Dana Gelb Safran, ScD

The presence of sustained relationships between physicians and patients is a defining characteristic of primary care.¹ Family physicians use these relationships to acquire the depth of medical and personal knowledge about a patient that is essential to primary care practice.² It is also the reason some physicians choose this area of medicine.

A substantial body of empirical research points to the value of continuity in the physician-patient relationship, particularly in primary care. The benefits of continuity have been shown to accrue in the form of cost savings, improved health outcomes, and greater satisfaction for patients and physicians.^3-15 Yet little empirical research exists to indicate the amount of physician switching that occurs in primary care or the reasons for it.

In 1976 Kastler and colleagues¹⁶ examined the association between patients’ assessments of their care and their “doctor shopping” behavior. They found that patients’ evaluations of both interpersonal and structural features of care were significantly associated with the likelihood of voluntarily changing physicians. Those authors did not attempt to determine the relative importance of the 2 domains with respect to physician switching. The cross-sectional design precluded the study from determining which factor (if either) prospectively predicted switching.

Marquis and coworkers¹⁷ studied the sequencing of the satisfaction-disenrollment relationship using longitudinal data from the RAND Health Insurance Experiment (HIE). The HIE data showed that patients’ general satisfaction with their medical care significantly predicted physician switching over the following year. However, the HIE data did not afford the ability to differentiate among the many components of patient satisfaction and to discern which aspects specifically drive disenrollment.

Thus, little is known about the relative importance of the many factors that shape patients’ overall satisfaction with their physician and the extent to which performance on any of these ultimately drives a patient’s decision to leave a physician’s practice. Moreover, these earlier studies pre-date the recent surge in managed care enrollment and in consumerism among patients, both of which are presumed to be having a substantial impact on the rates of physician switching and the reasons for it. The generalizability of earlier findings to the present circumstances is unclear.

Methods

Our longitudinal observational study includes a population of insured adults who were employed by the Commonwealth of Massachusetts at baseline (1996), completed a self-administered questionnaire at baseline and follow-up (1999), and reported having a regular personal physician at baseline. Between January 1996 and April 1996 the baseline questionnaire was administered to a random sample of commonwealth employees who subscribed to any of 12 health plans available to employees, their dependents, and retirees. A 68.5% response rate was achieved (n=7204) using a standard 3-stage mail survey protocol with limited telephone follow-up of nonrespondents (mail responses=6810; telephone responses=394). Further details of the baseline sampling and data collection methods are documented elsewhere.

Follow-up data collection occurred precisely 3 years after baseline (January 1999-April 1999). Respondents who identified a primary care physician at baseline and participated by mail were eligible for follow-up (n=6075). Data were obtained using a standard 3-stage mail survey protocol with a final targeted mailing to racial and ethnic minorities (n=311) and to those without a college diploma (n=521). The targeted mailings were done because these subgroups were found to be underrepresented among follow-up respondents near the conclusion of data collection, and their representation in the longitudinal sample was important to our objectives. A 69.4% response rate was achieved in follow-up (n=4108) after accounting for respondents who died (n=21), were too ill to participate (n=2), or could not be located by mail in 1999 (n=136). At baseline and follow-up, respondents were somewhat older than nonrespondents, more likely to be women and white, and less likely to be poor (Appendix, Table 1A.

The questionnaire administered to patients at both baseline and follow-up included 4 scales measuring features of the physician-patient relationship (quality of communication, interpersonal treatment, physician’s knowledge of the patient, and patient trust) and 4 scales measuring structural aspects of care (access to care, visit-based continuity, duration of primary care relationship, and integration of care). The 8 scales are part of the Primary Care Assessment Survey (PCAS), a validated questionnaire with measures corresponding to the defining features of primary care posited by the Institute of Medicine (IOM). All concepts are measured in the context of a specific physician-patient relationship and reference the entirety of that relationship (ie, they are not visit-specific). All PCAS scales are scored on a 0 to 100 scale, with higher scores indicating more of the referent attribute. Details of the development and psychometric performance of the PCAS scales are available elsewhere. The item content and reliability coefficient (Cronbach a)for each scale are summarized in the Appendix Table 2A. In addition to completing the PCAS items referencing their experiences with and assessments of their primary physicians, the respondents also provided their physicians’ names.

Using the physician-identifying information provided by the patients, we linked data from the Massachusetts Board of Registration in Medicine (BRM) to the study database. The BRM data provided the physician’s practice address and several characteristics of the physician’s training and practice. We linked with the BRM data by using a matching algorithm based on the spelling of the physician’s name as provided by the patient, the distance between the patient’s home ZIP code and the physician’s practice site (BRM database), and the physician’s medical specialty. At both baseline and follow-up, matches from the BRM data were identified for 94.0% of the patients who named a physician.

Identifying Voluntary Versus Involuntary Disenrollment

Patients were classified as having changed physicians during the study period if their follow-up questionnaire reported having been in their primary physician’s practice for less than 3 years and if the physician named at follow-up was different from the one named in 1996. Those who changed physicians were then classified as having switched either voluntarily or involuntarily. A switch was considered involuntary if: (1) the patient’s baseline physician was no longer listed as active in the Massachusetts BRM database (n=77), (2) the baseline physician had moved more than 10 miles (n=91), or (3) the patient had moved more than 15 miles from the baseline residence (n=62).

In addition we considered the possible involuntary nature of physician switches that occurred along with a change in health plan enrollment. Because the employer in our study did not force or even incentivize health plan changes during the study period (ie, there was a consistent offering of health plans and no notable changes in the employee contributions for coverage), respondents who were insured by the commonwealth throughout the study period did not incur any involuntary physician switching owing to employer-imposed health plan changes. Among respondents not insured by the commonwealth throughout the study (ie, respondents who left state employment [n=40] or deferred coverage [n=7]), there were 6 who changed physicians. Five of these did so while remaining in the same health plan and were thus coded as having voluntarily changed physicians. The remaining individual who both changed health plans and physicians was dropped from our analysis of voluntary disenrollment, since we were unable to ascertain whether the plan change forced a change in physician.

Statistical Analyses

We limited the analytic sample to patients who completed both the baseline and follow-up questionnaire, who identified a primary physician at baseline for whom a BRM database match was found, and who had either remained with their baseline physician throughout the study period or had voluntarily left the physician’s practice (n=3052). Patients who had involuntarily disenrolled from their baseline physician’s practice (n=230) were excluded. Their exclusion was necessary, since there was no way to determine whether those who involuntarily switched physicians would have otherwise voluntarily left their physician. The sociodemographic and health profile of the analytic sample (n=3052) did not differ from that of the complete 1999 sample (n=4108).

