Purpose: For patients with Multiple Sclerosis (MS), adherence to disease modifying therapy (DMT) is critical in altering the clinical progression of disease and preventing costly hospitalizations; therefore it is important to identify patients with an increased likelihood of becoming non-adherent to DMT.
Method: We used a traditional binary logistic regression technique combined with a new grouping strategy to develop a predictive model that rank orders MS patients from highest to lowest likelihood of becoming non-adherent to DMT in the next six months. Several input factors, such as demographic and personal information, prescription benefit plan sponsor information, prescription attributes, and previous prescription information were used to build the predictive model. There were 11,742 cases in the population. This model was developed on the overall population (experienced and naïve to therapy). Due to a relatively small population sample, a random 50% of the population was selected for model development and 50% was used for validation. Robustness across time was also tested on a non-contemporaneous sample of 12,946.
Result: We tested the robustness of the final model on contemporaneous and non-contemporaneous validation data sets. The performance of the model was good and consistent across development and validation data sets. The results were better than those reported for other adherence predictive models; the c-statistic was 0.83 and the KS statistic was 0.529. The lift chart showed that 30% of the population contained 57% of the non-adherent members.
Conclusion: We successfully developed a new approach to identify, in advance, which MS patients are at risk of non-adherence to DMT. By allowing healthcare providers to focus their interventions on MS patients most likely to have an adherence problem and to intervene before medication non-adherence occurs, the model may play a key role in better managing potential clinical and economic outcomes.