# Fatty Liver Disease Status Task Card Task id: `fatty_liver_disease_status` Evidence status: grounded Evidence reviewed: 2026-07-28 Benchmark status: complete ## Target Participant-level binary classification of fatty liver disease status from the curated MAFLD phenotype table. Implementation facts: - target dataset release: `anat_curated_phenotype` - target table: `anat.curated_phenotype.mafld` - target column read: `curated_phenotype`. **This is hardcoded in the loader** (an internal review record), not taken from the Task - `target_field` on the class is `mafld__curated_phenotype` and is **inert on this release** - it is read only by the alternate `frozen_v1_2` route (`_curated_phenotype_status.py:336`). Do not treat it as the operative field - positive label: `MAFLD/NASH` - negative label: `Normal liver` - excluded states: `Suspected MAFLD/NASH / Intermediate risk`, missing - default research stage: `00_00_visit` - default covariates: age, sex, BMI from baseline anthropometrics - primary metric: ROC AUC ## Why This Matters Fatty liver disease is a cardiometabolic liver condition selected in DS-1584. It complements `fli`, which is a continuous derived Fatty Liver Index, rather than duplicating it. ## Baselines And Ceilings The required floor is a no-modality-feature classifier baseline: `NoFeaturesPredictor` contributes no modality features, and `DemographicClassifierFloorStrategy` fits age, sex, and BMI. Useful future comparators include liver enzymes, FLI/FIB scores, lipid panels, gut microbiome species, and MMFM participant embeddings. ## Benchmark-Track Evidence literature review prioritized Nightingale metabolomics/lipoproteins and DXA regional body composition for binary MASLD prediction beyond age/sex/BMI. Unlike FLI, this target has no explicit triglyceride/BMI/GGT/waist formula, so the fixed 27-field Nightingale panel and six-field DXA panel are eligible cross-modal inputs. The label is still a curated phenotype rather than biopsy or a single direct imaging measurement. ## V0 Benchmark Execution The frozen same-visit V1→V1 matrix completed on 2026-07-29. All six rows are locally successful and remotely `FINISHED` in operational source. - Nightingale: n=6,025; test=920; best AUROC=0.816 (CPU TabSwift). Paired linear and GBDT gains over age/sex/BMI were both +0.009, with intervals crossing zero. - DXA: n=7,207; test=1,083; best AUROC=0.867 (CPU TabSwift). Paired gains were +0.045 [0.027, 0.063] for logistic regression and +0.053 [0.032, 0.077] for GBDT. - Full rows and provenance: deep dive and aggregate review. ## Caveats - Labels come from the curated MAFLD panel, not adjudicated biopsy or elastography. - Suspected/intermediate-risk MAFLD is excluded from this binary case/control slice. - Supporting liver-enzyme columns inside the curated phenotype table are curation snapshots, not live biomarker values. - BMI is a strong clinical floor feature and may explain much of the label. - Alcohol and medication context are not represented in the current binary definition. ## Metrics Primary: ROC AUC. Secondary: average precision, accuracy, balanced accuracy, Brier score, positive rate, split sizes. ## References - DS-1584 Yeela-selected condition curation: `pha_condition_projection/DS-1584/condition_candidates_detailed.csv`. **Does not resolve** in this repository or in HPP documentation / HPP documentation / HPP documentation; kept because it is the only pointer to how this condition was selected, and marked so it is not read as a repo path. - Existing `fli` Task for a derived continuous fatty-liver index. - HPP refreshed curated-phenotype docs for `anat.curated_phenotype.mafld`. - Masoodi et al. (2021), metabolomics/lipidomics in NAFLD: . - Pang et al. (2022), adiposity, metabolomics, and NAFLD risk: . ## Open Questions - Should this Task be paired with an imaging- or FLI-defined fatty-liver target before making model-value claims? - Should the suspected/intermediate-risk label become positive in a broader MAFLD-risk Task?