Multiple logistic regression methods were used to evaluate interpersonal and structural features of care, as measured by the baseline PCAS scales, as predictors of voluntary disenrollment from a physician’s practice. All scales were standardized ([X1-mean]/standard deviation) to permit direct comparison of results across scales. First, the 8 PCAS scales were tested individually as predictors of voluntary disenrollment. Testing scales independently in this way is useful in cases such as this where moderate to high correlations exist among some scales. Although the majority of PCAS scale correlations are small, higher correlations exist among some scales (r=0.40-0.86). We applied the Bonferroni correction for multiple comparisons to this set of analyses.

Next, we modeled voluntary disenrollment as a function of the 4 relationship-quality scales together and tested for the equality of their effects (odd ratio [OR]) using a chi-square test. We repeated this using the 4 measures of structural features of care. Finally, using factor analysis methods (principal factor), we explored the potential for defining a single factor denoting relationship quality and a single factor denoting structural features of care. The 4 scales denoting structural features of care failed to generate an acceptable factor (range of factor loadings=0.20 [relationship duration] to 0.67 [access to care]), so this factor was dropped. The relationship-quality factor was retained (range of factor loadings=0.84 [knowledge of patient] to 0.92 [communication]) and tested in multiple logistic regression along with each of the 4 structure-of-care measures. A chi-square test was used to test the equivalence of the effects (OR) associated with the relationship quality factor and each of the 4 structure-of-care scales.

All regression models controlled for patients’ baseline sociodemographic profile (age, sex, race, years of education, household income), baseline health status (physical functioning, mental functioning, number of primary care sensitive conditions [PCSC], and number of primary care insensitive conditions [PCIC]), and baseline utilization (number of ambulatory visits in the previous 6 months). Physical and mental functioning were measured with data from the Medical Outcomes Study Short Form-12 (SF-12) Health Survey, which was included in the patient questionnaire. The numbers of primary care sensitive and insensitive conditions were classified using patients’ baseline reports about 20 chronic medical conditions with high prevalence among adults in the United States. The classification of PCSC and PCIC was defined by 9 generalist physicians, blind to the study objectives, who were asked to identify those conditions for which good primary care management could substantially affect outcomes (PCSC) and those for which it could not (PCIC). PCSC included hypertension, recent myocardial infarction, congestive heart failure, diabetes, angina, migraines, seasonal allergies, asthma, ulcers, arthritis, cancer, back pain, weight problem, and depression. PCIC included blindness, deafness, liver disease, insomnia, nonseasonal allergies (eg, dust, food, pets), and limb paralysis or amputation.

We assessed the goodness-of-fit of the final models using the Hosmer and Lemeshow method. For each scale, the P on the chi-square test statistic was greater than .05, indicating that the model fit the data well.

Results

Slightly more than one fourth of the patients in the longitudinal study panel changed physicians during the 3-year follow-up period (n=899), while approximately three fourths remained with their baseline physician throughout the study (n=2383). Of those who changed physicians, most changed voluntarily (n=669), but some changed involuntarily (n=230) because the physician had moved, retired, died, or the patient had moved a substantial distance. Table 1 shows the unadjusted sociodemographic, health, and utilization characteristics of the analytic sample, comparing those who voluntarily changed physicians with those who remained with their baseline physician throughout our study. Voluntary disenrollees were younger and more likely to be women and nonwhite than those who stayed with their baseline physician (P <.01). There were no differences in the baseline health status or outpatient utilization of the 2 groups.

Table 2 presents the results of the regression analyses examining the 8 PCAS scales as individual predictors of voluntary disenrollment (column 1) and the results of a multivariable model, including the composite relationship-quality factor (RQ) and the 4 structure-of-care scales as predictors of voluntary disenrollment (column 2, columns 3-7). When all scales were modeled independently (column 1), each was a significant predictor of voluntary disenrollment (P <.001), with somewhat larger effects associated with the relationship quality scales (OR=1.49-1.56) than the structure-of-care scales (OR=1.29-1.44). Pairwise tests of the ORs associated with each of the 4 relationship quality scales indicated that they were statistically equivalent in their ability to predict voluntary disenrollment. When the 4 indicators of relationship quality were included together in a multiple regression model, a chi-square test of their effects (OR) revealed the 4 to be statistically equivalent predictors of voluntary disenrollment. Similarly, in a model including the 4 structure-of-care scales, chi-square testing showed these 4 variables to have statistically equivalent effects. With the exception of sex, patient characteristics (sociodemographics, health, utilization) did not significantly predict voluntary disenrollment in any of these models. The gender effect had marginal significance in most cases (.05

Table 2 (column 2) shows the results of modeling voluntary disenrollment as a function of both relationship quality and structure-of-care together. In that multivariable model, the composite relationship quality factor (RQ) emerged as the leading predictor of voluntary disenrollment (OR=1.59; P <.001). This OR signifies that a standard deviation (SD) decline in relationship quality was associated with a 59% increase in the odds of voluntary disenrollment. The results indicate that after accounting for patients’ baseline characteristics (sociodemographic, health, and utilization) and the 4 structural features of care, patients with relationship quality scores in the 5th percentile in 1996 were 3 times more likely to voluntarily disenroll from their physician’s practice than those with 95th percentile relationship quality scores (37.8% vs 12.2%). The 2 measures of continuity also significantly predicted disenrollment in the multivariable model (visit-based continuity: OR=1.14, P=.03; relationship duration: OR=1.16, P=.01). Access to care predicted disenrollment with marginal significance (OR=1.14; P=.08), and integration did not significantly predict disenrollment (P=.59) in this model. None of the patient characteristics (sociodemograhics, health, utilization) significantly predicted disenrollment in the presence of these 5 quality-of-care measures.

Discussion

In our observational study of insured employed adults, 20% of the patients voluntarily left their primary care physician’s practice over a 3-year period. Another 5% left involuntarily, owing to factors that forced a change (eg, the physician moved, retired, or died). For the average full-time primary physician, this translates into approximately 400 patients voluntarily leaving the practice over a given 3-year period and another 100 leaving involuntarily. Rates of involuntary switching are almost certainly higher among physicians whose patients face more employer-imposed disruptions than occurred in our study population.

Our data indicate that the quality of the physician-patient relationship significantly predicts patients’ loyalty. With patient characteristics and structural features of care taken into account, those with the poorest-quality physician-patient relationships in 1996 were 3 times more likely to leave the physician’s practice over the ensuing 3 years than those with the highest-quality relationships.

Structural features of care also emerged as important determinants of patients’ disenrollment decisions. When considered independently of relationship quality, each of the 4 structural elements of care significantly predicted voluntary disenrollment. With relationship quality taken into account, continuity of care (both relationship duration and visit-based continuity) remained significant predictors of disenrollment, while access to care and integration of care did not. The results suggest that although these patients put a high priority on being given timely and convenient access to their physician’s office, the issue of who they are given access to and the quality of their connection with that clinician mattered more.

