Doubly Robust Functional Representation Learning for Longitudinal Causal Inference with Irregular Histories
DR-FRL is built to make doubly robust causal estimation work on irregular, informative patient-history fragments.
The paper proposes a cross-fitted workflow that encodes point-cloud-like histories into states targeted to the estimating equation, then checks them with EIF-focused validation, calibration, overlap, tail, and ablation diagnostics. Its claim is conditional: if the learned state preserves the nuisance information needed by the EIF, representation error stays in the usual second-order remainder. Simulations favor the method under high-dimensional functional confounding, informative measurement, weak support, and heavy-tailed pseudo-outcomes. In a VitalDB audit, it also reports a negative finding: scalar lab summaries already captured much of the endpoint-relevant information for the ICU-disposition endpoint. ArXiv · AI/CL/LG's note
The paper proposes a cross-fitted workflow that encodes point-cloud-like histories into states targeted to the estimating equation, then checks them with EIF-focused validation, calibration, overlap, tail, and ablation diagnostics. Its claim is conditional: if the learned state preserves the nuisance information needed by the EIF, representation error stays in the usual second-order remainder. Simulations favor the method under high-dimensional functional confounding, informative measurement, weak support, and heavy-tailed pseudo-outcomes. In a VitalDB audit, it also reports a negative finding: scalar lab summaries already captured much of the endpoint-relevant information for the ICU-disposition endpoint. ArXiv · AI/CL/LG's note
score 4