P5947 - Development and Validation of a Simplified Risk Model to Predict High Liver Fat Using NHANES Elastography Data

Jinye Liu, DO¹, Hassam Ali, MD², Shiza Sarfraz, MD¹, Saeed S. Graham, MD³, Abdulazeez Swaiti, MD¹, Rami Basmaci, MD¹, Clarimar Diaz-Lopez, MD⁴, Parnita Kesar, MD¹, Vinay Jahagirdar, MD⁵, Dushyant S. Dahiya, MD⁶, Umar Hayat, MD⁷, Eslam Ali, MD¹
1East Carolina University Medical Center, Greenville, NC; 2East Carolina University/Brody School of Medicine, Greenville, NC; 3ECU Department of Internal Medicine, Greenville, NC; 4East Carolina University Medical Center, Greenville, FL; 5Virginia Commonwealth University Medical Center, Richmond, VA; 6University of Kansas School of Medicine, Kansas City, KS; 7Geisinger Wyoming Valley Medical Center, Wilkes-Barre, PA

Introduction: Hepatic steatosis, the hallmark of metabolic dysfunction-associated steatotic liver disease (MASLD), is increasingly prevalent. Early risk stratification can facilitate targeted screening and prevention. We aimed to develop and validate a simplified, interpretable logistic regression model for predicting high liver fat content using U.S. population-based data.

Methods: We analyzed NHANES 2021–2022 data on 2,151 adults with valid transient elastography and laboratory results. High liver fat was defined as CAP ≥275 dB/m. Candidate predictors included demographics, BMI category, diabetes, CRP, liver stiffness (LSM), and lifestyle variables. We trained full logistics LASSO logistic, and random forest models using 5-fold cross-validation. A simplified logistic model using 8 top predictors was derived. Model performance was evaluated using AUC, calibration, and diagnostic metrics at optimal thresholds based on Youden’s Index.

Results: High liver fat was present in 44% of participants. The AUC for the full logistic model was 0.825, LASSO 0.831, and random forest 0.821. The simplified model retained excellent discrimination (AUC = 0.795) with the following formula:

logit(p) = 0.3083 − 2.4617×BMI_Normal + 1.2299×BMI_Obesity III − 0.9276×BMI_Overweight + 0.9701×BMI_Obesity II + 0.6942×Diabetes − 0.5378×Female − 0.0093×CAP + 0.0022×LSM

At an optimized threshold of 0.45, the simplified model achieved sensitivity 66.4%, specificity 77.0%, PPV 64.6%, NPV 78.4%, and F1-score 65.5%. Calibration plots demonstrated strong agreement between predicted and observed probabilities (Figures 1–2).

Discussion: We present a simplified logistic model that retains high predictive performance using only 8 clinical variables. It offers an accessible, interpretable screening tool for MASLD risk stratification, with potential integration into primary care workflows and large-scale public health applications.



Figure: Figure 1. ROC curve comparing the performance of the simplified logistic regression model for predicting high liver fat. The model achieved an AUC of 0.795. A red dot indicates the optimal cutoff point (threshold = 0.45) based on Youden’s Index.


Figure: Figure 2. Calibration plot of the simplified logistic model, showing excellent agreement between predicted probabilities and observed outcomes across deciles of risk. The model was well-calibrated for clinical prediction.

Disclosures:
Jinye Liu indicated no relevant financial relationships.
Hassam Ali indicated no relevant financial relationships.
Shiza Sarfraz indicated no relevant financial relationships.
Saeed Graham indicated no relevant financial relationships.
Abdulazeez Swaiti indicated no relevant financial relationships.
Rami Basmaci indicated no relevant financial relationships.
Clarimar Diaz-Lopez indicated no relevant financial relationships.
Parnita Kesar indicated no relevant financial relationships.
Vinay Jahagirdar indicated no relevant financial relationships.
Dushyant Dahiya indicated no relevant financial relationships.
Umar Hayat indicated no relevant financial relationships.
Eslam Ali indicated no relevant financial relationships.

Jinye Liu, DO¹, Hassam Ali, MD², Shiza Sarfraz, MD¹, Saeed S. Graham, MD³, Abdulazeez Swaiti, MD¹, Rami Basmaci, MD¹, Clarimar Diaz-Lopez, MD⁴, Parnita Kesar, MD¹, Vinay Jahagirdar, MD⁵, Dushyant S. Dahiya, MD⁶, Umar Hayat, MD⁷, Eslam Ali, MD¹. P5947 - Development and Validation of a Simplified Risk Model to Predict High Liver Fat Using NHANES Elastography Data, ACG 2025 Annual Scientific Meeting Abstracts. Phoenix, AZ: American College of Gastroenterology.