|Home | About | Journals | Submit | Contact Us | Français|
To determine whether an established patient satisfaction scale commonly used in the primary care setting is sufficiently sensitive to identify racial/ethnic differences in satisfaction that may exist; to compare a composite indicator of overall patient satisfaction with a 4-item satisfaction scale that measures only the quality of the direct physician-patient interaction.
Real-time survey of patients during a primary care office visit.
Private medical offices in a generally affluent area of northern California.
Five hundred thirty-seven primary care patients selected at random from those entering a medical office.
Patient satisfaction using 1) a composite, 9-item satisfaction scale (VSQ-9); and 2) a 4-item subset of that scale that measures only satisfaction with direct physician care.
The 9-item, composite scale identified no significant difference in patient satisfaction between white and nonwhite patients, after controlling for patient demographics and other aspects of the visit. The 4-item, physician-specific scale indicated that nonwhite patients were less satisfied than white patients with their direct interaction with the physicians included in the study (P ≤ .01).
Measurements of patient satisfaction that use multi-item, composite indicators should also include focused comparisons of satisfaction directly with the care provided by the physician. In measurements of patient satisfaction, patient race/ethnicity should be included as an explanatory variable. The results also confirm earlier findings that factors external to the direct physician-patient interaction can have substantial effects on patients’ perceptions of that interaction.
In its 2002 report Unequal Treatment, the Institute of Medicine concluded that “racial and ethnic minority patients are found to receive a lower quality and intensity of healthcare and diagnostic services across a wide range of procedures and disease areas” (p. 61).1 The report specifically included the quality of primary care services within its analysis. This conclusion echoes earlier summaries of patient satisfaction studies.2,3 A number of more recent studies also identify racial and ethnic differences in patients’ satisfaction with the quality of care.4–7
As Unequal Treatment suggests, disparities in the experience of the clinical encounter can be influenced by the attitudes, expectations, and behavior of both patients and physicians. Racial discordance (i.e., the situation in which the doctor and patient are from different racial or ethnic groups) can create a “social distance” between doctor and patient that reduces patients’ perceptions of the quality of care and impairs their level of trust in the medical care system.8
While patient distrust may contribute to the lower levels of satisfaction reported among African American and other nonwhite patients,9–11 recent evidence also suggests that physicians may approach the clinical encounter with preconceived and sometimes stereotypical ideas and attitudes that may impact their care.12,13 For example, a study of 193 physicians suggested that those physicians perceived their black patients in more negative terms than their white patients, even after controlling for patients’ income and education levels.14
If the goal of eliminating racial and ethnic disparities in health care quality is to be attained, a crucial first step is to have quality assessment instruments that will identify such differences when they exist. The concept of “quality” in medical care has at least three aspects: structure, process, and outcome.15,16 Quality instruments that measure patients’ perceptions of the care process have become widely used.17 Commonly used scales that measure patients’ perceptions of quality include the Consumer Assessment of Health Plans Study (CAHPS),18 administered as part of the Health Plan Employer Data and Information Set (HEDIS),19 and the 9-Item Visit Satisfaction Questionnaire (VSQ-9), developed for the Medical Outcomes Study (MOS).20,21
While CAHPS is often used to measure the quality of care received from a health plan,22 the VSQ-9 provides a measurement specifically of a patient's perception of the quality of a single office visit with a physician or other provider.23,24 The reliability and relative ease of administration of the VSQ-9 has led to it being adopted by a number of medical groups and health services researchers to measure patient satisfaction with care.25,26 The American Medical Group Association (AMGA), a national association of large medical groups that, in aggregate, provide care to more than 50 million patients, has adapted the VSQ-9 as its recommended patient satisfaction instrument.27
As the VSQ-9 patient satisfaction scale is now widely used to measure this aspect of health care quality, it is essential to determine whether that scale, when used to calculate a composite score, is sufficiently sensitive in measuring racial and ethnic differences. An analysis of data from the original MOS identified racial and ethnic differences in overall patient satisfaction using a composite score from the VSQ-9 that combined different aspects of the clinical encounter.28 However, our previous report of a study of 291 patients from a single, large (>150-physician) medical group in California, also using the composite score, did not identify racial/ethnic differences in satisfaction.29 It is the purpose of this study to test whether that absence of an identified difference in satisfaction between white and nonwhite patients might be due to the manner in which the VSQ-9 score was calculated. We addressed this question by gathering data on an additional 246 patients from a variety of medical group settings. Using the combined data from a total of 537 patients, we first test for differences in satisfaction levels between white and nonwhite patients using the composite scale that includes questions pertaining to access to care as well as the quality of the physician/patient interaction. We then reanalyze the data, separating out only those questions from the VSQ-9 that pertain to the direct physician/patient interaction.