Our findings are consistent with those reported more than 2 decades ago by Kasteler and colleagues,¹⁶ who found both interpersonal quality of care and structural features of care to be significantly associated with voluntary physician switching in a cross-sectional study. Our study has the advantage of longitudinal data through which the sequencing of effects is clear. In addition, our study advances beyond earlier studies that evaluated a single patient-based measure of care in predicting disenrollment.^17,25 Marquis and coworkers¹⁷ showed that patients’ general satisfaction with their physician predicted disenrollment from the physician’s practice over the following year. Thom and colleagues²⁵ found that patients’ trust in their physicians significantly predicted disenrollment over the next 6 months. Our study includes measures of 8 characteristics that encompass the defining features of primary care as posited by the IOM1 and others,^26-29 with several features for which the relationship to disenrollment have not been previously studied. Our study contributes evidence concerning both the absolute and relative importance of interpersonal and structural features of care as predictors of patients’ loyalty to their primary care physician’s practice.

Limitations

Our study is limited to a population of adults in Massachusetts who were employed and insured at baseline. Rates of involuntary physician switching in this population were likely lower than would be observed in other employed populations (particularly in competitive health care markets) for the reasons mentioned (ie, benefit policies that minimized employer-imposed disruption of employees’ health care arrangements). However, the observed rates of voluntary physician switching and the predictors of voluntary disenrollment should not be affected. Those findings may be presumed to generalize more broadly.

A second limitation is the absence of information about salient health events that occurred between the baseline and follow-up phases of our study. For patients who incurred a serious episode of illness, information about the intervening health events and their experiences with their physician during that time might have enhanced our understanding of the factors that influenced their decisions about whether to remain in that physician’s practice.

Similarly, the study lacked detailed indicators of the technical quality of care provided and therefore could not assess the role that technical quality—and patients’ perceptions of it—play in shaping patients’ loyalty to their physician.

Finally, our study could not fully account for one potential source of involuntary disenrollment: patients leaving practices because the physicians no longer accepted their health plan. However, rates of physician turnover during the study period were no more than 5% in any of the health plans studied and were substantially lower in most.²¹ Thus, our findings are unlikely to have been substantively altered by a detailed accounting of this form of involuntary physician switching.

Conclusions

Previous empirical research has underscored the importance of physician-patient relationship quality by demonstrating its association with important outcomes, including adherence to medical advice,^19,30-32 satisfaction with care,^19,33-35 and litigation against physicians.^36-38 However, few studies have had the benefit of longitudinal data with which to verify the sequencing of effects between relationship quality and outcomes.

In our study the strength of physician-patient relationships in primary care—as indicated by patients’ trust in their physician, their assessments of how well the physician knows them, and the quality of communication and interpersonal treatment—was the leading predictor of patients’ loyalty to their primary physician’s practice. Continuity of care also significantly predicted voluntary disenrollment. The findings are noteworthy against a backdrop of health care delivery changes nationwide that many describe as threatening the therapeutic alliance between the physician and the patient.^29,39-42

The recent IOM report on the future of primary care called attention to the importance of the physician-patient relationship in primary care, asserting that primary care is predicated on sustained clinician-patient partnerships and on a whole-person orientation.1 In our study, longitudinal data demonstrate that the strength of connection between a patient and his or her primary care physician significantly predicts the likelihood of that patient remaining in that physician’s practice (vs voluntarily leaving) over the next several years. In an era marked by increasing pressure on clinicians and health care organizations to attend to such factors as market share, productivity, and efficiency, these findings point to a set of attributes that might otherwise be overlooked. They suggest that medical practices and health plans cannot afford to ignore that the essence of medical care delivery involves the interaction of one human being with another.

Acknowledgments

This research was supported by grant number R01 HS08841 from the Agency for Healthcare Research and Quality (formerly the Agency for Health Care Policy and Research) and by grant number 035321 from the Robert Wood Johnson Foundation. We are indebted to Dolores Mitchell, executive director of the Massachusetts Group Insurance Commission, whose commitment and participation have made this study possible. We also gratefully acknowledge Brian Clarridge, PhD, and his colleagues at The Center for Survey Research, University of Massachusetts, for their technical expertise and commitment to excellence in obtaining the data for our study. Finally, we acknowledge each of the health plans involved in the study, and particularly the health plan executives who served on our Advisory Committee and those who provided interviews, for generously giving their time, insights, and information that was critical to our research.

The presence of sustained relationships between physicians and patients is a defining characteristic of primary care.¹ Family physicians use these relationships to acquire the depth of medical and personal knowledge about a patient that is essential to primary care practice.² It is also the reason some physicians choose this area of medicine.

A substantial body of empirical research points to the value of continuity in the physician-patient relationship, particularly in primary care. The benefits of continuity have been shown to accrue in the form of cost savings, improved health outcomes, and greater satisfaction for patients and physicians.^3-15 Yet little empirical research exists to indicate the amount of physician switching that occurs in primary care or the reasons for it.

In 1976 Kastler and colleagues¹⁶ examined the association between patients’ assessments of their care and their “doctor shopping” behavior. They found that patients’ evaluations of both interpersonal and structural features of care were significantly associated with the likelihood of voluntarily changing physicians. Those authors did not attempt to determine the relative importance of the 2 domains with respect to physician switching. The cross-sectional design precluded the study from determining which factor (if either) prospectively predicted switching.

Marquis and coworkers¹⁷ studied the sequencing of the satisfaction-disenrollment relationship using longitudinal data from the RAND Health Insurance Experiment (HIE). The HIE data showed that patients’ general satisfaction with their medical care significantly predicted physician switching over the following year. However, the HIE data did not afford the ability to differentiate among the many components of patient satisfaction and to discern which aspects specifically drive disenrollment.

Thus, little is known about the relative importance of the many factors that shape patients’ overall satisfaction with their physician and the extent to which performance on any of these ultimately drives a patient’s decision to leave a physician’s practice. Moreover, these earlier studies pre-date the recent surge in managed care enrollment and in consumerism among patients, both of which are presumed to be having a substantial impact on the rates of physician switching and the reasons for it. The generalizability of earlier findings to the present circumstances is unclear.

Methods

Our longitudinal observational study includes a population of insured adults who were employed by the Commonwealth of Massachusetts at baseline (1996), completed a self-administered questionnaire at baseline and follow-up (1999), and reported having a regular personal physician at baseline. Between January 1996 and April 1996 the baseline questionnaire was administered to a random sample of commonwealth employees who subscribed to any of 12 health plans available to employees, their dependents, and retirees. A 68.5% response rate was achieved (n=7204) using a standard 3-stage mail survey protocol with limited telephone follow-up of nonrespondents (mail responses=6810; telephone responses=394). Further details of the baseline sampling and data collection methods are documented elsewhere.

