|
Abstract:
|
Applying a combination of data mining techniques and statistical methods to claims data can help identify causal relationships between medications and the incidence of related side effects. Given the unclear association between use of anti-dementia medications and the risk for cardiovascular adverse events, we leveraged 2007-2013 Medicare claims data and identified 30433 patients newly diagnosed with Alzheimer’s Disease (AD) who later initiated anti-dementia therapy. We used feature selection techniques followed by step-wise logistic regressions to identify risk factors for side effects. For the pre-specified 10381 potential factors, including patient demographics, ICD-9 diagnosis codes and therapeutic classes of patient-used medications, our combined model of factor screening, ranking and step-wise logistic regression classification successfully identified 55 risk factors for any adverse cardiovascular event. We obtained a model performance with a C-statistics of 0.67 and an accuracy of 0.74. Subgroup analyses studying specific adverse cardiovascular events by implementing same model also yielded similar outcomes, providing insights into patient safety related to AD drug use.
|
