Abstract
To assess the effects of physician–patient racial concordance and continuity of care on hypertension outcomes, we described patterns of care for hypertension; we used cross-tabulations and repeated measures (generalized estimating equations) analyses with panel survey data from elderly persons interviewed and examined in 1987 and 1990. Continuity of care was associated with recognition of hypertension, receipt of medication, and lower incidence of undetected hypertension. Physician race had little effect, but continuity is important for successful management of hypertension in older persons.
Despite progress in hypertension management, African American persons1,2 have lower rates of recognition, treatment, and control of hypertension than do White persons.3,4 Elderly persons have similar hypertension treatment rates but poorer control than do younger persons. Demographic dissimilarities underlie doctor–patient communication difficulties affecting health outcomes,5–11 whereas patient–provider racial concordance correlates with patient participation in care, satisfaction, and treatment adherence.12–19 Stability in doctor–patient relationships correlates with patient satisfaction and access to care.1 This study assessed how physician–patient continuity and racial concordance5,20 affect hypertension diagnosis and medication use in White and African American elderly patients.
The Piedmont Health Survey of the Elderly conducted in-home interviews and recorded blood pressure readings in 4162 persons aged 65 years or older in 1986 to 1987 (approximately 80% response) and followed up 3536 surviving older persons in 1990.21,22 Our subsample (1834 African American individuals; 1533 White individuals) excluded respondents who lacked critical survey responses (n = 25) or who named unidentifiable, out-of-state, or nonphysician practitioners (n = 45) or non-White, non–African American physicians (n = 99).
Named physicians were matched to licensure files. Anonymous physicians’ race, age, gender, graduation year, and specialty were linked to the Piedmont Health Survey of the Elderly files that had respondents’ care site location, demographics, trichotomized self-reported health (“poor” or “fair” vs combined “excellent” and “good”), chronic illness indices23 (hypertension, diabetes, heart disease, stroke, cancer), and dichotomized Katz scale.24 Physician affiliation was (1) discontinuous (naming no physician at least once), (2) switching physicians (naming different physicians at each survey), or (3) continuous (naming same physician both times). A 4-valued racial concordance measure compared physician with patient race. Methods for measuring hypertension-related outcomes are described in Table 1.25
Descriptive comparisons used χ2 and t tests. For each repeated outcome, a multivariate linear model was fit with generalized estimating equations, allowing assessment of the effects of multiple predictors across time for each analysis.26,27 Initial analyses tested associations between outcomes and respondent–physician racial dyads and continuity of care; subsequent models controlled for respondent and physician characteristics. Analyses of 2-way interactions between care source, racial dyad, and continuity of care aimed to detect subgroup effects. Subject clustering within physicians was assessed by alternating logistic regression28 to detect patterns of physician clustering of repeated binary outcomes within subjects. Clustering within physicians showed weak or no statistical significance and was not reported.
We incorporated Piedmont Health Survey of the Elderly weights into multivariate analyses when possible, but weighting had to include respondents not meeting inclusion criteria. Some strata lacked variation in physician characteristics or had only 1 physician yielding apparent “missing” cases in analyses, affecting more than 31% of the baseline sample. Hence we report full final models run without survey weights; we adjusted for sample design; showed adjusted odds ratios, significance levels, and confidence intervals in a table; and used footnotes for significant covariates. Given numerous statistical tests, P<.01 was considered statistically significant, with .01<P<.05 considered a trend. We used SAS software (SAS Institute Inc, Cary, NC).
The cohort size declined mostly through mortality (Table 1). Surviving respondents lost spouses; had higher income, better health status, and higher illness scores; and had more nursing home use and functional declines than they did at baseline. More older persons reported regular care sources in 1990; fewer named local private physicians who were younger and trained more recently than 1987. More than 10% named no physician at either survey; 24% named none at 1 survey. Twenty-five percent switched physicians; 38% had the same physician each time. African American physician–patient dyads decreased over time, replaced by White physician–African American patient dyads. Severe hypertension was comparable at each survey.