Follow-up data collection occurred precisely 3 years after baseline (January 1999-April 1999). Respondents who identified a primary care physician at baseline and participated by mail were eligible for follow-up (n=6075). Data were obtained using a standard 3-stage mail survey protocol with a final targeted mailing to racial and ethnic minorities (n=311) and to those without a college diploma (n=521). The targeted mailings were done because these subgroups were found to be underrepresented among follow-up respondents near the conclusion of data collection, and their representation in the longitudinal sample was important to our objectives. A 69.4% response rate was achieved in follow-up (n=4108) after accounting for respondents who died (n=21), were too ill to participate (n=2), or could not be located by mail in 1999 (n=136). At baseline and follow-up, respondents were somewhat older than nonrespondents, more likely to be women and white, and less likely to be poor (Appendix, Table 1A.

The questionnaire administered to patients at both baseline and follow-up included 4 scales measuring features of the physician-patient relationship (quality of communication, interpersonal treatment, physician’s knowledge of the patient, and patient trust) and 4 scales measuring structural aspects of care (access to care, visit-based continuity, duration of primary care relationship, and integration of care). The 8 scales are part of the Primary Care Assessment Survey (PCAS), a validated questionnaire with measures corresponding to the defining features of primary care posited by the Institute of Medicine (IOM). All concepts are measured in the context of a specific physician-patient relationship and reference the entirety of that relationship (ie, they are not visit-specific). All PCAS scales are scored on a 0 to 100 scale, with higher scores indicating more of the referent attribute. Details of the development and psychometric performance of the PCAS scales are available elsewhere. The item content and reliability coefficient (Cronbach a)for each scale are summarized in the Appendix Table 2A. In addition to completing the PCAS items referencing their experiences with and assessments of their primary physicians, the respondents also provided their physicians’ names.

Using the physician-identifying information provided by the patients, we linked data from the Massachusetts Board of Registration in Medicine (BRM) to the study database. The BRM data provided the physician’s practice address and several characteristics of the physician’s training and practice. We linked with the BRM data by using a matching algorithm based on the spelling of the physician’s name as provided by the patient, the distance between the patient’s home ZIP code and the physician’s practice site (BRM database), and the physician’s medical specialty. At both baseline and follow-up, matches from the BRM data were identified for 94.0% of the patients who named a physician.

Identifying Voluntary Versus Involuntary Disenrollment

Patients were classified as having changed physicians during the study period if their follow-up questionnaire reported having been in their primary physician’s practice for less than 3 years and if the physician named at follow-up was different from the one named in 1996. Those who changed physicians were then classified as having switched either voluntarily or involuntarily. A switch was considered involuntary if: (1) the patient’s baseline physician was no longer listed as active in the Massachusetts BRM database (n=77), (2) the baseline physician had moved more than 10 miles (n=91), or (3) the patient had moved more than 15 miles from the baseline residence (n=62).

In addition we considered the possible involuntary nature of physician switches that occurred along with a change in health plan enrollment. Because the employer in our study did not force or even incentivize health plan changes during the study period (ie, there was a consistent offering of health plans and no notable changes in the employee contributions for coverage), respondents who were insured by the commonwealth throughout the study period did not incur any involuntary physician switching owing to employer-imposed health plan changes. Among respondents not insured by the commonwealth throughout the study (ie, respondents who left state employment [n=40] or deferred coverage [n=7]), there were 6 who changed physicians. Five of these did so while remaining in the same health plan and were thus coded as having voluntarily changed physicians. The remaining individual who both changed health plans and physicians was dropped from our analysis of voluntary disenrollment, since we were unable to ascertain whether the plan change forced a change in physician.

Statistical Analyses

We limited the analytic sample to patients who completed both the baseline and follow-up questionnaire, who identified a primary physician at baseline for whom a BRM database match was found, and who had either remained with their baseline physician throughout the study period or had voluntarily left the physician’s practice (n=3052). Patients who had involuntarily disenrolled from their baseline physician’s practice (n=230) were excluded. Their exclusion was necessary, since there was no way to determine whether those who involuntarily switched physicians would have otherwise voluntarily left their physician. The sociodemographic and health profile of the analytic sample (n=3052) did not differ from that of the complete 1999 sample (n=4108).

Multiple logistic regression methods were used to evaluate interpersonal and structural features of care, as measured by the baseline PCAS scales, as predictors of voluntary disenrollment from a physician’s practice. All scales were standardized ([X1-mean]/standard deviation) to permit direct comparison of results across scales. First, the 8 PCAS scales were tested individually as predictors of voluntary disenrollment. Testing scales independently in this way is useful in cases such as this where moderate to high correlations exist among some scales. Although the majority of PCAS scale correlations are small, higher correlations exist among some scales (r=0.40-0.86). We applied the Bonferroni correction for multiple comparisons to this set of analyses.

Next, we modeled voluntary disenrollment as a function of the 4 relationship-quality scales together and tested for the equality of their effects (odd ratio [OR]) using a chi-square test. We repeated this using the 4 measures of structural features of care. Finally, using factor analysis methods (principal factor), we explored the potential for defining a single factor denoting relationship quality and a single factor denoting structural features of care. The 4 scales denoting structural features of care failed to generate an acceptable factor (range of factor loadings=0.20 [relationship duration] to 0.67 [access to care]), so this factor was dropped. The relationship-quality factor was retained (range of factor loadings=0.84 [knowledge of patient] to 0.92 [communication]) and tested in multiple logistic regression along with each of the 4 structure-of-care measures. A chi-square test was used to test the equivalence of the effects (OR) associated with the relationship quality factor and each of the 4 structure-of-care scales.

All regression models controlled for patients’ baseline sociodemographic profile (age, sex, race, years of education, household income), baseline health status (physical functioning, mental functioning, number of primary care sensitive conditions [PCSC], and number of primary care insensitive conditions [PCIC]), and baseline utilization (number of ambulatory visits in the previous 6 months). Physical and mental functioning were measured with data from the Medical Outcomes Study Short Form-12 (SF-12) Health Survey, which was included in the patient questionnaire. The numbers of primary care sensitive and insensitive conditions were classified using patients’ baseline reports about 20 chronic medical conditions with high prevalence among adults in the United States. The classification of PCSC and PCIC was defined by 9 generalist physicians, blind to the study objectives, who were asked to identify those conditions for which good primary care management could substantially affect outcomes (PCSC) and those for which it could not (PCIC). PCSC included hypertension, recent myocardial infarction, congestive heart failure, diabetes, angina, migraines, seasonal allergies, asthma, ulcers, arthritis, cancer, back pain, weight problem, and depression. PCIC included blindness, deafness, liver disease, insomnia, nonseasonal allergies (eg, dust, food, pets), and limb paralysis or amputation.

