# Association Between CBC‐Derived Inflammatory Indicators and 28‐Day Mortality in Patients With Coronary Heart Disease and Diabetes Mellitus: A Cohort Study From the MIMIC‐IV Database

## TL;DR

This study finds that blood count-based inflammation indicators can predict 28-day mortality in patients with heart disease and diabetes, offering a low-cost way to assess risk.

## Contribution

The study identifies six CBC-derived inflammatory indices as independent predictors of mortality in comorbid CHD-DM patients.

## Key findings

- All six CBC-derived indices showed dose-dependent associations with 28-day mortality.
- PLR and MLR were ranked as the top predictive indices by SHAP analysis.
- The indices demonstrated good discrimination and excellent calibration in both training and validation cohorts.

## Abstract

Coronary heart disease (CHD) remains the leading global cause of death; concomitant diabetes mellitus (DM) doubles short‐term mortality, largely through chronic low‐grade inflammation. Inexpensive, bedside complete blood count (CBC)‐derived inflammatory indices (NLR, MLR, PLR, SII, SIRI, and AISI) predict outcomes in CHD or DM alone, but their utility in comorbid patients is unclear.

Retrospective cohort study of CHD‐DM patients from Medical Information Mart for Intensive Care‐IV (MIMIC‐IV; 2008–2022), split into training (2017–2022, n = 1607) and validation (2008–2016, n = 1145) sets. Six indices (NLR, MLR, PLR, SII, SIRI, and AISI) were calculated from initial ICU CBC (48‐h mean for sensitivity analysis). Primary outcome: 28‐day mortality, Cox regression, restricted cubic splines (RCSs), receiver operating characteristic (ROC) curves, calibration plots, and SHAP‐based variable importance were used.

Mortality was 10.6% (training) and 11.9% (validation). All indices showed independent, dose‐dependent associations with mortality (e.g., training MLR per 1‐SD: HR = 1.49, 95% CI = 1.37–1.61), discrimination was good (training AUC 0.767, C‐index 0.752; validation AUC 0.755, C‐index 0.746), and calibration was excellent. Spearman correlation showed moderate‐to‐strong interindex correlations (e.g., MLR‐SIRI: r = 0.84). SHAP analysis ranked PLR and MLR as top predictive indices. Sensitivity analysis confirmed robustness.

Six CBC‐derived indices independently predict 28‐day mortality in critically ill CHD‐DM patients, with PLR and MLR showing superior predictive weight, and can be used for rapid, cost‐free bedside risk stratification.