the youngest group-aged 27 through 44 years-those who were least satisfied with their pay (OR, 4.62; P < .0001) and with their relationships with their communities (OR, 2.79; P < .0001) more often anticipated leaving their jobs than those of median satisfaction Table 3). For no facets of work among younger physicians were those who were most satisfied less likely to anticipate leaving.

For physicians aged 45 through 54 years, anticipated job departure was more common for those least satisfied with 5 areas of work, including again those least satisfied with pay (OR, 11.60; P < .0001) and with their community relationships (OR, 2.11; P < .01). One unexpected association was noted for these midcareer physicians: those in the highest quartile of satisfaction with their pay more often anticipated leaving their practices than those with midrange pay satisfaction scores (OR, 3.36; P < .01).

In the oldest physician group, those least satisfied with their community relationships (OR, 3.31; P < .01) and least satisfied with the administrative aspects of their work (OR, 2.64; P < .01) more often anticipated leaving their jobs. Fewer of those most satisfied with patient care issues foresaw leaving (OR, 0.30; P < .01). Unexpectedly, older physicians who were least satisfied with their relationships with other physicians were less likely to anticipate leaving their practices (OR, 0.36; P < .01).

Discussion

Our study has demonstrated 3 aspects of how physicians’ regard for their work relates to their plans to leave their jobs: (1) physicians’ plans to leave correlate with their satisfaction with only some aspects of their work; (2) the aspects of work for which satisfaction is associated with plans for leaving differ somewhat for generalists and specialists, and for physicians at various stages of their careers; and (3) anticipated job departure is more common among physicians who are relatively dissatisfied with any of a variety of aspects of their work, whereas satisfaction higher than median levels generally does not protect physicians from thoughts of leaving their jobs. To our knowledge, these last 2 features of the relationship between satisfaction and anticipated departure have not been demonstrated previously.

Even when physicians are displeased with a given aspect of their work, they are not always inclined to leave their jobs. For example, physicians in this and other studies^2,31 were least satisfied with the administrative requirements of their work, yet dissatisfaction in this area was not associated with the departure plans of most physician groups. In some cases, a dissatisfying issue is simply not important enough to individuals to warrant a job change. Physicians also have options to leaving, such as problem-solving with managers and coworkers, and taking comfort in the more satisfying aspects of their work.³² Thus, to know when dissatisfaction leads to job changes, one must actually test for statistical associations between the two.

In tests of this kind in the present study, we found 14 instances in which physicians in the lowest satisfaction quartiles were more likely to anticipate leaving than were those of average satisfaction. We found only 2 situations in which physicians in the highest satisfaction quartiles were less likely to anticipate leaving. To promote retention, these data suggest that physicians and their employers should avoid relative dissatisfaction in particular. These data also suggest that building ever-increasing higher levels of satisfaction generally does not prevent turnover. Thus, to foster workforce stability, practice managers should address aspects of physicians’ work or jobs causing them dissatisfaction, or more specifically, satisfaction lower than that of physicians working elsewhere. For aspects of work about which physicians are already satisfied, managers need not foster even higher satisfaction.

Satisfaction with specific aspects of work and turnover

Relative dissatisfaction with one’s community was associated with departure plans for all physician groups. An individual’s relationship with a community-measured in this study through questions about feeling “at home,” a sense of belonging, respected, and strongly connected²³-is basic to human existence,^33,34 but its importance has been underappreciated in physicians’ lives and careers.^33,35 Practices might promote workforce stability by providing opportunities and incentives for employed physicians to participate in local and state professional and community service organizations. Expectations can be set and time allocated for physicians to work with school health programs and volunteer clinics, speak to community groups, and become involved in other ways that build a sense of membership and contribution. These activities will be new for some physicians, will not always come naturally, and may need to be taught and encouraged.

Relative dissatisfaction with pay was associated with departure plans for all of this study’s physician groups, except for physicians older than 55 years. Further and unexpectedly, midcareer physicians who were particularly satisfied with their pay were also more likely to anticipate leaving. Perhaps midcareer physicians who are and feel well paid also more often feel that they can afford to retire or pursue second careers. We have no data to test this explanation. Similarly, we cannot know why older physicians who were least satisfied with their collegial relationships less often planned to leave their jobs. We wondered if this surprising finding was a result of confounding by physicians’ employment in group vs solo practices; we reasoned that among older physicians, those in solo practice may not get along as well with local physicians, but for unrelated reasons may be more attached to their practices and thus less likely to retire or otherwise plan to leave. This possibility proved not to be the case when we added an indicator of solo vs group practice to the statistical model for the older physician group.

How physicians felt about their relationships with patients and about their autonomy was not related to their intentions to leave their practices. Perhaps these once-central features of doctoring have lost their importance to physicians, or maybe physicians believe they will not find better relationships or more autonomy elsewhere, and thus see no benefit in changing jobs.

Age, satisfaction, and anticipated job changes

As in previous studies,^36,37 the older physicians of this study were generally more satisfied than the younger physicians. Older physicians, nevertheless, more often anticipated leaving their jobs. We suspect that in many cases an older physician’s thoughts of leaving a position are related to their retirement plans, but our data did not allow us to explore this relationship in detail. We also suspect that physicians’ satisfaction with their jobs influences their plans not only for leaving that particular job but also their plans for leaving the practice of medicine entirely; however, satisfaction with somewhat different aspects of work may influence each of the 2 types of career moves. For example, relative dissatisfaction with relationships with patients may push a physician toward retirement even though it did not seem to promote thoughts of a job change for this study’s physicians.

Common beliefs about physicians’ careers would have predicted the “J”-shape relationship between anticipated job departure rates and increasing physician age Figure 1). Young physicians fresh from training are thought to change jobs once or twice during an initial period of professional experimentation, then settle into suitable practice situations and raise their families. Later in life, job departures again become more common when children leave home, work takes on new meaning in middle age, and physicians become interested in and able to retire or pursue new jobs or second careers. Data from this study generally confirm these common notions, although the nadir in anticipated job departures occurred later than we expected.

Limitations

This study’s 52% response rate, although typical for physician studies,³⁸ may have yielded results not representative of physicians as a whole. To assess this possibility, we compared survey data from this study’s early and late respondents and found meaningful time trends in only 4 of 140 questionnaire items, providing some evidence that nonrespondents-if assumed to be in effect very late respondents-were similar to respondents. Nevertheless, to help adjust for any response bias we weighted analyses.

Statistical associations may not reflect causal relationships in a cross-sectional study such as this. Further, this study’s findings may have differed if actual job departure was available as an outcome measure, even though anticipated departure is a validated and frequently used proxy indicator of workers who will leave their jobs.^26-28 A measure of convergent validity was demonstrated in this study, as 16 of the 18 statistical associations between anticipated departure and satisfaction were in the expected direction, and age fluctuated with anticipated departure as we had expected it would with actual departure.

Conclusions and implications

These findings suggest that physicians who are relatively dissatisfied with specific aspects of their work more often plan to leave their jobs, and that these particular work aspects differ somewhat with age and specialty. Practice managers and employers concerned with maintaining a stable physician workforce should address relative dissatisfaction among physicians-particularly with income and community relationships-but need not build high levels of satisfaction. Physicians looking for stable employment should seek positions that they feel offer appropriate compensation and are located in communities to which they can connect. Once there, they should devote time and energy building ties in their communities.

Further research must confirm whether retention indeed improves by remediating relative dissatisfaction with the areas of physicians’ work flagged as important in this study. It also remains to be seen whether other outcomes of physicians’ work, such as the quality of care they provide, also relate primarily to relative dissatisfaction rather than satisfaction, and, specifically, to dissatisfaction with the same aspects of work seeming to affect job departure plans.

Acknowledgments

The authors thank the physician-fellows of UNC’s Primary Care Research Fellowship and the JFP’s reviewers for their thoughtful suggestions for strengthening this report. We also thank the many physician respondents who shared their time and revealed to us important aspects of their lives.

Until recently, the satisfaction of physicians was studied only among those working in special settings, such as in rural and urban underserved areas. In these settings, a less than satisfying work environment sometimes leads physicians to leave their jobs, with subsequent costs for needy communities.^1,2 Now, with US physicians often employed by others, physicians’ autonomy constrained, medicine managed as a business, and incomes falling for some specialties, the satisfaction of rank and file physicians in all locations and specialties can no longer be assumed.^3-5 Indeed, unhappiness is increasing among physicians, with growing numbers considering job changes, nonclinical work, and early retirement.^6-8 Some physicians regret ever pursuing a medical career.^8-10

Dissatisfaction among physicians is a concern even if medical work still pays handsomely and medical schools still have an abundance of applicants. Physicians’ satisfaction relates to their job effectiveness, including the quality of care they render, their patients’ compliance with recommended care, patients’ satisfaction with care, and patients’ health outcomes.^6,9,11-15 The most obvious consequence of a dissatisfying job, however, is that physicians will quit to work elsewhere, disrupting patient-physician continuity and creating organizational instability and replacement costs, estimated at nearly $250,000 per physician lost.^16-18 A job change also brings personal and financial costs to physicians and their families.

Although common experience and prior studies show that physicians’ job satisfaction generally influences whether they leave their jobs, little more is known about how satisfaction relates to job changes for physicians. From studies of other professions, we know that satisfaction is a multidimensional concept in which individuals can be satisfied with some aspects of work while dissatisfied with others.^16,19 Further, we know that workers’ satisfaction with only some aspects of work correlates with turnover.^16,20 For example, in 1 recent study rural generalist physicians who were satisfied with their local communities remained longer in their jobs, whereas their satisfaction with most other aspects of work, including with autonomy and doctor-patient relationships, was unrelated to their retention.² The effects of workers’ satisfaction and dissatisfaction may also require separate assessment, because greater dissatisfaction with a given aspect of work may lead a worker to leave a job, but increasingly higher levels of satisfaction with the same issue may not lead to ever-increasing chances of retention, or vice versa.^15,21

The goal of this study was to better understand the relationship between physicians’ satisfaction with various aspects of their work and whether or not they have thoughts or plans to leave their jobs. We assessed whether anticipated job departure related to satisfaction at both the upper and lower ends of its range; that is, whether particularly high satisfaction levels relative to other physicians protected against job change plans and significant relative dissatisfaction promoted such plans. We also assessed how the satisfaction/job-change relationship varied for physician groups with differing needs and values, specifically for generalists compared with specialists, and for younger, midcareer, and older physicians.

