What is a complete draft protocol for a retrospective cohort study of adult patients with metabolic‑dysfunction‑associated steatotic liver disease (MASLD), including inclusion and exclusion criteria, baseline data collection, imaging assessments, primary and secondary outcomes, statistical analysis plan, sample‑size calculation, data management, and anticipated challenges?

Complete Draft Protocol for Retrospective Cohort Study of MASLD

This protocol outlines a comprehensive retrospective cohort study examining disease progression, metabolic risk factors, and clinical outcomes in adult patients with metabolic dysfunction-associated steatotic liver disease (MASLD), incorporating the 2024 EASL-EASD-EASO clinical practice guidelines 1 and recent real-world evidence.

---

Study Design & Objectives

Primary Objective: To characterize the natural history of MASLD progression to end-stage liver disease (ESLD), including compensated cirrhosis, decompensated cirrhosis, hepatocellular carcinoma (HCC), and liver transplantation.

Secondary Objectives:

• Identify metabolic risk factors independently associated with MASLD incidence and progression
• Evaluate non-invasive testing patterns and their association with clinical outcomes
• Assess healthcare resource utilization and costs across disease stages
• Examine cardiovascular and hepatic event rates

---

Study Population

Inclusion Criteria

Adults ≥18 years with MASLD diagnosis defined by:

• ICD-10 codes for hepatic steatosis with metabolic dysfunction (presence of ≥1 cardiometabolic risk factor)
• Imaging evidence of steatosis (ultrasound, CT, MRI, or controlled attenuation parameter) OR elevated liver enzymes with metabolic risk factors
• Minimum 12 months of continuous enrollment before index date
• At least one outpatient visit with sufficient clinical/laboratory data

Exclusion Criteria

Exclude patients with:

• Significant alcohol consumption (>20g/day females, >30g/day males) documented within 6 months of diagnosis
• Other chronic liver diseases: viral hepatitis (HBV, HCV), autoimmune hepatitis, primary biliary cholangitis, primary sclerosing cholangitis, hemochromatosis, Wilson disease, alpha-1 antitrypsin deficiency
• HIV infection
• Active malignancy (except non-melanoma skin cancer) within 5 years
• Prior liver transplantation before index date
• Pregnancy during study period
• Schizophrenia or active substance use disorders 2
• Insufficient follow-up data (<6 months)

---

Baseline Data Collection

Demographics & Clinical Characteristics

• Age, sex, race/ethnicity (stratify by White, Black, Asian, Hispanic) 3
• Geographic region (Northeast, North Central, South, West) 4
• Body mass index (BMI) - strongest independent risk factor in most ethnic groups 5, 3
• Smoking status

Metabolic Risk Factors (All Seven Independent Factors) 3

1. Obesity (BMI ≥30 kg/m²) - strongest predictor in Asians, Whites, Hispanics, especially age <50
2. Type 2 diabetes mellitus - document HbA1c, fasting glucose
3. Hypertension - strongest predictor in Black patients 3
4. Hyperlipidemia - LDL-C, HDL-C, triglycerides 5
5. Obstructive sleep apnea
6. Hypothyroidism
7. Type 1 diabetes (document separately, not included in multivariable models) 3

Laboratory Parameters at Baseline

• Liver enzymes: ALT, AST, alkaline phosphatase, GGT
• Metabolic panel: Fasting plasma glucose, HbA1c, lipid profile (LDL-C, HDL-C, triglycerides)
• Renal function: Serum creatinine, calculate serum uric acid to creatinine ratio (SUA/Cr) 5
• Hematology: White blood cell count, platelet count 5
• Liver function: Albumin, bilirubin, INR

Non-Invasive Fibrosis Assessment 1

Calculate FIB-4 score at baseline and annually:

• FIB-4 = (Age × AST) / (Platelet count × √ALT)
• Interpretation:
  • <1.3 (or <2.0 if age >65): Low risk, repeat in 2-3 years
  • 1.3-2.67: Intermediate risk, proceed to elastography OR 1-year lifestyle intervention then retest

  • 2.67: High risk, proceed to elastography

Document elastography when available:

• Vibration-controlled transient elastography (VCTE) with controlled attenuation parameter
• Magnetic resonance elastography (MRE)
• Record liver stiffness measurements (kPa)

