Predictive value of near-term prediction models for severe immune-related adverse events in malignant tumor PD-1 inhibitor therapy

Immune-related adverse events (irAEs) impact outcomes, with most research focusing on early prediction (baseline data), rather than near-term prediction (one cycle before the occurrence of irAEs and the current cycle). We aimed to explore the near-term predictive value of neutrophil/lymphocyte ratio (NLR), platelet/lymphocyte ratio (PLR), absolute eosinophil count (AEC) for severe irAEs induced by PD-1 inhibitors. Data were collected from tumor patients treated with PD-1 inhibitors. NLR, PLR, and AEC data were obtained from both the previous and the current cycles of irAEs occurrence. A predictive model was developed using elastic net logistic regression Cutoff values were determined using Youden's Index. The predicted results were compared with actual data using Bayesian survival analysis. A total of 138 patients were included, of whom 47 experienced grade 1-2 irAEs and 18 experienced grade 3-5 irAEs. The predictive model identified optimal alpha and lambda through 10-fold cross-validation. The Shapiro-Wilk test, Kruskal-Wallis test and logistic regression showed that only current cycle data were meaningful. The NLR was statistically significant in predicting irAEs in the previous cycle. Both NLR and AEC were significant predictors of irAEs in the current cycle. The model achieved an area under the ROC curve (AUC) of 0.783, with a sensitivity of 77.8% and a specificity of 80.8%. A probability >= 0.1345 predicted severe irAEs. The model comprising NLR, AEC, and sex may predict the irAEs classification in the current cycle, offering a near-term predictive advantage over baseline models and potentially extending the duration of immunotherapy for patients.