Physician, Heal Thyself?Regular Source of Care and Use of Preventive Health Services Among Physicians

Background  Little is known about the regular source of care (RSOC) among physicians, a group whose self-care may reflect the attitudes and recommendations they convey to their patients.

Methods  We performed a cohort study of physicians who graduated from the Johns Hopkins School of Medicine from 1948 through 1964 to identify predictors of not having an RSOC, and to determine whether not having an RSOC was associated with subsequent receipt of preventive services. The RSOC was assessed in a 1991 survey; use of cancer screening tests and the influenza vaccine was assessed in 1997.

Results  The response rate in 1991 was 77% (915 respondents); 35% (312) had no RSOC. Internists (odds ratio [OR], 3.26; 95% confidence interval [CI], 1.58-6.74), surgeons (OR, 2.42; 95% CI, 1.17-5.02), and pathologists (OR, 5.46; 95% CI, 2.09-14.29) were significantly more likely to not have an RSOC than pediatricians. Not having an RSOC was inversely related to the belief that health is determined by health professionals (OR, 0.45; 95% CI, 0.29-0.68) and directly related to the belief that chance (OR, 1.90; 95% CI, 1.28-2.82) determines health. Not having an RSOC in 1991 predicted not being screened for breast, colon, and prostate cancer, as well as not receiving an influenza vaccine at 6 years of follow-up.

Conclusions  A large percentage of physicians in our sample had no RSOC, and this was associated with both medical specialty and beliefs about control of health outcomes. Not having an RSOC was significantly associated with failure to use preventive services several years later.

APPROXIMATELY 14% to 18% of Americans do not have a regular source of care (RSOC), a characteristic that has been associated with decreased access to health services.^1-3 Although lack of insurance is associated with not having an RSOC, 2 large studies have suggested that the most common reason for lacking an RSOC may be the belief that it is unnecessary.^2,3 One frequently used instrument for assessing health beliefs is the Multidimensional Health Locus of Control (MHLC) scale.^4,5 The MHLC measures the extent to which individuals believe that health-related events are under their own control (internal), controlled by physicians and other health professionals (powerful others), or are a matter of fate and luck (chance).^4,5 The association between MHLC and use of an RSOC has not been previously assessed.

Little is known about the use of an RSOC among physicians, despite the fact that many patients look to their physician for advice about medical care. Prior work has suggested that physicians' own health habits influence the preventive health counseling they provide to their patients.^6-9 Other studies have suggested that physician specialty may influence the approach to patient care and use of preventive services.^7,9 Another reason that physicians are an important group to study is that they have access to excellent health care as well as a higher than average educational and socioeconomic status. Thus, studies of health-related beliefs and behavior among physicians are less likely to be confounded by these factors and might elucidate other risk factors associated with having no RSOC.

We therefore studied a cohort of physicians to answer the following research questions: (1) What proportion of physicians do not have an RSOC? (2) Is health locus of control or medical specialty associated with not having an RSOC? (3) Does not having an RSOC predict lower utilization of preventive health services among physicians?

This was a prospective cohort study using data derived from the Johns Hopkins Precursors Study. Initiated by Caroline Bedell Thomas, MD, in 1946, this cohort includes 1337 members of the Johns Hopkins University School of Medicine graduating classes of 1948 through 1964. Participants are asked to complete mailed surveys annually, and the response rate averages 86% over every 5-year period. Our sample included the 915 respondents (77% of those eligible) to the 1991 questionnaire about having an RSOC. There was no significant difference between these respondents and nonrespondents with regard to sex or medical specialty. The mean age of both the respondent and nonrespondent groups was 61 years. About 78% of the respondents to the 1991 survey also returned the 1995 and 1997 surveys.

The 1991 questionnaire included a multiple-choice question about the source of medical care. Possible responses were the following: (1) no RSOC; (2) self-treated; (3) physician in own group or clinic organization; (4) physician independent of group practice; and (5) "other" source of care. We defined individuals not having an RSOC as those who indicated either no RSOC or self-treated.

As described elsewhere, physicians' specialties were classified as internal medicine, surgery (including obstetrics and gynecology), pediatrics, psychiatry, or pathology.¹⁰ Subspecialists were classified according to their primary fields; physicians who could not be classified into any of the above categories were classified as other.

The 18-item MHLC scale was administered as part of the 1995 survey.⁴ Responses were on a 6-point Likert scale and ranged from strongly disagree (1) to strongly agree (6). Three domains (internal, chance, and powerful others) are defined, consisting of 6 items each. Individual domain scores range from a minimum of 6 to a maximum of 36.