Methods

Subjects, survey mailings, and response rates

A sampling frame, previously described,^22,23 was constructed from the American Medical Association’s Physician Masterfile. The target population was the 171,252 civilian physicians listed as participating in direct patient care (rather than in training, administration, teaching, inactive, etc), including generalists (family physicians, general internists, and general pediatricians; general practitioners were not included) and specialists (medical and pediatric nonsurgical subspecialists) in the United States. We stratified the frame into geographic regions of high and low penetration of managed care, non-Hispanic white vs other physician ethnicity, and the 5-targeted specialty groupings.

We drew a random sample of 5704 physicians, oversampling smaller strata. Questionnaires and up to 3 follow-up mailings were sent to the sample from January through September 1997. A total of 2325 physicians responded, 52% when corrected for noncontact and ineligibility.²⁴ Based on information on the Masterfile, response rates did not differ by age, for those in office vs hospital-based practices, or for those in communities of high vs low managed care penetration. However, response rates were higher for subspecialists than for generalists, US medical graduates than for international graduates, and those for whom race-ethnicity was listed as black, “missing,” or “other” than for those listed as Hispanic, non-Hispanic white, or Asian. Statistical weighting in analyses adjusted for group contact and response rate differences.

We dropped the 218 respondents who worked less than 25 hours per week in activities related to patient care and another 169 physicians who primarily practiced outside of their specialty, such as general internists who worked as endocrinologists or oncologists. Analyses were based on the remaining 1939 physicians.

Questionnaire design and analyses

The questionnaire included 36 declarative statements about physicians’ views and regard for various features of their work and practices, to which respondents provided Likert-scaled responses from “strongly agree” to “strongly disagree.”²² Representative statements were, “I feel a strong personal connection with my patients” and “My total compensation package is fair.” These items were drawn from previous studies of physician satisfaction, from issues raised in physician focus groups we assembled, and through the suggestions of experts. We then used exploratory factor analysis with a pilot physician sample, followed by confirmatory factor analysis with a validation physician sample to group these items into 10 scales reflecting physicians’ satisfaction with 10 aspects of their work.^22,23 Scale alpha coefficients of reliability ranged from 0.65 to 0.77.

Physicians’ anticipated departure from their jobs was measured through a questionnaire item asking “What is the likelihood that you will leave your current practice situation within two years?”, with allowable responses of “none,” “slight,” “moderate,” “likely,” and “definitely.” Social psychology research has substantiated the use of people’s expressed intentions as measures of their future behavior.²⁵ Prior studies^26-28 have used anticipated departure and anticipated retention as expeditious proxy indicators for actual job departure and retention, with correlation coefficients between workers’ plans to leave and subsequent departure typically measured at about 0.5. In the only validating physician study, rural physicians in Western Australia much more often left their practices if they had predicted 10 years earlier that they would (odds ratio [OR], 3.3).²⁹ Shorter-term predictions, such as the 2-year time horizon used in this study, should be even more accurate.

This study’s dichotomous outcome indicator-whether physicians anticipated leaving their jobs within 2 years-defined planned “leavers” as those who indicated a moderate, likely, or definite likelihood of leaving their practices; planned “retainees” were those who envisioned only a slight or no chance of leaving. This cut point was selected for both its face validity-falling at the scale point at which leaving became a real possibility for subjects and no longer an unlikely event-and because it yielded an adequately balanced split in outcome values.

In analyses, levels of satisfaction with the 10 aspects of work and anticipated job departure were described for generalists, specialists, and physicians in 3 age groups. We compared groups on satisfaction levels and anticipated departure rates with t tests and 1-way analysis of variance.

We used logistic regression analysis to model separately the anticipated departure of each of the 5 groups. We entered satisfaction levels for each of the 10 areas of work into the models as 2 dichotomous (dummy) variables, yielding 20 satisfaction indicator variables. Ten of these indicators reflected values in the top quartile of satisfaction (ie, top quartile vs lower 3 quartiles) in each area of work and another 10 indicators reflected values in the lowest quartile of satisfaction (bottom quartile vs upper 3 quartiles). Cut points for the middle vs high satisfaction groups for the 10 areas of work ranged from 3.0 to 4.25 on the satisfaction scale (in which 1 indicated dissatisfaction; 3, neutral; and 5, satisfaction); for the middle vs lower satisfaction group, cut points ranged from 2.0 to 3.33. The inclusion of dichotomous indicators of high and low satisfaction for each area of work treated the middle 2 satisfaction quartiles statistically as the omitted, comparison category. This analytic approach modeled the effects on anticipated departure of relatively high and low satisfaction levels for each area of work, compared with more typical, midrange satisfaction levels, simultaneously controlling for relative satisfaction in the other 9 areas of work. We also included control variables for sex, age, and whether physicians owned their practices.

The logistic models for the 5 physician groups were repeated with the more parsimonious forward stepwise variable selection approach to test the robustness of the model findings and as a functional colinearity check.³⁰ With only 1 exception of the 100 satisfaction-departure associations tested across the 5 physician groups, the findings were identical to those of the full logistic models. We reported only the full logistic models.

The statistical significance level was set at .01, to partially adjust for this study’s numerous comparisons.

Results

Unweighted characterizations of the 1939 eligible respondent physicians found that 70% were male and 72% non-Hispanic white, with a median age of 46 years (range, 27-88 years). Most were married (84%) and had children (85%). There were 459 family physicians (24%), 375 general internists (19%), 494 general pediatricians (25%), and 611 pediatric and medical subspecialists (32%). Eighteen percent worked in solo practices, 36% worked in small groups, and the rest worked in larger groups. Most (57%) owned their practices. Physicians had worked in their current practices for a median duration of 9 years when surveyed (range, 0-58 years).

Satisfaction scores

As a whole, the generalist and specialist physicians reported midrange satisfaction levels in the 10 areas of work assessed Table 1). For 2 work facets-administrative responsibilities and having adequate personal time-the average satisfaction scores for both groups indicated overall dissatisfaction. Physicians indicated greatest satisfaction with their intrinsic rewards from patient care, specifically their personal connection with and gratitude from patients.

The mean satisfaction scores of generalists and specialists were comparable for 7 of the 10 aspects of practice. In the remaining 3 areas, the generalist-specialist differences in satisfaction were small. Satisfaction scores varied more often with age, with physicians 55 years and older more satisfied than younger physicians in 8 of the 10 areas queried.

Anticipated job turnover

In total, 27.1% of physicians foresaw leaving their practices within the next 2 years: 13.8% believed leaving was likely or definite and 13.3% believed they had a moderate chance of leaving. Generalist and specialist physicians perceived similar chances of leaving (26.7% vs 30.0%, respectively; P = .31), with and without adjusting for age and sex. In contrast, physicians of various age groups anticipated different likelihoods of leaving Figure 1), with a 29% rate for the youngest group, a nadir of 22% to 23% for physicians aged 45 to 54 years, and again higher rates for older physicians, reaching 63% for those 70 years and older (P < .001).

Those who anticipated leaving their jobs within 2 years frequently also anticipated leaving the practice of clinical medicine entirely within 5 years, including 84% of leavers older than 54 years, 50% of leavers 45 to 54 years, and even 33% of leavers younger than 45 years.

Satisfaction with work and anticipated job turnover

Overall observations. In analyses that controlled for satisfaction with other facets of work and for several physician characteristics, we found that physicians’ regard for only some aspects of work was associated with their anticipated departure from their jobs (Table 2 and Table 3). Further, we found that physicians who were relatively dissatisfied with a variety of aspects of their work more often foresaw leaving their jobs than those with median satisfaction levels. In contrast, intentions to leave were generally just as likely for those of high and median satisfaction levels.Figure 2 shows the typical relationship between satisfaction and anticipated departure. In this example, as in most instances, the transition from lower to higher likelihood of anticipated departure occurs only when physicians’ feelings drop below neutral into dissatisfied levels.

Generalists and specialists. Among generalists and compared with those of average satisfaction (the omitted group in the statistical models), leaving one’s job was anticipated more often by those less satisfied with their pay (OR, 3.65; P < .0001), with their relationships with their communities (OR, 2.26; P < .0001), and with their relationships with the nonphysician staff in their offices (OR, 1.59; P < .01) Table 2). Only for 1 area of work, patient care issues (eg, feeling overwhelmed with patients’ demands and feeling that time constraints prevented good relationships with patients), were generalists in the highest satisfaction quartile less likely to anticipate leaving their practices than generalists with midrange satisfaction scores (OR, 0.60; P < .01).

For specialists, those reporting lowest satisfaction with their pay (OR, 3.24; P = .01) and community relationships (OR, 3.89; P < .01) anticipated leaving more often than those with midrange scores Table 2). For no aspects of work were the most satisfied specialists less likely to anticipate leaving than specialists of average satisfaction.

Younger, midcareer, and older physicians. Among physicians in the youngest group-aged 27 through 44 years-those who were least satisfied with their pay (OR, 4.62; P < .0001) and with their relationships with their communities (OR, 2.79; P < .0001) more often anticipated leaving their jobs than those of median satisfaction Table 3). For no facets of work among younger physicians were those who were most satisfied less likely to anticipate leaving.

For physicians aged 45 through 54 years, anticipated job departure was more common for those least satisfied with 5 areas of work, including again those least satisfied with pay (OR, 11.60; P < .0001) and with their community relationships (OR, 2.11; P < .01). One unexpected association was noted for these midcareer physicians: those in the highest quartile of satisfaction with their pay more often anticipated leaving their practices than those with midrange pay satisfaction scores (OR, 3.36; P < .01).

In the oldest physician group, those least satisfied with their community relationships (OR, 3.31; P < .01) and least satisfied with the administrative aspects of their work (OR, 2.64; P < .01) more often anticipated leaving their jobs. Fewer of those most satisfied with patient care issues foresaw leaving (OR, 0.30; P < .01). Unexpectedly, older physicians who were least satisfied with their relationships with other physicians were less likely to anticipate leaving their practices (OR, 0.36; P < .01).

Discussion

Our study has demonstrated 3 aspects of how physicians’ regard for their work relates to their plans to leave their jobs: (1) physicians’ plans to leave correlate with their satisfaction with only some aspects of their work; (2) the aspects of work for which satisfaction is associated with plans for leaving differ somewhat for generalists and specialists, and for physicians at various stages of their careers; and (3) anticipated job departure is more common among physicians who are relatively dissatisfied with any of a variety of aspects of their work, whereas satisfaction higher than median levels generally does not protect physicians from thoughts of leaving their jobs. To our knowledge, these last 2 features of the relationship between satisfaction and anticipated departure have not been demonstrated previously.