We assessed the goodness-of-fit of the final models using the Hosmer and Lemeshow method. For each scale, the P on the chi-square test statistic was greater than .05, indicating that the model fit the data well.

Results

Slightly more than one fourth of the patients in the longitudinal study panel changed physicians during the 3-year follow-up period (n=899), while approximately three fourths remained with their baseline physician throughout the study (n=2383). Of those who changed physicians, most changed voluntarily (n=669), but some changed involuntarily (n=230) because the physician had moved, retired, died, or the patient had moved a substantial distance. Table 1 shows the unadjusted sociodemographic, health, and utilization characteristics of the analytic sample, comparing those who voluntarily changed physicians with those who remained with their baseline physician throughout our study. Voluntary disenrollees were younger and more likely to be women and nonwhite than those who stayed with their baseline physician (P <.01). There were no differences in the baseline health status or outpatient utilization of the 2 groups.

Table 2 presents the results of the regression analyses examining the 8 PCAS scales as individual predictors of voluntary disenrollment (column 1) and the results of a multivariable model, including the composite relationship-quality factor (RQ) and the 4 structure-of-care scales as predictors of voluntary disenrollment (column 2, columns 3-7). When all scales were modeled independently (column 1), each was a significant predictor of voluntary disenrollment (P <.001), with somewhat larger effects associated with the relationship quality scales (OR=1.49-1.56) than the structure-of-care scales (OR=1.29-1.44). Pairwise tests of the ORs associated with each of the 4 relationship quality scales indicated that they were statistically equivalent in their ability to predict voluntary disenrollment. When the 4 indicators of relationship quality were included together in a multiple regression model, a chi-square test of their effects (OR) revealed the 4 to be statistically equivalent predictors of voluntary disenrollment. Similarly, in a model including the 4 structure-of-care scales, chi-square testing showed these 4 variables to have statistically equivalent effects. With the exception of sex, patient characteristics (sociodemographics, health, utilization) did not significantly predict voluntary disenrollment in any of these models. The gender effect had marginal significance in most cases (.05

Table 2 (column 2) shows the results of modeling voluntary disenrollment as a function of both relationship quality and structure-of-care together. In that multivariable model, the composite relationship quality factor (RQ) emerged as the leading predictor of voluntary disenrollment (OR=1.59; P <.001). This OR signifies that a standard deviation (SD) decline in relationship quality was associated with a 59% increase in the odds of voluntary disenrollment. The results indicate that after accounting for patients’ baseline characteristics (sociodemographic, health, and utilization) and the 4 structural features of care, patients with relationship quality scores in the 5th percentile in 1996 were 3 times more likely to voluntarily disenroll from their physician’s practice than those with 95th percentile relationship quality scores (37.8% vs 12.2%). The 2 measures of continuity also significantly predicted disenrollment in the multivariable model (visit-based continuity: OR=1.14, P=.03; relationship duration: OR=1.16, P=.01). Access to care predicted disenrollment with marginal significance (OR=1.14; P=.08), and integration did not significantly predict disenrollment (P=.59) in this model. None of the patient characteristics (sociodemograhics, health, utilization) significantly predicted disenrollment in the presence of these 5 quality-of-care measures.

Discussion

In our observational study of insured employed adults, 20% of the patients voluntarily left their primary care physician’s practice over a 3-year period. Another 5% left involuntarily, owing to factors that forced a change (eg, the physician moved, retired, or died). For the average full-time primary physician, this translates into approximately 400 patients voluntarily leaving the practice over a given 3-year period and another 100 leaving involuntarily. Rates of involuntary switching are almost certainly higher among physicians whose patients face more employer-imposed disruptions than occurred in our study population.

Our data indicate that the quality of the physician-patient relationship significantly predicts patients’ loyalty. With patient characteristics and structural features of care taken into account, those with the poorest-quality physician-patient relationships in 1996 were 3 times more likely to leave the physician’s practice over the ensuing 3 years than those with the highest-quality relationships.

Structural features of care also emerged as important determinants of patients’ disenrollment decisions. When considered independently of relationship quality, each of the 4 structural elements of care significantly predicted voluntary disenrollment. With relationship quality taken into account, continuity of care (both relationship duration and visit-based continuity) remained significant predictors of disenrollment, while access to care and integration of care did not. The results suggest that although these patients put a high priority on being given timely and convenient access to their physician’s office, the issue of who they are given access to and the quality of their connection with that clinician mattered more.

Our findings are consistent with those reported more than 2 decades ago by Kasteler and colleagues,¹⁶ who found both interpersonal quality of care and structural features of care to be significantly associated with voluntary physician switching in a cross-sectional study. Our study has the advantage of longitudinal data through which the sequencing of effects is clear. In addition, our study advances beyond earlier studies that evaluated a single patient-based measure of care in predicting disenrollment.^17,25 Marquis and coworkers¹⁷ showed that patients’ general satisfaction with their physician predicted disenrollment from the physician’s practice over the following year. Thom and colleagues²⁵ found that patients’ trust in their physicians significantly predicted disenrollment over the next 6 months. Our study includes measures of 8 characteristics that encompass the defining features of primary care as posited by the IOM1 and others,^26-29 with several features for which the relationship to disenrollment have not been previously studied. Our study contributes evidence concerning both the absolute and relative importance of interpersonal and structural features of care as predictors of patients’ loyalty to their primary care physician’s practice.

Limitations

Our study is limited to a population of adults in Massachusetts who were employed and insured at baseline. Rates of involuntary physician switching in this population were likely lower than would be observed in other employed populations (particularly in competitive health care markets) for the reasons mentioned (ie, benefit policies that minimized employer-imposed disruption of employees’ health care arrangements). However, the observed rates of voluntary physician switching and the predictors of voluntary disenrollment should not be affected. Those findings may be presumed to generalize more broadly.

A second limitation is the absence of information about salient health events that occurred between the baseline and follow-up phases of our study. For patients who incurred a serious episode of illness, information about the intervening health events and their experiences with their physician during that time might have enhanced our understanding of the factors that influenced their decisions about whether to remain in that physician’s practice.

Similarly, the study lacked detailed indicators of the technical quality of care provided and therefore could not assess the role that technical quality—and patients’ perceptions of it—play in shaping patients’ loyalty to their physician.

Finally, our study could not fully account for one potential source of involuntary disenrollment: patients leaving practices because the physicians no longer accepted their health plan. However, rates of physician turnover during the study period were no more than 5% in any of the health plans studied and were substantially lower in most.²¹ Thus, our findings are unlikely to have been substantively altered by a detailed accounting of this form of involuntary physician switching.

