Identifying routine clinical predictors of non-adherence to second line therapies in Type 2 diabetes: a retrospective cohort analysis in a large primary care database

Aims: Non-adherence to medication is a major problem for patients with diabetes leading to poor response to therapy. Many factors associated with poor adherence have been identified, but their combined predictive ability has not been assessed. We investigated whether combinations of routinely available clinical features can predict which patients are likely to be non-adherent. Materials and methods: 67882 patients with prescription records for their first and second oral glucose lowering therapies were identified from electronic healthcare records (Clinical Practice Research Datalink (CPRD)). Non-adherence was defined as a medical possession ratio (MPR) ≤80%. Potential predictors were examined including age at diagnosis, sex, BMI, duration of diabetes, HbA1c, Charlson Index and other recent prescriptions. Results: Routine clinical features were poor at predicting non-adherence to the first diabetes therapy (c-statistic=0.601 for all in combined model). Non-adherence to the second drug was better predicted for all combined factors (c=0.715) but this improvement was predominantly a result of including adherence to the first drug (c=0.695 for this alone). Patients with MPR≤80% on their first drug were 3.6 (95% CI 3.3,3.8) times more likely to be non-adherent on their second drug (32% v 9%). Conclusions: Although certain clinical features are associated with poor adherence, their performance for predicting who is likely to be non-adherent, even when combined, is weak. The strongest predictor of adherence to second-line therapy is adherence to the first therapy. Examining previous prescription records could offer a practical way for clinicians to identify potentially non-adherent patients and is an area warranting further research.