Development and Validation of a Prediction Model for Pulmonary Arterial Hypertension in Diabetic Patients: A Retrospective Study of 57,375 Patients

Figures

• 

  Figure 1. Data preprocessing and patient selection flowchart. A total of 107,111 initial records were identified. Hospitalization ID was used to exclude duplicate entries from the same patient, only the first eligible measurement was retained when repeated values were present, and variables were standardized for analysis. A total of 57,375 patients formed the overall descriptive cohort. For prediction modeling, 47,042 records from admissions before January 1, 2022 were screened, yielding 10,458 eligible patients in the training cohort, whereas 60,069 records from January 1, 2022 to July 31, 2025 were screened, yielding 14,060 eligible patients in the temporal validation cohort.

• 

  Figure 2. Univariable logistic regression analysis of nine prespecified clinical risk domains for PAH in the pre‐2022 training cohort. Odds ratios and 95% confidence intervals are shown for PAH‐related pressure gradient, cardiac dysfunction, cardiac cavity and vascular dilatation, coagulation dysfunction, hemoglobin and bilirubin abnormalities, hepatic and renal injury, cardiovascular thrombotic disease, chronic airway disease, and all‐cause mortality. The vertical reference line indicates an odds ratio of 1.0. These exploratory associations were used to inform subsequent multivariable modeling; mortality‐related information was not included in the final baseline nomogram. **p < 0.01, and ***p < 0.001.

• 

  Figure 3. Multivariable logistic regression analysis of baseline predictors associated with PAH in patients with T2DM. Adjusted odds ratios and 95% confidence intervals are presented for the clinical, biochemical, and echocardiographic variables evaluated during model development. Predictor selection was performed in the pre‐2022 training cohort using forward stepwise selection based on the Akaike information criterion. Variables occurring after the index clinical assessment, including mortality‐related outcomes, were excluded from the baseline nomogram. The vertical reference line indicates an odds ratio of 1.0. *p < 0.05, **p < 0.01, and ***p < 0.001.

• 

  Figure 4. Receiver operating characteristic curves for eight individual risk domains and the combined multivariable sensitivity model in the pre‐2022 training cohort after exclusion of the PAH‐related pressure‐gradient domain. The AUCs were 0.514 for hepatic and renal injury, 0.538 for cardiovascular thrombotic disease, 0.568 for chronic airway disease, 0.581 for all‐cause mortality, 0.606 for hemoglobin and bilirubin abnormalities, 0.627 for cardiac dysfunction, 0.697 for coagulation dysfunction, 0.819 for cardiac cavity and vascular dilatation, and 0.823 for the combined model. The dashed diagonal line indicates random classification (AUC = 0.500).

• 

  Figure 5. Nomogram for predicting PAH in patients with T2DM. The nomogram incorporates eight baseline predictors: COPD, congenital heart disease, aortic valve velocity, pulmonary valve velocity, tricuspid regurgitation velocity, d‐dimer, total bilirubin, and left ventricular ejection fraction. The illustrated patient has a total score of 134 points, corresponding to an estimated PAH probability of 31%. The nomogram C‐index was 0.666 (95% CI, 0.631–0.699), and the Hosmer‐Lemeshow test yielded p = 0.453. Because the nomogram includes tricuspid regurgitation velocity, it should be interpreted as an echocardiography‐informed risk‐stratification tool rather than a pre‐echocardiographic screening model.

• 

  Figure 6. Receiver operating characteristic curves of the pressure‐gradient‐excluded eight‐domain combined sensitivity model in the training and temporal validation cohorts. The training cohort comprised patients admitted from January 1, 2018 to December 31, 2021 and had an AUC of 0.823. The fixed model was applied without refitting to patients admitted from January 1, 2022 to July 31, 2025, yielding an AUC of 0.780. The dashed diagonal line indicates random classification (AUC = 0.500).