Predicting the prevalence of peripheral arterial diseases: modelling and validation in different cohorts

Abstract. Background: To develop models for prevalence estimation of peripheral arterial disease (PAD) and to validate them in an external cohort. Methods: Model training cohort was a population based cross-sectional survey. Age, sex, smoking status, body mass index, total cholesterol (TC), high density lipoprotein (HDL), TC/HDL ratio, low density lipoprotein, fasting glucose, diabetes, hypertension, pulse pressure, and stroke history were considered candidate predicting variables. Ankle brachial index ≤ 0.9 was defined as the presence of peripheral arterial disease. Logistic regression method was used to build the prediction models. The likelihood ratio test was applied to select predicting variables. The bootstrap method was used for model internal validation. Model performance was validated in an external cohort. Results: The final models included age, sex, pulse pressure, TC/HDL ratio, smoking status, diabetes, and stroke history. Area under receiver operating characteristics (AUC) with 95% confidence interval (CI) of the final model from the training cohort was 0.74 (0.70, 0.77). Model validation in another cohort revealed AUC (95% CI) of 0.72 (0.70, 0.73). P value of Hosmer-Lemeshow’s model goodness of fit test was 0.75 indicating good model calibration. Conclusions: The developed model yielded a moderate usefulness for predicting the prevalence of PAD in general population.