{
  "abstract": "Background Oestrogen is hepatoprotective, yet sex-specific fibrosis risk architectures in MASLD remain poorly characterised and FIB-4 — derived from mixed-sex cohorts — may perform inequitably across the female lifespan. We applied machine learning with SHAP explainability to characterise fibrosis risk drivers across the menopausal transition and quantify FIB-4 performance disparities by sex and reproductive stage.Methods We analysed 15,587 adults from NHANES 2017-2023 with elastography-confirmed fibrosis (LSM ≥7.0 kPa). Participants were stratified into five sex-age groups: men <50, men ≥50, premenopausal (<50), perimenopausal (50-59), and postmenopausal women (≥60). XGBoost classifiers incorporating 16 clinical and metabolic variables were trained with stratified 5-fold cross-validation per stratum. SHAP values identified dominant fibrosis drivers per group, with post-hoc isotonic calibration applied.Results Fibrosis prevalence rose sharply across the menopausal transition: 8.7% premenopausal, 18.4% perimenopausal, and 17.1% postmenopausal women, versus 13.6% and 23.0% in younger and older men respectively.FIB-4 demonstrated near-random discrimination across all groups (AUC 0.501-0.563), with the most extreme failure in premenopausal women (AUC 0.511), flagging only 2.5% at the >1.3 threshold despite 8.7% true fibrosis prevalence. Machine learning substantially outperformed FIB-4 in all groups, with the largest advantage in premenopausal women (ΔAUC +0.315; ML AUC 0.832), the greatest ML advantage observed across any subgroup in this analysis (IDDF2026-ABS-0278 Figure 1. ML reveals sex specific fibrosis risk architectures and exposes critical FIB-4 failure in women).SHAP analysis revealed a fundamental shift in fibrosis risk architecture across the transition: premenopausal fibrosis was almost entirely adiposity-driven (BMI mean SHAP 0.657, the highest single-feature value across all analyses) (IDDF2026-ABS-0278 Figure 2. SHAP feature); the perimenopausal period showed emergence of visceral adiposity, diabetes, and HDL dysregulation as co-dominant drivers; postmenopausal fibrosis showed a hepatocellular injury pattern with AST, platelets, and insulin replacing BMI as primary determinants. All five ML models demonstrated near-perfect calibration. (IDDF2026-ABS-0278 Figure 3. ML advantage over FIB-4 and FIB-4 flag rate, IDDF2026-ABS-0278 Figure 4. Hepatic steatosis and fibrosis prevalence by age and sex, IDDF2026-ABS-0278 Figure 5. Disproportionate impact on NH black and NH Asian populations)Conclusions Menopause represents a critical hepatic inflection point at which fibrosis prevalence doubles and its biological drivers fundamentally reorganise. FIB-4 is essentially uninformative in premenopausal women, missing the vast majority of true cases using physiologically inappropriate thresholds derived from male-predominant cohorts. Machine learning with sex-aware feature weighting offers the largest clinical gains precisely where FIB-4 fails most. These findings support sex-specific and reproductive-stage-aware fibrosis risk stratification in MASLD clinical practice.Abstract IDDF2026-ABS-0278 Figure 1Abstract IDDF2026-ABS-0278 Figure 2Abstract IDDF2026-ABS-0278 Figure 3Abstract IDDF2026-ABS-0278 Figure 4Abstract IDDF2026-ABS-0278 Figure 5",
  "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-0278 Are we missing fibrosis in women? A machine learning interrogation of liver screening",
  "uid": "ac5447ad-a5ed-5171-89ca-835c185323ea"
}