Reporting Guidelines

Reporting Guidelines

Overview

Reporting guidelines ensure transparent, complete, and reproducible research communication. Journals increasingly require adherence to specific guidelines. This reference covers guidelines most relevant to EHR-based clinical research.

Which Guideline to Use?

For most MIMIC/eICU research: STROBE + RECORD (observational) or TRIPOD (prediction)

STROBE: Observational Studies

STrengthening the Reporting of OBservational studies in Epidemiology

Website: strobe-statement.org

Checklist Summary

STROBE Common Gaps in EHR Research

RECORD: Routinely Collected Health Data

REporting of studies Conducted using Observational Routinely-collected health Data

Website: record-statement.org

RECORD extends STROBE specifically for database studies (EHR, claims, registries).

Additional RECORD Items

RECORD-Specific Reporting for MIMIC/eICU

TRIPOD: Prediction Models

Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis

Website: tripod-statement.org

TRIPOD Checklist Summary

TRIPOD Study Types

TRIPOD Critical Items for ML Models

TRIPOD+AI: AI/ML Prediction Models

Extension for studies using artificial intelligence/machine learning methods.

Additional TRIPOD+AI Items

STARD: Diagnostic Accuracy

STAndards for Reporting of Diagnostic accuracy studies

Use when evaluating a test/marker's ability to distinguish disease states.

Key STARD Elements

Reporting Metrics

For Association Studies (STROBE)

Example:

Mortality was higher in the exposed group (adjusted HR 1.45, 95% CI 1.12–1.88, p=0.005) after adjustment for age, sex, SOFA score, and Charlson comorbidity index.

For Prediction Models (TRIPOD)

Example:

The model demonstrated good discrimination (AUROC 0.82, 95% CI 0.78–0.86) and adequate calibration (slope 0.94, intercept -0.12). Decision curve analysis showed net benefit over treating all/none for threshold probabilities between 10% and 40%.

Tables and Figures Checklist

Table 1: Baseline Characteristics

• Stratified by exposure/outcome group
• Continuous: mean (SD) or median (IQR) based on distribution
• Categorical: n (%)
• Missing data: n (%) per variable
• Standardized mean differences (for propensity studies)

Flow Diagram

• Initial population screened
• Excluded at each step with counts and reasons
• Final analytic sample
• Lost to follow-up (if applicable)

Results Table

• Crude and adjusted estimates
• 95% confidence intervals
• P-values
• Number of events and total N

Prediction Model

• Calibration plot
• ROC curve with AUC and CI
• Model coefficients or code availability
• Performance in validation set

Journal-Specific Requirements

Many journals require:

• Completed checklist uploaded as supplementary material
• Statement of adherence in methods section
• Data availability statement
• Code availability statement

Check target journal's author guidelines.

Common Reporting Deficiencies

Quick Reference: Minimum Reporting

Observational Study (STROBE + RECORD)

1. Study design in title
2. Database name, version, time period
3. Eligibility criteria with codes
4. Flow diagram with numbers
5. Variable definitions (codes)
6. Missing data extent and handling
7. Crude and adjusted estimates with CIs
8. Sensitivity analyses
9. Limitations section addressing bias

Prediction Model (TRIPOD)

1. Development or validation study in title
2. Outcome and predictor definitions with timing
3. Sample size and events per variable
4. Missing data handling
5. Model specification (full coefficients or code)
6. Internal validation method
7. Discrimination (AUROC with CI)
8. Calibration (plot + metrics)
9. External validation if applicable
10. Limitations and clinical applicability

Resources

EQUATOR Network (equator-network.org): Central repository for all reporting guidelines.