Alternative blood biomarkers:

• Enhanced Liver Fibrosis (ELF) score
• APRI score
• NAFLD Fibrosis Score (NFS)

Imaging Assessments

• Ultrasound findings (echogenicity, hepatomegaly)
• CT or MRI findings when available (not for routine steatosis assessment per guidelines) 1
• Document presence/absence of cirrhosis, portal hypertension features

Liver Biopsy Data (When Available)

Note: Only 4% of MASLD patients undergo biopsy; rates declining (9.4% to 3.6% over 5 years) 4. Biopsy patients represent a selected high-risk group with 9× higher HCC risk, 3× higher cirrhosis risk, and 78% higher decompensation risk 4.

When biopsy performed, document:

• NAFLD Activity Score (NAS): steatosis grade (0-3), lobular inflammation (0-3), ballooning (0-2)
• Fibrosis stage (F0-F4)
• Presence of steatohepatitis (MASH)
• Biopsy indication and timing

Cardiovascular Comorbidities

• Coronary artery disease
• Prior myocardial infarction
• Heart failure
• Cerebrovascular disease
• Peripheral vascular disease

---

Follow-Up & Outcome Assessment

Primary Outcomes (Liver-Related Events)

Time to first ESLD event (median 10.6 months in progressors, IQR 3.8-19.7) 6:

1. Compensated cirrhosis (ICD-10 codes, imaging, or biopsy confirmation)
2. Decompensated cirrhosis - most frequent first event 6:
   • Ascites
   • Hepatic encephalopathy
   • Variceal hemorrhage
   • Spontaneous bacterial peritonitis
3. Hepatocellular carcinoma (ICD-10 codes, imaging, or pathology)
4. Liver transplantation

Progression definition: Development of any ESLD event in patients without baseline ESLD (expected rate ~15.9%) 6

Secondary Outcomes

Cognitive outcomes (assess at 5,10,15,20-year intervals) 2:

• Mild cognitive impairment (MCI) - higher odds with MASLD/MASH (OR 1.37) 2
• Vascular dementia - paradoxically lower odds (OR 0.86) 2
• Alzheimer disease - paradoxically lower odds (OR 0.83) 2

Cardiovascular events:

• Coronary artery disease (expected 17.1% vs 9.4% controls) 3
• Myocardial infarction (expected 7.1% vs 4.2% controls) 3
• Stroke
• Cardiovascular death

Hepatic events:

• Cirrhosis development (expected 7.5% vs 1.1% controls) 3
• HCC (expected 0.5% vs 0.1% controls) 3
• Liver-related mortality

Laboratory progression:

• Elevated ALT (expected 27.7% vs 10.1% controls) 3
• Elevated AST (expected 18.0% vs 6.4% controls) 3
• Elevated alkaline phosphatase (expected 19.8% vs 13.1% controls) 3
• Serial FIB-4 increases (C-statistic >0.80 for one-stage fibrosis progression) 1

All-cause mortality

---

Statistical Analysis Plan

Sample Size Calculation

Based on recent real-world data 6, 4:

• Expected MASLD prevalence: 6.0% in general population 3
• Expected progression rate: 15.9% over median 10.6 months 6
• Anticipated effect sizes:
  • Cirrhosis incidence: 90.44 per 1,000 person-years (progressors) vs 29.75 (non-progressors) 4
  • HCC incidence: 2.18 per 1,000 person-years (progressors) vs 0.22 (non-progressors) 4

Minimum sample size:

• Training cohort: 40,000-60,000 patients (based on 7, 5)
• Validation cohort: 15,000-25,000 patients (30-40% of training set)
• Power: 80% to detect HR ≥1.5 for primary outcomes, α=0.05
• Account for 20% loss to follow-up

Descriptive Statistics

• Continuous variables: mean (SD) or median (IQR) based on distribution
• Categorical variables: frequencies and percentages
• Stratify by baseline ESLD status, age groups (<50, ≥50), sex, race/ethnicity

Propensity Score Matching

Match MASLD patients to controls (1:5 ratio) 4 on:

