RECURRENCE AFTER RECOVERY IN UNIPOLAR MDD - AN OBSERVATIONAL FOLLOW-UP-STUDY OF CLINICAL PREDICTORS AND SOMATIC TREATMENT AS A MEDIATING FACTOR

This paper presents exploratory statistical analyses of the clinical p redictors of the first recurrence after recovery from the intake episo de of unipolar RDC Major Depressive Disorder and the effects of natura lly occurring somatic treatment decisions. After recovering, 359 unipo lar patients were followed for affective recurrence for up to five yea rs. The median length of the well-interval was 20 months, with 22% pro bability of sustaining recovery for five years. Duration of the index episode after intake, number of prior episodes, secondary subtype and endogenous subtype had significant predictive value for recurrence. A long index episode after intake appears to predict increased recurrenc e risk consistently, but the influence of the other predictors diminis hes as the remission lengthens. Of patients whose index episode lasted at least six months after they sought treatment and who recovered and remained well for two years, only one-third escape recurrence over th e next two year. This suggests there may be common factors linking a l ong episode and a short well interval. Equal proportions (about one-th ird) of patient began the interval with no antidepressant somatotherap y (ADS) and stayed off, received ADS continuously or dropped ADS durin g the well interval. Only 3% started ADS during the well interval. By two years, only 25% of patients still in remission were still on some ADS. Among patients who started with ADS, dropping ADS during the firs t six months at risk was associated with an average 2.5:1 increase in the weekly odds of recurrence; a larger odds ratio at the start declin ed to even odds after six months. Risk was inversely related to 'inten sity' of ADS, with lowest recurrence associated with continued high le vels of ADS. Patients with no ADS at the start of the well-interval ha d as low a recurrence rate as patients who continued ADS at high level s. To adjust for confounding of dropped ADS with prognosis, we used po st-stratification and covariance adjustment for the propensity to rece ive ADS, a technique for causal inference in observational studies. We believe that these non-experimental results complement and generalize the results of recent controlled clinical trials.