No racial differences were evident in age, gender, employment, or disease severity. Fewer African American individuals were married, and, as a group, they had less education, income, and private insurance and more Medicaid. Self-reported health improved, whereas impairment increased for both groups, but racial disparities persisted. Racial groups had parity in “usual source of care” in 1990, but White patients were more likely seeing nearby private physicians; public sources cared for 1 in 3 African American patients and only 1 in 10 White patients. More African American individuals than White individuals lacked regular physicians at both surveys (14.9% vs 5.5%) or named a physician only once (27.5% vs 20.3%). Conversely, more White patients than African American patients had the same physician across surveys (46.7% vs 30.4%). More African American persons reported that a physician had told them they had high blood pressure. Adverse racial differences were largest for severe hypertension, widening between surveys.
Table 2 shows no significant effects on measured hypertension. There was a tendency for those with discontinuity in care to have had undetected hypertension more often than did those with continuity of care. Compared with White patients with White physicians, African American patients had a lower incidence of undetected hypertension (and of severe hypertension) regardless of physician race. Elderly persons with discontinuous care were more likely to have undetected severe hypertension, but those naming generalist physicians also may have been at risk for having severe hypertension previously undetected. Those who lacked or had switched physicians received fewer hypertension diagnoses and, if diagnosed, took fewer medications compared with those keeping the same physician.
Interaction tests suggested that African American respondents who switched physicians may have been more likely to be taking hypertension medications if their new physician was White (P< .02). African American patients whose usual care sources were public clinics and who had African American physicians may have been more likely to have been taking hypertension medications than were African American patients using White physicians or private practitioners (P< .03). Those experiencing discontinuity in physician care and whose usual care sources were public clinics were more likely to have been taking medication (P< .001) than were those who had discontinuous care but from a private practitioner.
Unlike cross-sectional, retrospectively self-reported “usual person and place” surveys, we measured longitudinal patient–physician relationships with 2 temporally separated respondent reports, minimizing error in physician characteristics by combining survey and license data. Study limitations include small numbers in one southern state, which omitted non–African American, non-White physicians and patients. Cumulative reduction in cases for multivariate analyses came from a few missing values in many predictors.
Consistent with other chronic disease studies,29 continuity of care entailed better outcomes. Ongoing physician affiliation improved hypertension detection and medication use once diagnosed. Rates of detection in individuals changing physicians sometimes were midway between those without physicians and those keeping the same physician. African American individuals’ elevated hypertension diagnosis risk was unaffected by physicians’ race, suggesting widespread awareness of African American persons’ worse cardiovascular disease prognoses.4,30 African American patients had a lower risk of having undetected severe (stage 2) hypertension, but elderly patients lacking physicians had a higher hypertension risk. Patient–physician racial concordance effects seemed contextually conditioned (e.g., African American patients using public sources of care may use medication more often if their physician is African American, whereas African American patients who switched physicians may use medication more often if their new physician is White).
Regular access to a usual care source and sustained affiliation with a physician can improve the management of hypertension in older African American and White patients. Because African American Medicare beneficiaries are cared for by a subset of African American physicians often in challenging practice situations,31 better understanding of hypertension care may require more longitudinal study of physician availability and the dynamics of physician selection in addition to racial concordance and continuity of care.
