OBJECTIVES:  Retrospective evaluations conducted on claims databases have been used to detect medication non-adherence and identify risk-factors associated with non-adherence. The aim of the study was to construct predictive models of COPD medication non-adherence, using demographics, comorbid conditions, and COPD treatment claims data. METHODS: Using the Truven commercial and claims encounter database, this retrospective longitudinal study included 44,393 COPD patients initiating maintenance therapy over a 12 month identification period (Jan, 2011 to Dec, 2011). Beneficiaries were followed over two year rolling index period through Dec, 2013 to assess medication adherence using proportion of days covered (PDC). Risk factors were selected using logistic regression model employing a backward elimination process to develop the final model. Sensitivity, specificity, false positive and false negative rates were estimated for the final model. Model performance was also described using c-statistic, percent concordant and discordant pairs.  To estimate the predictive validity of the final model, variables were added to a GLM model and predicted adherence rates were estimated assuming a normal distribution of PDC scores. An inclusion criterion for p-value was 0.35 to allow for clinically important variables to be included for estimating the final model. RESULTS: The final model included 13 variables which were below the inclusion p-value criteria of p<0.35 and had a c-statistic of 0.799. The newly developed model had a specificity of 81.70% and a sensitivity of 75.20% with a 79.60% concordance in the final model. The model also demonstrated concordant validity with an error rate of <5%. The model predicted that 55.84% individuals are likely to be adherent as compared to actual adherence of 61.85%.  CONCLUSIONS: Using information available from healthcare claims data only, predictive models can reliably identify COPD medication non-adherence. Prior medication adherence was the best predictor of the future medication adherence among COPD patients.