Even when physicians are displeased with a given aspect of their work, they are not always inclined to leave their jobs. For example, physicians in this and other studies^2,31 were least satisfied with the administrative requirements of their work, yet dissatisfaction in this area was not associated with the departure plans of most physician groups. In some cases, a dissatisfying issue is simply not important enough to individuals to warrant a job change. Physicians also have options to leaving, such as problem-solving with managers and coworkers, and taking comfort in the more satisfying aspects of their work.³² Thus, to know when dissatisfaction leads to job changes, one must actually test for statistical associations between the two.

In tests of this kind in the present study, we found 14 instances in which physicians in the lowest satisfaction quartiles were more likely to anticipate leaving than were those of average satisfaction. We found only 2 situations in which physicians in the highest satisfaction quartiles were less likely to anticipate leaving. To promote retention, these data suggest that physicians and their employers should avoid relative dissatisfaction in particular. These data also suggest that building ever-increasing higher levels of satisfaction generally does not prevent turnover. Thus, to foster workforce stability, practice managers should address aspects of physicians’ work or jobs causing them dissatisfaction, or more specifically, satisfaction lower than that of physicians working elsewhere. For aspects of work about which physicians are already satisfied, managers need not foster even higher satisfaction.

Satisfaction with specific aspects of work and turnover

Relative dissatisfaction with one’s community was associated with departure plans for all physician groups. An individual’s relationship with a community-measured in this study through questions about feeling “at home,” a sense of belonging, respected, and strongly connected²³-is basic to human existence,^33,34 but its importance has been underappreciated in physicians’ lives and careers.^33,35 Practices might promote workforce stability by providing opportunities and incentives for employed physicians to participate in local and state professional and community service organizations. Expectations can be set and time allocated for physicians to work with school health programs and volunteer clinics, speak to community groups, and become involved in other ways that build a sense of membership and contribution. These activities will be new for some physicians, will not always come naturally, and may need to be taught and encouraged.

Relative dissatisfaction with pay was associated with departure plans for all of this study’s physician groups, except for physicians older than 55 years. Further and unexpectedly, midcareer physicians who were particularly satisfied with their pay were also more likely to anticipate leaving. Perhaps midcareer physicians who are and feel well paid also more often feel that they can afford to retire or pursue second careers. We have no data to test this explanation. Similarly, we cannot know why older physicians who were least satisfied with their collegial relationships less often planned to leave their jobs. We wondered if this surprising finding was a result of confounding by physicians’ employment in group vs solo practices; we reasoned that among older physicians, those in solo practice may not get along as well with local physicians, but for unrelated reasons may be more attached to their practices and thus less likely to retire or otherwise plan to leave. This possibility proved not to be the case when we added an indicator of solo vs group practice to the statistical model for the older physician group.

How physicians felt about their relationships with patients and about their autonomy was not related to their intentions to leave their practices. Perhaps these once-central features of doctoring have lost their importance to physicians, or maybe physicians believe they will not find better relationships or more autonomy elsewhere, and thus see no benefit in changing jobs.

Age, satisfaction, and anticipated job changes

As in previous studies,^36,37 the older physicians of this study were generally more satisfied than the younger physicians. Older physicians, nevertheless, more often anticipated leaving their jobs. We suspect that in many cases an older physician’s thoughts of leaving a position are related to their retirement plans, but our data did not allow us to explore this relationship in detail. We also suspect that physicians’ satisfaction with their jobs influences their plans not only for leaving that particular job but also their plans for leaving the practice of medicine entirely; however, satisfaction with somewhat different aspects of work may influence each of the 2 types of career moves. For example, relative dissatisfaction with relationships with patients may push a physician toward retirement even though it did not seem to promote thoughts of a job change for this study’s physicians.

Common beliefs about physicians’ careers would have predicted the “J”-shape relationship between anticipated job departure rates and increasing physician age Figure 1). Young physicians fresh from training are thought to change jobs once or twice during an initial period of professional experimentation, then settle into suitable practice situations and raise their families. Later in life, job departures again become more common when children leave home, work takes on new meaning in middle age, and physicians become interested in and able to retire or pursue new jobs or second careers. Data from this study generally confirm these common notions, although the nadir in anticipated job departures occurred later than we expected.

Limitations

This study’s 52% response rate, although typical for physician studies,³⁸ may have yielded results not representative of physicians as a whole. To assess this possibility, we compared survey data from this study’s early and late respondents and found meaningful time trends in only 4 of 140 questionnaire items, providing some evidence that nonrespondents-if assumed to be in effect very late respondents-were similar to respondents. Nevertheless, to help adjust for any response bias we weighted analyses.

Statistical associations may not reflect causal relationships in a cross-sectional study such as this. Further, this study’s findings may have differed if actual job departure was available as an outcome measure, even though anticipated departure is a validated and frequently used proxy indicator of workers who will leave their jobs.^26-28 A measure of convergent validity was demonstrated in this study, as 16 of the 18 statistical associations between anticipated departure and satisfaction were in the expected direction, and age fluctuated with anticipated departure as we had expected it would with actual departure.

Conclusions and implications

These findings suggest that physicians who are relatively dissatisfied with specific aspects of their work more often plan to leave their jobs, and that these particular work aspects differ somewhat with age and specialty. Practice managers and employers concerned with maintaining a stable physician workforce should address relative dissatisfaction among physicians-particularly with income and community relationships-but need not build high levels of satisfaction. Physicians looking for stable employment should seek positions that they feel offer appropriate compensation and are located in communities to which they can connect. Once there, they should devote time and energy building ties in their communities.

Further research must confirm whether retention indeed improves by remediating relative dissatisfaction with the areas of physicians’ work flagged as important in this study. It also remains to be seen whether other outcomes of physicians’ work, such as the quality of care they provide, also relate primarily to relative dissatisfaction rather than satisfaction, and, specifically, to dissatisfaction with the same aspects of work seeming to affect job departure plans.

Acknowledgments

The authors thank the physician-fellows of UNC’s Primary Care Research Fellowship and the JFP’s reviewers for their thoughtful suggestions for strengthening this report. We also thank the many physician respondents who shared their time and revealed to us important aspects of their lives.

Until recently, the satisfaction of physicians was studied only among those working in special settings, such as in rural and urban underserved areas. In these settings, a less than satisfying work environment sometimes leads physicians to leave their jobs, with subsequent costs for needy communities.^1,2 Now, with US physicians often employed by others, physicians’ autonomy constrained, medicine managed as a business, and incomes falling for some specialties, the satisfaction of rank and file physicians in all locations and specialties can no longer be assumed.^3-5 Indeed, unhappiness is increasing among physicians, with growing numbers considering job changes, nonclinical work, and early retirement.^6-8 Some physicians regret ever pursuing a medical career.^8-10

Dissatisfaction among physicians is a concern even if medical work still pays handsomely and medical schools still have an abundance of applicants. Physicians’ satisfaction relates to their job effectiveness, including the quality of care they render, their patients’ compliance with recommended care, patients’ satisfaction with care, and patients’ health outcomes.^6,9,11-15 The most obvious consequence of a dissatisfying job, however, is that physicians will quit to work elsewhere, disrupting patient-physician continuity and creating organizational instability and replacement costs, estimated at nearly $250,000 per physician lost.^16-18 A job change also brings personal and financial costs to physicians and their families.

Although common experience and prior studies show that physicians’ job satisfaction generally influences whether they leave their jobs, little more is known about how satisfaction relates to job changes for physicians. From studies of other professions, we know that satisfaction is a multidimensional concept in which individuals can be satisfied with some aspects of work while dissatisfied with others.^16,19 Further, we know that workers’ satisfaction with only some aspects of work correlates with turnover.^16,20 For example, in 1 recent study rural generalist physicians who were satisfied with their local communities remained longer in their jobs, whereas their satisfaction with most other aspects of work, including with autonomy and doctor-patient relationships, was unrelated to their retention.² The effects of workers’ satisfaction and dissatisfaction may also require separate assessment, because greater dissatisfaction with a given aspect of work may lead a worker to leave a job, but increasingly higher levels of satisfaction with the same issue may not lead to ever-increasing chances of retention, or vice versa.^15,21

The goal of this study was to better understand the relationship between physicians’ satisfaction with various aspects of their work and whether or not they have thoughts or plans to leave their jobs. We assessed whether anticipated job departure related to satisfaction at both the upper and lower ends of its range; that is, whether particularly high satisfaction levels relative to other physicians protected against job change plans and significant relative dissatisfaction promoted such plans. We also assessed how the satisfaction/job-change relationship varied for physician groups with differing needs and values, specifically for generalists compared with specialists, and for younger, midcareer, and older physicians.

Methods

Subjects, survey mailings, and response rates

A sampling frame, previously described,^22,23 was constructed from the American Medical Association’s Physician Masterfile. The target population was the 171,252 civilian physicians listed as participating in direct patient care (rather than in training, administration, teaching, inactive, etc), including generalists (family physicians, general internists, and general pediatricians; general practitioners were not included) and specialists (medical and pediatric nonsurgical subspecialists) in the United States. We stratified the frame into geographic regions of high and low penetration of managed care, non-Hispanic white vs other physician ethnicity, and the 5-targeted specialty groupings.

We drew a random sample of 5704 physicians, oversampling smaller strata. Questionnaires and up to 3 follow-up mailings were sent to the sample from January through September 1997. A total of 2325 physicians responded, 52% when corrected for noncontact and ineligibility.²⁴ Based on information on the Masterfile, response rates did not differ by age, for those in office vs hospital-based practices, or for those in communities of high vs low managed care penetration. However, response rates were higher for subspecialists than for generalists, US medical graduates than for international graduates, and those for whom race-ethnicity was listed as black, “missing,” or “other” than for those listed as Hispanic, non-Hispanic white, or Asian. Statistical weighting in analyses adjusted for group contact and response rate differences.

We dropped the 218 respondents who worked less than 25 hours per week in activities related to patient care and another 169 physicians who primarily practiced outside of their specialty, such as general internists who worked as endocrinologists or oncologists. Analyses were based on the remaining 1939 physicians.

Questionnaire design and analyses

The questionnaire included 36 declarative statements about physicians’ views and regard for various features of their work and practices, to which respondents provided Likert-scaled responses from “strongly agree” to “strongly disagree.”²² Representative statements were, “I feel a strong personal connection with my patients” and “My total compensation package is fair.” These items were drawn from previous studies of physician satisfaction, from issues raised in physician focus groups we assembled, and through the suggestions of experts. We then used exploratory factor analysis with a pilot physician sample, followed by confirmatory factor analysis with a validation physician sample to group these items into 10 scales reflecting physicians’ satisfaction with 10 aspects of their work.^22,23 Scale alpha coefficients of reliability ranged from 0.65 to 0.77.

