A logistic Generalized Additive Model using three covariates derived from heart rate and systolic arterial pressure achieved an AUC of 0.931 for distinguishing Type 2 Diabetes Mellitus patients from healthy individuals.
Observational (n=68)
Does a feature selection method based on logistic GAMs using cardiovascular time series measures accurately diagnose Type 2 Diabetes Mellitus?
A novel feature selection method based on logistic GAMs using cardiovascular time series measures demonstrates high accuracy in distinguishing between Type 2 Diabetes Mellitus patients and healthy individuals.
Abstract Feature selection has emerged as a powerful tool for evaluating the relevance of variables in classification models. In high-dimensional datasets, not all features contribute meaningfully to accurate classification, and their inclusion can negatively affect model performance. Often, only a small subset of features is truly informative; therefore, identifying and removing irrelevant variables is crucial. In this paper, we propose a novel feature selection method based on logistic Generalized Additive Models (GAMs). Model performance was evaluated using the Bayesian Information Criterion (BIC) and the Area Under the ROC - Receiver Operating Characteristic- Curve (AUC). To demonstrate the effectiveness of this approach, we applied it to data from a clinical study aimed at diagnosing Type 2 Diabetes Mellitus (T2DM), distinguishing between diabetic and healthy individuals. The regression model incorporated features derived from Heart Period (HP) and Systolic Arterial Pressure (SAP) time series, as well as Baroreflex Sensitivity (BRS) measures, collected under two conditions: at rest and after active standing. The best-performing models selected between three and five covariates. Among the top 20 models, AUC values ranged from 0.931 to 0.918. The best 3-covariate model achieved a sensitivity of 0.816 with a false positive rate of 0.184. These results highlight the potential of this statistical methodology to enhance the discrimination between T2DM patients and control subjects.
Roca-Pardinas et al. (Wed,) conducted a observational in Type 2 Diabetes Mellitus (n=68). Logistic Generalized Additive Model (GAM) with 3 covariates vs. Healthy subjects was evaluated on Area Under the ROC Curve (AUC) for T2DM diagnosis (95% CI 0.879-0.961). A logistic Generalized Additive Model using three covariates derived from heart rate and systolic arterial pressure achieved an AUC of 0.931 for distinguishing Type 2 Diabetes Mellitus patients from healthy individuals.