• Age (±2 years)
• Sex
• Race/ethnicity
• Baseline comorbidities (diabetes, hypertension, obesity)
• Index year
• Geographic region

Use nearest-neighbor matching with caliper width 0.2 standard deviations of logit propensity score

Univariable & Multivariable Analysis

Cox proportional hazards regression for time-to-event outcomes:

• Assess proportional hazards assumption using Schoenfeld residuals
• Report hazard ratios (HR) with 95% confidence intervals
• Adjust for:
  • Age, sex, race/ethnicity
  • BMI, metabolic risk factors
  • Baseline FIB-4 category
  • Geographic region
  • Comorbidity burden (Charlson Comorbidity Index)

Modified Poisson regression for progression predictors 6:

• Calculate risk ratios for binary outcomes
• Robust variance estimation

Logistic regression for cross-sectional outcomes:

• Report odds ratios with 95% CI
• Assess model discrimination (C-statistic/AUROC)
• Calibration curves for model fit

Subgroup Analyses

Pre-specified subgroups:

1. By race/ethnicity (White, Black, Asian, Hispanic) - different MRF rankings 3
2. By age (<50 years vs ≥50 years) - obesity stronger in younger patients 3
3. By sex (male vs female)
4. By baseline FIB-4 category (<1.3,1.3-2.67, >2.67)
5. By diabetes status (present vs absent)
6. By biopsy status (performed vs not performed) - critical given selection bias 4
7. By baseline ESLD status (present vs absent)

Test for interaction using likelihood ratio tests

Sensitivity Analyses

• Complete case analysis vs multiple imputation for missing data
• Varying lookback periods (6,12,24 months)
• Excluding patients with <12 months follow-up
• Competing risk analysis for mortality
• Time-varying covariates for metabolic risk factors

Model Development & Validation

For prediction models (if developing risk scores):

• Split data 70:30 training:validation 7, 5
• Feature selection using LASSO regression 7
• Evaluate multiple algorithms: logistic regression, random forest, XGBoost, neural networks 7
• Prioritize interpretability - select LASSO logistic regression with top 10 features 7
• Time-dependent AUROC at 1,2,3,4 years (expected 0.78-0.83) 5
• C-index for discrimination (expected 0.78-0.79) 5
• Calibration plots
• Decision curve analysis for clinical utility
• Apply fairness adjustments using equal opportunity postprocessing to reduce disparities across racial/ethnic subgroups 7

---

Data Management

Data Sources

• Electronic health records (EHR) from multi-center healthcare organizations
• Administrative claims databases (e.g., Optum MarketScan, TriNetX)
• Laboratory information systems
• Radiology/imaging databases
• Pathology databases (for biopsy results)

Data Quality Assurance

• Validate ICD-10 codes with chart review (random sample of 5-10%)
• Cross-reference diagnoses with laboratory/imaging data
• Identify and exclude duplicate records
• Flag implausible values (e.g., BMI <10 or >80, age <18)
• Document missing data patterns

Variable Definitions

Standardize all definitions:

• MASLD: ICD-10 K76.0, K75.81 + ≥1 metabolic risk factor + exclusion of other liver diseases
• Cirrhosis: ICD-10 K74.x with supporting imaging/laboratory/clinical evidence
• HCC: ICD-10 C22.0 with imaging or pathology confirmation
• Decompensation: ICD-10 codes for ascites (R18.x), encephalopathy (K72.x), variceal bleeding (I85.x)

Missing Data Handling

• Report extent and patterns of missingness
• Multiple imputation by chained equations (MICE) for <20% missing
• Sensitivity analysis comparing complete case vs imputed datasets
• Do not impute outcome variables

Data Security & Privacy

• De-identified dataset per HIPAA regulations
• Secure data storage with encrypted access
• Limited access to authorized personnel only
• IRB approval or waiver (retrospective, de-identified data)

---

Anticipated Challenges & Mitigation Strategies

Challenge 1: Low Biopsy Rates & Selection Bias

Problem: Only 4% undergo biopsy; biopsy patients have 9× higher HCC risk, 3× higher cirrhosis risk 4

Mitigation:

