STROBE, STARD and TRIPOD+AI: Selecting the Reporting Guideline Appropriate to the Study Type

STROBE governs observational studies, STARD governs diagnostic accuracy studies, and TRIPOD+AI governs studies that develop or validate a clinical prediction model, whether the model is built with regression or with machine learning. Selection among them begins with identifying which of these three study types applies.

STROBE for observational studies

The STROBE Statement, from von Elm and colleagues for the STROBE Initiative, is a 22-item checklist covering cohort, case-control and cross-sectional studies, organised across title/abstract, introduction, methods, results, discussion and a funding item1. It is the general-purpose reporting guideline for observational epidemiology and applies when the study observes exposures and outcomes as they occur, without an intervention assigned by the investigator.

STARD for diagnostic accuracy studies

STARD, in its current 2015 update from Bossuyt and colleagues, is purpose-built for studies evaluating how well a test identifies a target condition2. It requires reporting specific to that task (the reference standard used to establish the true condition status, the flow of participants through index testing and reference-standard verification, and a cross-tabulation of index-test results against the reference standard), none of which maps onto STROBE’s general observational-study structure. A diagnostic-accuracy study reported against STROBE instead of STARD will typically omit precisely this test-performance detail.

TRIPOD+AI for prediction models based on regression or machine learning

TRIPOD+AI, from Collins and colleagues, updates and explicitly replaces the original 2015 TRIPOD statement, extending it to cover prediction models built with either regression or machine-learning methods3. It provides a 27-item main checklist (with 52 sub-items), flags each item as applying to model development, external validation, or both, and adds explicit attention to fairness and open-science practices such as registration and code sharing. TRIPOD+AI is a reporting guideline, not a quality-appraisal tool — PROBAST is the companion tool used to assess a prediction-model study's risk of bias, and the two are typically applied together rather than as substitutes for one another.

Key points

Practical application

Frequently asked questions

@misc{strobe_stard_tripod_ai_2026,
  author       = {{Pubrica Academy}},
  title        = {STROBE, STARD and TRIPOD+AI: Selecting the Reporting Guideline Appropriate to the Study Type},
  howpublished = {Pubrica},
  year         = {2026},
  month        = {oct},
  url          = {https://pubrica.com/academy/concepts-definitions/strobe-stard-tripod-ai/},
  note         = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}

Does the study assign an intervention, or only observe exposure and outcome? If observe only → consider STROBE.
Is the study evaluating a diagnostic test against a reference standard? → STARD 2015.
Does the study develop or validate a model that predicts an outcome, by any method? → TRIPOD+AI (and PROBAST for risk-of-bias appraisal).