Conclusions

Previous empirical research has underscored the importance of physician-patient relationship quality by demonstrating its association with important outcomes, including adherence to medical advice,^19,30-32 satisfaction with care,^19,33-35 and litigation against physicians.^36-38 However, few studies have had the benefit of longitudinal data with which to verify the sequencing of effects between relationship quality and outcomes.

In our study the strength of physician-patient relationships in primary care—as indicated by patients’ trust in their physician, their assessments of how well the physician knows them, and the quality of communication and interpersonal treatment—was the leading predictor of patients’ loyalty to their primary physician’s practice. Continuity of care also significantly predicted voluntary disenrollment. The findings are noteworthy against a backdrop of health care delivery changes nationwide that many describe as threatening the therapeutic alliance between the physician and the patient.^29,39-42

The recent IOM report on the future of primary care called attention to the importance of the physician-patient relationship in primary care, asserting that primary care is predicated on sustained clinician-patient partnerships and on a whole-person orientation.1 In our study, longitudinal data demonstrate that the strength of connection between a patient and his or her primary care physician significantly predicts the likelihood of that patient remaining in that physician’s practice (vs voluntarily leaving) over the next several years. In an era marked by increasing pressure on clinicians and health care organizations to attend to such factors as market share, productivity, and efficiency, these findings point to a set of attributes that might otherwise be overlooked. They suggest that medical practices and health plans cannot afford to ignore that the essence of medical care delivery involves the interaction of one human being with another.

Acknowledgments

This research was supported by grant number R01 HS08841 from the Agency for Healthcare Research and Quality (formerly the Agency for Health Care Policy and Research) and by grant number 035321 from the Robert Wood Johnson Foundation. We are indebted to Dolores Mitchell, executive director of the Massachusetts Group Insurance Commission, whose commitment and participation have made this study possible. We also gratefully acknowledge Brian Clarridge, PhD, and his colleagues at The Center for Survey Research, University of Massachusetts, for their technical expertise and commitment to excellence in obtaining the data for our study. Finally, we acknowledge each of the health plans involved in the study, and particularly the health plan executives who served on our Advisory Committee and those who provided interviews, for generously giving their time, insights, and information that was critical to our research.

The presence of sustained relationships between physicians and patients is a defining characteristic of primary care.¹ Family physicians use these relationships to acquire the depth of medical and personal knowledge about a patient that is essential to primary care practice.² It is also the reason some physicians choose this area of medicine.

A substantial body of empirical research points to the value of continuity in the physician-patient relationship, particularly in primary care. The benefits of continuity have been shown to accrue in the form of cost savings, improved health outcomes, and greater satisfaction for patients and physicians.^3-15 Yet little empirical research exists to indicate the amount of physician switching that occurs in primary care or the reasons for it.

In 1976 Kastler and colleagues¹⁶ examined the association between patients’ assessments of their care and their “doctor shopping” behavior. They found that patients’ evaluations of both interpersonal and structural features of care were significantly associated with the likelihood of voluntarily changing physicians. Those authors did not attempt to determine the relative importance of the 2 domains with respect to physician switching. The cross-sectional design precluded the study from determining which factor (if either) prospectively predicted switching.

Marquis and coworkers¹⁷ studied the sequencing of the satisfaction-disenrollment relationship using longitudinal data from the RAND Health Insurance Experiment (HIE). The HIE data showed that patients’ general satisfaction with their medical care significantly predicted physician switching over the following year. However, the HIE data did not afford the ability to differentiate among the many components of patient satisfaction and to discern which aspects specifically drive disenrollment.

Thus, little is known about the relative importance of the many factors that shape patients’ overall satisfaction with their physician and the extent to which performance on any of these ultimately drives a patient’s decision to leave a physician’s practice. Moreover, these earlier studies pre-date the recent surge in managed care enrollment and in consumerism among patients, both of which are presumed to be having a substantial impact on the rates of physician switching and the reasons for it. The generalizability of earlier findings to the present circumstances is unclear.

Methods

Our longitudinal observational study includes a population of insured adults who were employed by the Commonwealth of Massachusetts at baseline (1996), completed a self-administered questionnaire at baseline and follow-up (1999), and reported having a regular personal physician at baseline. Between January 1996 and April 1996 the baseline questionnaire was administered to a random sample of commonwealth employees who subscribed to any of 12 health plans available to employees, their dependents, and retirees. A 68.5% response rate was achieved (n=7204) using a standard 3-stage mail survey protocol with limited telephone follow-up of nonrespondents (mail responses=6810; telephone responses=394). Further details of the baseline sampling and data collection methods are documented elsewhere.

Follow-up data collection occurred precisely 3 years after baseline (January 1999-April 1999). Respondents who identified a primary care physician at baseline and participated by mail were eligible for follow-up (n=6075). Data were obtained using a standard 3-stage mail survey protocol with a final targeted mailing to racial and ethnic minorities (n=311) and to those without a college diploma (n=521). The targeted mailings were done because these subgroups were found to be underrepresented among follow-up respondents near the conclusion of data collection, and their representation in the longitudinal sample was important to our objectives. A 69.4% response rate was achieved in follow-up (n=4108) after accounting for respondents who died (n=21), were too ill to participate (n=2), or could not be located by mail in 1999 (n=136). At baseline and follow-up, respondents were somewhat older than nonrespondents, more likely to be women and white, and less likely to be poor (Appendix, Table 1A.

The questionnaire administered to patients at both baseline and follow-up included 4 scales measuring features of the physician-patient relationship (quality of communication, interpersonal treatment, physician’s knowledge of the patient, and patient trust) and 4 scales measuring structural aspects of care (access to care, visit-based continuity, duration of primary care relationship, and integration of care). The 8 scales are part of the Primary Care Assessment Survey (PCAS), a validated questionnaire with measures corresponding to the defining features of primary care posited by the Institute of Medicine (IOM). All concepts are measured in the context of a specific physician-patient relationship and reference the entirety of that relationship (ie, they are not visit-specific). All PCAS scales are scored on a 0 to 100 scale, with higher scores indicating more of the referent attribute. Details of the development and psychometric performance of the PCAS scales are available elsewhere. The item content and reliability coefficient (Cronbach a)for each scale are summarized in the Appendix Table 2A. In addition to completing the PCAS items referencing their experiences with and assessments of their primary physicians, the respondents also provided their physicians’ names.

Using the physician-identifying information provided by the patients, we linked data from the Massachusetts Board of Registration in Medicine (BRM) to the study database. The BRM data provided the physician’s practice address and several characteristics of the physician’s training and practice. We linked with the BRM data by using a matching algorithm based on the spelling of the physician’s name as provided by the patient, the distance between the patient’s home ZIP code and the physician’s practice site (BRM database), and the physician’s medical specialty. At both baseline and follow-up, matches from the BRM data were identified for 94.0% of the patients who named a physician.

