Utility based approach in individualized optimal dose selection using machine learning methods

Abstract

The goal in personalized medicine is to individualize treatment using patient characteristics and improve health outcomes. Selection of optimal dose must balance the effect of dose on both treatment efficacy and toxicity outcomes. We consider a setting with one binary efficacy and one binary toxicity outcome.The goal is to find the optimal dose for each patient using clinical features and biomarkers from available dataset. We propose to use flexible machine learning methods such as random forest and Gaussian process models to build models for efficacy and toxicity depending on dose and biomarkers. A copula is used to model the joint distribution of the two outcomes and the estimates are constrained to have non-decreasing dose-efficacy and dose-toxicity relationships.Numerical utilities are elicited from clinicians for each potential bivariate outcome. For each patient, the optimal dose is chosen to maximize the posterior mean of the utility function. We also propose alternative approaches to optimal dose selection by adding additional toxicity based constraints and an approach taking into account the uncertainty in the estimation of the utility function. The proposed methods are evaluated in a simulation study to compare expected utility outcomes under various estimated optimal dose rules. Gaussian process models tended to have better performance than random forest. Enforcing monotonicity during modeling provided small benefits. Whether and how, correlation between efficacy and toxicity, was modeled, had little effect on performance.The proposed methods are illustrated with a study of patients with liver cancer treated with stereotactic body radiation therapy. K E Y W O R D SGaussian process, random forest, utility matrix INTRODUCTIONPrecision medicine is an approach for the treatment of a disease that takes into account individual variability. In many situations, the clinical decision is whether to give a particular treatment or a standard treatment to the patient, and the statistical goal is to identify the subgroup of patients likely to derive benefit from the treatment compared to standard orThis is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Adaptive phase I–II clinical trial designs identifying optimal biological doses for targeted agents and immunotherapies

“…Besides, many other phase I-II clinical trial designs have been proposed to handle more complicated clinical settings for targeted agents and immunotherapies such as the late-onset outcomes, 24,47,48 drug-drug combination, [49][50][51][52] dose schedule, [53][54][55][56][57][58] and personalized medicine. 40,45,46,[59][60][61][62][63][64][65][66] The phase I-II clinical trial designs belong to the class of seamless designs and are dedicated to the early stages of drug development. Many other types of seamless designs, such as the phase II-III design, have also been developed, focusing on the later stage of the drug development, such as the treatment effect confirmation and validation.…”

Artificial Intelligence Uncertainty Quantification in Radiotherapy Applications - A Scoping Review

“…The uncertainty in DL-based dose prediction models could be critical, as it could determine when model-generated plans should be directly accepted, or if manual interventions from physicians and physicists are required to improve plan quality [80]. Surprisingly, in our review there were relatively few manuscripts directly investigating model UQ in dose prediction applications [56,65,72,80,95]. This scarcity is mirrored in outcome prediction research, where only a few studies explored dose-related toxicities [42,70,72,78], as opposed to broader oncologic outcomes like survival.…”

“…Surprisingly, in our review there were relatively few manuscripts directly investigating model UQ in dose prediction applications [56,65,72,80,95]. This scarcity is mirrored in outcome prediction research, where only a few studies explored dose-related toxicities [42,70,72,78], as opposed to broader oncologic outcomes like survival. Naturally, a major challenge in outcome-related research stems from the limited availability of training samples.…”