Construction of a near-term predictive model for irAEs induced by PD-1 inhibitors.

3147 Background: Immune checkpoint inhibitors have opened a new chapter in cancer therapy, but the incidence of irAEs caused by them is high, and severe irAEs can be fatal. The current research on irAEs is almost focused on early predictions, and there is a lack of near-term predictions (the cycle before the occurrence of irAEs). Absolute eosinophil count (EO#) has been reported to be associated with immune-related pneumonia, but its association with other systemic irAEs requires further exploration. The aim of this study was to explore the near-term predictive value of neutrophil/lymphocyte (NLR), platelet/lymphocyte (PLR), and EO# for PD-1 inhibitor-induced irAEs. Methods: The data are from tumor patients who received PD-1 inhibitor therapy in our department from July 2019 to May 2021. A total of 146 cases were included, of which 56 had irAEs. The data of NLR, PLR and EO# in the cycle before the occurrence of irAEs (the median number of cycles was the second cycle) were collected, and the data of the second cycle was used as the control for patients without irAEs group. Logistic method was used to analyze the correlation between NLR, PLR and EO# and irAEs, and a predictive model was constructed. The sensitivity and specificity of the model were evaluated by ROC curve. This study was registered on Chinese Clinical Trail Registry (ChiCTR2100049849). Results: A total of 146 tumor patients were included, of which 56 developed at least one irAEs. Grade 1-2 irAEs occurred in 39 cases, grade 3-4 in 12 cases (including cardiac, liver, lung and skin toxicity), grade 5 in 2 cases(including cardiac and lung toxicity), and ungraded in 3 cases. The data of the cycle before the occurrence of irAEs were analyzed. Univariate analysis showed that NLR (odds ratio [OR], 1.4, p< 0.05) and EO# (OR, 12.6, p< 0.05) were associated with irAEs, and multivariate analysis suggested NLR (OR, 1.7, p< 0.001) and EO# (OR, 20.4, p< 0.05) were independent risk factors for irAEs. The prediction model composed of NLR, PLR and EO# had a correct rate of 76.7% (AUC = 0.752) in predicting the occurrence of irAEs in the near-term cycle, with a sensitivity of 51.8% and a specificity of 92.2%; the correct rate of predicting irAEs of grade 3 and above was as high as 91.9% (AUC = 0.778), the sensitivity was 14.3% and the specificity was 99.2%. Conclusions: The model composed of NLR, PLR and EO# may predict the occurrence of irAEs in the near-term cycle, especially the prediction of irAEs above grade 3, which can provide early warning for the occurrence of irAEs. Clinical trial information: ChiCTR2100049849.