Integrating Serum Biomarkers and Dose–volume Metrics to Predict Radiation Pneumonitis

Abstract

Published reports have correlated the development of radiation pneumonitis (RP) among NSCLC patients to various biomarkers, dose–volume (D-V) metrics, and anatomic factors. This study aims to improve predictive models for the development of RP by integrating multiple serum biomarkers and D-V metrics in patients receiving radiotherapy for NSCLC. In the present study, we examine the role serum biomarkers play in predicting pneumonitis and the potential for combining these markers with anatomic information gathered from dose–volume assessment. We analyzed a cohort of 19 inoperable Stage III NSCLC patients accrued prospectively, who received radiotherapy treatment with a mean follow-up of 1 year. Three events were identified as RP (RTOG Grade ≥3). Serum was collected from the patients immediately prior to RT, during (at 3 weeks), at the conclusion of RT, as well as at 3-months post-RT. Concentrations of three potential radiation pneumonitis biomarkers, namely angiotensin converting enzyme (ACE), transforming growth factor-β (TGF-β), and interleukin-6 (IL-6), were determined using enzyme-linked immunoassays. Previously reported D-V metrics, including mean dose to both lungs minus the GTV and GTV location in the superior-inferior direction, were also extracted from the patients' radiation therapy planning CT scans. Models based on logistic regression, combining the candidate biomarkers and a previous dose–volume model based on the mean dose and GTV position (Bradley et al., IJROBP 2007), were evaluated. Statistic association was measured using Spearman's rank correlation (Rs). None of the markers (biologic or dosimetric) was individually significant on univariate analysis. With multivariate logistic regression model building, however, models based on adding either normalized midtherapy ACE (Rs = 0.560; p = 0.008) or IL-6 levels (Rs = 0.730; p = 0.013) to the previously reported mean lung dose and GTV position metrics (the Bradley model) were significant. A model based on the midtherapy TGF-β concentration, added to the Bradley model, but was less predictive (Rs = 0.346; p = 0.067). Improved significance was achieved by including both midtherapy ACE and IL-6 values with the Bradley model (Rs = 0.845; p = 0.004). On leave-one-out cross validation, the predictive Rs improved from 0.18 to 0.34 by adding ACE and IL-6 to the Bradley model. In our preliminary analysis, we show that physical and biologic markers can be combined to improve the prediction of RP, with an 89% improvement in prediction ability when multiple biomarkers (IL-6 and ACE) are combined with dose–volume information. However, analysis on larger datasets will be required to elucidate and verify these observations.