Abstract
Background
Myocardial infarction (MI) continues to be one of the leading causes of morbidity and death worldwide. The international normalized ratio (INR) to albumin ratio (PTAR), calculated as INR divided by serum albumin, is a composite biomarker reflecting both nutritional status and coagulation abnormalities. While PTAR has shown prognostic relevance across various severe clinical scenarios, its role in forecasting mortality among critically ill MI patients remains insufficiently defined. The objective of this study was to examine the relationship between PTAR and all-cause mortality, and to evaluate its predictive capability through machine learning models.
Methods
Data from 1616 critically ill MI patients were retrieved from Version 3.1 of the MIMIC-IV database. The primary and secondary outcomes were 30- and 360-day all-cause hospital mortality. The relationship between PTAR and mortality was evaluated using multivariate Cox regression models, and nonlinear associations were further explored through restricted cubic spline (RCS) analysis. Survival analysis was performed using the Kaplan-Meier method. Feature variables were selected through the Boruta algorithm, and eight machine learning models were constructed to evaluate predictive performance.
Results
Elevated PTAR levels were independently associated with increased risks of mortality at both 30 and 360 days. In Cox models with full adjustment, participants in the highest PTAR tertile showed a substantially higher risk of mortality relative to those in the lowest category (HR = 2.09, 95% CI: 1.53–2.86 for 30-day mortality; HR = 2.15, 95% CI: 1.69–2.73 for 360-day mortality). RCS analysis demonstrated that the mortality risk increased in a nonlinear fashion with rising PTAR values. Subgroup analyses revealed consistent predictive value across all subgroups. The Boruta algorithm ranked PTAR among the top predictors of mortality. Comprehensively considering the AUC value, calibration curve, and decision curve, the predictive model using LightGBM demonstrated the best performance (AUC = 0.7861).
Conclusion
Our study revealed a strong positive association between the PTAR and the mortality risk in critically ill patients with MI. Higher PTAR is associated with greater mortality risk. Predictive models based on machine learning demonstrated good performance. These findings suggest that PTAR may serve as a potential predictor of adverse outcomes in critically ill MI patients.