人體測量指標在預測低睪固酮中的優越性—一項機器學習研究

吳志緯¹、何承勳¹

¹新光吳火獅紀念醫院外科部泌尿科

Superiority of Anthropometric Measures over Symptom Scores in Predicting Low Testosterone — A Machine Learning-Based Analysis

Chih-Wei Wu¹, Chen-Hsun Ho¹

¹Divisions of Urology, Department of Surgery, Shin Kong Wu Ho-Su Memorial Hospital, Taipei, Taiwan

Purpose:

To evaluate and compare the predictive performance of clinical symptom scores, anthropometric data, and baseline comorbidities for identifying low testosterone in aging men using various machine learning algorithms.

Materials and Methods:

A clinical dataset of 1,011 individuals was cleaned to 1,007 complete cases for analysis. Low testosterone (Low T) was defined as serum total testosterone < 350 ng/dL. Features were categorized into three domains: (1) Symptoms (AMS and ADAM scales), (2) Anthropometric data (Age, BMI, Waist Circumference), and (3) Baseline diseases. Predictive power was assessed using the Area Under the ROC Curve (AUC) via 5-fold cross-validation. Multiple models, including Logistic Regression, Random Forest, and Gradient Boosting, were implemented to identify the optimal integrated predictive framework.

Results:

The prevalence of low testosterone in the cohort was 38.4%. Comparative analysis of the feature domains revealed that anthropometric data demonstrated the highest predictive capability (AUC: 0.622), significantly outperforming clinical symptom scores (AUC: 0.528) and baseline diseases (AUC: 0.528). Logistic Regression was identified as the most stable and performant integrated model. Multivariate analysis demonstrated that abdominal girth (waist circumference) was the only significant independent predictor for low T (OR: 1.039, 95% CI: 1.010–1.068, p = 0.007). Conversely, total scores from the AMS and ADAM scales showed no statistically significant association with biochemical testosterone levels.

Conclusions:

Objective anthropometric measurements, particularly waist circumference, provide superior predictive value for low testosterone compared to subjective symptom questionnaires. These findings suggest that clinical screening strategies should prioritize metabolic indicators over conventional symptom-based assessments to improve the identification of biochemical hypogonadism.