Predictors of a Medical-Offset Effect Among Patients Receiving Antidepressant Therapy

Study subjects were identified by using the computerized claims-processing system of a large private health insurer in New England with a membership of approximately 1.8 million persons. We scanned all paid pharmacy claims between July 1, 1991, and June 30, 1993, to identify each plan member who received a prescription for a selective serotonin reuptake inhibitor (SSRI) or a tricyclic antidepressant. To ensure that each patient was being treated for depression, we excluded those who did not have at least one provider encounter within the 6 months before the earliest-observed prescription (“index date”) with a listed diagnosis of depression (ICD-9-CM 296.2, 296.3, 300.4, 309.0, 309.1, or 311). We further excluded patients who had received any antidepressants—including SSRIs, tricyclic antidepressants, and monoamine oxidase inhibitors—in the 6 months preceding the index date to ensure that each subject was initiating a new course of therapy. Finally, to ensure that the claims histories were complete, we excluded all patients who were eligible for Medicare benefits anytime during the study period or who had not been continuously enrolled in the plan during the 1-year periods before and after the index date.

For each patient, we tabulated the costs of medical services provided in 1) physicians' offices, 2) hospital outpatient clinics, 3) hospital inpatient units, and 4) all of these settings combined. Costs were calculated by summing patient copayments and amounts reimbursed by the insurer for services rendered during the 1-year periods before and after the index date (pretreatment and follow-up, respectively). Since our interest was in changes in the costs of medical services only (i.e., exclusive of mental health treatment), all encounters in which a diagnosis of a mental disorder (ICD-9-CM 290–319) was rendered or in which the specialty of the provider or institution was mental health were excluded from the cost calculations.

For each cost measure, we calculated the change between the 1-year pretreatment and follow-up periods (i.e., as follow-up cost minus pretreatment cost). To control for the effect of inflation, we used the medical care component of the consumer price index to deflate costs in the follow-up period to levels prevailing in the pretreatment period. We then stratified patients according to the direction and magnitude of the change in costs between these periods. The subjects for whom costs increased were first distinguished from the subjects for whom costs decreased, and then, within each of these groups, the subjects for whom the change in costs exceeded the median increase or decrease, respectively, were distinguished from those for whom the change was less than this amount. All subjects for whom costs did not change were excluded from analyses of that measure. This method resulted in four ordered patient strata for each cost measure, subjects who had 1) large reductions in costs, 2) small reductions in costs, 3) small increases in costs, and 4) large increases in costs.

For each cost measure, a variety of factors were compared across the four patient strata, including patient demographic characteristics, the prevalence of selected diseases that have been identified as important comorbid conditions for depression management (on the basis of guidelines for depression diagnosis and management from the Agency for Health Care and Policy Research [6]), various patterns of antidepressant use that have been reported to be associated with changes in the costs of mental health services among patients with depression (7) (i.e., early discontinuation, switching or augmentation, upward titration, partial compliance, 6-month stability), the type of medication received (SSRI versus tricyclic antidepressant), and the numbers of visits to mental health providers.

We used techniques of ordinal regression analysis to estimate odds ratios and 95% confidence intervals for each of these factors within a multivariate context (8). For each cost measure, a regression model was estimated that included the patient stratification variable as the dependent variable (equal to 1, 2, 3, or 4, corresponding to the four ordered strata described in the preceding section) and all of the factors of interest as explanatory variables. For binary factors (e.g., presence or absence of diabetes), the resulting odds ratios depict the relative likelihood of being in lower-ordered versus higher-ordered strata in the presence of the factor of interest, for any given bifurcation of patients (i.e., stratum 1 versus all higher-ordered ones, strata 1 and 2 versus all higher-ordered ones, etc.) (8). For continuous factors (e.g., patient age), the odds ratios depict the relative likelihood associated with a one-unit increase (e.g., 1 more year of age) of being in lower-ordered versus higher-ordered strata (8). We used the odds ratios to assess the relative likelihood of experiencing a large reduction in cost in the presence of the factor of interest.