The sampling design of the Physician Worklife Survey (PWS) has been previously described in detail. Briefly, a national probability sample of physicians in family practice, general internal medicine, pediatrics, internal medicine subspecialties, or pediatric subspecialties was drawn from the AMA Masterfile. The sampling frame was stratified on the basis of indicators of physician race (white or missing vs. all other), physician specialty (as categorized above), and penetration of managed care in the state of registry (state in highest quartile of proportion of physicians with managed care contracts vs. all other states). These categories created 20 strata, which were disproportionately sampled to produce a final sample of 5,704 physicians. Physicians in states with high levels of managed care penetration, those practicing subspecialties, and those from racial and ethnic minority groups were oversampled. The survey was mailed up to four times, resulting in 2,326 usable responses. Taking into account an estimated noncontact rate of 18%, this corresponds to an adjusted response rate of 52%. Further assessment of the correlation between time to response and 140 variables collected on the survey identified only four variables with Spearman correlation coefficients greater than 0.10. This suggests minimal differences between early and late responders in most characteristics and presumably between responders and non-responders. However, white physicians averaged a much shorter time to return than did physicians of any other racial/ethnic category. After an initial lower response rate from minority physicians, nationally recognized minority physicians appealed to minority physicians to return their surveys. This may have improved the minority physician response rate.
The current analyses exclude respondents who either did not report their race or ethnicity (n=13) or who reported an ethnic group other than white, black, Asian or Pacific Islander, or Hispanic (n=60). Similarly, respondents with missing scores on the global job satisfaction scale (n=20), the global career satisfaction scale (n=14), or the stress scale (n=2) were excluded. These eliminations resulted in a final sample size for these analyses of 2,217 physicians.
Measures of Professional Satisfaction
Development and validation of the PWS items and scales has been described elsewhere. The PWS was an eight-page, 150-item mail survey that measured physician practice characteristics and professional satisfaction. Survey items were developed from existing satisfaction measures, physician focus groups (including focus groups with physicians from racial and ethnic minority groups), and analysis of open-ended responses from a survey of physicians in large group practices. The resulting items were refined by an expert panel and then tested on a pilot group of 2,000 physicians. Factor and reliability analyses resulted in a final list of 36 items measuring 10 domains of satisfaction: autonomy; relationships with patients; relationships with colleagues; relationships with staff; patient care issues; personal time; community; income; administration; and resources. In a parallel process, global measures of job satisfaction (five items) and career satisfaction (four items) were developed. Each individual item used a 1-5 Likert response format, and the sums were scaled to a 1-5 range in our analyses. Although each item on the job satisfaction scales was ordinal, the combined scales have 20-25 possible response values and were therefore treated as continuous. Internal consistencies for the 10 satisfaction domains ranged from 0.65 to 0.77, and for the global satisfaction measures 0.86 to 0.88. The 10 domains accounted for 58% of the variance seen in global job satisfaction. The index was a four-item version of the previously validated Perceived Stress Scale and had an internal consistency reliability of 0.75 in this sample.
All respondents were asked to identify their own racial or ethnic group, by choosing one race or ethnic group from 10 possible categories: (white American-born, white other, black African-American, black African descent, Asian or Pacific Islander, Hispanic Puerto Rican, Hispanic Mexican, Hispanic other, Native American or Alaskan Native, or other). Because of sample size limitations, we could not utilize all of these categories in the analysis. Primary analyses used data consolidated into only four racial/ethnic categories: black (African-American and African), white (American born and other), Hispanic (Puerto Rican, Mexican, and other), and Asian or Pacific Islander. cialis canadian pharmacy
The primary outcome measures for our analyses included global career satisfaction, global job satisfaction, and stress. We also examined differences by physician race and ethnicity separately for the 10 specific domains of professional satisfaction.
Chi-squared tests or analyses of variance were used to assess differences in demographic characteristics (age, sex, and marital status), and practice setting. Based on regression models adjusted for survey design, combined F-tests were used to assess differences in patient panel characteristics (percent female, elderly, speaking little or no English, with complex/ numerous medical problems, with complex/numerous psychosocial problems, with substance abuse problems, on Medicaid, uninsured, and patient race/ethnic identity) across physician racial/ethnic categories. We tested for racial differences in professional satisfaction using analysis of variance methods for bivariate analyses and linear regression for multivariate analyses. To provide an adjusted estimate of the effect of physician race on job satisfaction, career satisfaction, and stress, we calculated regression models examining the independent association of physician race with each of these dependent variables, after adjusting for physician demographic characteristics. Demographic covariates included age (continuous), marital status (married versus all other), sex, and annual income category (<$100,000, $100,000-$149,000, $150,000-$249,000, and $250,000 or more). We hypothesized that the effect of race and ethnicity on professional satisfaction may operate in part through mediating variables such as practice characteristics (practice setting and specialty) and patient characteristics. To explore this possibility, we calculated regression models that added physician practice characteristics (specialty and practice setting), and patient panel characteristics (percent of patients female, elderly, with limited English, with complex medical problems, with complex psychosocial problems, with substance abuse problems, on Medicaid, or uninsured) as independent variables. All regression models were adjusted to account for stratified sampling in the survey design and differing response rates. For multiple regression models, independent variables with missing values were assigned the mean value, and an indicator variable for such imputations was included in the models. Covariates with missing values included age (n=17 missing), income (n=305), practice type (n=19), fraction of patients using Medicaid (n=367), fraction of patients uninsured (n=421), and other patient panel characteristics (n=30).
All analyses were conducted using the SAS system (Cary, NC) version 8.2 or Stata version 8.0 (College Station, TX).