Physicians’ anticipated departure from their jobs was measured through a questionnaire item asking “What is the likelihood that you will leave your current practice situation within two years?”, with allowable responses of “none,” “slight,” “moderate,” “likely,” and “definitely.” Social psychology research has substantiated the use of people’s expressed intentions as measures of their future behavior.²⁵ Prior studies^26-28 have used anticipated departure and anticipated retention as expeditious proxy indicators for actual job departure and retention, with correlation coefficients between workers’ plans to leave and subsequent departure typically measured at about 0.5. In the only validating physician study, rural physicians in Western Australia much more often left their practices if they had predicted 10 years earlier that they would (odds ratio [OR], 3.3).²⁹ Shorter-term predictions, such as the 2-year time horizon used in this study, should be even more accurate.

This study’s dichotomous outcome indicator-whether physicians anticipated leaving their jobs within 2 years-defined planned “leavers” as those who indicated a moderate, likely, or definite likelihood of leaving their practices; planned “retainees” were those who envisioned only a slight or no chance of leaving. This cut point was selected for both its face validity-falling at the scale point at which leaving became a real possibility for subjects and no longer an unlikely event-and because it yielded an adequately balanced split in outcome values.

In analyses, levels of satisfaction with the 10 aspects of work and anticipated job departure were described for generalists, specialists, and physicians in 3 age groups. We compared groups on satisfaction levels and anticipated departure rates with t tests and 1-way analysis of variance.

We used logistic regression analysis to model separately the anticipated departure of each of the 5 groups. We entered satisfaction levels for each of the 10 areas of work into the models as 2 dichotomous (dummy) variables, yielding 20 satisfaction indicator variables. Ten of these indicators reflected values in the top quartile of satisfaction (ie, top quartile vs lower 3 quartiles) in each area of work and another 10 indicators reflected values in the lowest quartile of satisfaction (bottom quartile vs upper 3 quartiles). Cut points for the middle vs high satisfaction groups for the 10 areas of work ranged from 3.0 to 4.25 on the satisfaction scale (in which 1 indicated dissatisfaction; 3, neutral; and 5, satisfaction); for the middle vs lower satisfaction group, cut points ranged from 2.0 to 3.33. The inclusion of dichotomous indicators of high and low satisfaction for each area of work treated the middle 2 satisfaction quartiles statistically as the omitted, comparison category. This analytic approach modeled the effects on anticipated departure of relatively high and low satisfaction levels for each area of work, compared with more typical, midrange satisfaction levels, simultaneously controlling for relative satisfaction in the other 9 areas of work. We also included control variables for sex, age, and whether physicians owned their practices.

The logistic models for the 5 physician groups were repeated with the more parsimonious forward stepwise variable selection approach to test the robustness of the model findings and as a functional colinearity check.³⁰ With only 1 exception of the 100 satisfaction-departure associations tested across the 5 physician groups, the findings were identical to those of the full logistic models. We reported only the full logistic models.

The statistical significance level was set at .01, to partially adjust for this study’s numerous comparisons.

Results

Unweighted characterizations of the 1939 eligible respondent physicians found that 70% were male and 72% non-Hispanic white, with a median age of 46 years (range, 27-88 years). Most were married (84%) and had children (85%). There were 459 family physicians (24%), 375 general internists (19%), 494 general pediatricians (25%), and 611 pediatric and medical subspecialists (32%). Eighteen percent worked in solo practices, 36% worked in small groups, and the rest worked in larger groups. Most (57%) owned their practices. Physicians had worked in their current practices for a median duration of 9 years when surveyed (range, 0-58 years).

Satisfaction scores

As a whole, the generalist and specialist physicians reported midrange satisfaction levels in the 10 areas of work assessed Table 1). For 2 work facets-administrative responsibilities and having adequate personal time-the average satisfaction scores for both groups indicated overall dissatisfaction. Physicians indicated greatest satisfaction with their intrinsic rewards from patient care, specifically their personal connection with and gratitude from patients.

The mean satisfaction scores of generalists and specialists were comparable for 7 of the 10 aspects of practice. In the remaining 3 areas, the generalist-specialist differences in satisfaction were small. Satisfaction scores varied more often with age, with physicians 55 years and older more satisfied than younger physicians in 8 of the 10 areas queried.

Anticipated job turnover

In total, 27.1% of physicians foresaw leaving their practices within the next 2 years: 13.8% believed leaving was likely or definite and 13.3% believed they had a moderate chance of leaving. Generalist and specialist physicians perceived similar chances of leaving (26.7% vs 30.0%, respectively; P = .31), with and without adjusting for age and sex. In contrast, physicians of various age groups anticipated different likelihoods of leaving Figure 1), with a 29% rate for the youngest group, a nadir of 22% to 23% for physicians aged 45 to 54 years, and again higher rates for older physicians, reaching 63% for those 70 years and older (P < .001).

Those who anticipated leaving their jobs within 2 years frequently also anticipated leaving the practice of clinical medicine entirely within 5 years, including 84% of leavers older than 54 years, 50% of leavers 45 to 54 years, and even 33% of leavers younger than 45 years.

Satisfaction with work and anticipated job turnover

Overall observations. In analyses that controlled for satisfaction with other facets of work and for several physician characteristics, we found that physicians’ regard for only some aspects of work was associated with their anticipated departure from their jobs (Table 2 and Table 3). Further, we found that physicians who were relatively dissatisfied with a variety of aspects of their work more often foresaw leaving their jobs than those with median satisfaction levels. In contrast, intentions to leave were generally just as likely for those of high and median satisfaction levels.Figure 2 shows the typical relationship between satisfaction and anticipated departure. In this example, as in most instances, the transition from lower to higher likelihood of anticipated departure occurs only when physicians’ feelings drop below neutral into dissatisfied levels.

Generalists and specialists. Among generalists and compared with those of average satisfaction (the omitted group in the statistical models), leaving one’s job was anticipated more often by those less satisfied with their pay (OR, 3.65; P < .0001), with their relationships with their communities (OR, 2.26; P < .0001), and with their relationships with the nonphysician staff in their offices (OR, 1.59; P < .01) Table 2). Only for 1 area of work, patient care issues (eg, feeling overwhelmed with patients’ demands and feeling that time constraints prevented good relationships with patients), were generalists in the highest satisfaction quartile less likely to anticipate leaving their practices than generalists with midrange satisfaction scores (OR, 0.60; P < .01).

For specialists, those reporting lowest satisfaction with their pay (OR, 3.24; P = .01) and community relationships (OR, 3.89; P < .01) anticipated leaving more often than those with midrange scores Table 2). For no aspects of work were the most satisfied specialists less likely to anticipate leaving than specialists of average satisfaction.

Younger, midcareer, and older physicians. Among physicians in the youngest group-aged 27 through 44 years-those who were least satisfied with their pay (OR, 4.62; P < .0001) and with their relationships with their communities (OR, 2.79; P < .0001) more often anticipated leaving their jobs than those of median satisfaction Table 3). For no facets of work among younger physicians were those who were most satisfied less likely to anticipate leaving.

For physicians aged 45 through 54 years, anticipated job departure was more common for those least satisfied with 5 areas of work, including again those least satisfied with pay (OR, 11.60; P < .0001) and with their community relationships (OR, 2.11; P < .01). One unexpected association was noted for these midcareer physicians: those in the highest quartile of satisfaction with their pay more often anticipated leaving their practices than those with midrange pay satisfaction scores (OR, 3.36; P < .01).

In the oldest physician group, those least satisfied with their community relationships (OR, 3.31; P < .01) and least satisfied with the administrative aspects of their work (OR, 2.64; P < .01) more often anticipated leaving their jobs. Fewer of those most satisfied with patient care issues foresaw leaving (OR, 0.30; P < .01). Unexpectedly, older physicians who were least satisfied with their relationships with other physicians were less likely to anticipate leaving their practices (OR, 0.36; P < .01).

Discussion

Our study has demonstrated 3 aspects of how physicians’ regard for their work relates to their plans to leave their jobs: (1) physicians’ plans to leave correlate with their satisfaction with only some aspects of their work; (2) the aspects of work for which satisfaction is associated with plans for leaving differ somewhat for generalists and specialists, and for physicians at various stages of their careers; and (3) anticipated job departure is more common among physicians who are relatively dissatisfied with any of a variety of aspects of their work, whereas satisfaction higher than median levels generally does not protect physicians from thoughts of leaving their jobs. To our knowledge, these last 2 features of the relationship between satisfaction and anticipated departure have not been demonstrated previously.

Even when physicians are displeased with a given aspect of their work, they are not always inclined to leave their jobs. For example, physicians in this and other studies^2,31 were least satisfied with the administrative requirements of their work, yet dissatisfaction in this area was not associated with the departure plans of most physician groups. In some cases, a dissatisfying issue is simply not important enough to individuals to warrant a job change. Physicians also have options to leaving, such as problem-solving with managers and coworkers, and taking comfort in the more satisfying aspects of their work.³² Thus, to know when dissatisfaction leads to job changes, one must actually test for statistical associations between the two.

In tests of this kind in the present study, we found 14 instances in which physicians in the lowest satisfaction quartiles were more likely to anticipate leaving than were those of average satisfaction. We found only 2 situations in which physicians in the highest satisfaction quartiles were less likely to anticipate leaving. To promote retention, these data suggest that physicians and their employers should avoid relative dissatisfaction in particular. These data also suggest that building ever-increasing higher levels of satisfaction generally does not prevent turnover. Thus, to foster workforce stability, practice managers should address aspects of physicians’ work or jobs causing them dissatisfaction, or more specifically, satisfaction lower than that of physicians working elsewhere. For aspects of work about which physicians are already satisfied, managers need not foster even higher satisfaction.

Satisfaction with specific aspects of work and turnover

Relative dissatisfaction with one’s community was associated with departure plans for all physician groups. An individual’s relationship with a community-measured in this study through questions about feeling “at home,” a sense of belonging, respected, and strongly connected²³-is basic to human existence,^33,34 but its importance has been underappreciated in physicians’ lives and careers.^33,35 Practices might promote workforce stability by providing opportunities and incentives for employed physicians to participate in local and state professional and community service organizations. Expectations can be set and time allocated for physicians to work with school health programs and volunteer clinics, speak to community groups, and become involved in other ways that build a sense of membership and contribution. These activities will be new for some physicians, will not always come naturally, and may need to be taught and encouraged.

