Systematic analysis of biological endpoint variability and implications for quantitative modeling of the FLASH sparing effect

TL;DR

This paper reviews preclinical FLASH radiotherapy studies to identify inconsistencies in biological endpoints and suggests ways to improve data comparability for better modeling.

Contribution

The paper systematically quantifies endpoint variability in FLASH studies and identifies a subset of comparable endpoints for modeling.

Findings

Only four endpoints were found to be amenable to cross-study analysis.

High collinearity among parameters limited logistic regression analysis.

Standardization of endpoints and methodologies is urgently needed in FLASH research.

Abstract

•Systematically reviews in vivo FLASH studies and quantifies endpoint variability.•Identifies comparable endpoints suitable for cross-study analysis and modeling.•Reveals key knowledge gaps and strategies to improve data comparability.

Systematically reviews in vivo FLASH studies and quantifies endpoint variability.

Identifies comparable endpoints suitable for cross-study analysis and modeling.

Reveals key knowledge gaps and strategies to improve data comparability.

Ultra-high dose rate (UHDR) radiotherapy shows promise in sparing normal tissue while maintaining tumor control. However, heterogeneity across preclinical experiments limits cross-study comparability and the development of predictive models. By employing both qualitative and quantitative analyses, we aim to identify variability in endpoints, clarify inconsistencies, and highlight knowledge gaps critical for future quantitative modeling.

We reviewed in-vivo proton and electron FLASH murine experiments (1966–2024) and extracted commonly reported endpoints. A subset of comparable endpoints—crypt cell ratio and survival (abdominal), maximum moist desquamation (skin), and Novel-Object-Recognition tests (brain)—was analyzed using qualitative comparison and logistic regression to assess parameters potentially driving the FLASH effect.

Fifty papers were reviewed, revealing substantial differences in endpoint selection, timing, and methodology. Most endpoints were not directly comparable across studies, with only four endpoints amenable to cross-study analysis. Logistic regression was constrained by high collinearity among parameters, reflecting both experimental heterogeneity and limited harmonization.

Our review highlights the urgent need for standardization in biological endpoints and methodologies in FLASH research. By quantifying variability in preclinical FLASH data and identifying comparable endpoints, our work provides essential guidance for future modeling efforts, bridging biological evidence and quantitative approaches for treatment planning applications.

Figures