Skeletal Muscle Index As a Predictor of Clinical Outcomes In Head and Neck Radiotherapy: Preliminary Results of a Retrospective cohort study
Abstract
Purpose
Low muscle quantity, defined by computed tomography (CT)-based skeletal muscle index (SMI), is emerging as a predictor of clinical outcomes in patients with HNC. This study aims to evaluate whether CT-defined SMI is associated with all-cause mortality in patients with HNC.
Methods
This is an ongoing retrospective study of patients with HNC treated with definitive or adjuvant RT from January 2014 until October 2023. A pre-trained deep-learning model was used to segment and quantify the cross-sectional area (cm2) of the skeletal muscle at the third cervical level (C3), which was then divided by the square of the patient’s height in meters to derive the SMI. Demographic-, disease-, and treatment-related variables were extracted from medical records. Univariate Cox regression analyses were performed to assess the associations between baseline characteristics and all-cause mortality.
Results
In this preliminary analysis, n=50 patients (mean age: 61) were included, of whom 10 were females, and 35 had received definitive RT. Of those 50 patients, 14 died. Cox regression identified no statistically significant predictors of all-cause mortality, likely attributable to limited statistical power at this stage. However, notable trends emerged. Higher BMI showed a borderline association with improved survival (HR 0.93, 95% CI 0.85-1.01, p=0.07), while lower SMI per 5 units trended toward an increased risk of mortality (HR 1.93, 95% CI 0.76-4.88, p=0.16). Age, disease stage, and tobacco use were not significant predictors in this limited sample. As this is an ongoing study, continued data collection and expansion of the cohort size are anticipated to provide the necessary power to validate these associations in future analyses.
Conclusion
The findings of this study will shed light on the prognostic value of radiographic SMI for clinical outcomes in patients with HNC, which may aid risk stratification while enabling more personalized monitoring strategies and improved adaptive RT workflows.