Construction and validation of a prediction model for developing type 2 diabetes mellitus in patients with chronic obstructive pulmonary disease.

BACKGROUND

Type 2 diabetes mellitus (T2DM) is a common comorbidity of chronic obstructive pulmonary disease (COPD), which significantly increases the risk of rehospitalization and mortality in patients with COPD. Therefore, the purpose of this study was to identify the influencing factors of COPD complicated by T2DM and to construct a visualized disease prediction model.

METHOD

We included the medical records of 1,773 patients with COPD treated at Quzhou People's Hospital from 2020 to 2023. Subjects were randomly divided into a training set (n = 1,241) and a test set (n = 532) in a 7:3 ratio. Variable selection was performed using the least absolute shrinkage and selection operator (LASSO), Pearson correlation, and multicollinearity diagnostics. Variables were then refined through backward stepwise selection based on the Akaike Information Criterion (AIC) to construct a nomogram. The accuracy of the nomogram was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and the Hosmer-Lemeshow test (H-L test). The clinical utility of the model was evaluated using decision analysis curves (DCA). Additionally, k-fold cross-validation (k = 10) was performed to rigorously assess model stability and mitigate the risk of overfitting. A sex-stratified subgroup analysis was also conducted to address potential sex-related bias.

RESULTS

The prevalence of T2DM in COPD patients was 27.13%. Seven independent predictors of COPD complicated by T2DM were identified: arterial partial pressure of carbon dioxide (PCO) (OR = 1.04, 95%CI: 1.02-1.05), neutrophil number (NEUT) (OR = 1.15, 95%CI: 1.10-1.19), C-reactive protein (CRP) (OR = 1.01, 95%CI: 1.01-1.02), erythrocyte sedimentation rate (ESR) (OR = 1.03, 95%CI: 1.02-1.05), bilirubin (OR = 0.92, 95%CI: 0.88-0.96), triglyceride (TG) (OR = 1.33, 95%CI: 1.13-1.56), and body mass index (BMI) (OR = 1.16, 95%CI: 1.11-1.20). The model demonstrated good predictive performance, with a C-index of 0.78. The area under the curve (AUC) values were 0.79 (95%CI: 0.76-0.81) for the training set and 0.80 (95%CI: 0.76-0.84) for the test set, consistent with the k-fold cross-validation average AUC of 0.79 (95%CI: 0.76-0.81). Calibration curves and the H-L test (0.05) indicated good agreement between predicted and observed outcomes. DCA curves demonstrated clinical utility across threshold probabilities. Subgroup analysis showed robust performance in both male (0.82, 95%CI: 0.77-0.86) and female (0.71, 95%CI: 0.60-0.83) groups, with no significant difference in discriminatory ability (DeLong = 0.101).

CONCLUSION

In this study, we developed and internally validated a visualized prediction model for early identification of T2DM risk in patients with COPD. This tool may facilitate targeted prevention strategies by identifying high-risk populations. While the model demonstrated good performance, external validation is still required to confirm its generalizability.