A focused review of statistical practices for relating radiation dose-volume exposure and toxicity

Abstract

Purpose: Relating dose-volume histogram (DVH) information to patient outcomes is critical for outcomes research in radiation oncology, but this is statistically challenging. We performed this focused review of DVH toxicity studies to characterize current statistical approaches and determine the need for updated reporting recommendations.

Methods and materials: We performed a focused MEDLINE search to identify studies published in 5 radiation oncology specialty journals that associated dosimetry with toxicity outcomes in humans receiving radiotherapy between 2015 and 2021. Elements abstracted from each manuscript included the study outcome, organs-at-risk (OARs) considered, DVH parameters analyzed, summary of the analytic approach, use of multivariable statistics, goodness-of-fit reporting, completeness of model reporting, assessment of multicollinearity, adjustment for multiple comparisons, and methods for dichotomizing variables. Each study was also assessed for sufficient reporting to allow for replication of results.

Results: The MEDLINE search returned 2,300 studies for review and 325 met the inclusion criteria for the analysis. DVH variables were dichotomized using cut points in 154 (47.4%) studies. Logistic regression (55.4% of studies) was the most common statistical method used to relate DVH to toxicity outcomes, followed by Cox regression (20.6%) and linear regression (12.0%). Multivariable statistical tests were performed in 226 (69.5%) studies; of these, the possibility of multicollinearity was addressed in 47.8% and model goodness-of-fit were reported in 32.6%. The threshold for statistical significance was adjusted to account for multiple comparisons in 41 of 196 (17.1%) studies that included multiple statistical comparisons. Twenty-eight (8.6%) studies were classified as missing details necessary to reproduce the study results.

Conclusions: Current practices of statistical reporting in DVH outcomes suggest that studies may be vulnerable to threats against internal and external validity. Recommendations for reporting are provided herein to guard against such threats and to promote cohesiveness among radiation oncology outcomes researchers.