Two reporting guidelines get confused constantly because they sit next to each other in a clinical study’s statistical toolkit: STARD (Standards for Reporting of Diagnostic Accuracy Studies) and TRIPOD+AI (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis, updated for AI/ML). Both are hosted in the EQUATOR Network’s Reporting Guidelines Library, both are frequently required by journals at submission, and both cover studies that produce a number meant to help classify or predict a patient’s condition — which is exactly why authors mix them up. They are not interchangeable. This page explains what each one is, what its checklist actually asks for, and — the question that actually matters when a deadline is looming — which one (or both) your study needs.
STARD vs. TRIPOD+AI at a Glance
| STARD 2015 | TRIPOD+AI | |
|---|---|---|
| Full name | Standards for Reporting of Diagnostic Accuracy Studies | Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis + Artificial Intelligence |
| What it reports | A study that evaluates how accurately an index test classifies patients against a reference standard (sensitivity, specificity, and related accuracy measures) | A study that develops, validates, or updates a multivariable model predicting an individual’s diagnosis or prognosis, using regression or machine-learning methods |
| Checklist length | 30 items | 27 items |
| Primary citation | Bossuyt PM, Reitsma JB, Bruns DE, et al., for the STARD Group. STARD 2015. BMJ 2015;351:h5527 | Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement. BMJ 2024;385:e078378 |
| Supersedes | STARD 2003 | TRIPOD 2015 (22-item checklist) |
| Typical study design | Cross-sectional or cohort study comparing one or more index tests to a reference standard | Model development study, external validation study, or a study updating/extending an existing model |
Last verified 2026-08-16 against the EQUATOR Network’s guideline pages and the TRIPOD Statement website; see sources linked throughout.
What Is STARD (the STARD Checklist)?
STARD — Standards for Reporting of Diagnostic Accuracy Studies — is the reporting guideline for studies that ask: how well does this test classify patients who do or don’t have a target condition? The current version, STARD 2015, is a 30-item checklist published simultaneously across several journals, with the primary citation Bossuyt et al., “STARD 2015: An Updated List of Essential Items for Reporting Diagnostic Accuracy Studies,” BMJ 2015;351:h5527 (also published in Radiology and Clinical Chemistry the same year). It replaced the original STARD 2003 checklist.
STARD applies to studies with a specific structural shape: one or more index tests (the test being evaluated) are compared against a reference standard (the best available method for establishing whether the target condition is truly present), in a defined patient sample, to generate accuracy measures — sensitivity, specificity, predictive values, likelihood ratios, or an ROC/AUC summary. If your study fits that shape, STARD is very likely the guideline your target journal expects. CASRAI’s guide to sensitivity, specificity, and the 2×2 table covers the statistics STARD reporting is built around, including PPV/NPV, likelihood ratios, and ROC curves.
The STARD checklist is organized into the same broad sections most CONSORT-family guidelines use: title/abstract/keywords; introduction (study objectives and hypotheses); methods (study design, participants, test methods, reference standard, sample size, statistical analysis); results (participant flow, baseline characteristics, test results, accuracy estimates, adverse events); and discussion. A standard STARD flow diagram, tracking participants from enrollment through to final classification, is expected alongside the checklist itself.
What Is TRIPOD+AI (the TRIPOD Checklist)?
TRIPOD+AI is the current reporting guideline for studies of multivariable prediction models — models that combine several predictors to estimate an individual’s probability of a diagnosis (a condition present now) or a prognosis (an outcome in the future). It is a 27-item checklist, published as Collins GS, Moons KGM, Dhiman P, et al., “TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods,” BMJ 2024;385:e078378.
TRIPOD+AI supersedes the original TRIPOD 2015 statement (a 22-item checklist, Collins GS, Reitsma JB, Altman DG, Moons KGM, published in 2015). The “+AI” extension exists because the original TRIPOD items assumed a regression-based model and didn’t adequately cover machine-learning-specific reporting needs: how the model was trained and tuned, how hyperparameters were selected, how code and data availability are disclosed, and how model performance was assessed for overfitting, calibration, and fairness across subgroups. Per its developers, TRIPOD+AI is deliberately written to apply to both regression models and machine-learning/AI models, so it is now the single guideline covering the full spectrum of prediction-model studies rather than a fork.
TRIPOD+AI applies across three distinct study types, and authors need to identify which one theirs is before starting the checklist: model development (building a new model from a dataset, with or without internal validation), external validation (testing an already-published model’s performance in a new, independent dataset), and model updating or extension (recalibrating or extending an existing model with new data or predictors). The checklist items differ slightly by type — a validation study, for instance, doesn’t need to report model-building steps that never happened.