Relative dissatisfaction with pay was associated with departure plans for all of this study’s physician groups, except for physicians older than 55 years. Further and unexpectedly, midcareer physicians who were particularly satisfied with their pay were also more likely to anticipate leaving. Perhaps midcareer physicians who are and feel well paid also more often feel that they can afford to retire or pursue second careers. We have no data to test this explanation. Similarly, we cannot know why older physicians who were least satisfied with their collegial relationships less often planned to leave their jobs. We wondered if this surprising finding was a result of confounding by physicians’ employment in group vs solo practices; we reasoned that among older physicians, those in solo practice may not get along as well with local physicians, but for unrelated reasons may be more attached to their practices and thus less likely to retire or otherwise plan to leave. This possibility proved not to be the case when we added an indicator of solo vs group practice to the statistical model for the older physician group.

How physicians felt about their relationships with patients and about their autonomy was not related to their intentions to leave their practices. Perhaps these once-central features of doctoring have lost their importance to physicians, or maybe physicians believe they will not find better relationships or more autonomy elsewhere, and thus see no benefit in changing jobs.

Age, satisfaction, and anticipated job changes

As in previous studies,^36,37 the older physicians of this study were generally more satisfied than the younger physicians. Older physicians, nevertheless, more often anticipated leaving their jobs. We suspect that in many cases an older physician’s thoughts of leaving a position are related to their retirement plans, but our data did not allow us to explore this relationship in detail. We also suspect that physicians’ satisfaction with their jobs influences their plans not only for leaving that particular job but also their plans for leaving the practice of medicine entirely; however, satisfaction with somewhat different aspects of work may influence each of the 2 types of career moves. For example, relative dissatisfaction with relationships with patients may push a physician toward retirement even though it did not seem to promote thoughts of a job change for this study’s physicians.

Common beliefs about physicians’ careers would have predicted the “J”-shape relationship between anticipated job departure rates and increasing physician age Figure 1). Young physicians fresh from training are thought to change jobs once or twice during an initial period of professional experimentation, then settle into suitable practice situations and raise their families. Later in life, job departures again become more common when children leave home, work takes on new meaning in middle age, and physicians become interested in and able to retire or pursue new jobs or second careers. Data from this study generally confirm these common notions, although the nadir in anticipated job departures occurred later than we expected.

Limitations

This study’s 52% response rate, although typical for physician studies,³⁸ may have yielded results not representative of physicians as a whole. To assess this possibility, we compared survey data from this study’s early and late respondents and found meaningful time trends in only 4 of 140 questionnaire items, providing some evidence that nonrespondents-if assumed to be in effect very late respondents-were similar to respondents. Nevertheless, to help adjust for any response bias we weighted analyses.

Statistical associations may not reflect causal relationships in a cross-sectional study such as this. Further, this study’s findings may have differed if actual job departure was available as an outcome measure, even though anticipated departure is a validated and frequently used proxy indicator of workers who will leave their jobs.^26-28 A measure of convergent validity was demonstrated in this study, as 16 of the 18 statistical associations between anticipated departure and satisfaction were in the expected direction, and age fluctuated with anticipated departure as we had expected it would with actual departure.

Conclusions and implications

These findings suggest that physicians who are relatively dissatisfied with specific aspects of their work more often plan to leave their jobs, and that these particular work aspects differ somewhat with age and specialty. Practice managers and employers concerned with maintaining a stable physician workforce should address relative dissatisfaction among physicians-particularly with income and community relationships-but need not build high levels of satisfaction. Physicians looking for stable employment should seek positions that they feel offer appropriate compensation and are located in communities to which they can connect. Once there, they should devote time and energy building ties in their communities.

Further research must confirm whether retention indeed improves by remediating relative dissatisfaction with the areas of physicians’ work flagged as important in this study. It also remains to be seen whether other outcomes of physicians’ work, such as the quality of care they provide, also relate primarily to relative dissatisfaction rather than satisfaction, and, specifically, to dissatisfaction with the same aspects of work seeming to affect job departure plans.

Acknowledgments

The authors thank the physician-fellows of UNC’s Primary Care Research Fellowship and the JFP’s reviewers for their thoughtful suggestions for strengthening this report. We also thank the many physician respondents who shared their time and revealed to us important aspects of their lives.

In 1995, in recognition of the need for trained family medicine researchers,^1-6 the North American Primary Care Research Group (NAPCRG) Committee on Building Research Capacity launched the Family Medicine Grant Generating Project (GGP) to help family medicine researchers successfully apply for their first major research grant. Between 1995 and the time of the evaluation, 23 faculty participants from 21 family medicine departments enrolled in this year-long “fellowship without walls” designed to permit participants to remain housed in their home department while working on their research proposal. Participants attended 3 GGP sessions held in conjunction with the annual meeting of the Society of Teachers of Family Medicine (STFM), and the NACPRG and the Primary Care Research Methods and Statistics meetings over the course of 1 year. Additionally, participants worked via e-mail, telephone, and postal mail with GGP faculty and research mentors during the year. Many participants met in person with mentors.

Although the program is referred to as a “fellowship” and participants as “fellows,” this is neither a traditional family medicine “fellowship” in which participants are housed for a designated time period (usually 2 years), nor an innovative fellowship in which individuals spend concentrated time at another institution. Traditional fellowships usually offer courses for academic credit in a variety of topics such as statistics and epidemiology and typically result in a master’s degree. GGP, however, is designed to train fellows in the design and writing of competitive external grant proposals: it does not offer courses or a degree. Over the fellowship year, fellows attend sessions that focus on the various tools, techniques, and methods required to produce a competitive proposal. These sessions include time devoted to concept paper development, conducting a literature review, development of specific aims, and elaboration of various research methodologies. GGP uses the term “fellowship” to identify the program as a special 1-year period with at least 25% of the time devoted to developing “grantsmanship skills” and producing a competitive proposal. This program is the only one of its kind in family medicine and therefore cannot be compared with more traditional fellowships programs. This article reports on the evaluation results.

Methods

From December 1999 through April 2000 we mailed surveys to 23 former GGP participants. Respondents were asked to evaluate the GGP’s components (statistics, concept paper, grant writing, mentoring) on a scale from “not at all helpful” to “very helpful.” Several open-ended questions were asked about their program expectations; what they considered GGP’s strengths and weaknesses; how participation contributed to their academic career and helped increase research capacity; the barriers encountered in achieving their research goals; the skills learned; suggestions for improving the program; and advice they would give to future applicants. Respondents were asked to indicate the grants/contracts that they submitted after their fellowship. In addition, demographic information was obtained on age, employment position, date of initiation of GGP, the year they graduated from medical school, and the year they participated in the GGP program.

Eighteen (78%) of the participants returned a completed survey. Reasons for not completing the survey, given either by the participants who chose not to complete the survey, or by their departments, included the participant now working outside research; the participant on leave; the participant in the middle of writing a research grant (R01) and therefore not having time to complete the survey; and lack of response (only 2 participants did not respond after 3 attempts to contact them). Of those who did not complete the survey, 1 was from the first-year class, 2 were from the second-year class, and 2 were from the third-year class.

Results

The average age of respondents was 43.8 years (range, 37-53 years). Most were at the assistant professor level. An additional 4 were at the associate professor level, 2 were research directors, 1 was a professor, 1 was a clinician investigator, and 1 was an interim chair. A total of 12 of the respondents were physicians, 5 had a PhD, and 1 had a Doctor of Public Health degree; 4 of the physicians also had a Masters in Public Health and 2 had a PhD. The average number of years since obtaining their terminal degree was 15.1 years (range, 4-26 years). Most respondents (11) had their terminal degree 15 years or more.

Evaluation of research activities

A total of 58 grants/contracts were submitted by the respondents, representing approximately US$19.3 million. These figures represent an underestimate because 1 first-year class respondent only mentioned having submitted more than 10 grants over the last 3 years ranging from $50,000 to more than $1 million, but did not elaborate further. Due to this lack of specification, this information could not be entered into the subsequent analysis. Seven (12%) of those submitted were training grants, representing approximately $1.7 million. Most (32, or 55%) of the submitted grants/contracts were to nonfederal organizations, whereas 19 (or -33%) were to federal agencies. Although 22 (38%) of the submitted grants/contracts were for less than $50,000, 11 (19%) (excluding training grants) were for more than $300,000. Nineteen (33%) of the submitted grants/contracts were funded including 4 training grants, representing approximately $1 million excluding training grants ($2.3 million including training grants).

Although 22 (38%) of the grant submissions were not funded (representing approximately $6.2 million), another 17 (29%) were pending, representing approximately $10.8 million, including training grants. Based on prior experience, approximately 46% of these pending grants may receive funding, which could result in an additional $4.8 million.

Table 1 shows the number of grants/contracts submitted, funded, not funded, and pending as Principal Investigator (PI) or Co-PI by class year of the GGP fellowship. Most (42, 72%) of the grants/contracts submitted were by the first- or second-year class fellows with 12 grants pending and 16 funded. The Figure 1 shows the number of grants/contracts submitted by submission date and fellowship class year. The number of submissions increased within 1 to 2 years of the fellowship.

Although the sample size was too small for statistical evaluation, some trends are noted. Only 3 of the 18 respondents indicated that they had not submitted a grant or contract as PI or Co-PI, whereas 1 respondent indicated being an evaluator or research director on 6 funded grants. The average number of grants submitted by participants was 4.4. Participants who had submitted a grant tended to be older and to have held their terminal degree for a longer time. Thirty-seven of the total grant/contract submissions were by respondents older than the mean age of 43 years, and 41 of the total grant/contract submissions were by respondents who had received their terminal degree more than 15 years ago. The latter respondents also accounted for 15 of the 19 funded grants. No difference was noted in the rate of grant submission by terminal degree type.

Several respondents also indicated the following grant roles: qualitative analyst on 2 grants; research writer, administrator, project director, or evaluator on 8 grants; co-investigator on 2 grants; and consultant on 4 grants. Twelve of these grants were funded, with 4 training grants representing approximately $1.7 million and 8 research grants representing approximately $1.3 million.

