Treatment Responder Classification with Abstention

Treatment responder classification seeks to learn a rule to classify individuals who will benefit from the treatment. This paper studies a new scenario in treatment responder classification when abstention is allowed, i.e., practitioners can opt out of making uncertain classification on some individuals for further investigation. By revealing the implicit relation between causal misclassification risk with abstention and Conditional Value at Risk (CVaR), we develop a doubly robust method named TRECA to learn the classification rule under loose convergence conditions on nuisance parameters, and further extend it to deal with possible violation on key assumptions such as monotonicity and unconfoundedness. Rigorous theories and extensive experiments on two real-world datasets demonstrate the theoretical and experimental guarantee on our methods in learning treatment responders classification rules with low regret at the cost of limited abstention.

In the task of classifying responders to the treatment, mis-classification risk varies among individuals. Safety concern arises when assigning either treatment or control to units with high risks. Our paper addresses this problem by proposing a treatment responder classification framework with abstention (rejection) option. A rejection rule is learned to abstain making decision on high-risk individuals under constraint on rejection rate, while a prediction rule is jointly learned to classify responders on remaining samples. The framework is developed by revealing an insightful relation between misclassification loss with abstention and Conditional Value at Risk (CVaR), a value widely studied in risk-control studies, where we leverage its properties to design a doubly robust estimation on the loss function. We also discuss the rationality and extension of our method to important assumptions. This work is supported by rigorous theoretical guarantees and comprehensive real-world experiments.