The risks of risk assessment: causal blind spots when using prediction models for treatment decisions

2024-02-27

AI Summary

Clinical prediction models are frequently developed from observational data that include early treatments, rendering them vulnerable to confounding, selection bias, mediation effects, and dynamic treatment regimes—collectively termed “causal blind spots”—which lead to miscalibrated risk estimates and suboptimal clinical decisions. This paper formally defines “causal blind spots” for the first time and demonstrates that conventional modeling strategies—treating treatment as a covariate, stratifying by treatment, or omitting treatment—are all unreliable. We propose an intervention-oriented framework centered on the *interventional prediction estimand*, integrating causal diagrams, do-calculus, and potential outcomes theory. This framework mandates embedding causal inference into both model development and validation. By shifting predictive modeling from associative pattern recognition to causal intervention modeling, our approach provides a principled foundation for revising clinical prediction guidelines to ensure causal validity, thereby enhancing the scientific rigor and safety of treatment decisions.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Prediction models are increasingly proposed for guiding treatment decisions, but most fail to address the special role of treatments, leading to inappropriate use. This paper highlights the limitations of using standard prediction models for treatment decision support. We identify `causal blind spots' in three common approaches to handling treatments in prediction modelling: including treatment as a predictor, restricting data based on treatment status and ignoring treatments. When predictions are used to inform treatment decisions, confounders, colliders and mediators, as well as changes in treatment protocols over time may lead to misinformed decision-making. We illustrate potential harmful consequences in several medical applications. We advocate for an extension of guidelines for development, reporting and evaluation of prediction models to ensure that the intended use of the model is matched to an appropriate risk estimand. When prediction models are intended to inform treatment decisions, prediction models should specify upfront the treatment decisions they aim to support and target a prediction estimand in line with that goal. This requires a shift towards developing predictions under the specific treatment options under consideration (`predictions under interventions'). Predictions under interventions need causal reasoning and inference techniques during development and validation. We argue that this will improve the efficacy of prediction models in guiding treatment decisions and prevent potential negative effects on patient outcomes.

Problem

• Identify causal blind spots in treatment prediction models
• Address misinterpretation risks in clinical decision-making
• Advocate causal reasoning for treatment risk estimation

Innovation

• Causal reasoning in prediction model development
• Prediction under specific treatment interventions
• Guidelines extension for model evaluation