Combining multi-level evidence for medical record retrieval

The increasing prevalence of electronic health records containing rich information about a patient's health and physical condition has the potential to transform research in health and medicine. In this work, we present a health record search system for finding patients matching certain inclusion criteria (specified as keyword queries) for clinical studies. In particular, our system aggregates multi-level evidence and combines proven statistical IR models, both in an innovative way, and achieves a 20% MAP (mean average precision) improvement over a strong baseline. Moreover, our cross-validation results show that the overall performance of our system is comparable to other top-performing systems on the same task.