Note. NS = not significant; ADL = activity of daily living; NA = not applicable. aP values from χ2 and t tests for African American and White comparisons for 1987. bP values from χ2 and t tests for African American and White comparisons for 1990. cFamily practice, general practice, internal medicine, geriatrics dWhen these data were collected in 1986 and 1990, the prevailing guidelines, the 1984 Joint National Committee on Detection, Evaluation, and Treatment of High Blood Pressure, recommended a modest treatment regimen for the elderly. This recommendation was that among those with existing systolic blood pressure higher than 160 mm Hg, drug treatment, even in the presence of nonpharmacological therapy, should be considered on an individual basis. Thus, although the clinical guidelines were less aggressive during the time of the data collection compared with the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC-7) recommendations, disparities in care among racial groups would have the same implications in both periods. Blood pressure, as measured in both 1986 and 1990, was the average of 2 consecutive readings but was collapsed into 4 ordered JNC-7 categories: (1) normal, (2) prehypertensive, (3) mild (stage 1), and (4) severe (stage 2). Hypertension was categorized as stage 1 (systolic = 140–159 mm or diastolic = 90–99 mm) or stage 2 (systolic ≥ 160 mm or diastolic ≥ 100 mm). All other readings were categorized as either prehypertensive (systolic = 120–139 mm and diastolic = 80–89 mm) or “normal” if readings were below those levels. e χ2 test for stage 2 hypertension versus all other states combined. f“Yes” and “suspect or possible” answers to the question “Has a doctor ever told you that you have high blood pressure?” at either survey were coded as self-reported hypertension and prompted inquiry about the subject’s taking blood pressure medication. Undetected hypertension was ascribed to respondents whose field readings were consistent with hypertension but who did not report at either survey that a physician had said that they had high blood pressure. The subset with high enough readings were considered to have undetected stage 2 (i.e., severe) hypertension. Note. OR = odds ratio; CI = confidence interval. aFor racial dyad, the omitted reference category for this 4-valued variable is “White patient with White physician.” The other 3 values are shown in column headings in the table. bFor continuity of care, the omitted reference category is same physician named in 1987 and 1990 surveys (i.e., continuous care). The other 2 values are shown in column headings in the table. cSignificant covariates predicting measured hypertension include the passing of time (i.e., second survey; OR = 1.06; P < .001); education (OR = 1.04/y; P < .001); being currently employed (OR = 1.31; P < .05); receiving Medicaid (OR = 1.27; P < .05); and having a stroke (OR = 0.66; P < .01). dSignificant covariates predicting undetected hypertension include being male (OR = 2.04; P < .001); being older (OR = 1.03/y; P < .01); claiming fair self-rated health (OR = 0.73; P < .05); and not having heart disease (OR = 0.65; P < .001), stroke (OR = 0.48; P < .001), or cancer (OR = 0.60; P < .001). eSignificant covariates predicting undetected severe hypertension include being male (OR = 2.89; P < .001); being older (OR = 1.04/y; P < .05); having fair (OR = 0.60; P < .05) or poor (OR = 0.32; P < .01) self-reported health; having stroke history (OR = 0.37; P < .01); and receiving care by a generalist physician (OR = 0.55; P < .05). fSignificant covariates predicting self-reported hypertension include the passing of time (i.e., second survey; OR = 1.04; P < .001); being male (OR = 0.44; P < .001); being younger (OR = 0.97/y; P < .001); having elevated blood pressure measurement (OR = 3.96; P < .001); having heart problems (OR = 1.62; P < .001) or stroke (OR = 2.36; P < .001); and having fair (OR = 1.50; P < .001) or poor (OR = 1.44; P < .05) health status. g Significant covariates predicting use of hypertension medication include being male (OR = 0.70; P < .05) and having income less than $1000 per year (OR = 2.45; P < .01). *P < .001; **P < .01; ***P < .05.
