OBJECTIVES: The Center for Medicaid and Medicare Services (CMS) has created plan Star ratings that indicate the quality of Medicare plans.  In 2012, CMS added three pharmacy measures that focus on member medication adherence, i.e. oral diabetes medications, hypertension medication (ACEI or ARB), and cholesterol medication (statins).  To proactively identify patients at risk for non-adherence, a multi-variate regression prediction model was developed to create individual persistency risk scores.  METHODS:  The predictive model is created using prescription drug and medical claims from a large managed care database. Medicare and commercially insured patients over age 55 from 2008-2010 who are new to the Star rating medication categories are included. Patients included in the model have a full 18 months of continuous enrollment in the health plan (6 month drug naïve period, 12 months of follow up). The predictors are created from the 6 month pre period and include: a) socio-economic factors; b) medical characteristics (e.g. Charlson Comorbidity Index); and c) drug characteristics (i.e. drug cost and past chronic drug adherence). RESULTS: Multivariable analysis of study outcomes will be conducted using appropriate regression models based on the distribution of the measure. A logistic regression model will be estimated (=1 for at least 80% PDC, =0 for non-compliance). Results of a logistic regression will be presented as odds ratios associated with each independent variable.  The parameter estimates from the above econometric model will be retained and used to estimate the probability of non-compliance on a new set of patients.  To test the accuracy of the predictive model, we will choose a random sample of patients new to these medications in 2011, as exhibited by the average PDC in each risk group (high, medium, low). CONCLUSIONS: An adherence predictive model can be useful to identify patients who may benefit from a drug adherence intervention program.