Predicting Outcomes of Primary Care Patients with Major Depression
Development of a Depression Prognosis Index
Published in: Psychiatric Services, v. 58, no. 8, Aug. 2007, p. 1049-1056
Posted on RAND.org on January 01, 2007
OBJECTIVE: Depression research and practice focus increasingly on diverse patient populations with varying probabilities of response to clinical care. Prognostic indices use preexisting patient characteristics to estimate the probability of subsequent negative clinical outcomes and are useful tools for improving the study and care of diverse populations. Few such measures, however, have been developed for mental health conditions. This study developed and validated a depression prognosis measure for primary care patients with major depression. METHODS: Consecutive patients in 108 primary care practices were screened for depression, and 1,471 with major depression were enrolled. A Depression Prognosis Index (DPI) predicting persistent depression six months after baseline was developed for a random one-third subsample and validated with the remaining two-thirds. Models included prior treatment, demographic characteristics, comorbidities, and other physical, psychological, and social predictors. RESULTS: Sixty-four percent to 65% of patients classified by baseline DPI score as being in the sample quartile with the worst prognosis had probable major depression six months later, compared with 14% to 15% in the best-prognosis quartile. The DPI had an R(2) of .40 in the development sample and .27 in the validation sample. Important predictors included severity of depression symptoms at baseline, social support, common physical symptoms, and having completed three months of antidepressants at sample entry. CONCLUSIONS: The ability of the DPI to predict six-month outcomes compares favorably to that of prognostic indices for general medical problems. These results validate the DPI and provide conceptual guidance for further development of depression risk stratification instruments for clinical and research use.