The prognostic value of the PALBI score for mortality prediction in stroke: an internal and external validation study

Recent studies show that there's a link between liver problems and how well someone does after having a stroke. The platelet-albumin-bilirubin (palbi) score is a new liver function evaluation method which has been proved useful in predicting hepatic condition. But it is still unknown whether it can predict the prognosis of stroke people. This study aimed to investigate the association between the PALBI score and clinical outcomes in stroke patients. We screened out severely ill stroke patients in MIMIC-IV by ICD-9/10 criteria, then divided them into PALBI score quartiles. The primary outcome was 1-year mortality, and the secondary outcome was in-hospital mortality. Cox proportional hazards regression and logistic regression analyzed associations with 1-year and in-hospital mortality, respectively. RCS curves showed the relationship between dose and response, and K-M survival curves were drawn. Subgroup analysis and interaction tests were performed for robustness. Random Forest and SVM machine learning models were built to rank feature importance and compare discrimination ability with the standard logistic regression model. Also, an external validation was done with the eICU Collaborative Research Database (eICU-CRD) to verify the results. A total of 2,265 patients were included (57.75% male). 1 year mortality rate was 35.01%, and mortality rate during hospital stay was 15.94%. Multivariate Cox regression revealed that a higher PALBI score had a significant impact on an increased risk of 1 year overall mortality (aHR: 1.42, 95%CI: 1.23-1.65, p < 0.001). Multivariate logistic regression showed that higher PALBI scores correlated with a greater chance of dying during a hospital stay (aOR: 1.35, 95% CI: 1.03-1.77, p = 0.01). RCS analysis shows that the risk of death increases gradually as the level of PALBI rises. Machine learning models consistently ranked the PALBI score as the second most important predictor, following the Charlson Comorbidity Index, and its inclusion improved the AUC of the Random Forest model from 0.707 to 0.724 in the validation set. Using eICU-CRD for external validation found a strong connection between high PALBI scores and higher in-hospital mortality rates (continuous variable, aOR: 1.35; 95% CI: 1.107-1.663; P = 0.003). Our findings indicate that there is a significant relationship between the PALBI score and 1-year overall mortality as well as in-hospital mortality for people with stroke. The same results in MIMIC-IV and eICU database show that the PALBI score may have some use clinically to find out which stroke patient has worse results.