What is it about?
Establishing the benefit of a real-world clinical intervention involves working with time-to-death, re-hospitalization, or a composite of multiple such outcomes. This requires decoupling the effects of confounding physiological characteristics that affect baseline survival rates from the effects of the interventions being assessed. In this paper, we present a latent variable approach to model heterogeneous treatment effects by proposing that an individual can belong to one of latent clusters with distinct response characteristics. We show that this latent structure can mediate the base survival rates and helps determine the effects of an intervention. We demonstrate the ability of our approach to discover actionable phenotypes of individuals based on their treatment response on multiple large landmark randomized clinical trials.
Featured Image
Photo by Clint Adair on Unsplash
Why is it important?
Estimation of treatment benefit at a population level is important as it informs current clinical guidelines and practices, however the effect of any intervention is rarely uniform across any population under observation. The advent of precision medicine aims to address these differences in treatment effect by applying individualized treatments designed based on each patient’s individuals characteristics. This strategy assumes there are differences in treatment effects that may be explained by varying demographic factors, baseline physiology, or prior medical history. The crux of precision medicine thus entails careful phenotyping of individuals that may receive augmented benefit from a treatment versus those who would not benefit (or worse, suffer adverse effects). These individual patient factors can then be accounted for when framing clinical guidelines based on randomized trials.
Perspectives
✓ We propose a deep latent variable approach to recover subgroups of patients that respond differentially to an intervention, in the presence of censored outcomes. ✓ We present conditions in which the counterfactuals are identifiable using observational data under the proposed model, along with an efficient approach for learning and inference. ✓ We demonstrate the proposed approach applied to multiple large landmark clinical trials that were originally carried out to assess the efficacy of medical interventions to reduce risk of adverse cardiovascular outcomes among hypertensive and diabetic patients, and discover clinically actionable counterfactual phenotypes. The proposed approach, Cox Mixtures with Heterogeneous Effects has been released as part of the open-source package, auton-survival and is available at autonlab.github.io/auton-survival/models/cmhe.
Chirag Nagpal
Carnegie Mellon University
Read the Original
This page is a summary of: Counterfactual Phenotyping with Censored Time-to-Events, August 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3534678.3539110.
You can read the full text:
Resources
Contributors
The following have contributed to this page