Evaluation of GGP components

Table 2 shows the evaluation of each GGP component. Most respondents indicated that the concept paper and the grant-writing components were very helpful. Additional components indicated as very helpful included networking, mock reviews, consultant mentoring, and peer groups. Compared with the concept paper and grant-writing components, the statistics and the mentoring components were rated as less helpful. Expectations of the GGP

Respondents were asked to describe their major expectations of the GGP. Most respondents replied that they expected to receive help in grant writing, to better understand the grant application process, and to develop helpful relationships with experienced researchers.

All of the respondents indicated that most of their expectations were met. However, respondents were mixed regarding the mentoring process. Although 8 indicated that the mentoring was very helpful, some respondents had difficulty in identifying a mentor or were never successful in finding a mentor, whereas others indicated problems with long distance mentoring.

Strengths of the GGP

Respondents indicated that program strengths include enhancing the value of family medicine research, receiving guidance and feedback, learning from the experiences of their peers, and developing grant-writing skills. Respondents specifically mentioned the focus on the NIH grant process and the mock study section as strengths.

With respect to how the GGP contributed to their career, 7 of the respondents said that the program helped them connect with other researchers and gave them confidence in grant writing. In other instances, the GGP program helped respondents to achieve a more secure position, obtain a promotion, start a career, or change academic departments/institutions. As one respondent mentioned, “I think more critically and more daringly . . . I [was] promoted after that from assistant to associate.”

Most respondents indicated that the GGP program enabled them to mentor, teach, and encourage other faculty members in their department in their grant-writing efforts. As one respondent mentioned, the GGP experience ”. . . has helped me to bridge the gap between clinicians and researchers.”

Challenges of the GGP

The most specific challenge of the GGP program mentioned by respondents was the variable commitment of mentors and the lack of having a mentor with expertise in a respondent’s area of research. The major barriers to respondents toward achieving their research goals were (1) the lack of a supportive research environment including a lack of financial support, few peers engaged in research, a lack of local mentoring, and infrastructural and administrative barriers (eg, no links to statisticians, confused priorities within department, and lack of diversity and foresight in department); (2) the lack of time to do research while participating in clinical duties and academic activities; and (3) the lack of information about funding opportunities at their institution. One respondent mentioned not having any external barriers but that the GGP experience “. . . did cause me to reevaluate and ultimately redefine my research goals.”

When asked what skills they learned during the GGP that helped them address these barriers, several respondents indicated (1) time management, including how to realistically organize the research process; (2) the ability to make connections with outside mentors and consultants; and (3) confidence development, such as contacting a program specialist despite poor information from their institution. However, respondents who mentioned the barrier of a nonsupportive research environment indicated that skills to address this barrier were not taught in the GGP program. Recommendations for improving the GGP

Suggestions for improving the GGP program included having 2 GGP entry levels (for beginning researchers and experienced researchers), continuing networking opportunities for “graduates,” addressing the need for female mentors, following up with consultations and mentoring, and having a mentor with expertise in participants’ area of research.

With respect to advice to future applicants, respondents mentioned being more aggressive in working with a mentor, having reasonable expectations, having long-term support from the department, using time efficiently to network, and reflecting on one’s motivation toward an intensive research career. As one participant said, “I would support the application 100% and notify them that it is one of the best investments of time they can make for their academic career.”

Discussion

The respondents’ comments made clear that the GGP is an “excellent place to start a career.” Most respondents appeared to regard the program as well worth the time and effort invested. Aside from some adjustments, such as consistency in mentoring and accommodating the research focus and skill level of participants, it is “. . . very important to continue [the] effort to train new researchers.” Interestingly, the GGP had a substantial impact on the extent to which many participants are now teaching and mentoring others to help develop the skills necessary for meaningful research.

Moreover, within 1 to 2 years of the program participants achieved a remarkable track record of grant submissions. Although for some participants “. . . breaking in [getting grant awards] still seems mysterious and forbidding,” they appreciated the opportunity to learn how to write a coherent concept paper, how to apply “good methods,” the importance of a track record, and to learn from experienced researchers.

As a result in large part of this evaluation, funding for the GGP has been continued with support from NAPCRG, the American Academy of Family Physicians, the STFM, the College of Family Physicians of Canada, and the home departments of participating academic departments and community-based residency programs.

Family medicine is poised to generate new knowledge through the development of its own research agenda-one that focuses on integrated, relationship-centered patient, family, and community health care.⁷ Young researchers, however, need to be nurtured, mentored, and organizationally supported.⁸ Research initiatives such as the GGP represent a collective enterprise, a building of coalitions toward the development of a culture of family medicine research. Continued research success requires commitments from both individuals and departments if family medicine research is to grow and prosper.

In 1995, in recognition of the need for trained family medicine researchers,^1-6 the North American Primary Care Research Group (NAPCRG) Committee on Building Research Capacity launched the Family Medicine Grant Generating Project (GGP) to help family medicine researchers successfully apply for their first major research grant. Between 1995 and the time of the evaluation, 23 faculty participants from 21 family medicine departments enrolled in this year-long “fellowship without walls” designed to permit participants to remain housed in their home department while working on their research proposal. Participants attended 3 GGP sessions held in conjunction with the annual meeting of the Society of Teachers of Family Medicine (STFM), and the NACPRG and the Primary Care Research Methods and Statistics meetings over the course of 1 year. Additionally, participants worked via e-mail, telephone, and postal mail with GGP faculty and research mentors during the year. Many participants met in person with mentors.

Although the program is referred to as a “fellowship” and participants as “fellows,” this is neither a traditional family medicine “fellowship” in which participants are housed for a designated time period (usually 2 years), nor an innovative fellowship in which individuals spend concentrated time at another institution. Traditional fellowships usually offer courses for academic credit in a variety of topics such as statistics and epidemiology and typically result in a master’s degree. GGP, however, is designed to train fellows in the design and writing of competitive external grant proposals: it does not offer courses or a degree. Over the fellowship year, fellows attend sessions that focus on the various tools, techniques, and methods required to produce a competitive proposal. These sessions include time devoted to concept paper development, conducting a literature review, development of specific aims, and elaboration of various research methodologies. GGP uses the term “fellowship” to identify the program as a special 1-year period with at least 25% of the time devoted to developing “grantsmanship skills” and producing a competitive proposal. This program is the only one of its kind in family medicine and therefore cannot be compared with more traditional fellowships programs. This article reports on the evaluation results.

Methods

From December 1999 through April 2000 we mailed surveys to 23 former GGP participants. Respondents were asked to evaluate the GGP’s components (statistics, concept paper, grant writing, mentoring) on a scale from “not at all helpful” to “very helpful.” Several open-ended questions were asked about their program expectations; what they considered GGP’s strengths and weaknesses; how participation contributed to their academic career and helped increase research capacity; the barriers encountered in achieving their research goals; the skills learned; suggestions for improving the program; and advice they would give to future applicants. Respondents were asked to indicate the grants/contracts that they submitted after their fellowship. In addition, demographic information was obtained on age, employment position, date of initiation of GGP, the year they graduated from medical school, and the year they participated in the GGP program.

Eighteen (78%) of the participants returned a completed survey. Reasons for not completing the survey, given either by the participants who chose not to complete the survey, or by their departments, included the participant now working outside research; the participant on leave; the participant in the middle of writing a research grant (R01) and therefore not having time to complete the survey; and lack of response (only 2 participants did not respond after 3 attempts to contact them). Of those who did not complete the survey, 1 was from the first-year class, 2 were from the second-year class, and 2 were from the third-year class.

Results

The average age of respondents was 43.8 years (range, 37-53 years). Most were at the assistant professor level. An additional 4 were at the associate professor level, 2 were research directors, 1 was a professor, 1 was a clinician investigator, and 1 was an interim chair. A total of 12 of the respondents were physicians, 5 had a PhD, and 1 had a Doctor of Public Health degree; 4 of the physicians also had a Masters in Public Health and 2 had a PhD. The average number of years since obtaining their terminal degree was 15.1 years (range, 4-26 years). Most respondents (11) had their terminal degree 15 years or more.

Evaluation of research activities

A total of 58 grants/contracts were submitted by the respondents, representing approximately US$19.3 million. These figures represent an underestimate because 1 first-year class respondent only mentioned having submitted more than 10 grants over the last 3 years ranging from $50,000 to more than $1 million, but did not elaborate further. Due to this lack of specification, this information could not be entered into the subsequent analysis. Seven (12%) of those submitted were training grants, representing approximately $1.7 million. Most (32, or 55%) of the submitted grants/contracts were to nonfederal organizations, whereas 19 (or -33%) were to federal agencies. Although 22 (38%) of the submitted grants/contracts were for less than $50,000, 11 (19%) (excluding training grants) were for more than $300,000. Nineteen (33%) of the submitted grants/contracts were funded including 4 training grants, representing approximately $1 million excluding training grants ($2.3 million including training grants).

Although 22 (38%) of the grant submissions were not funded (representing approximately $6.2 million), another 17 (29%) were pending, representing approximately $10.8 million, including training grants. Based on prior experience, approximately 46% of these pending grants may receive funding, which could result in an additional $4.8 million.

Table 1 shows the number of grants/contracts submitted, funded, not funded, and pending as Principal Investigator (PI) or Co-PI by class year of the GGP fellowship. Most (42, 72%) of the grants/contracts submitted were by the first- or second-year class fellows with 12 grants pending and 16 funded. The Figure 1 shows the number of grants/contracts submitted by submission date and fellowship class year. The number of submissions increased within 1 to 2 years of the fellowship.

Although the sample size was too small for statistical evaluation, some trends are noted. Only 3 of the 18 respondents indicated that they had not submitted a grant or contract as PI or Co-PI, whereas 1 respondent indicated being an evaluator or research director on 6 funded grants. The average number of grants submitted by participants was 4.4. Participants who had submitted a grant tended to be older and to have held their terminal degree for a longer time. Thirty-seven of the total grant/contract submissions were by respondents older than the mean age of 43 years, and 41 of the total grant/contract submissions were by respondents who had received their terminal degree more than 15 years ago. The latter respondents also accounted for 15 of the 19 funded grants. No difference was noted in the rate of grant submission by terminal degree type.

Several respondents also indicated the following grant roles: qualitative analyst on 2 grants; research writer, administrator, project director, or evaluator on 8 grants; co-investigator on 2 grants; and consultant on 4 grants. Twelve of these grants were funded, with 4 training grants representing approximately $1.7 million and 8 research grants representing approximately $1.3 million.

Evaluation of GGP components

Table 2 shows the evaluation of each GGP component. Most respondents indicated that the concept paper and the grant-writing components were very helpful. Additional components indicated as very helpful included networking, mock reviews, consultant mentoring, and peer groups. Compared with the concept paper and grant-writing components, the statistics and the mentoring components were rated as less helpful. Expectations of the GGP

Respondents were asked to describe their major expectations of the GGP. Most respondents replied that they expected to receive help in grant writing, to better understand the grant application process, and to develop helpful relationships with experienced researchers.

