Rapid Lymphocyte Decline Predicts Development of Severe Radiation-Induced Oral Mucositis

Abstract

<h3>Purpose/Objective(s)</h3> Radiation-induced oral mucositis (RIOM) is a major dose-limiting toxicity in head and neck cancer (HNC) and causes considerable financial burden (estimated at $17,244 per patient). When severe, RIOM is debilitating and can progress to dysphagia, aspiration pneumonia, increased hospitalization and treatment interruptions. Early prediction of patients at risk for severe RIOM (during the first few weeks of RT) provides opportunity for early interventions. This study investigates the role of early on-treatment bloodwork in predicting the eventual development of severe mucositis. <h3>Materials/Methods</h3> We conducted a retrospective analysis of prospectively acquired data for 434 HNC patients who underwent RT ± systemic therapy at our institution from 2015—2019. We collected laboratory data at baseline and at week 3 of RT, including absolute lymphocyte count (ALC), absolute neutrophil count (ANC) and C-reactive protein (CRP). We also recorded the biologically effective dose (BED₁₀) that was delivered. Additional metrics included % ALC decrease (baseline – nadir ALC) and ANC/ALC ratio. Patients were randomly partitioned into training and validation datasets in a 70%/30% split (n = 306/128). In the training dataset, we used univariate and multivariate logistic regression models to identify significant predictors for severe mucositis (defined as CTCAE grade 3 and 4), and established optimal cutoff for each significant covariate via the Youden Index. Then, we used an independent validation dataset to determine sensitivity and specificity associated with these cutoff values. <h3>Results</h3> In total, 279 (64.2%) patients developed severe mucositis, with a median onset at 4.9 weeks after initiation of RT. To achieve early prediction for at-risk patients, we analyzed laboratory measurements during week 3 of RT. After accounting for BED, in the training dataset, we identified %ALC decrease as the only significant predictor in both our univariate (OR 4.10, 95% CI 1.20 – 10.36) and multivariate models (OR 2.98, 95% CI 1.10-8.30) (Table 1). With an optimal cut-off value at 60% ALC decrease, the sensitivity and specificity in the validation dataset were 66% and 74%, respectively, with an AUC of 0.72. CRP demonstrated moderate, albeit not significant, effect size despite limited data. <h3>Conclusion</h3> High-grade RIOM adversely affects patient quality-of-life and poses a high financial burden. Our study demonstrates the practical potential of early %ALC decrease in predicting patients at risk for later development of severe RIOM, which may facilitate early interventions to limit the development of severe RIOM.