The 3 locus of control domains were scored as dichotomous variables. For each domain, individuals whose scores were in the upper quartile were categorized as being high prior to examining the data. Responses in the lower 3 quartiles were the comparison group. The domains were not mutually exclusive; a respondent who scored in the upper quartile for the internal locus of control, for example, might also score highly on the chance domain.

To control for the presence and severity of various comorbid illnesses, the Charlson comorbidity index was used for risk adjustment.^11,12 This is a validated method of using data obtained directly from patient records to predict short- and long-term mortality. Patients without comorbidities were assigned a score of 0, and risk of death increased in a stepwise manner, with scores above 5 having the highest risk. We used diagnoses made in 1988 or earlier for constructing the comorbidity index because this was the most recent year that comprehensive data on all comorbidities was available for the entire cohort.

We also assessed covariates that might be related to not having an RSOC or use of preventive health services. Age in 1991 was categorized into quartiles. Parental incidence of cancer was identified beginning in medical school and ascertained throughout the follow-up period. A history of cancer in either parent or in both parents was also used as a covariate in the analysis. Health status was ascertained using the 36-item short-form health survey (SF-36%) in 1992.¹³ Summary scores for mental and physical health were calculated using previously described techniques,¹³ and the resulting scores were categorized as quartiles.

Use of cancer screening tests and influenza vaccination was assessed in 1997. Respondents were asked whether they had received various tests and whether they were performed to investigate specific signs or symptoms or for asymptomatic screening. We excluded people who were tested to evaluate a specific sign or symptom, as well as those with a history of the relevant cancer for each screening test (ie, women with a history of breast cancer were excluded from the mammography analysis).

We calculated the proportion of eligible respondents who received each test within the recommended time. In general, we followed the guidelines recommended by the United States Preventive Services Task Force (USPSTF).¹⁴ For example, screening for colon cancer was defined as either fecal occult blood testing within the prior 2 years or colonoscopy or barium enema within the prior 5 years.

Screening for breast cancer was defined as mammography within the prior 2 years. All female respondents, who ranged in age from 52 to 71 years, were included in this portion of the analysis. Although definitive evidence for the efficacy of the serum prostate-specific antigen (PSA) is lacking, prostate cancer screening was defined as having had a PSA test within the prior 2 years. Digital rectal examination can be used as a screening test for prostate cancer, colorectal cancer, or both. Because we could not determine the screening indication for rectal examinations, we did not include this procedure in our analysis of prostate cancer screening. All respondents were deemed eligible for influenza vaccination; recipients were defined as those who received the vaccine after September of the prior year.

For each cancer screening test, "screened" patients were defined as those who were tested within the appropriate time, as listed above. The number of screened patients was divided by the number of patients eligible for screening to calculate the screening rate.

The dependent variable for the first portion of the analysis was not having an RSOC. Bivariate associations with not having an RSOC were assessed using the χ² test. Logistic regression was used to determine whether medical specialty or locus of control was significantly related to not having an RSOC, independent of age, functional status, Charlson comorbidity index, and physical and mental health summary scores derived from the SF-36.

The dependent variable for the second portion of the analysis was receipt of preventive services in 1997. A χ² test was used to determine whether having an RSOC was associated with receiving influenza vaccine or screening for prostate, colon, or breast cancer, when applicable. Logistic regression was then used to adjust for potential confounders such as age, parental cancer history, Charlson comorbidity index, medical specialty, and functional status.

Overall, 312 respondents (34%; Table 1) reported that they had no RSOC in 1991; 252 (28%) had no regular medical care, while 60 (7%) treated themselves. Approximately 43% of our sample had an independent physician as their RSOC, while 18% saw someone in their own group; 5% indicated other. The mean age of the respondents in 1991 was 61 years; 8% were women. The most common specialties were surgery (including obstetrics and gynecology and surgical subspecialties; 36% of respondents%) and internal medicine (35%). The sample contained few people with comorbidities. The Charlson comorbidity index scores ranged from 0 to 4, and most had scores of either 0 (n = 589; 64%) or 1 (n = 252; 28%).

Neither age nor sex was significantly associated with use of an RSOC (Table 1). However, the likelihood of not having an RSOC varied markedly by medical specialty, ranging from 21% among psychiatrists to 46% among pathologists. Respondents who scored highly on the "powerful others" domain, indicating a strong belief that their physicians and other health professionals had a significant influence on their health status, were much less likely to not have an RSOC. Approximately 21% of the 204 patients who scored in the highest quartile in this domain lacked an RSOC, while 38% of the patients who scored in the lower 3 quartiles lacked an RSOC.

In contrast, respondents who scored highly in either the "chance" or "internal" domains were significantly more likely to not have an RSOC (43% and 40%, respectively) than their counterparts who scored lower (31% and 32%, respectively; Table 1). Patients with fewer comorbidities were more likely to not have an RSOC (P = .03). However, there was no significant relationship between RSOC and either the physical (P = .09) or mental function (P = .44) summary scores of the SF-36.