This study is an extension of an earlier report of 291 primary care patient visits at a single, large, multispecialty medical group that participates in a variety of managed care plans.29 The study combines data from those visits with 246 additional visits at other medical groups in the same geographic area. In addition to the large group originally studied, we approached 10 small primary care groups, ranging in size from 2 to 8 physicians. Seven of the 10 groups agreed to participate in the study; all physicians in the participating groups agreed to have their patients included in the study. All were private practice groups with no direct academic affiliation. As obstetric/gynecology visits were included in the original sample, one of the new groups included was a strictly ObGyn group with 5 physicians. The other 7 groups were either general internal medicine or family practice. Of the approximately 65 physicians included in this study, 4 were Asian American and the remainder were white. There were no African-American physicians either on the primary care staff of the large medical group or in primary care practice in the office clusters from which we selected our small practices. The physicians were approximately evenly split between men and women.
At each medical group a trained surveyor stood at the entrance to the office and, using a randomization protocol described elsewhere,30 approached an entering patient. After obtaining consent for participation (patient participation rate was 82%), the surveyor accompanied the patient through all aspects of the visit except entering the examination area, asking survey questions as described below. If the patient to be seen was under 18 years of age, parental consent was obtained and survey questions were addressed to the parent.
The dependent variable for this study is patient satisfaction with the subject office visit as measured by the VSQ-9 scale. The 9 questions included in the scale are shown in Table 1. Questions 1 to 4 pertain to satisfaction with access to care; 5 to 8 pertain to satisfaction with the direct encounter with the physician; 9 pertains to an overall assessment of satisfaction. All responses are measured on a 5-point scale, ranging from poor = 1 to excellent = 5. In using the scale, one can either report a mean per-item score, or the distribution of scores among the 5 categories of response.31 For this study, we report the mean, per-item response.
The surveyor asked questions 1 to 3 while the patient was in the waiting room waiting to be seen. We asked question 4, rating satisfaction with the time spent waiting at the office, as two separate questions: satisfaction with time spent waiting in the waiting room before being called in to see the physicians, and satisfaction with time spent in the examination room before the physician actually entered. We took the mean of these responses to compute a single score for this question. We asked questions 5 to 8 as soon as the patient returned to the waiting area from the examination area, and question 9 as the patient exited the facility. We have reported elsewhere the results of our test as to whether the presence of the interviewer influenced patients’ responses, creating a “Hawthorne effect.”30 To do this we identified matched, case-control patients seeing the same physician for the first 100 patients from the initial sample of 291 patients, and administered a follow-up phone survey to both groups. There was no difference in satisfaction between the patients who interacted with a surveyor at the time of the original visit and those who did not.
Because organizations theory suggests that the quality of the interaction between patient and nonphysician staff may affect the patient's perception of the quality of the direct physician/patient encounter,32 we included these measures as independent variables. As the patient interacted with individual staff members, the surveyor in our study asked the patient to rate the quality of that specific interaction. We grouped staff members encountered into nursing staff and nonnursing staff, and calculated a mean response for each group.
The principal hypothesis of this study is that the association between patients’ race or ethnicity and their satisfaction ratings using the combined, 9-item scale will be different than their ratings of their direct interaction with the physician. Accordingly, we calculated a separate score for satisfaction with the direct physician encounter by taking the mean of the responses to questions 5 to 8. Using confirmatory factor analysis, we were able to show that this 4-item indicator is highly reliable (Cronbach's α = .88). Further analysis suggested that questions 1 to 4 cannot reliably be combined into a indicator of access/convenience of care (Cronbach's α = .56).
