Relationship between Pan-immuno-inflammatory value and hospitalized all-cause mortality in ischemic stroke patients: a retrospective cohort study and predictive modeling

Background Systemic inflammation and immune response are major factors in the development and progression of ischemic stroke (IS). Numerous studies have shown how Pan-immuno-inflammatory value (PIV) affects the chance of dying from a serious disease. However, the value of PIV in IS patients in the ICU remains unclear. The objective of this study was to explore the correlation between PIV and IS, and to construct a machine learning (ML) model for in-hospital mortality risk in IS patients using variables related to PIV.Methods In the present study, patients who had been diagnosed with IS and admitted to the ICU were retrospectively pulled from the publicly accessible MIMIC-IV v3.0 database. The primary result was defined as in-hospital mortality. To investigate the association between Log2-transformed PIV and clinical outcomes in IS patients, a Cox proportional hazards regression using with restricted cubic splines (RCS) was undertaken. The optimum model within the validation cohort was chosen based on accuracy and area under the curve (AUC). Furthermore, the SHAP method was utilized to determine the significance of model features and assess the influence of the top three characteristics on model predictions.Results The research included 2,223 participants with IS. The connection between the probability of in-hospital mortality in IS and Log2PIV was nonlinear. Among the 14 ML algorithms, the GBDT model has higher prediction accuracy, better clinical decision-making performance, and better overall performance. Furthermore, the SHAP algorithm analysis revealed that Log2-PIV, Hemoglobin, and LODS were the three clinical characteristics that most significantly influenced the GBDT model's outputs.Conclusion The study findings revealed a U-shaped association between Log2-PIV and the risk of in-hospital mortality from IS. Furthermore, the GBDT model emerged as the most effective predictor, enabling clinicians to pinpoint high-risk patients and take proactive measures to minimize mortality.