Abstract:

Objective To establish and validate an early risk warning model for the potential high-risk population of chronic obstructive pulmonary disease (COPD) progressing to type II respiratory failure. Methods Stratified random sampling method was used to divide 297 COPD patients who met the inclusion criteria after acute hospitalization into stable stage into the modeling set (n=208) and the validation set (n=89) according to the ratio of 7∶3. The patients in the modeling set were divided into the progressive group (n=81) and the non-progressive group (n=127) according to whether the re-acute exacerbation progressed to type II respiratory failure. The clinical data of the progressive group and the non-progressive group were compared. The random forest was used to preliminarily screen the predictive feature variables. Least absolute shrinkage and selection operator（LASSO） regression was used to further compress and screen the important predictive feature variables, and the nomogram early warning recognition model was constructed and verified. Results There were no significant differences in age, gender, body mass index (BMI), past history, stable treatment plan, course of disease, Global Initiative for Chronic Obstructive Lung Disease (GOLD) lung function classification, number of acute exacerbations in the past year, sarcopenia and laboratory examination indexes [white blood cells, hemoglobin, platelets, neutrophils, eosinophils（EOS), albumin and pulmonary surfactant protein-D（SP-D)] between the modeling set and the validation set. The age, GOLD lung function classification, proportion of patients with acute exacerbation ≥ 2 times in the past year, proportion of patients with sarcopenia, EOS and SP-D were higher in the progressive group than those in the non-progressive group, the course of disease was longer than that of the non-progressive group, and the albumin level was lower than that in the non-progressive group (P<0.05). Based on this random forest, the top six important characteristic variables were course of disease, GOLD lung function classification, the number of acute exacerbations in the past year, sarcopenia, EOS and SP-D. After further compression by LASSO, GOLD lung function classification, the number of acute exacerbations in the past year, sarcopenia, EOS and SP-D were finally determined to be important predictors of COPD progression to type Ⅱ respiratory failure (P<0.05). Based on this, the consistency index（C-index）of the early warning identification model was 0.904. Receiver operating characteristic (ROC) curve showed that the area under the curve of the model was 0.904（95%CI: 0.860-0.948) and 0.924（95%CI: 0.861-0.986） in the modeling set and the validation set, respectively. The results of calibration curve and decision curve showed that the model had good calibration and clinical applicability. Conclusion The nomogram early warning recognition model based on predictors has good predictive performance and clinical applicability.

Key words: pulmonary disease, chronic obstructive, respiratory insufficiency, nomograms, influencing factors, risk prediction model