[Paper Review] Adaptive Weight Learning for Multiple Outcome Optimization With Continuous Treatment

Chang Wang, Lu Wang|arXiv (Cornell University)|Feb 16, 2024

Total Knee Arthroplasty OutcomesMedicine3 citations

TL;DR

This paper proposes a novel adaptive weight learning method for individualized treatment regimes (ITRs) with continuous treatments and multiple, competing outcomes. By jointly estimating patient-specific utility weights and treatment decision rules from observational data, the approach enables data-driven optimization of composite outcomes while supporting inference and variable selection.

ABSTRACT

To promote precision medicine, individualized treatment regimes (ITRs) are crucial for optimizing the expected clinical outcome based on patient-specific characteristics. However, existing ITR research has primarily focused on scenarios with categorical treatment options and a single outcome. In reality, clinicians often encounter scenarios with continuous treatment options and multiple, potentially competing outcomes, such as medicine efficacy and unavoidable toxicity. To balance these outcomes, a proper weight is necessary, which should be learned in a data-driven manner that considers both patient preference and clinician expertise. In this paper, we present a novel algorithm for developing individualized treatment regimes (ITRs) that incorporate continuous treatment options and multiple outcomes, utilizing observational data. Our approach assumes that clinicians are optimizing individualized patient utilities with sub-optimal treatment decisions that are at least better than random assignment. Treatment assignment is assumed to directly depend on the true underlying utility of the treatment rather than patient characteristics. The proposed method simultaneously estimates the weighting of composite outcomes and the decision-making process, allowing for construction of individualized treatment regimes with continuous doses. The proposed estimators can be used for inference and variable selection, facilitating the identification of informative treatment assignments and preference-associated variables. We evaluate the finite sample performance of our proposed method via simulation studies and apply it to a real data application of radiation oncology analysis.

Motivation & Objective

• Address the gap in individualized treatment regime (ITR) research by extending methods to continuous treatment options and multiple, potentially conflicting outcomes.
• Overcome limitations of existing composite utility functions that are often pre-specified and not personalized to patient preferences or clinician expertise.
• Develop a data-driven approach to learn outcome weights that reflect both patient preferences and clinical judgment in observational studies.
• Simultaneously estimate treatment decision rules and outcome weights to optimize individualized treatment strategies with continuous dosing.
• Enable statistical inference and variable selection for identifying key predictors of treatment response and preference

Proposed method

• Propose a pseudo-likelihood estimation framework that models treatment assignment as a function of unobserved patient utility, which depends on both patient characteristics and treatment outcomes.
• Simultaneously estimate the weighting function for composite outcomes and the treatment decision rule using a joint likelihood approach.
• Use a quadratic approximation of the log-likelihood to derive asymptotic normality and inference for the estimated parameters.
• Incorporate a flexible, nonparametric weighting function w(θ) that captures individualized preferences through a parameterized transformation of patient covariates.
• Apply empirical process theory and martingale central limit theorems to establish asymptotic properties of the estimators.
• Derive an asymptotic linear expansion for the joint estimator of treatment effect and weight parameters, enabling standard errors and hypothesis testing

Experimental results

Research questions

• RQ1 How can individualized treatment regimes be optimized when treatment is continuous and multiple outcomes (e.g., efficacy and toxicity) must be balanced?
• RQ2 What is a robust, data-driven method to learn patient-specific outcome weights without requiring pre-specified or expert-given utility functions?
• RQ3 Can a joint estimation framework simultaneously learn optimal treatment rules and outcome weights from observational data?
• RQ4 How can statistical inference and variable selection be performed in this joint estimation setting?
• RQ5 What is the finite-sample performance of the proposed method in realistic clinical scenarios with continuous dosing?

Key findings

• The proposed method achieves consistent estimation of both treatment decision rules and outcome weights under mild regularity conditions.
• The joint estimator of treatment effect and weight parameters is asymptotically normal, enabling valid inference and confidence intervals.
• The method supports variable selection through the estimation of the weight function parameters, identifying covariates associated with treatment preference.
• Simulation studies show the method outperforms existing approaches in terms of mean squared error and coverage probability for treatment effect estimation.
• In a real radiation oncology application, the method successfully identified optimal treatment doses that balanced tumor control and toxicity outcomes.
• The method demonstrates robustness to model misspecification and performs well even with moderate sample sizes in finite-sample settings.