The results of the multivariate analysis are outlined in Table 2. The relationship between medical specialty and not having an RSOC persisted after adjusting for locus of control and comorbidity index. Compared with pediatricians, pathologists were more than 5 times as likely to not have an RSOC (odds ratio [OR], 5.46; 95% confidence interval [CI], 2.09-14.29). Internists, surgeons, and physicians practicing other specialties were also significantly more likely than pediatricians to not have an RSOC (ORs, 3.26, 2.42, and 3.00, respectively). Psychiatrists were not significantly different from pediatricians with regard to having an RSOC.

Locus of control continued to be associated with not having an RSOC after adjusting for covariates (Table 2). Respondents who scored in the highest quartile in the chance domain, indicating a strong belief that chance or fate determines their health outcomes, were about twice as likely to not have an RSOC as their counterparts who scored in the lower 3 quartiles (OR, 1.90; 95% CI, 1.28-2.82). People who scored highly in the internal domain, indicating a strong belief that health-related outcomes are under their own control, were also significantly more likely to lack an RSOC. Conversely, those in the highest quartile for powerful others (indicating a strong belief in health professionals) were about half as likely to not have an RSOC as their counterparts in the lower 3 quartiles.

The presence of chronic disease was also significant, as respondents with comorbidity index scores of 2 or higher were significantly less likely to have no RSOC (OR, 0.40; 95% CI, 0.18-0.90) than those with a score of 0. Age, sex, and physical and mental functional status were not significantly associated with RSOC in multivariate analysis.

Table 3 lists the use of preventive services based on responses to the 1997 survey and stratified by RSOC in 1991. Overall, 66% of the sample had received some form of colorectal cancer screening (fecal occult blood testing, endoscopy, or barium enema%) as recommended by the USPSTF guidelines. The use of mammography and PSA among eligible candidates was 74% and 76%, respectively. Approximately 71% of the sample had received an influenza vaccination.

After adjusting for age, sex, comorbidity score, SF-36 physical and mental health summary scores, medical specialty, and parental cancer history, there was still a significant relationship between not having an RSOC and receipt of these preventive services. Respondents without an RSOC in 1991 were significantly less likely to have undergone screening for colon cancer (OR, 0.23; 95% CI, 0.15-0.38), breast cancer with mammography (OR, 0.16; 95% CI, 0.03-0.84), and prostate cancer with PSA (OR, 0.28; 95% CI, 0.17-0.46%) than their counterparts with an RSOC. They were also less likely to have received an influenza vaccination (OR, 0.44; 95% CI, 0.30-0.64).

Studies of the general population have documented that having an RSOC is strongly associated with higher levels of patient satisfaction and use of preventive services, and lower use of emergency department visits and hospital length of stay.^15-18 Accordingly, many states have attempted to provide Medicaid patients with an RSOC with the intention of improving quality and decreasing costs.¹⁹ Despite the well-described benefits, approximately one third of the physicians in our sample lacked an RSOC. This is a far higher proportion than has been estimated for the general population, where previous studies have reported that approximately 15% lack an RSOC; the Healthy People 2000 goal was for only 5% of the population to lack an RSOC.^1-3

Understanding physician use of an RSOC is particularly important because previous work has demonstrated that physicians' own health habits and attitudes influence the counseling and care they provide to patients.²⁰ Physicians with poor health practices regarding smoking, alcohol intake, exercise, or seat-belt use are less likely to counsel patients about those habits.^7,9,21 Our results are consistent with earlier studies reporting that 35% to 56% of physicians did not have a personal physician.^22-24 However, these cross-sectional studies did not investigate determinants of lacking an RSOC or whether there was an adverse effect on health care.

Our findings support the hypothesis that people's beliefs about factors influencing their health play a major role in their use of an RSOC.^5,25 Respondents who were more fatalistic about their health, scoring in the highest quartile of the chance locus of control domain, were almost twice as likely to lack an RSOC. In contrast, those who expressed faith in the ability of physicians and other authority figures to influence their health outcomes were far less likely to lack an RSOC. This is consistent with data from the 1995 Behavioral Risk Factor Surveillance System,² a random telephone survey of adults in the United States. In this sample, the most frequently reported reason for not having an RSOC was the feeling that it was unnecessary. The Robert Wood Johnson Foundation's 1986 access to health care telephone survey³ yielded similar findings: of the 16% of the respondents who lacked an RSOC, the most common reason cited was not wanting one. Future work should explore strategies for educating patients about the importance of regular medical care.