Identifying Voluntary Versus Involuntary Disenrollment

Patients were classified as having changed physicians during the study period if their follow-up questionnaire reported having been in their primary physician’s practice for less than 3 years and if the physician named at follow-up was different from the one named in 1996. Those who changed physicians were then classified as having switched either voluntarily or involuntarily. A switch was considered involuntary if: (1) the patient’s baseline physician was no longer listed as active in the Massachusetts BRM database (n=77), (2) the baseline physician had moved more than 10 miles (n=91), or (3) the patient had moved more than 15 miles from the baseline residence (n=62).

In addition we considered the possible involuntary nature of physician switches that occurred along with a change in health plan enrollment. Because the employer in our study did not force or even incentivize health plan changes during the study period (ie, there was a consistent offering of health plans and no notable changes in the employee contributions for coverage), respondents who were insured by the commonwealth throughout the study period did not incur any involuntary physician switching owing to employer-imposed health plan changes. Among respondents not insured by the commonwealth throughout the study (ie, respondents who left state employment [n=40] or deferred coverage [n=7]), there were 6 who changed physicians. Five of these did so while remaining in the same health plan and were thus coded as having voluntarily changed physicians. The remaining individual who both changed health plans and physicians was dropped from our analysis of voluntary disenrollment, since we were unable to ascertain whether the plan change forced a change in physician.

Statistical Analyses

We limited the analytic sample to patients who completed both the baseline and follow-up questionnaire, who identified a primary physician at baseline for whom a BRM database match was found, and who had either remained with their baseline physician throughout the study period or had voluntarily left the physician’s practice (n=3052). Patients who had involuntarily disenrolled from their baseline physician’s practice (n=230) were excluded. Their exclusion was necessary, since there was no way to determine whether those who involuntarily switched physicians would have otherwise voluntarily left their physician. The sociodemographic and health profile of the analytic sample (n=3052) did not differ from that of the complete 1999 sample (n=4108).

Multiple logistic regression methods were used to evaluate interpersonal and structural features of care, as measured by the baseline PCAS scales, as predictors of voluntary disenrollment from a physician’s practice. All scales were standardized ([X1-mean]/standard deviation) to permit direct comparison of results across scales. First, the 8 PCAS scales were tested individually as predictors of voluntary disenrollment. Testing scales independently in this way is useful in cases such as this where moderate to high correlations exist among some scales. Although the majority of PCAS scale correlations are small, higher correlations exist among some scales (r=0.40-0.86). We applied the Bonferroni correction for multiple comparisons to this set of analyses.

Next, we modeled voluntary disenrollment as a function of the 4 relationship-quality scales together and tested for the equality of their effects (odd ratio [OR]) using a chi-square test. We repeated this using the 4 measures of structural features of care. Finally, using factor analysis methods (principal factor), we explored the potential for defining a single factor denoting relationship quality and a single factor denoting structural features of care. The 4 scales denoting structural features of care failed to generate an acceptable factor (range of factor loadings=0.20 [relationship duration] to 0.67 [access to care]), so this factor was dropped. The relationship-quality factor was retained (range of factor loadings=0.84 [knowledge of patient] to 0.92 [communication]) and tested in multiple logistic regression along with each of the 4 structure-of-care measures. A chi-square test was used to test the equivalence of the effects (OR) associated with the relationship quality factor and each of the 4 structure-of-care scales.

All regression models controlled for patients’ baseline sociodemographic profile (age, sex, race, years of education, household income), baseline health status (physical functioning, mental functioning, number of primary care sensitive conditions [PCSC], and number of primary care insensitive conditions [PCIC]), and baseline utilization (number of ambulatory visits in the previous 6 months). Physical and mental functioning were measured with data from the Medical Outcomes Study Short Form-12 (SF-12) Health Survey, which was included in the patient questionnaire. The numbers of primary care sensitive and insensitive conditions were classified using patients’ baseline reports about 20 chronic medical conditions with high prevalence among adults in the United States. The classification of PCSC and PCIC was defined by 9 generalist physicians, blind to the study objectives, who were asked to identify those conditions for which good primary care management could substantially affect outcomes (PCSC) and those for which it could not (PCIC). PCSC included hypertension, recent myocardial infarction, congestive heart failure, diabetes, angina, migraines, seasonal allergies, asthma, ulcers, arthritis, cancer, back pain, weight problem, and depression. PCIC included blindness, deafness, liver disease, insomnia, nonseasonal allergies (eg, dust, food, pets), and limb paralysis or amputation.

We assessed the goodness-of-fit of the final models using the Hosmer and Lemeshow method. For each scale, the P on the chi-square test statistic was greater than .05, indicating that the model fit the data well.

Results

Slightly more than one fourth of the patients in the longitudinal study panel changed physicians during the 3-year follow-up period (n=899), while approximately three fourths remained with their baseline physician throughout the study (n=2383). Of those who changed physicians, most changed voluntarily (n=669), but some changed involuntarily (n=230) because the physician had moved, retired, died, or the patient had moved a substantial distance. Table 1 shows the unadjusted sociodemographic, health, and utilization characteristics of the analytic sample, comparing those who voluntarily changed physicians with those who remained with their baseline physician throughout our study. Voluntary disenrollees were younger and more likely to be women and nonwhite than those who stayed with their baseline physician (P <.01). There were no differences in the baseline health status or outpatient utilization of the 2 groups.

Table 2 presents the results of the regression analyses examining the 8 PCAS scales as individual predictors of voluntary disenrollment (column 1) and the results of a multivariable model, including the composite relationship-quality factor (RQ) and the 4 structure-of-care scales as predictors of voluntary disenrollment (column 2, columns 3-7). When all scales were modeled independently (column 1), each was a significant predictor of voluntary disenrollment (P <.001), with somewhat larger effects associated with the relationship quality scales (OR=1.49-1.56) than the structure-of-care scales (OR=1.29-1.44). Pairwise tests of the ORs associated with each of the 4 relationship quality scales indicated that they were statistically equivalent in their ability to predict voluntary disenrollment. When the 4 indicators of relationship quality were included together in a multiple regression model, a chi-square test of their effects (OR) revealed the 4 to be statistically equivalent predictors of voluntary disenrollment. Similarly, in a model including the 4 structure-of-care scales, chi-square testing showed these 4 variables to have statistically equivalent effects. With the exception of sex, patient characteristics (sociodemographics, health, utilization) did not significantly predict voluntary disenrollment in any of these models. The gender effect had marginal significance in most cases (.05

