Early prediction of skeletal muscle loss using longitudinal clinical data in patients with gastric cancer after radical gastrectomy and adjuvant chemotherapy: a retrospective cohort study

Gastric cancer patients frequently experience skeletal muscle loss during the perioperative and adjuvant treatment period, which has been associated with poorer treatment tolerance and adverse clinical outcomes. Early identification of patients at high risk of skeletal muscle loss may allow timely supportive intervention, but repeated computed tomography assessment is not always practical in routine care. This study aimed to develop an interpretable machine learning model based on routinely available clinical data for early prediction of significant skeletal muscle loss in patients with gastric cancer. This single-center retrospective study screened 362 patients who underwent radical gastrectomy followed by adjuvant chemotherapy, of whom 292 were finally included. Significant skeletal muscle loss was defined as a decrease of at least 5% in skeletal muscle index between the baseline scan performed before surgery and the follow-up scan obtained 3 months after initiation of adjuvant chemotherapy. Candidate predictors included demographic, clinicopathological, laboratory, tumor marker, and inflammatory or nutritional variables, together with their early postoperative dynamic changes. Six machine learning models were developed and compared. Among the evaluated models, the multilayer perception showed the best overall performance in the validation set, with an area under the receiver operating characteristic curve of 0.757 and an area under the precision-recall curve of 0.745. At the selected decision threshold of 0.45, this model achieved an accuracy of 0.693, a recall of 0.833, and a specificity of 0.525. Compared with the model using baseline variables alone, the model incorporating longitudinal dynamic features showed better overall performance. Model interpretation suggested that prediction of skeletal muscle loss was mainly related to nutritional reserve, operation-related burden, and inflammatory or metabolic status. These findings indicate that routinely available preoperative and early postoperative clinical data can support early prediction of subsequent skeletal muscle loss in gastric cancer. This approach may help identify high-risk patients earlier and facilitate individualized nutritional support and supportive care during treatment.