Prediction-Constrained Topic Models for Antidepressant Recommendation

TL;DR

This paper introduces a prediction-constrained supervised topic model that balances interpretability and predictive accuracy, demonstrated through improved depression medication recommendations from electronic health records.

Contribution

It proposes a novel training framework for supervised LDA that effectively balances data explanation and label prediction, addressing a key asymmetry in prior methods.

Findings

Improved depression medication prediction accuracy.

Enhanced interpretability of clinical topics.

Outperforms previous supervised topic models and logistic regression.

Abstract

Supervisory signals can help topic models discover low-dimensional data representations that are more interpretable for clinical tasks. We propose a framework for training supervised latent Dirichlet allocation that balances two goals: faithful generative explanations of high-dimensional data and accurate prediction of associated class labels. Existing approaches fail to balance these goals by not properly handling a fundamental asymmetry: the intended task is always predicting labels from data, not data from labels. Our new prediction-constrained objective trains models that predict labels from heldout data well while also producing good generative likelihoods and interpretable topic-word parameters. In a case study on predicting depression medications from electronic health records, we demonstrate improved recommendations compared to previous supervised topic models and high- dimensional logistic regression from words alone.