All of the respondents indicated that most of their expectations were met. However, respondents were mixed regarding the mentoring process. Although 8 indicated that the mentoring was very helpful, some respondents had difficulty in identifying a mentor or were never successful in finding a mentor, whereas others indicated problems with long distance mentoring.

Strengths of the GGP

Respondents indicated that program strengths include enhancing the value of family medicine research, receiving guidance and feedback, learning from the experiences of their peers, and developing grant-writing skills. Respondents specifically mentioned the focus on the NIH grant process and the mock study section as strengths.

With respect to how the GGP contributed to their career, 7 of the respondents said that the program helped them connect with other researchers and gave them confidence in grant writing. In other instances, the GGP program helped respondents to achieve a more secure position, obtain a promotion, start a career, or change academic departments/institutions. As one respondent mentioned, “I think more critically and more daringly . . . I [was] promoted after that from assistant to associate.”

Most respondents indicated that the GGP program enabled them to mentor, teach, and encourage other faculty members in their department in their grant-writing efforts. As one respondent mentioned, the GGP experience ”. . . has helped me to bridge the gap between clinicians and researchers.”

Challenges of the GGP

The most specific challenge of the GGP program mentioned by respondents was the variable commitment of mentors and the lack of having a mentor with expertise in a respondent’s area of research. The major barriers to respondents toward achieving their research goals were (1) the lack of a supportive research environment including a lack of financial support, few peers engaged in research, a lack of local mentoring, and infrastructural and administrative barriers (eg, no links to statisticians, confused priorities within department, and lack of diversity and foresight in department); (2) the lack of time to do research while participating in clinical duties and academic activities; and (3) the lack of information about funding opportunities at their institution. One respondent mentioned not having any external barriers but that the GGP experience “. . . did cause me to reevaluate and ultimately redefine my research goals.”

When asked what skills they learned during the GGP that helped them address these barriers, several respondents indicated (1) time management, including how to realistically organize the research process; (2) the ability to make connections with outside mentors and consultants; and (3) confidence development, such as contacting a program specialist despite poor information from their institution. However, respondents who mentioned the barrier of a nonsupportive research environment indicated that skills to address this barrier were not taught in the GGP program. Recommendations for improving the GGP

Suggestions for improving the GGP program included having 2 GGP entry levels (for beginning researchers and experienced researchers), continuing networking opportunities for “graduates,” addressing the need for female mentors, following up with consultations and mentoring, and having a mentor with expertise in participants’ area of research.

With respect to advice to future applicants, respondents mentioned being more aggressive in working with a mentor, having reasonable expectations, having long-term support from the department, using time efficiently to network, and reflecting on one’s motivation toward an intensive research career. As one participant said, “I would support the application 100% and notify them that it is one of the best investments of time they can make for their academic career.”

Discussion

The respondents’ comments made clear that the GGP is an “excellent place to start a career.” Most respondents appeared to regard the program as well worth the time and effort invested. Aside from some adjustments, such as consistency in mentoring and accommodating the research focus and skill level of participants, it is “. . . very important to continue [the] effort to train new researchers.” Interestingly, the GGP had a substantial impact on the extent to which many participants are now teaching and mentoring others to help develop the skills necessary for meaningful research.

Moreover, within 1 to 2 years of the program participants achieved a remarkable track record of grant submissions. Although for some participants “. . . breaking in [getting grant awards] still seems mysterious and forbidding,” they appreciated the opportunity to learn how to write a coherent concept paper, how to apply “good methods,” the importance of a track record, and to learn from experienced researchers.

As a result in large part of this evaluation, funding for the GGP has been continued with support from NAPCRG, the American Academy of Family Physicians, the STFM, the College of Family Physicians of Canada, and the home departments of participating academic departments and community-based residency programs.

Family medicine is poised to generate new knowledge through the development of its own research agenda-one that focuses on integrated, relationship-centered patient, family, and community health care.⁷ Young researchers, however, need to be nurtured, mentored, and organizationally supported.⁸ Research initiatives such as the GGP represent a collective enterprise, a building of coalitions toward the development of a culture of family medicine research. Continued research success requires commitments from both individuals and departments if family medicine research is to grow and prosper.

In 1995, in recognition of the need for trained family medicine researchers,^1-6 the North American Primary Care Research Group (NAPCRG) Committee on Building Research Capacity launched the Family Medicine Grant Generating Project (GGP) to help family medicine researchers successfully apply for their first major research grant. Between 1995 and the time of the evaluation, 23 faculty participants from 21 family medicine departments enrolled in this year-long “fellowship without walls” designed to permit participants to remain housed in their home department while working on their research proposal. Participants attended 3 GGP sessions held in conjunction with the annual meeting of the Society of Teachers of Family Medicine (STFM), and the NACPRG and the Primary Care Research Methods and Statistics meetings over the course of 1 year. Additionally, participants worked via e-mail, telephone, and postal mail with GGP faculty and research mentors during the year. Many participants met in person with mentors.

Although the program is referred to as a “fellowship” and participants as “fellows,” this is neither a traditional family medicine “fellowship” in which participants are housed for a designated time period (usually 2 years), nor an innovative fellowship in which individuals spend concentrated time at another institution. Traditional fellowships usually offer courses for academic credit in a variety of topics such as statistics and epidemiology and typically result in a master’s degree. GGP, however, is designed to train fellows in the design and writing of competitive external grant proposals: it does not offer courses or a degree. Over the fellowship year, fellows attend sessions that focus on the various tools, techniques, and methods required to produce a competitive proposal. These sessions include time devoted to concept paper development, conducting a literature review, development of specific aims, and elaboration of various research methodologies. GGP uses the term “fellowship” to identify the program as a special 1-year period with at least 25% of the time devoted to developing “grantsmanship skills” and producing a competitive proposal. This program is the only one of its kind in family medicine and therefore cannot be compared with more traditional fellowships programs. This article reports on the evaluation results.

Methods

From December 1999 through April 2000 we mailed surveys to 23 former GGP participants. Respondents were asked to evaluate the GGP’s components (statistics, concept paper, grant writing, mentoring) on a scale from “not at all helpful” to “very helpful.” Several open-ended questions were asked about their program expectations; what they considered GGP’s strengths and weaknesses; how participation contributed to their academic career and helped increase research capacity; the barriers encountered in achieving their research goals; the skills learned; suggestions for improving the program; and advice they would give to future applicants. Respondents were asked to indicate the grants/contracts that they submitted after their fellowship. In addition, demographic information was obtained on age, employment position, date of initiation of GGP, the year they graduated from medical school, and the year they participated in the GGP program.

Eighteen (78%) of the participants returned a completed survey. Reasons for not completing the survey, given either by the participants who chose not to complete the survey, or by their departments, included the participant now working outside research; the participant on leave; the participant in the middle of writing a research grant (R01) and therefore not having time to complete the survey; and lack of response (only 2 participants did not respond after 3 attempts to contact them). Of those who did not complete the survey, 1 was from the first-year class, 2 were from the second-year class, and 2 were from the third-year class.

Results

The average age of respondents was 43.8 years (range, 37-53 years). Most were at the assistant professor level. An additional 4 were at the associate professor level, 2 were research directors, 1 was a professor, 1 was a clinician investigator, and 1 was an interim chair. A total of 12 of the respondents were physicians, 5 had a PhD, and 1 had a Doctor of Public Health degree; 4 of the physicians also had a Masters in Public Health and 2 had a PhD. The average number of years since obtaining their terminal degree was 15.1 years (range, 4-26 years). Most respondents (11) had their terminal degree 15 years or more.

Evaluation of research activities

A total of 58 grants/contracts were submitted by the respondents, representing approximately US$19.3 million. These figures represent an underestimate because 1 first-year class respondent only mentioned having submitted more than 10 grants over the last 3 years ranging from $50,000 to more than $1 million, but did not elaborate further. Due to this lack of specification, this information could not be entered into the subsequent analysis. Seven (12%) of those submitted were training grants, representing approximately $1.7 million. Most (32, or 55%) of the submitted grants/contracts were to nonfederal organizations, whereas 19 (or -33%) were to federal agencies. Although 22 (38%) of the submitted grants/contracts were for less than $50,000, 11 (19%) (excluding training grants) were for more than $300,000. Nineteen (33%) of the submitted grants/contracts were funded including 4 training grants, representing approximately $1 million excluding training grants ($2.3 million including training grants).

Although 22 (38%) of the grant submissions were not funded (representing approximately $6.2 million), another 17 (29%) were pending, representing approximately $10.8 million, including training grants. Based on prior experience, approximately 46% of these pending grants may receive funding, which could result in an additional $4.8 million.

Table 1 shows the number of grants/contracts submitted, funded, not funded, and pending as Principal Investigator (PI) or Co-PI by class year of the GGP fellowship. Most (42, 72%) of the grants/contracts submitted were by the first- or second-year class fellows with 12 grants pending and 16 funded. The Figure 1 shows the number of grants/contracts submitted by submission date and fellowship class year. The number of submissions increased within 1 to 2 years of the fellowship.

Although the sample size was too small for statistical evaluation, some trends are noted. Only 3 of the 18 respondents indicated that they had not submitted a grant or contract as PI or Co-PI, whereas 1 respondent indicated being an evaluator or research director on 6 funded grants. The average number of grants submitted by participants was 4.4. Participants who had submitted a grant tended to be older and to have held their terminal degree for a longer time. Thirty-seven of the total grant/contract submissions were by respondents older than the mean age of 43 years, and 41 of the total grant/contract submissions were by respondents who had received their terminal degree more than 15 years ago. The latter respondents also accounted for 15 of the 19 funded grants. No difference was noted in the rate of grant submission by terminal degree type.

Several respondents also indicated the following grant roles: qualitative analyst on 2 grants; research writer, administrator, project director, or evaluator on 8 grants; co-investigator on 2 grants; and consultant on 4 grants. Twelve of these grants were funded, with 4 training grants representing approximately $1.7 million and 8 research grants representing approximately $1.3 million.

Evaluation of GGP components

Table 2 shows the evaluation of each GGP component. Most respondents indicated that the concept paper and the grant-writing components were very helpful. Additional components indicated as very helpful included networking, mock reviews, consultant mentoring, and peer groups. Compared with the concept paper and grant-writing components, the statistics and the mentoring components were rated as less helpful. Expectations of the GGP

Respondents were asked to describe their major expectations of the GGP. Most respondents replied that they expected to receive help in grant writing, to better understand the grant application process, and to develop helpful relationships with experienced researchers.

