An Electronic Health Record Model for Predicting Risk of Hepatic Fibrosis in Primary Care Patients

Area of Science:

• Hepatology
• Medical Informatics
• Biostatistics

Background:

• Nonalcoholic fatty liver disease (NAFLD) poses a challenge for primary care, requiring identification of patients at high risk for significant liver disease.
• Obesity and diabetes disproportionately affect NAFLD prevalence and progression.
• Accurate risk stratification is essential for timely intervention and management.

Purpose of the Study:

• To develop a risk stratification tool for NAFLD using structured electronic health record (EHR) data.
• To identify key predictors of hepatic fibrosis in populations with high rates of obesity and diabetes.
• To assess the utility of standard clinical thresholds versus optimized cut-offs for risk prediction.

Main Methods:

• Utilized data from 344 participants undergoing Fibroscan for liver fat and stiffness measurement (LSM).
• Employed multivariable logistic regression and random forest classification to identify risk factors for hepatic fibrosis (LSM > 7 kPa) and significant fibrosis (LSM > 8 kPa).
• Included predictors such as age, gender, diabetes, hypertension, FIB-4 index, BMI, LDL, HDL, and triglycerides from EHR data.

Main Results:

• Three variables—BMI, FIB-4, and diabetes—consistently predicted fibrosis in both modeling approaches.
• A model using Youden's index cut-offs for BMI and FIB-4 achieved an AUC of 0.75 for predicting any hepatic fibrosis.
• Models using standard clinical thresholds showed lower discriminatory ability but higher positive predictive value (PPV) for fibrosis prediction.

Conclusions:

• Standard clinical thresholds for risk factors may require modification for improved predictive accuracy in high-risk populations.
• The derived risk stratification tool, utilizing EHR data, shows promise for identifying NAFLD patients needing further evaluation.
• Further validation is needed to optimize the tool for diverse patient groups with metabolic comorbidities.