Medical specialty also seemed to have a significant effect on the use of an RSOC in our sample: internists, pathologists, surgeons, and physicians in other specialties were significantly more likely to have no RSOC. This may reflect differences in training or beliefs about the importance of preventive services, although these findings were independent of difference in beliefs concerning control of health outcomes. Physicians who were not within these specialty groups also had an elevated risk of having no RSOC. These results are difficult to interpret because this "other specialty" group was very heterogeneous in their professional activities. Although internists in our sample were somewhat more likely to report self-care than were physicians in other specialties (data not shown), they were still significantly more likely to have no RSOC than pediatricians, even when respondents who reported self-care were excluded from the analysis. Other factors were related to having an RSOC in our sample. Healthier patients, defined as having less comorbidity, were less likely to have an RSOC. This was expected because patients with a greater need for medical services are more likely to encounter a primary care provider and initiate a longitudinal relationship.

Our data demonstrate that not having an RSOC was significantly associated with not receiving indicated preventive services years later. These findings were particularly surprising in a sample of physicians, who are not only aware of the existence, purpose, and probably the efficacy of various preventive services, but also have the ability to access these health services without having a personal physician. It is likely that the association between RSOC and preventive services is even stronger in nonphysicians, a group who does not enjoy such unfettered access. Thus, screening for not having an RSOC may allow identification of patients at risk for not receiving preventive services, independent of insurance status, educational level, medical knowledge, age, comorbidities, and functional status.

The association between RSOC and use of preventive services does not necessarily imply causation. Although studies have suggested that patients with an RSOC have a higher use of preventive services,^1,2 it is unclear whether this is due directly to the influence of an RSOC or whether those with an RSOC have prevention-prone attitudes and request more services. Our data suggest that people with an RSOC do have different attitudes toward health than those without—at least with regard to locus of control. However, even when we adjusted for locus of control beliefs, having an RSOC was still significantly associated with subsequent use of preventive services (data not shown). This suggests that the presence of an RSOC may contribute to receiving health services independently of these beliefs.

While studies have investigated physicians' personal use of preventive services,^3,4 we were able to take advantage of additional clinical information about relevant symptoms, signs, and cancer history to exclude tests that were not obtained for asymptomatic screening. Use of mammography and influenza vaccine was higher among respondents in our sample (74% and 71%, respectively) than the goals for Healthy People 2000⁵ (60% each for mammography and influenza vaccine of eligible individuals). However, since studies have shown that physicians' own screening preferences influence the screening advice they offer patients, this low degree of use among physicians is still suboptimal.⁶ Physician organizations should encourage physicians to have an RSOC, and the importance of obtaining appropriate health care should also be stressed to medical students.

Serum PSA measurement was the most frequently used screening test in our sample, even though it is not recommended by the USPSTF.¹⁴ Respondents were significantly less likely to have been screened for colorectal cancer, a practice that has been recommended by the USPSTF.¹⁴ This discrepancy may reflect the fact that PSA measurement is easier to undergo than colorectal cancer screening.

This study has several limitations. First, physician behavior may not correlate with behavior in the general population, in that physicians generally enjoy a higher socioeconomic and health status. Patients in our sample were relatively healthy, with little comorbidity. Second, since this study includes only individuals who graduated from a single medical school with a strong academic orientation, the results may not be applicable to all physicians. Use of screening tests was self-reported, and respondents may have underestimated the time since the most recent screening test. However, it is unlikely that this recall would be different across the subgroups tested. Third, relatively few women were included, limiting our power to detect sex differences. Fourth, the association between RSOC and use of preventive services may not be causal. However, we did have 6-year follow-up data after ascertainment of RSOC, and we were also able to account for locus of control. Future studies should explore the relation between other health-related attitudes and use of RSOC and preventive services. Finally, some physicians may have acquired an RSOC between the initial assessment in 1991 and the 1997 follow-up questionnaire. However, this misclassification would have biased our results toward the null, resulting in underestimating the association between RSOC and use of preventive services.

In conclusion, many physicians in our sample had no RSOC, and this adversely affected their use of preventive services. This underscores the importance of continuity of care, which is increasingly threatened by frequent employer-driven changes in insurance plans. More importantly, our data reemphasize that potential access and realized access are not synonymous; even in a sample of people who probably could have accessed health care if they had desired, use of an RSOC and preventive services was suboptimal. Health-related beliefs such as locus of control can have a profound effect on how people relate to the health care system.

Accepted for publication June 14, 2000.

This work was supported by National Institutes of Health grant AG01760, Bethesda, Md, and the Robert Wood Johnson Clinical Scholars Program, Johns Hopkins School of Medicine, Baltimore, Md.

Reprints: Michael J. Klag, MD, MPH, 2024 E Monument St, Baltimore, MD 21287 (e-mail: mklag@welchlink.welch.jhu.edu).