Neural networks provide a powerful framework for learning the association between input and response variables and making accurate predictions and offer promise in using the rapidly growing volume of health care data to surface causal relationships that cannot necessarily be tested in randomized clinical trials. In pursuit of models whose predictive power comes maximally from causal variables, we propose a novel causal regularizer based on the assumption about independence of different steps of data generation process. We use the causal regularizer to steer deep neural network architectures towards causally-interpretable solutions. We perform a large-scale analysis of electronic health records. Our causally-regularized algorithm outperforms its L1-regularized counterpart both in predictive performance as well as causal relevance. Finally, we show that the proposed causal regularizer can be used together with representation learning algorithms to yield up to 20% improvement in the causality score of the generated multivariate hypotheses.

该研究使用因果正则化方法来提高预测模型的因果可解释性，针对大规模的电子健康记录数据进行了分析，结果显示相较于L1正则化模型在因果准确性方面表现出色且有竞争力的预测性能，并提高了多层感知器在探测多元因果关系上的表现约20%。