The independent variables included in the study are:
In measuring continuity of care, we also asked patients both about the number of visits and the length of time they had been under the care of the physician they were seeing that day. Neither variable had a significant association with reported satisfaction, and so neither is included in the model we report here.
While we noted the ethnicity and gender of the physicians included in our study, we did not attempt to control for these factors in our analysis. As the physicians in the sample were 92% white, the patient's race/ethnicity appears to be a close approximation of racial/ethnic concordance between physician and patient. Also, as the physicians were approximately equally split between men and women, and as the literature suggests that gender concordance is not a major factor in patients’ perceptions of the quality of the interaction with a primary care physician,14 we did not include this variable in our analysis.
We use linear regression analysis (SPSS-10; SPSS Incorporated, Chicago, Ill) to test for associations between the independent variables and the 2 dependent variables: mean 9-item satisfaction, and mean physician-specific satisfaction. In both models, we include only those independent variables found to have a significant association with the full, 9-item scale. In a separate analysis, we tested for an association between excluded variables and the physician-specific scale, and found none.
We created dummy variables from the categorical variables as follows.
A total of 537 patients completed at least part of the survey. As shown in Table 2, the patients included in this study represent an unusual study population, in that they have both high levels of education and high family income. Two thirds of the patients had graduated from college, with more than half of those having gone on to do graduate work. Consistent with this educational level, nearly three quarters of the patients reported annual family income of $60,000 or more. These findings are not surprising, however, as the study was done in a geographic region of northern California that has one of the highest educational and income levels in the country. In addition, the study population is skewed from the general population in the area in that all the medical groups included in the study require either proof of insurance or advance payment before accepting new patients.
Of the 537 patients in the study, one or more variables were missing from 57 cases, or 11%. Of these, 10 cases were missing data required to compute the satisfaction scales used as our dependent variable. For the additional 47 cases, the most frequent missing variable was the subject's reported satisfaction with the nonphysician staff (n = 31). To test for the effect of missing data, we replaced missing variables with mean values (continuous) or median values (discrete) of all missing variables, and reanalyzed the data. The overall results of the analysis were not changed by using these imputed data for the missing cases.
Table 3 shows the results of the two separate regressions conducted in this analysis. The results include both the unstandardized (b) as well as the standardized (β) regression coefficients. The first question to ask is whether nonwhite race/ethnicity shows an association with the full, 9-item satisfaction score after controlling for the other independent variables in the model. As shown in column 1, with an adjusted r-square of .36, the variables included account for 36% of the variance in mean satisfaction. The standardized regression coefficients (β), the absolute value of which provides a comparison of the relative magnitude of the effects of the different variables, indicates that age, overall level of health, seeing a physician in one of the smaller offices, and quality of the interaction with the nonphysician staff have the strongest associations with patient satisfaction. The patient's level of education had a smaller but significant effect. Patient gender did not have a significant effect, so is not included in the model. A separate analysis tested for different outcomes for children (age <18, for whom the parent was surveyed) and adults, and found none.
The analysis using this model did not identify a significant difference between white and nonwhite patients in their overall satisfaction with their visit. This finding duplicates that reported in our initial study.30 If the 9-item satisfaction scale is a valid measure of racial/ethnic differences in satisfaction, this finding would suggest that white and nonwhite patients in the community studied report the same level of satisfaction with care.
Column 2 of Table 3 shows the association between the above variables and the 4-item, physician-specific scale. As with the earlier model, the variables included account for a substantial amount of the variance (25%) in this satisfaction scale. Age, education, and health status continue to have significant, though somewhat weaker, effects. The size of the facility in which the visit took place no longer shows an effect, suggesting that patients’ perceptions of physician quality is independent of the organizational size of the practice setting after controlling for other aspects of the visit. The patients’ satisfaction with the nonphysician staff continues to show a strong association with their perception of the quality of the direct physician encounter, even though most (but not all) of the staff interactions measured by these variables took place before the patient was seen by the physician.
In this second model, measuring only satisfaction with the direct physician encounter and controlling for demographic variables and aspects of the visit not involving the physician, nonwhite patients rate the quality of their interaction with the physician significantly lower than white patients seeing the same groups of physicians. The standardized regression coefficients (β) indicate that the relative magnitude of this association is stronger than that for education, health status, and continuity of previous care, and comparable to the effect of age.
