{
  "abstract": "Background Existing non-invasive liver scores use fixed formulas derived from small cohorts and limited variable sets. No study has trained a machine learning model across the full panel of routinely available blood analytes to derive a simplified liver screening score validated against elastography-confirmed outcomes. We developed and validated the NHANES Liver Screening Score (NLSS) using 32 analytes from standard blood panels.Methods We analysed 15,587 adults from NHANES 2017-2023 with valid transient elastography. Steatosis was defined as CAP ≥248 dB/m and significant fibrosis as LSM ≥8.0 kPa. XGBoost models trained on 32 analytes identified top predictive features via SHAP importance values. An 8-variable logistic regression score (NLSS) was derived using only universally available analytes (≥90% population coverage) and compared head-to-head against FIB-4 (Fibrosis-4 Index), NFS (NAFLD Fibrosis Score), HSI (Hepatic Steatosis Index), and FLI (Fatty Liver Index) on identical elastography-validated populations.Results The full ML model achieved AUC 0.855 for steatosis and 0.809 for fibrosis ( IDDF2026-ABS-0226 Figure 1. AUC comparison). The simplified 8-variable NLSS retained most performance (steatosis AUC 0.830, fibrosis AUC 0.795), outperforming all comparators across both outcomes (IDDF2026-ABS-0226 Figure 4. NLSS steatosis and fibrosis). NLSS surpassed FIB-4 (steatosis AUC 0.539, fibrosis 0.615), NFS (0.642, 0.715), and HSI (0.797, 0.716) for fibrosis detection in the full cohort of 15,587 subjects (IDDF2026-ABS-0226 Figure 5. ROC comparison fibrosis). FLI was competitive for steatosis (AUC 0.823) but required fasting triglycerides, limiting applicability to only 31% of subjects, versus universal applicability of NLSS (IDDF2026-ABS-0226 Figure 6. ROC comparison steatosis).SHAP analysis identified GGT, uric acid, and WBC as novel contributors not captured by existing scores. NLSS-steatosis was dominated by waist circumference, GGT, ALT, and inverse HDL, reflecting visceral adiposity and hepatic oxidative stress (IDDF2026-ABS-0226 Figure 3. Feature importance steatosis). NLSS-fibrosis was driven by AST, age, waist circumference, and inverse platelet count, with AST substantially outweighing ALT as a fibrosis-specific marker (IDDF2026-ABS-0226 Figure 2. Feature importance fibrosis). Both models demonstrated near-perfect calibration across the full predicted probability range.Conclusions An 8-variable score derived from routine blood analytes outperforms all existing non-invasive liver screening tools for both steatosis and fibrosis detection in a nationally representative elastography-validated cohort of 15,587 adults. GGT, uric acid, and WBC contribute an independent predictive signal beyond existing scores. NLSS is implementable using standard CBC and CMP results, offering a practical population-level liver disease screening tool.Abstract IDDF2026-ABS-0226 Figure 1Abstract IDDF2026-ABS-0226 Figure 2Abstract IDDF2026-ABS-0226 Figure 3Abstract IDDF2026-ABS-0226 Figure 4Abstract IDDF2026-ABS-0226 Figure 5Abstract IDDF2026-ABS-0226 Figure 6",
  "authors": [
    {
      "affiliations": [
        "New York Medical College, United States"
      ],
      "name": "Jeril Lasington"
    },
    {
      "affiliations": [
        "Rutgers University, United States"
      ],
      "name": "Lawin Steve Mathew Lasington"
    },
    {
      "affiliations": [
        "Boston University, United States"
      ],
      "name": "Swamynathan Umamaheshwaran"
    }
  ],
  "title": "IDDF2026-ABS-0226 Hidden in the blood panel: a machine learning approach to population-level liver disease detection",
  "uid": "b4d57c74-df3f-5379-a565-31cebbb07a37"
}