Biomarker screening and diagnostic model construction for radiation-induced heart disease based on metabolomics

Radiation-induced heart disease (RIHD) is a serious complication following radiotherapy for thoracic malignancies. Early diagnosis is challenging due to the lack of specific biomarkers. This study aimed to screen potential plasma biomarkers for RIHD using metabolomics and construct high-accuracy diagnostic models. A total of 57 RIHD patients and 22 healthy controls (HC) were enrolled. Untargeted metabolomic profiling of plasma samples was performed using gas chromatography–mass spectrometry (GC–MS). Differential metabolites were screened by combining multivariate statistical analysis and univariate analysis. Metabolic pathway enrichment analysis was conducted. Three machine learning models—Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and Random Forest (RF)—were constructed based on the identified biomarkers, and their diagnostic performance was evaluated. A total of 44 differential metabolites (6 upregulated, 38 downregulated) were identified between the RIHD and HC groups. These metabolites were primarily involved in phenylalanine/tyrosine/tryptophan biosynthesis, arginine biosynthesis, the TCA cycle, and glycine/serine/threonine metabolism. Variable importance analysis based on random forest algorithms identified key metabolites such as lysine and cysteine as significant contributors to group separation. The Random Forest diagnostic model demonstrated the best performance in the test set, achieving an area under the curve (AUC) of 0.872 (95% CI: 0.666–1), with a sensitivity of 1.00 and a specificity of 0.76. This study reveals distinct plasma metabolic profile disturbances in RIHD patients and identifies key metabolite biomarkers through variable importance analysis. The RF diagnostic model, constructed using these key metabolites, shows high accuracy and potential for clinical application, offering a novel strategy for the early diagnosis of RIHD.