Prediction of adherence to treatment with statins and anti-platelet drugs in first-year post-stroke patients: Validation of beta-regression models.

The statin adherence prediction model explained 67% of variance in 1-year adherence in the validation cohort.

• R² was 0.67 for the statin model
• The statin cohort included 2369 patients
• The model used first 90-day Proportion of Days Covered (PDC) as a single predictor for 1-year adherence
• Adherence was measured using PDC derived from prescription-filling data

The antiplatelet adherence prediction model explained 56% of variance in 1-year adherence in the validation cohort.

• R² was 0.56 for the antiplatelet model
• The antiplatelet cohort included 2147 patients
• The model used first 90-day PDC as a single predictor for 1-year adherence
• Adherence was measured using PDC derived from prescription-filling data

The statin model slightly underestimated observed 1-year adherence by a mean difference of 3.7 percentage points.

• The difference between the mean observed and the mean predicted PDC was -3.7% for the statin model
• A calibration slope of 1.06 was observed for the statin model, close to the ideal value of 1
• The negative difference indicates the model predicted slightly lower adherence than was observed

The antiplatelet model slightly underestimated observed 1-year adherence by a mean difference of 2.5 percentage points.

• The difference between the mean observed and the mean predicted PDC was -2.5% for the antiplatelet model
• A calibration slope of 0.96 was observed for the antiplatelet model, close to the ideal value of 1
• The negative difference indicates the model predicted slightly lower adherence than was observed

The previously published beta regression models demonstrated good predictive performance when validated on a new external population of post-stroke patients.

• Model performance was assessed using R², difference between mean observed and mean predicted PDC, and calibration slope
• Calibration slopes of 1.06 and 0.96 for statin and antiplatelet models respectively indicate good calibration
• The study population consisted of patients post-stroke or transient ischemic attack
• The authors concluded the models 'may be used for early identification of patients at high risk for low 1-year adherence within 90 days post-stroke'

Very few medication adherence prediction models for post-stroke patients exist and most have not been validated using external data prior to this study.

• The authors noted that 'very few medication adherence prediction models are available and have not been validated using external data'
• Stroke is described as 'the third most common cause of disability and the second most common cause of death worldwide'
• Greater levels of medication adherence after stroke or transient ischemic attack are associated with improved survival