Wave 1 (n = 4136) Wave 2 (n = 3536) African American (n = 2261) White (n = 1875) African American (n = 1943) White (n = 1593) Probability 1987a Probability 1990b Demographic characteristics Male 34.9 35.0 33.4 33.8 NS NS Age, y, mean 73.6 73.5 76.4 76.2 NS NS Years of education, mean 7.3 10.0 7.4 10.1 .0001 .0001 Married 33.8 43.7 31.1 40.0 .0001 .0001 Currently working 11.5 12.2 8.7 9.3 NS NS Income categories, $ 0–3999 30.6 11.1 19.5 7.3 .0001 .0001 4000–6999 43.4 29.5 48.7 26.7 7000–14 999 18.6 28.3 22.6 30.2 ≥ 15 000 7.4 31.2 9.2 35.9 Medicaid insurance 11.0 3.3 18.1 6.3 .0001 .0001 Medi-gap insurance 32.7 73.7 31.0 66.7 .0001 .0001 Resided in rural area 56.1 46.6 57.4 46.8 .0001 .0001 Health and functional status Self-reported health status Excellent or good 48.6 58.3 52.4 60.4 .0001 .0001 Fair 36.3 28 36.1 29.4 Poor 15.1 13.7 11.5 10.2 Severity of illness categories Good 23.0 24.1 30.8 30.5 NS NS Fair 51.5 51.2 43.7 42.3 Poor 25.5 24.8 25.5 27.3 ≥ 1 ADL limitation 13.5 10.7 21.9 18.7 .0062 .0207 Diagnosed health condition Heart condition 13.1 17.3 15.8 18.6 .0001 .0001 Diabetes 24.2 15.3 26.0 16.9 .0001 .0001 Stroke 9.6 8.1 10.4 9.7 .0245 .0675 Cancer 7.5 16.8 8.1 19.2 .0001 .0001 Use of health services Ever in a nursing home 1.1 2.9 5.5 7.8 .0001 .0064 Lived in same county where care provided 56.4 75.2 48.1 63.7 .0001 .0001 No usual source for care 5.3 4.0 3.2 2.6 .0394 NS Received care in public clinic or hospital or emergency department 34.9 9.3 34.7 10.4 .0001 .0001 Received care in private office or hospital 59.6 86.4 53.6 77.7 .0001 .0001 Physician characteristics Male 88.5 94.6 87.9 94.0 .0001 .0001 Age, y, mean 49.9 52.7 47.3 48.9 .0001 .0015 ≥ 65 13.5 12.9 9.6 6.2 36–64 73.6 82.1 71.5 82.3 ≤ 35 12.9 15.0 18.9 11.5 Years since medical school graduation, mean 22.7 26.9 19.8 22.7 .0001 .0001 Generalistc 63.3 72.5 53.8 65.1 .0001 .0001 Continuity of care No physician in 1987 or 1990 NA NA 14.9 5.5 NA .0001 Physician in 1987 or 1990 but not both NA NA 27.5 20.3 NA .0001 Same physician in 1987 and 1990 NA NA 30.4 46.7 NA .0001 Dependent variables Measured blood pressured Normal 8.6 9.2 10.7 10.3 Prehypertensive 35.0 37.0 38.2 42.3 Stage 1 34.4 34.3 30.1 30.9 Stage 2 (severe)e 22.0 19.6 21.0 16.6 .0605 .0023 Told about high blood pressure by physicianf 63.2 50.4 63.7 53.5 .0001 .0001 Taking high blood pressure medicationf 80.9 82.9 80.2 76.0 NS .0312 Racial Dyada African American Physician White Physician Continuity of Careb Regression Model: Dependent Variable No. African American Respondent OR (95% CI) White Respondent OR (95% CI) African American Respondent OR (95% CI) No Named Physician: 1987, 1990, or Both OR (95% CI) Different Physician in 1987 and 1990 OR (95% CI) Ordinal logistic regression: measured hypertension levelsc 2075 0.97 (0.80, 1.19) 1.14 (0.52, 2.48) 0.89 (0.74, 1.07) 1.06 (0.84, 1.33) 1.00 (0.84, 1.20) Binary logistic regression: undetected hypertensiond 1332 0.43* (0.30, 0.60) 0.47 (0.14, 1.63) 0.49* (0.36, 0.68) 1.47*** (1.02, 2.13) 1.31 (0.98, 1.76) Binary logistic regression: undetected severe hypertensione 596 0.28* (0.14, 0.54) 0.35 (0.04, 3.50) 0.44** (0.24, 0.78) 2.46** (1.30, 4.66) 1.22 (0.73, 2.04) Binary logistic regression: ever told of high blood pressuref 2016 1.87* (1.43, 2.44) 1.03 (0.42, 2.50) 2.00* (1.56, 2.56) 0.71* (0.53, 0.95) 0.74* (0.59, 0.93) Binary logistic regression: currently taking hypertension medicationg 1999 1.41 (0.95, 2.07) 1.53 (0.46, 5.05) 1.05 (0.74, 1.49) 0.44* (0.29, 0.66) 0.64* (0.47, 0.88)
Original data collection was sponsored by the Established Populations for Epidemiologic Studies of the Elderly, conducted by the Duke University Center for Aging and Human Development (contract no. N01-AG-12102 and grant no. R01 AG 12765, National Institute on Aging). Analyses were supported by the Agency for Health Care Research and Quality’s Center of Excellence on Overcoming Racial Health Disparities at the Cecil G. Sheps Center for Health Services Research (PO1 HS10861). This study was also supported, in part, by the National Center on Minority Health and Health Disparities (1 P60 MD00239 and 1 R24 MD000167) and the Agency for Health Care Research and Quality (1 R24 HS13353).
Physician data were extracted by the North Carolina Health Professions Data System with the permission of the North Carolina Medical Board.
The authors express their appreciation to Carol Porter for programming assistance and to Gerda Fillen-baum and Donald Pathman, who provided important information for preparation of this brief. Special thanks to Larry Logan and Donna Curasi for editorial assistance.
Human Participant Protection The University of North Carolina Committee on the Protection of Human Subjects approved the research.
