Metabolomic and machine learning identify predictive plasma biomarkers for sintilimab-induced rash

Sintilimab is a crucial immunotherapy for lung cancer, but adverse events like rash often lead to treatment interruption or discontinuation, compromising patient outcomes. The precise mechanisms driving these sintilimab-induced rashes are poorly understood, hindering proactive management. Identifying predictive biomarkers is critical to personalize treatment strategies, mitigate side effects, and improve therapeutic adherence, ultimately enhancing the efficacy of immunotherapy in this vulnerable patient population. Current clinical markers are insufficient for early prediction.

Researchers enrolled 55 lung cancer patients receiving sintilimab, comprising 32 who developed rash and 23 matched controls without rash. Blood samples were collected before sintilimab infusion and at rash onset. Untargeted metabolomic analysis of plasma was performed using ultra-high-performance liquid chromatography-tandem mass spectrometry (UHPLC-MS/MS). Differential metabolites were identified via pathway enrichment, univariate analysis (AUC ≥0.800), and SHAP analysis to pinpoint predictive biomarkers.

No significant demographic differences were observed between groups. However, the rash group showed significantly elevated total bile acids, glucose (GLU), and basophil percentage (BAS%), alongside reduced AST/ALT ratio, alkaline phosphatase, lactate dehydrogenase, phosphorus (P), neutrophil count (NEU), and high-sensitivity C-reactive protein (hsCRP) (all P < 0.05). Metabolomic analysis identified 92 differentially expressed metabolites. Pathway enrichment highlighted alterations in oxytocin signaling, GnRH signaling, platelet activation, FcγR-mediated phagocytosis, retrograde endocannabinoid signaling, pantothenate and CoA biosynthesis, FcεRI signaling, and aldosterone synthesis and secretion. > Univariate analysis identified 25 metabolites with high predictive value (AUC ≥0.800), and SHAP analysis pinpointed 20 metabolites, with 5 overlapping: N,N,N-trimethyl-L-histidine, laurolactam, 2-naphthalenesulfonic acid, and limonenecarboxyl.