• Do not rely on biopsy for case definition - use clinical/laboratory/imaging criteria 1
• Stratify all analyses by biopsy status
• Conduct sensitivity analyses excluding biopsy patients
• Use propensity score matching to balance biopsy vs non-biopsy groups
• Acknowledge limitation that histologic MASH cannot be confirmed in most patients 1

Challenge 2: Heterogeneous Non-Invasive Testing Patterns

Problem: Only 7.5% undergo NIT at diagnosis; usage varies by practice setting 1

Mitigation:

• Calculate FIB-4 for all patients with available data (age, AST, ALT, platelets)
• Document NIT availability and timing
• Analyze NIT use as time-varying covariate
• Stratify outcomes by baseline FIB-4 category
• Acknowledge that ~10% with advanced fibrosis may be missed by FIB-4 <1.3 1

Challenge 3: Competing Mortality Risk

Problem: MASLD patients have paradoxically lower VD/AD but higher MCI, suggesting survivor bias 2

Mitigation:

• Competing risk regression (Fine-Gray models) for liver outcomes
• Separate analysis of all-cause mortality
• Stratify by age and comorbidity burden
• Acknowledge potential for competing risks in interpretation

Challenge 4: Racial/Ethnic Disparities in Risk Factors

Problem: MRF rankings differ by ethnicity (obesity strongest in Asians/Whites/Hispanics; hypertension strongest in Blacks) 3

Mitigation:

• Mandatory stratification by race/ethnicity in all multivariable models
• Test for interaction between race/ethnicity and each MRF
• Apply fairness adjustments in prediction models 7
• Report subgroup-specific effect estimates
• Ensure adequate representation of minority groups

Challenge 5: Evolving Nomenclature

Problem: Transition from NAFLD to MASLD terminology (2023-2024)

Mitigation:

• Include both NAFLD and MASLD ICD-10 codes
• Document date of diagnosis relative to nomenclature change
• Sensitivity analysis by diagnosis era
• Use MASLD terminology throughout manuscript

Challenge 6: Confounding by Indication

Problem: Sicker patients more likely to receive intensive monitoring, NIT, or biopsy

Mitigation:

• Propensity score methods for treatment/testing comparisons
• Instrumental variable analysis if appropriate instruments identified
• Multivariable adjustment for disease severity markers
• Transparent reporting of confounding limitations

Challenge 7: Loss to Follow-Up

Problem: Patients may transfer care, disenroll from insurance, or be lost to follow-up

Mitigation:

• Require minimum 6-12 months continuous enrollment
• Censor at last known contact date
• Inverse probability of censoring weighting
• Sensitivity analysis varying follow-up requirements
• Report median follow-up duration with IQR

Challenge 8: Misclassification of Alcohol Use

Problem: Alcohol consumption often underreported or undocumented

Mitigation:

• Use multiple data sources (ICD codes, clinical notes, substance use screening)
• Exclude patients with documented alcohol use disorders
• Sensitivity analysis with stricter alcohol exclusion criteria
• Acknowledge residual misclassification as limitation

---

Timeline & Feasibility

Month 1-2: Protocol finalization, IRB submission, data access agreements
Month 3-4: Data extraction, cleaning, variable creation
Month 5-6: Descriptive statistics, cohort characterization
Month 7-9: Primary analyses (Cox models, progression predictors)
Month 10-11: Subgroup analyses, sensitivity analyses
Month 12: Model validation, manuscript preparation

Expected sample size: 50,000-100,000 MASLD patients based on recent studies 7, 6, 4, 3
Expected follow-up: Median 2-5 years (sufficient for 15.9% progression rate) 6

---

Ethical Considerations

• IRB approval or waiver for retrospective, de-identified data 2
• Compliance with HIPAA, GDPR (if applicable)
• No direct patient contact or intervention
• Results will inform clinical practice and guideline development
• Transparent reporting of conflicts of interest
• Data sharing per journal/funder requirements

---

Clinical Implications

This study will:

• Validate the 2024 EASL-EASD-EASO guideline recommendations in real-world practice 1
• Identify high-risk subgroups requiring intensive monitoring
• Inform resource allocation for NIT and specialty referral
• Support development of EHR-integrated risk prediction tools 7
• Guide primary care screening strategies
• Provide natural history data for clinical trial design
• Highlight disparities requiring targeted interventions 7, 3