A Prediction Model for Metabolic Abnormality Clustering Based on Abdominal Body Composition Parameters From...

To investigate the association of photon-counting computed tomography (PCCT)-derived abdominal body composition parameters with metabolic abnormality clustering, and to develop and internally validate clinical, imaging, and combined models for risk assessment MATERIALS AND METHODS: We retrospectively enrolled 116 patients undergoing abdominal PCCT, including 61 with metabolic abnormality clustering and 55 without metabolic abnormality clustering. Metabolic abnormality clustering was defined using a study-specific criterion as the presence of ≥2 of the following: hyperglycemia, hypertension, hypertriglyceridemia, and low high-density lipoprotein cholesterol, for risk-oriented identification of concurrent metabolic abnormalities rather than diagnosis of metabolic syndrome. Clinical (age, sex, body mass index [BMI]) and imaging parameters (visceral fat area, visceral adipose tissue [VAT]/subcutaneous adipose tissue [SAT] ratio, skeletal muscle index, lumbar muscle density, and liver-to-spleen ratio) were collected. Univariable and multivariable logistic regression analyses were performed to identify factors independently associated with metabolic abnormality clustering. Three models were developed: a clinical model (age, sex, and BMI), an imaging model (VAT/SAT ratio), and a combined model incorporating the clinical variables and VAT/SAT ratio. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC), with pairwise comparisons by the DeLong test. Clinical utility was explored using decision curve analysis (DCA). Internal validation was performed using 1000 bootstrap resamples. Calibration was assessed using calibration curves, and a nomogram was constructed based on the combined model.

BMI (odds ratio [OR] = 1.476, 95% confidence interval [CI]: 1.263-1.779, p < 0.001) and VAT/SAT ratio (OR = 4.093, 95% CI: 1.487-13.159, p = 0.011) were independently associated with metabolic abnormality clustering in the combined model. AUCs were 0.812 for the clinical model, 0.754 for the imaging model, and 0.841 for the combined model. The combined model outperformed the imaging model (p = 0.022) but not the clinical model (p = 0.143). Although the combined model had the numerically highest AUC, its discrimination was not statistically superior to that of the clinical model. Exploratory DCA suggested numerically higher net benefit for the combined model within a limited range of threshold probabilities. At the optimal probability threshold of 0.565, sensitivity, specificity, positive predictive value, and negative predictive value were 70.5%, 87.3%, 86.0%, and 72.7%, respectively. Bootstrap internal validation yielded an optimism-corrected C-index of 0.81, a calibration slope close to 1, and a mean absolute error of 0.038. A nomogram based on age, sex, BMI, and VAT/SAT ratio was developed to estimate the probability of metabolic abnormality clustering.

The PCCT-derived VAT/SAT ratio was independently associated with metabolic abnormality clustering. A combined model incorporating age, sex, BMI, and VAT/SAT ratio achieved an AUC of 0.841, although its discrimination was not significantly better than that of the clinical model alone. Further multicenter studies with external validation are warranted before clinical application.