Predicting Depression Status in the Elderly: An Example from Medicare CAHPS Surveys
*Marc N. Elliott, The RAND Corporation
While depression in the elderly is of great importance in all aspects of healthcare, many datasets lack such an indicator for use as a predictor in analyses of patient experience or other outcomes. We develop a model to predict a senior’s probability of depression based on SF-12 items and commonly available demographic information such as age and gender. We define depression from ICD-9 codes in Medicare utilization files and predict this status in a series of logistic regression models from administrative records and survey responses of 271,479 beneficiaries responding to 2002-2004 CAHPS Medicare surveys. While lower MCS (mental) scores are the best predictor of depression, the relationship is nonlinear, and lower PCS (physical) scores, female gender, and greater age also matter. Moreover PCS, gender, and age are most important at low levels of MCS.