Machine learning-based causal models for predicting the response of individual patients to dexamethasone treatment as prophylactic antiemetic

Risk-based strategies are widely used for decision making in the prophylaxis of postoperative nausea and vomiting (PONV), a major complication of general anesthesia. However, whether risk is associated with individual treatment effect remains uncertain. Here, we used machine learning-based algorithms for estimating the conditional average treatment effect (CATE) (double machine learning [DML], doubly robust [DR] learner, forest DML, and generalized random forest) to predict the treatment response heterogeneity of dexamethasone, the first choice for prophylactic antiemetics. Electronic health record data of 2026 adult patients who underwent general anesthesia from January to June 2020 were analyzed. The results indicated that only a small subset of patients respond to dexamethasone treatment, and many patients may be non-responders. Estimated CATE did not correlate with predicted risk, suggesting that risk may not be associated with individual treatment responses. The current study suggests that predicting treatment responders by CATE models may be more appropriate for clinical decision making than conventional risk-based strategy.