Risk Aversion and Costs A Comparison of Family Physicians and General Internists
Complete survey information was available for 173 (63%) physicians. Those who did not complete the questionnaire were not statistically significantly different in age or sex from those who did. Physicians who completed the questionnaire were more likely to be family physicians (34% vs 21%, P = .02). Physicians who completed the questionnaire also had lower per-member observed expenditures ($1136 vs $1487, P = .02) and lower per-member expected expenditures ($1167 vs $1373, P = .02), but the ratio of observed to expected expenditures was not statistically significantly different (P = .80).
Observed Expenditures
Total expenditures per physician for each member who was enrolled for all of 1995 were calculated from the “allowed amount” variable in the claims files. The allowed amount is the sum of the amount paid, the copayment, the deductible, and the amount withheld for the risk pool. Since the allowed amounts varied across providers, we standardized prices using the claims data. For physician claims, our standardized prices were the average amounts allowed for each Current Procedural Terminology, 4th edition (CPT-4) code and provider specialty. For inpatient hospital claims, our standardized price was the average or allowed amounts by diagnosis-related group. For all other claims, our standardized price was the average of the amounts allowed by CPT-4 code, with separate facility and nonfacility categories.
We defined the total expenditures for each patient as the sum of the standardized prices for all services listed on the patient’s claims for the calendar year. And we calculated the observed expenditures per panel member for each physician using the mean of the total expenditures for each patient in the physician’s panel, with nonusers assigned 0 expenditures.
Case-Mix Adjustment
We developed an expected expenditures per panel member measure to adjust for possible case-mix differences between family physicians and internists. Following the standard method of Duan and colleagues,28 we used a 2-part model to derive anticipated expenditures per panel member: The first part predicted the probability that services were used, and the second predicted expenditures contingent on use.28 The product of those 2 predictions was considered to be the anticipated expenditures per panel member. Since little was known about panel members who did not use any services, the proportions of patients who used some services for 10 age/sex categories were used to define the predicted probability of using some services. Predicted expenditures among users were made on the basis of patient age, sex, and case-mix methodology using the ambulatory care groups system.29 The logarithm of expenditures contingent on some use of services was predicted with an ordinary least squares regression using ambulatory diagnostic groups, age group, and sex. We used ambulatory diagnostic groups instead of ambulatory care groups because they explained more variance. The logarithmic transformation of expenditures was used in the regression analyses because of the extremely skewed distribution of expenditures.
The result was retransformed into anticipated expenditures for those who used services with the “smearing” estimate recommended by Duan and coworkers.30 Smearing is recommended because simply retransforming the logarithms results in an underestimate of expenditures above the median.* The anticipated expenditures per panel member for each physician was calculated as the mean of the anticipated expenditures for each patient in the physician’s panel, including those who did and did not use services.
Analyses
We conducted the analyses in 3 parts. First, we sought to identify psychological factors that would distinguish family physicians from internists. Second, we examined whether there was a relationship between generated physician costs and specialty or psychological factors. Third, we explored the extent to which the psychological factors found to distinguish family physicians from internists also accounted for the differences in generated costs.
We used logistic regression to compare the demographic, practice, and psychometric profiles of family physicians and internists. In this analysis, physician specialty (family physician or internist) was the dependent variable and the demographic, practice, and psychometric scale scores were the independent variables. We used stepwise regression to determine which factors made a statistically significant contribution to distinguishing family physicians from internists. Second, we used physician-level ordinary least squares regression analyses to examine the relationships between observed logarithmic expenditures per panel member and specialty or the psychological factors discriminating between the 2 specialties. Covariate adjustment included logarithmic anticipated expenditures per panel member and significant physician demographic and practice variables. A final analysis with logarithmic observed expenditures as the dependent variable included both specialty and the psychological factors found to differ between the 2 specialties as independent variables. Covariate adjustment again included logarithmic anticipated expenditures and significant physician demographic and practice variables. We used logarithms of expenditures per panel member because both the total and anticipated expenditures exhibited skewed distributions that were normalized by logarithmic transformation. The expenditure analyses were weighted by the physician panel size.
