To construct a machine learning model based on serum biomarkers for predicting the risk of mechanical ventilation in patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD) and analyze its predictive performance.

A total of 62 AECOPD patients treated in our hospital from January to December 2023 were enrolled. Among them, 33 patients without mechanical ventilation were assigned to the control group, and 29 patients requiring mechanical ventilation were assigned to the observation group. An XGBoost model was developed to predict the risk of mechanical ventilation in AECOPD patients based on serum biomarkers. The predictive performance was evaluated using the receiver operating characteristic (ROC) curve, and the model′s predictive value was validated by comparing it with individual serum biomarkers.

High-sensitivity C-reactive protein(hs-CRP)46.30 (16.55, 104.20) mg/L, albumin(ALB)36.90 (31.05, 41.65) g/L, direct bilirubin(DBIL) 4.00 (3.05, 5.85) μmol/L, total cholesterol(TC) 3.83 (3.12, 4.30) mmol/L, andapolipoprotein A1(APOA1) (1.17±0.32) g/L in the observation group showed statistically significant differences compared to the control group in hs-CRP 7.30 (1.05, 18.25) mg/L), ALB 40.90 (36.85, 44.20) g/L, DBIL 2.90(2.45, 3.85) μmol/L), TC 4.33 (3.72, 4.90) mmol/L and APOA1 (1.46±0.29) g/L (P<0.05). The XGBoost model achieved an area under the ROC area under curve (AUC) of 0.831, sensitivity of 71.29%, specificity of 90.00%, accuracy of 75.81%, and an F1-score of 0.933, outperforming individual biomarkers such as hs-CRP, ALB, DBIL, TC, HDL, and APOA1. The calibration curve demonstrated good model fit with a calibration score of 0.075.

The machine learning model based on serum biomarkers provides a more accurate prediction of mechanical ventilation risk in AECOPD patients, aiding clinicians in developing more effective treatment strategies.