Clinical characteristics and a diagnostic model for high-altitude pulmonary edema in habitual low altitude dwellers

Abstract

Background: The fatal risk of high-altitude pulmonary edema (HAPE) is attributed to the inaccurate diagnosis and delayed treatment. This study aimed to identify the clinical characteristics and to establish an effective diagnostic nomogram for HAPE in habitual low altitude dwellers.

Methods: A total of 1,255 individuals of Han Chinese were included in the study on the Qinghai-Tibet Plateau at altitudes exceeding 3,000 m. LASSO algorithms were utilized to identify significant predictors based on Akaike's information criterion (AIC), and a diagnostic nomogram was developed through multivariable logistic regression analysis. Internal validation was conducted through bootstrap resampling. Model performance was evaluated using ROC curves and the Hosmer-Lemeshow test.

Results: The nomogram included eleven predictive factors and demonstrated high discrimination with an AUC of 0.787 (95% CI [0.757-0.817]) and 0.833 (95% CI [0.793-0.874]) in the training and validation cohorts, respectively. Calibration curves were assessed in both the training (P = 0.793) and validation datasets (P = 0.629). Confusion matrices revealed accuracies of 70.95% and 74.17% for the training and validation groups. Furthermore, decision curve analysis supported the use of the nomogram for patients with HAPE.

Conclusion: We propose clinical features and column charts based on hematological parameters and demographic variables, which can be conveniently used for the diagnosis of HAPE. In high-altitude areas with limited emergency environments, a diagnostic model can provide fast and reliable diagnostic support for medical staff, helping them make better treatment decisions.

Keywords: High-altitude pulmonary edema; Lasso regression; Nomogram; Validation.

Figure 1. Enrolled subjects and outcome of HAPE in training and validation datasets.

Figure 2. Predictor selection using the least absolute shrinkage and selection operator (LASSO) binary logistic regression model.

(A) Identifying the optimal penalising coefficient lambda (λ) in the LASSO model using tenfold cross-validation and the minimum criterion. (B) LASSO coefficient profiles of 11 variables. The value selected by 10-fold cross-validation was plotted as a vertical line. As λ decreased, the degree of model compression increased and the model’s ability to select important variables increased.

Figure 3. Nomogram to estimate the probability of HAPE.

Predictor scores associated with each patient variable are obtained and summarised on the upper scale. The percentage probability of HAPE is calculated from the sum of the scores projected onto the lower scale.

Figure 4. Receiver operating characteristic (ROC) curves of the nomograms in training and validation dataset.

(A) Training dataset. (B) Validation dataset. The nomogram had good discriminative power with area under ROC curve (95% confidence interval) of 0.787 (95% CI [0.757–0.817]) and 0.833 (95% CI [0.793–0.874]) in the training and validation dataset, respectively.

Figure 5. The calibration curve of nomogram for predicting HAPE in the training and validation dataset.

(A) Training dataset. (B) Validation dataset. Calibration focuses on the accuracy of the model in predicting absolute risk, i.e., the consistency between what the model predicts will happen and what actually happens. The y-axis is the actual rate of HAPE. The x-axis is the predicted probability of HAPE. The y-axis represents the actual rate of HAPE. The x-axis is the predicted probability of HAPE. In a well-calibrated nomogram, the scatter points should be aligned along a diagonal line at a 45 degree angle. P > 0.05 means no significant difference and the model is well calibrated. (C) Calibration plot of training dataset that the BS was calculated by bootstrap resampling of 100 replicates. (D) Decision curve analysis for the nomogram. The y-axis measures the net benefit. The black dotted line represents the nomogram. The solid black line represents the assumption that all patients have HAPE. The grey line represents the assumption that no patients have HAPE. The net benefit was calculated by subtracting the proportion of all patients with a false-positive result from the proportion with a true-positive result, weighted by the relative harm of not receiving treatment compared with the negative consequences of receiving unnecessary treatment (Vickers & Elkin, 2006). The decision curves showed that the use of the nomogram in the current study for predicting HAPE gave more benefit than either the “treat all patients” or the “treat none” regimen when the patient or physician threshold probability exceeded 10%.