All of the respondents indicated that most of their expectations were met. However, respondents were mixed regarding the mentoring process. Although 8 indicated that the mentoring was very helpful, some respondents had difficulty in identifying a mentor or were never successful in finding a mentor, whereas others indicated problems with long distance mentoring.

Strengths of the GGP

Respondents indicated that program strengths include enhancing the value of family medicine research, receiving guidance and feedback, learning from the experiences of their peers, and developing grant-writing skills. Respondents specifically mentioned the focus on the NIH grant process and the mock study section as strengths.

With respect to how the GGP contributed to their career, 7 of the respondents said that the program helped them connect with other researchers and gave them confidence in grant writing. In other instances, the GGP program helped respondents to achieve a more secure position, obtain a promotion, start a career, or change academic departments/institutions. As one respondent mentioned, “I think more critically and more daringly . . . I [was] promoted after that from assistant to associate.”

Most respondents indicated that the GGP program enabled them to mentor, teach, and encourage other faculty members in their department in their grant-writing efforts. As one respondent mentioned, the GGP experience ”. . . has helped me to bridge the gap between clinicians and researchers.”

Challenges of the GGP

The most specific challenge of the GGP program mentioned by respondents was the variable commitment of mentors and the lack of having a mentor with expertise in a respondent’s area of research. The major barriers to respondents toward achieving their research goals were (1) the lack of a supportive research environment including a lack of financial support, few peers engaged in research, a lack of local mentoring, and infrastructural and administrative barriers (eg, no links to statisticians, confused priorities within department, and lack of diversity and foresight in department); (2) the lack of time to do research while participating in clinical duties and academic activities; and (3) the lack of information about funding opportunities at their institution. One respondent mentioned not having any external barriers but that the GGP experience “. . . did cause me to reevaluate and ultimately redefine my research goals.”

When asked what skills they learned during the GGP that helped them address these barriers, several respondents indicated (1) time management, including how to realistically organize the research process; (2) the ability to make connections with outside mentors and consultants; and (3) confidence development, such as contacting a program specialist despite poor information from their institution. However, respondents who mentioned the barrier of a nonsupportive research environment indicated that skills to address this barrier were not taught in the GGP program. Recommendations for improving the GGP

Suggestions for improving the GGP program included having 2 GGP entry levels (for beginning researchers and experienced researchers), continuing networking opportunities for “graduates,” addressing the need for female mentors, following up with consultations and mentoring, and having a mentor with expertise in participants’ area of research.

With respect to advice to future applicants, respondents mentioned being more aggressive in working with a mentor, having reasonable expectations, having long-term support from the department, using time efficiently to network, and reflecting on one’s motivation toward an intensive research career. As one participant said, “I would support the application 100% and notify them that it is one of the best investments of time they can make for their academic career.”

Discussion

The respondents’ comments made clear that the GGP is an “excellent place to start a career.” Most respondents appeared to regard the program as well worth the time and effort invested. Aside from some adjustments, such as consistency in mentoring and accommodating the research focus and skill level of participants, it is “. . . very important to continue [the] effort to train new researchers.” Interestingly, the GGP had a substantial impact on the extent to which many participants are now teaching and mentoring others to help develop the skills necessary for meaningful research.

Moreover, within 1 to 2 years of the program participants achieved a remarkable track record of grant submissions. Although for some participants “. . . breaking in [getting grant awards] still seems mysterious and forbidding,” they appreciated the opportunity to learn how to write a coherent concept paper, how to apply “good methods,” the importance of a track record, and to learn from experienced researchers.

As a result in large part of this evaluation, funding for the GGP has been continued with support from NAPCRG, the American Academy of Family Physicians, the STFM, the College of Family Physicians of Canada, and the home departments of participating academic departments and community-based residency programs.

Family medicine is poised to generate new knowledge through the development of its own research agenda-one that focuses on integrated, relationship-centered patient, family, and community health care.⁷ Young researchers, however, need to be nurtured, mentored, and organizationally supported.⁸ Research initiatives such as the GGP represent a collective enterprise, a building of coalitions toward the development of a culture of family medicine research. Continued research success requires commitments from both individuals and departments if family medicine research is to grow and prosper.

The benefits of using electronic medical records (EMRs) instead of paper records have been well documented.^1-6 However, the current use (5% to 10%) falls very short of the 100% by the year 2000 recommended by a 1991 Institute of Medicine (IOM) report⁷; furthermore, the rate of EMR use has remained relatively unchanged (5% to 10%) over the past decade. Given this stagnant rate of growth despite the IOM’s support, it is important to analyze the needs and perceptions of physicians with regard to EMRs.

There are few articles on the use of EMRs in out-patient settings.^1-20 Only 2 studies have analyzed the perceived needs and preferences for use of EMRs and family physicians.^11,12 The first study found significant concerns about the ease of data entry, data confidentiality, data sharing, and initial EMR train-ing.¹¹ The other article rated user requirements in rank order, but was limited by a 24% response rate.¹² Most of the literature and information on the Internet about EMRs focuses on the needs and perceptions of current EMR users.^1-20

The chasm theory of marketing (Figure) states that the early market of new product adopters (the first 5% to 10%) often has different concerns than the mainstream market. This group leaps ahead of the mainstream, creating a chasm.²¹ Early market adopters are more willing to endure technical difficulties and poor service to make a product work if it promises a competitive advantage, while mainstream users are more likely to look for widely used products that are inexpensive and easily assimilated into their current work environment with little organizational discomfort.²¹ Products engineered for the needs of the early market are often too complex to meet the needs of mainstream users; therefore, many products are unable to cross the chasm from early markets to mainstream markets because they fail to meet the needs of the mainstream users.²¹

This theory may explain the slow adoption of EMRs by family physicians. The purpose of this study was to investigate differences in attitudes and beliefs about EMRs between current EMR users (early market) and nonusers (mainstream market).

FIGURE 1
Chasm theory

Methods

The study population comprised active members of the Indiana Academy of Family Physicians (IAFP) listed in the 2000–2001 IAFP Membership Database. IAFP members who were practicing outside of Indiana or who spent less than 4 hours per week seeing patients were excluded from the study, leaving a study population of 1398 participants.

The authors designed a 53-item questionnaire based on the principles outlined in the 1991 Institute of Medicine report.^7,18,19 Six physicians with expertise in medical informatics screened the questionnaire for content validity. Twelve academic family physicians reviewed the instrument for structure, clarity, and relevance to test face validity. Ten resident physicians generated a test–retest reliability rate of >80% for each item over a 2-week interval.

The questionnaire consisted of 3 parts. The first section included questions about physician demographics, use of computer aided technology, practice location and type, and volume of patients. The second section contained questions about respondents’ attitudes, beliefs, and concerns regarding current and emerging technologic issues related to EMRs using a 4-point Likert scale ranging from “strongly agree” to “strongly disagree.” The Likert Scale also contained a “Don’t Know” option. The third section included multiple-choice questions to evaluate specific computer technology needs and preferences of family physicians. The final item was an open-ended question inviting a written response about any issues the respondent felt were not adequately addressed by the questionnaire.

Questionnaires were mailed in January 2001, with a followup mailing to nonrespondents 3 weeks later.²² Of the 726 (51.7%) responses, 45 (6.2%) were excluded because they did not indicate user or nonuser EMR status and 63 (8.8%) were excluded because half or more of the questions were unanswered. There were 618 (44.2%) usable surveys. The questionnaires were scanned into an electronic database and verified for accuracy using TELEform® software; the data were analyzed using SPSS^® (Version 10.0). The Likert scales were collapsed to a dichotomous variable, “agree” (strongly agree and agree) and “disagree” (strongly disagree, disagree, and unsure) for this analysis. Differences in demographic characteristics and attitudes, beliefs, and concerns were tested for significance using chi-square tests and the z-test of proportions. Statistical significance was determined by P < .05.

Results

Demographic and practice characteristics of the respondents are presented in Table 1. Of the 618 respondents, 89 (14.4%) were EMR users. A comparison of the users and nonusers revealed that there were no statistically significant differences in age or sex. EMR nonusers were more likely to practice in a suburban or rural location and were more often in a solo or small practice (2–6 physicians). Nonusers also tended to see an average of 31 more patients per week (116 vs 85).

Table 2 presents responses to questions about respondents’ experience with computer technology. A significantly greater proportion of EMR users use the Internet at home (98.9% vs 89.3%), at work (88.6% vs 62.8%), e-mail at work (78.4% vs 48.2%), and personal digital assistants (PDAs) (49.4% vs 26.1%). Other differences in technology use were not statistically significant.

Responses to questions about perceived EMR need and usefulness are summarized in Table 3. Nonusers were significantly less likely than users to believe that (1) physicians should computerize their medical records (67.7% vs 92.1%); (2) current EMRs are useful (51.5% vs 92.0%); (3) EMRs will reduce their risk of making medical errors (56.6% vs 78.7%); and (4) EMRs will improve health care quality in their office (52.4% vs 80.9%) or in the United States overall (54.3% vs 78.4%). Most respondents (77.6%) expressed an interest in an EMR system that would connect all physician practices, laboratories, radiography facilities, and hospitals for the secure exchange of patient data (85.4% user vs 76.2% nonuser; P = .076).

Table 4 summarizes the attitudes and beliefs of respondents about previously reported potential obstacles to EMR use, including: (1) data input; (2) cost; and (3) confidentiality and security.^{1,11,12,14,17-20,23,24} Few respondents (55.1% users vs 13.4% nonusers) stated that it is easy to enter data into current EMRs. Many respondents in both groups (61.8% vs 68.0%) responded that EMRs are too costly. Users tend to consider a relatively higher price as affordable for setting up an EMR system (66% would pay more than $5000) and are willing to pay a slightly higher monthly fee for the ongoing use of an EMR (65% would pay more than $100 per month). Users consider EMRs as more secure (54.5% users vs 21.2% nonusers) and more confidential (62.5% users vs 21.9% nonusers) than paper records. Few respondents stated they would object to sharing their EMR data with other physicians (24.9% nonusers vs 13.6% users).

TABLE 1
Demographic characteristics of respondents

TABLE 2
Technology use by respondents

TABLE 3
Attitudes and beliefs regarding electronic medical records

TABLE 4
Possible barriers to electronic medical record use