Prediction Focused Topic Models for Electronic Health Records
This work addresses the challenge of interpretable feature extraction for prediction tasks in electronic health records, representing an incremental improvement over existing supervised topic modeling methods.
The paper tackles the problem of balancing prediction quality and topic coherence in supervised topic models for EHR data by introducing a prediction-focused topic model that retains only features improving prediction, resulting in more coherent topics while maintaining competitive predictions on EHR and movie review datasets.
Electronic Health Record (EHR) data can be represented as discrete counts over a high dimensional set of possible procedures, diagnoses, and medications. Supervised topic models present an attractive option for incorporating EHR data as features into a prediction problem: given a patient's record, we estimate a set of latent factors that are predictive of the response variable. However, existing methods for supervised topic modeling struggle to balance prediction quality and coherence of the latent factors. We introduce a novel approach, the prediction-focused topic model, that uses the supervisory signal to retain only features that improve, or do not hinder, prediction performance. By removing features with irrelevant signal, the topic model is able to learn task-relevant, interpretable topics. We demonstrate on a EHR dataset and a movie review dataset that compared to existing approaches, prediction-focused topic models are able to learn much more coherent topics while maintaining competitive predictions.