# Development and external validation of an admission-based model for in-hospital mortality in acute exacerbation of COPD: incremental prognostic value of type 2 diabetes mellitus

## TL;DR

This study shows that type 2 diabetes adds important risk information for predicting in-hospital death in patients with acute COPD flare-ups.

## Contribution

Demonstrates T2DM provides incremental prognostic value for AECOPD mortality beyond standard physiological markers in a validated model.

## Key findings

- T2DM patients had higher in-hospital mortality during AECOPD than non-diabetic patients.
- A six-variable model including T2DM showed strong discrimination (AUC 0.843) and good calibration after recalibration.
- The nomogram provides clinically meaningful risk stratification across 5–15% threshold probabilities.

## Abstract

Acute exacerbation of chronic obstructive pulmonary disease (AECOPD) carries substantial short-term mortality. Whether type 2 diabetes mellitus (T2DM) provides incremental prognostic information for in-hospital death during AECOPD, beyond acute physiological decompensation, remains incompletely defined.

We conducted a multicenter retrospective cohort study of consecutive inpatients admitted for AECOPD at three tertiary hospitals (June 2022–December 2024). A multivariable logistic model was developed in the training cohort using routinely available variables obtained early after admission, with feature selection by LASSO and performance evaluation by discrimination, calibration, and decision curve analysis. External validation was performed in an independent hospital cohort with intercept/slope recalibration when indicated.

Among 4,292 patients (training n=2,861; external n=1,431), patients with T2DM had higher in-hospital mortality than those without T2DM. In the multivariable model, T2DM contributed incremental prognostic information for in-hospital death (adjusted OR = 2.74, 95% CI 1.62–4.56), together with PaCO2, blood urea nitrogen, neutrophil-to-lymphocyte-to-albumin ratio, C-reactive protein, and age. The resulting six-variable nomogram showed strong discrimination (AUC = 0.843 training; 0.817 external), low overall prediction error (external Brier≈0.025), and clinically meaningful net benefit across 5–15% threshold probabilities; calibration in the external cohort was improved to near-ideal after recalibration.

In hospitalized AECOPD, T2DM provides clinically relevant incremental prognostic information for short-term in-hospital mortality within a parsimonious multivariable model. The nomogram may facilitate early risk stratification and support integrated respiratory and metabolic co-management.