Discussion

This study showed that longitudinal CT-derived biomarkers provided prognostic information in NSCLC patients treated with curative-intent radiotherapy. Severe skeletal muscle loss and LV myocardial remodelling, including both atrophy and hypertrophy, were associated with poorer survival. Low-dose thoracic dose-volume metrics were also associated with these adverse changes, suggesting potential nutritional and cardiopulmonary pathways that are discussed below.

Our results were broadly concordant with, and extend, existing literature. Al-Sawaf et al. [4] (n = 188) reported that >20% of visceral fat or > 10% of SKM loss between diagnosis and relapse was associated with the poorer OS in NSCLC patients. Chaunzwa et al. [19] (n = 1791) measured muscle and fat changes before treatment and 4–8 weeks after in metastatic NSCLC patients. They observed that SKM loss, but not fat loss, correlated with survival outcomes. Similarly, we found that SKM loss predicted worse OS and further showed a non-linear SKM muscle-loss–mortality relationship. Oesophageal V10Gy was associated with skeletal muscle loss, which may reflect an indirect nutritional pathway rather than a direct radiobiological effect on skeletal muscle. Low-dose oesophageal irradiation may contribute to oesophagitis, reduced intake, and subsequent muscle loss, supporting the hypothesis that treatment-related side effects may partly drive post-treatment body-composition changes [20].

Although L3 is the standard level for CT-based body composition assessment, we used L1 because routine thoracic follow-up CT scans did not consistently include L3 [21]. L1 has been used as a pragmatic alternative in lung-cancer cohorts for survival analyses [22], [23], [24]. Respiratory-phase differences may have introduced variability in automated L1 segmentation because dedicated respiratory-motion correction was not applied. However, estimating baseline-normalised velocity across multiple time points may reduce random inter-scan variability. Therefore, our findings should be interpreted as L1-derived longitudinal body-composition biomarkers.

For LV myocardial mass loss, aorta V5Gy was significant, whereas heart myocardium V10Gy showed only a non-significant trend (p = 0.09). Zhang et al. linked low-dose LV myocardial exposure, particularly V10Gy, to declines in left ventricular ejection fraction [25]. However, the biological plausibility of low-dose exposure to the aorta or myocardium causing ventricular atrophy remains unclear. These associations may reflect indirect radiation effects, vascular injury, systemic illness, or residual confounding, and should therefore be interpreted cautiously [26].

In the cardiac-hypertrophy subgroup, higher dose per fraction and increased right atrium V10Gy were associated with LV myocardial hypertrophy. This finding may be relevant because the cohort included SBRT patients treated with different fractionation schemes. These findings are consistent with preclinical data showing that a single high dose (up to 8Gy) can induce significant myocardial fibrosis in animal models, whereas the same total dose delivered in smaller fractions appears less harmful to cardiomyocytes [27]. However, this result should be interpreted cautiously because the cardiac-hypertrophy subgroup was small and the confidence intervals were wide. Two representative cases in Supplementary material H showed reduced left lung volume with cardiac enlargement or displacement after treatment. Therefore, right atrium V10Gy may reflect broader cardiopulmonary irradiation rather than directly causing LV myocardial hypertrophy. Some changes may also reflect cardiac motion or segmentation uncertainty because the CT scans were not ECG-gated.

Methodologically, patient-specific linear slopes and cubic spline Cox models accounted for variable scan timing and potential non-linear associations with survival. Routine ungated CT images increased clinical relevance but introduced measurement error. Patients with body-composition or myocardial change velocities close to the bootstrapped cut-off values should therefore be interpreted with caution, particularly when values fall within the confidence intervals of these thresholds. Additional clinical information, such as ECG findings, blood tests, and echocardiography, may support interpretation. Future validation using ECG-gated CT is required to assess individual-level agreement and the stability of these categories [28], [29], [30].

The prognostic value of the vmyo andvskm may have several clinical implications. These imaging biomarkers may reflect downstream effects of nutritional status, treatment-related side effects, physical activity, and cardiopulmonary stress. They may help identify patients who could benefit from early supportive interventions, including dietary counselling, exercise programmes, or targeted cardiac surveillance.

This study has several limitations. First, it was a single-centre retrospective analysis, which may limit generalisability. Second, important non-imaging factors were not available, including ECOG performance status, cardiac comorbidities, chemotherapy details, and immunotherapy use. These missing data were potential sources of bias because they may have influenced both survival and longitudinal changes in body composition or cardiac morphology. The relatively high mean age of the cohort may have introduced further bias, as older patients are more likely to have frailty, sarcopenia, cardiovascular disease, and other comorbidities at baseline. Third, follow-up CT scans occurred at variable times, and the requirement for multiple scans may have introduced survivor bias. Fourth, LAD coronary artery dose was not assessed because TotalSegmentator did not provide LAD contours, which limited assessment of cardiac substructure dose effects. Finally, the dose-based models showed only moderate discrimination, with AUC values of approximately 0.6–0.7. Therefore, the reported dose-volume metrics should be regarded as candidate planning parameters for further investigation rather than definitive clinical thresholds. Prospective multi-centre validation with systematic clinical data collection is needed.

This study shows that longitudinal CT-derived biomarkers provide prognostic information in NSCLC patients treated with curative-intent radiotherapy. Skeletal muscle loss and LV myocardial remodelling were associated with poorer survival. These routine CT-based measures may help identify vulnerable patients and generate dose-planning hypotheses, but prospective validation is needed before clinical implementation.