Learning to Select the Best Forecasting Tasks for Clinical Outcome Prediction

TL;DR

This paper introduces a meta-learning approach to select optimal forecasting tasks for clinical outcome prediction, improving patient risk prediction accuracy with limited data by leveraging unlabeled clinical trajectories.

Contribution

It proposes a self-supervised meta-learning method that directly optimizes patient representations for clinical outcome prediction, outperforming traditional supervised and pretraining methods.

Findings

Significantly improved risk prediction performance on MIMIC-III dataset.

Effective use of limited samples for fine-tuning patient representations.

Outperforms direct supervised learning and all-observation pretraining.

Abstract

We propose to meta-learn an a self-supervised patient trajectory forecast learning rule by meta-training on a meta-objective that directly optimizes the utility of the patient representation over the subsequent clinical outcome prediction. This meta-objective directly targets the usefulness of a representation generated from unlabeled clinical measurement forecast for later supervised tasks. The meta-learned can then be directly used in target risk prediction, and the limited available samples can be used for further fine-tuning the model performance. The effectiveness of our approach is tested on a real open source patient EHR dataset MIMIC-III. We are able to demonstrate that our attention-based patient state representation approach can achieve much better performance for predicting target risk with low resources comparing with both direct supervised learning and pretraining with all-observation trajectory forecast.