A recent report recommends that the quality of physician office care be measured nationally using a standardized, composite score that combines a series of indicators.33 The results of this study suggest that using such a composite score may not be sufficiently sensitive to identify racial or ethnic differences in patient satisfaction in certain settings. Among the patients in this study, representing a generally affluent, well-educated population with full health insurance coverage, nonwhites reported a lower level of satisfaction with their direct physician/patient interaction than white patients. However, these differences were not apparent when using the composite 9-item scale.
This study confirms that the interpersonal dynamics that occur between physician and patient play a role in creating the racial and ethnic differences in satisfaction that have been observed. From a quality improvement perspective, it is crucial to have measurement instruments that are sufficiently sensitive to racial and ethnic differences among patient populations if the national policy goal of eliminating racial and ethnic differences is to be achieved. For purposes of improving the quality of the physician-patient interaction, analysis should be done using those items within the scale that measure the quality of the physician/patient interaction directly, using other measures (e.g., waiting time, staff courtesy) as control variables.
It is not possible, however, to look for racial or ethnic differences in care if a medical group or organization does not include the patient's racial or ethnic group as part of the standard demographic patient database. As part of its National Standards for Culturally and Linguistically Appropriate Services in Health Care, the U.S. Department of Health and Human Services has adopted the following:
Standard 10. Health care organizations should ensure that data on the individual patient's/consumer's race, ethnicity, and spoken and written language are collected in health records, integrated into the organization's management information systems, and periodically updated.34
Whether medical care providers should routinely gather information regarding patient race or ethnicity has been a subject of recent debate. None of the medical groups included in this study includes that information in its database. Racial and ethnic differences in patient satisfaction and other measures of health care quality do not occur only among low-income patients. As this study suggests, those differences span the socioeconomic spectrum. In this study, in which we asked patients to identify their race or ethnicity as part of a publicly administered survey, 536 out of 537 patients provided this information. Concerns that patients will object to or be uncomfortable with efforts to gather data regarding race or ethnicity do not seem to be well founded.
Once racial or ethnic differences in care have been identified, the question arises as to the source(s) of those differences. While this study identifies such differences, it does not identify how the differences originate. Certainly there are at least two possibilities.
If the results of this study were due principally to patients’ preconceived attitudes toward the medical care system, we would expect nonwhite patients to be less satisfied with their interaction with the physician as well as the nonphysician staff. To test for this potentiality, we ran separate regression analyses looking for an association between the independent variables in our model and patients’ mean satisfaction with the nursing staff encountered, and their mean satisfaction with the nonnursing staff encountered. In neither case did any of the demographic variables, including patient's race/ethnicity, have any statistical association with these satisfaction outcomes. There are thus no indirect effects of demographic variables through satisfaction with the nonphysician staff. Among the patients included in this study, nonwhite patients were no different than white patients in their perceptions of the quality of their interactions with the nonphysician staff. These results suggest that the observed racial/ethnic differences in patient satisfaction with physician-provided care reflect differences in the way the physicians studied approach patients from different racial or ethnic groups. Whether this difference reflects bias on the part of the physician, or rather simply physicians’ unfamiliarity with the differing dynamics of physician-patient communication in the racially or ethnically discordant encounter, cannot be answered by this study, and deserves further research.
This study has a number of weaknesses that may affect our ability to generalize from these findings. Our study population is from a relatively small number of medical groups, from a relatively small geographic area. It is possible that, in a larger study population, the full 9-item satisfaction scale may have identified racial/ethnic differences. In addition, the highly skewed nature of the socioeconomic background of the patients studied makes it difficult to predict what results might be found in studies of patients more representative of the general population. Similarly, the inclusion of obstetrics/gynecology among the primary care specialties studied contributed to the skewed gender distribution of our study sample, with two-thirds of the patients surveyed being female. However, it is not uncommon for individual medical groups to conduct patient satisfaction surveys on study populations of approximately the same size as those reported here. In that situation, it would seem prudent to analyze those data in ways that would identify specific racial/ethnic differences in the direct physician/patient encounter.
This research was supported in part by grant 1 RO3 HS09350 from the U.S. Agency for Healthcare Research and Quality.