Causal Inference in Threshold Regression and the Neural Network Extension (TRNN)

The first-hitting-time based model conceptualizes a random process for subjects’ latent health status. time-to-event outcome is modeled as the first hitting time of to pre-specified threshold. Threshold regression with linear predictors has numerous benefits in causal survival analysis, such estimators’ collapsibility. We propose neural network extension threshold model. With flexibility networks, extended can efficiently capture complex relationships among and underlying processes while providing clinically meaningful interpretations, also tackle challenge high-dimensional inputs. proposed further be applied performing Q-model G-computation. More efficient estimations are expected given algorithm’s robustness. Simulations were conducted validate estimator collapsibility performance illustrated by using simulated real data from an observational study.