Table 2 (column 2) shows the results of modeling voluntary disenrollment as a function of both relationship quality and structure-of-care together. In that multivariable model, the composite relationship quality factor (RQ) emerged as the leading predictor of voluntary disenrollment (OR=1.59; P <.001). This OR signifies that a standard deviation (SD) decline in relationship quality was associated with a 59% increase in the odds of voluntary disenrollment. The results indicate that after accounting for patients’ baseline characteristics (sociodemographic, health, and utilization) and the 4 structural features of care, patients with relationship quality scores in the 5th percentile in 1996 were 3 times more likely to voluntarily disenroll from their physician’s practice than those with 95th percentile relationship quality scores (37.8% vs 12.2%). The 2 measures of continuity also significantly predicted disenrollment in the multivariable model (visit-based continuity: OR=1.14, P=.03; relationship duration: OR=1.16, P=.01). Access to care predicted disenrollment with marginal significance (OR=1.14; P=.08), and integration did not significantly predict disenrollment (P=.59) in this model. None of the patient characteristics (sociodemograhics, health, utilization) significantly predicted disenrollment in the presence of these 5 quality-of-care measures.

Discussion

In our observational study of insured employed adults, 20% of the patients voluntarily left their primary care physician’s practice over a 3-year period. Another 5% left involuntarily, owing to factors that forced a change (eg, the physician moved, retired, or died). For the average full-time primary physician, this translates into approximately 400 patients voluntarily leaving the practice over a given 3-year period and another 100 leaving involuntarily. Rates of involuntary switching are almost certainly higher among physicians whose patients face more employer-imposed disruptions than occurred in our study population.

Our data indicate that the quality of the physician-patient relationship significantly predicts patients’ loyalty. With patient characteristics and structural features of care taken into account, those with the poorest-quality physician-patient relationships in 1996 were 3 times more likely to leave the physician’s practice over the ensuing 3 years than those with the highest-quality relationships.

Structural features of care also emerged as important determinants of patients’ disenrollment decisions. When considered independently of relationship quality, each of the 4 structural elements of care significantly predicted voluntary disenrollment. With relationship quality taken into account, continuity of care (both relationship duration and visit-based continuity) remained significant predictors of disenrollment, while access to care and integration of care did not. The results suggest that although these patients put a high priority on being given timely and convenient access to their physician’s office, the issue of who they are given access to and the quality of their connection with that clinician mattered more.

Our findings are consistent with those reported more than 2 decades ago by Kasteler and colleagues,¹⁶ who found both interpersonal quality of care and structural features of care to be significantly associated with voluntary physician switching in a cross-sectional study. Our study has the advantage of longitudinal data through which the sequencing of effects is clear. In addition, our study advances beyond earlier studies that evaluated a single patient-based measure of care in predicting disenrollment.^17,25 Marquis and coworkers¹⁷ showed that patients’ general satisfaction with their physician predicted disenrollment from the physician’s practice over the following year. Thom and colleagues²⁵ found that patients’ trust in their physicians significantly predicted disenrollment over the next 6 months. Our study includes measures of 8 characteristics that encompass the defining features of primary care as posited by the IOM1 and others,^26-29 with several features for which the relationship to disenrollment have not been previously studied. Our study contributes evidence concerning both the absolute and relative importance of interpersonal and structural features of care as predictors of patients’ loyalty to their primary care physician’s practice.

Limitations

Our study is limited to a population of adults in Massachusetts who were employed and insured at baseline. Rates of involuntary physician switching in this population were likely lower than would be observed in other employed populations (particularly in competitive health care markets) for the reasons mentioned (ie, benefit policies that minimized employer-imposed disruption of employees’ health care arrangements). However, the observed rates of voluntary physician switching and the predictors of voluntary disenrollment should not be affected. Those findings may be presumed to generalize more broadly.

A second limitation is the absence of information about salient health events that occurred between the baseline and follow-up phases of our study. For patients who incurred a serious episode of illness, information about the intervening health events and their experiences with their physician during that time might have enhanced our understanding of the factors that influenced their decisions about whether to remain in that physician’s practice.

Similarly, the study lacked detailed indicators of the technical quality of care provided and therefore could not assess the role that technical quality—and patients’ perceptions of it—play in shaping patients’ loyalty to their physician.

Finally, our study could not fully account for one potential source of involuntary disenrollment: patients leaving practices because the physicians no longer accepted their health plan. However, rates of physician turnover during the study period were no more than 5% in any of the health plans studied and were substantially lower in most.²¹ Thus, our findings are unlikely to have been substantively altered by a detailed accounting of this form of involuntary physician switching.

Conclusions

Previous empirical research has underscored the importance of physician-patient relationship quality by demonstrating its association with important outcomes, including adherence to medical advice,^19,30-32 satisfaction with care,^19,33-35 and litigation against physicians.^36-38 However, few studies have had the benefit of longitudinal data with which to verify the sequencing of effects between relationship quality and outcomes.

In our study the strength of physician-patient relationships in primary care—as indicated by patients’ trust in their physician, their assessments of how well the physician knows them, and the quality of communication and interpersonal treatment—was the leading predictor of patients’ loyalty to their primary physician’s practice. Continuity of care also significantly predicted voluntary disenrollment. The findings are noteworthy against a backdrop of health care delivery changes nationwide that many describe as threatening the therapeutic alliance between the physician and the patient.^29,39-42

The recent IOM report on the future of primary care called attention to the importance of the physician-patient relationship in primary care, asserting that primary care is predicated on sustained clinician-patient partnerships and on a whole-person orientation.1 In our study, longitudinal data demonstrate that the strength of connection between a patient and his or her primary care physician significantly predicts the likelihood of that patient remaining in that physician’s practice (vs voluntarily leaving) over the next several years. In an era marked by increasing pressure on clinicians and health care organizations to attend to such factors as market share, productivity, and efficiency, these findings point to a set of attributes that might otherwise be overlooked. They suggest that medical practices and health plans cannot afford to ignore that the essence of medical care delivery involves the interaction of one human being with another.

Acknowledgments

This research was supported by grant number R01 HS08841 from the Agency for Healthcare Research and Quality (formerly the Agency for Health Care Policy and Research) and by grant number 035321 from the Robert Wood Johnson Foundation. We are indebted to Dolores Mitchell, executive director of the Massachusetts Group Insurance Commission, whose commitment and participation have made this study possible. We also gratefully acknowledge Brian Clarridge, PhD, and his colleagues at The Center for Survey Research, University of Massachusetts, for their technical expertise and commitment to excellence in obtaining the data for our study. Finally, we acknowledge each of the health plans involved in the study, and particularly the health plan executives who served on our Advisory Committee and those who provided interviews, for generously giving their time, insights, and information that was critical to our research.