Which Guideline Does Your Study Need?
- Use STARD if your study evaluates a single test (or compares a small number of tests) against a reference standard and reports sensitivity/specificity-type accuracy measures for a binary or ordinal classification.
- Use TRIPOD+AI if your study combines multiple predictors into a model — regardless of whether that model is a logistic regression equation or a trained machine-learning classifier — to generate an individual-level diagnostic or prognostic probability.
- Use both if your study does both things: for example, you develop a multivariable diagnostic model (TRIPOD+AI) and then evaluate that model’s classification accuracy against a reference standard in the same paper (STARD-style accuracy reporting). This overlap is common in AI-based diagnostic model papers, and some journals explicitly ask authors to complete both checklists in that situation.
- Check your target journal’s Instructions for Authors first. Journals that require diagnostic or prediction-model reporting almost always name the specific guideline and version they expect, and many require the completed checklist as a submitted supplementary file with page/line numbers filled in.
The EQUATOR Network’s registry also lists STARD-AI, a proposed extension specifically for AI-centred diagnostic accuracy studies. As of this page’s last verification it has not reached the same finalized, citable-statement status as STARD 2015 or TRIPOD+AI — check the EQUATOR Reporting Guidelines Library directly for its current status before relying on it for a submission.
Using the Checklists at Journal Submission
Both guidelines work the same way in practice:
- Download the current checklist (STARD 2015 or TRIPOD+AI) directly from the EQUATOR Network or the guideline’s own statement website — not a secondhand copy, since item wording and numbering have changed between versions.
- Complete it as a manuscript-tracking document: for each item, record the manuscript page and line number where it’s addressed.
- Submit the completed checklist as a supplementary file, exactly as instructed by the target journal — most journals in this space require it, not just recommend it.
- Include the corresponding flow diagram (STARD) or model-development/validation flowchart (TRIPOD+AI) as a manuscript figure, not only in the checklist.
Neither checklist replaces methodological rigor — they are reporting standards, not study-design or risk-of-bias tools. For diagnostic accuracy studies specifically, the companion risk-of-bias tool most journals and systematic reviewers expect is QUADAS-2; for prediction models, the equivalent is PROBAST. STARD and TRIPOD+AI tell you what to report; QUADAS-2 and PROBAST assess whether the study was designed well enough to trust.
How STARD and TRIPOD+AI Relate to Other Reporting Guidelines
Both belong to the same EQUATOR-coordinated family as the trial and observational-study guidelines CASRAI already covers: the CONSORT Statement for randomized trials and STROBE for observational studies follow the same essential-items-checklist model, just for different study designs. If you’re not certain which of the 400-plus guidelines in the EQUATOR Library applies to your study at all, start with CASRAI’s guide to finding the right EQUATOR reporting guideline before diving into STARD or TRIPOD+AI specifically. And because both guidelines are built around classification and predictive-accuracy statistics, it’s worth reviewing CASRAI’s sensitivity, specificity, and ROC curve guide alongside either checklist — several checklist items ask you to report exactly those numbers.
Frequently Asked Questions
What does STARD stand for?
STARD stands for Standards for Reporting of Diagnostic Accuracy Studies. The current version is STARD 2015, a 30-item checklist.
What is TRIPOD+AI, and is it different from TRIPOD?
TRIPOD+AI is the 2024 update to the original 2015 TRIPOD statement, expanding the original 22-item regression-focused checklist to 27 items that also cover machine-learning and AI-based prediction models. TRIPOD+AI supersedes TRIPOD 2015; new manuscripts should use TRIPOD+AI rather than the original.
Do I need both STARD and TRIPOD+AI for one study?
Only if your study both develops/validates a multivariable prediction model and reports that model’s diagnostic accuracy against a reference standard. A study that only evaluates a single existing test’s accuracy needs STARD alone; a study that only develops or validates a model without reporting classification accuracy against a reference standard needs TRIPOD+AI alone.
Is completing the checklist mandatory for journal submission?
It depends on the journal, but for diagnostic-accuracy and prediction-model manuscripts in medical and health-sciences journals, requiring a completed checklist as a supplementary submission file is now standard practice at most major titles. Check the specific journal’s Instructions for Authors.
Where can I download the official checklists?
Both are available free through the EQUATOR Network’s Reporting Guidelines Library, which links to each guideline’s own maintained website and the primary journal publication.
