Predicting drug cost under the Medicare Part D benefit

Objectives: In 2012, beneficiaries in every state have at least 25 different stand-alone prescription drug plans from which to choose to receive their prescription drug coverage. We sought to create a regression model to identify factors which help predict the estimated annual costs (EAC) of the lowest cost Part D plan for beneficiaries in 2012. Methods: Targeted community outreach events were held at 13 sites between October and December 2011 during which Medicare beneficiaries were provided Part D plan assistance. A survey was used to collect and record Part D plan cost data that was retrieved subsequent to a personalized plan search (conducted on www.medicare.gov) during each intervention. Additionally, beneficiary-specific data were collected. A linear regression model via the Stepwise method was created in which EAC was the dependent variable and potential cost drivers were independent predictors. Results: Data from 362 beneficiaries were used to create the regression model. Three factors were identified as significant predictors of EAC including number of prescription medications, subsidy status, and age. Low degrees of multicollinearity were found between variables comprising the final model. Additionally, the final model coefficient of determination revealed that 29.1% of the variance in EAC could be explained by the included independent variables. Conclusions: Although certain variables are reliable for predicting plan cost, most of the variance in the EAC of the lowest cost plan was unexplained. This further supports the beneficiary-specific nature of optimal Part D plan selection and reinforces the need for annual plan evaluation to minimize out-of-pocket costs.