A risk-predictive model for obstructive sleep apnea in patients with chronic obstructive pulmonary disease

Figure 1. Nomogram for predicting the risk of OSA in patients with COPD. The nomogram included six risk factors: age, NC, type 2 diabetes, mMRC questionnaire, SACS questionnaire, and CRP. The patient's status for each predictor is plotted on a horizontal scale, and vertical lines are drawn to obtain the corresponding points. After all points have been summed, the total score on the total point line is plotted and a vertical line is drawn down to the accompanying line labeled Risk of OSA. The point where this line crosses the accompanying line indicates the predicted probability of diagnosis. COPD, chronic obstructive pulmonary disease; OSA, obstructive sleep apnea syndrome; NC, neck circumference; mMRC, modified Medical Research Council; SACS, Sleep Apnea Clinical Score; CRP, C-reactive protein.

Figure 2. The variables filtering process of the LASSO regression. In order to avoid overfitting, the LASSO regression suggested including six variables when merging OSA was the endpoint. In the variable selection process, first of all, the univariate analysis was used to select potential factors. Then, based on these potential factors, the multivariable logistic regression model was constructed. In this study, the LASSO regression was only used to ensure that the multivariable logistic regression models were not overfitting rather than for variable selection and modeling. (A) Optimal parameter (lambda) selection in the LASSO logistic regression used 10-fold cross-validation via minimum criteria. The dotted vertical lines were drawn at the best values using the minimum criteria and 1 standard error of the minimum criteria (the 1-SE criteria). (B) LASSO coefficient profiles of the 34 features. A coefficient profile plot was produced against the log (lambda) sequence. LASSO, least absolute shrinkage and selection operator; SE, standard error.

Figure 3. ROC curves in both the training and validation cohorts (A). Training group (blue curve), AUC = 0.929 (95%CI 0.894–0.965). Validation group (red curve), AUC = 0.928 (95%CI 0.873–0.984) The AUCs for the OSA in COPD consecutive patients in the training and validation group exceed 0.7, which demonstrated that the nomogram could accurately predict the risk of OSA in consecutive patients with COPD. Calibration plot in the training group (B) and validation group (C). The y-axis represents the actual probability of patients with OSA as validated by the sleep monitoring study, and the x-axis represents the predicted risk of OSA by the risk nomogram. The dotted line represents a perfect prediction by an ideal model, while the blue line represents the performance of the risk nomogram. A closer fit of the blue line to the dotted line represents a better prediction. COPD, chronic obstructive pulmonary disease; OSA, obstructive sleep apnea syndrome; ROC, receiver operating characteristic curves; AUC, area under the ROC curve; CI, confidence interval.

Figure 4. Evaluation of the clinical utility of nomogram prediction models in the training and validation groups. (A) Decision curves in both the training and validation groups. (B) Clinical impact curve in the training group. (C) Clinical impact curve in the validation group.