Published  |  Last updated

Quick Answer: STARD vs TRIPOD comes down to what you are studying. Use STARD 2015 (30 items) when you measure how accurately a test identifies a condition against a reference standard.1 Use TRIPOD+AI (27 items) when you develop or evaluate a multivariable prediction model; it supersedes the 22-item TRIPOD 2015 checklist.2 For an AI-based test, add STARD-AI, which adds 18 new or modified items to STARD 2015.3

STARD vs TRIPOD at a Glance

The two guidelines answer different questions. STARD asks how well one test separates people with and without a condition. TRIPOD asks how well a model that combines several predictors estimates a person’s risk.1,2

STARD 2015TRIPOD+AI (2024)
Study typeDiagnostic accuracy of an index test against a reference standardDevelopment or evaluation of a prediction model, using regression or machine learning
Checklist size30 items27 main items, 52 subitems
Abstract guidanceItem 2 points to STARD for AbstractsTRIPOD+AI for Abstracts, 13 items
Participant flowItem 19: diagram expected in all reportsItem 20a: describe flow; a diagram may be helpful
Accuracy or performanceItem 23 cross-tabulation; item 24 accuracy with precision such as 95% CIsItem 12e measures and plots such as discrimination and calibration; item 23a estimates with CIs
FairnessNo fairness itemItem 14, and fairness is embedded across the checklist
Relation to earlier versionUpdate of STARD 2003Supersedes TRIPOD 2015, which should no longer be used

Swipe sideways to see every column.

Counts differ between sources because of sub-items. The TRIPOD 2015 paper describes a checklist of 22 items;4 the TRIPOD+AI paper describes TRIPOD 2015 as 37 checklist items, including 25 items for both development and validation studies, six for development only and six for validation only.2 An explanation and elaboration document accompanies TRIPOD 2015.5

Which Guideline Should You Use?

Match the guideline to the question your paper answers, not to the field you work in.

  1. 1One test against a reference standard. Use STARD 2015. Examples are a new biomarker, an imaging test or a point-of-care device compared with the best available reference.
  2. 2A model that combines several predictors. Use TRIPOD+AI, whether the model predicts a current condition (diagnostic) or a future outcome (prognostic).2
  3. 3An AI-based test. Use STARD-AI. The TRIPOD+AI paper listed it as in preparation in 2024; it was published in Nature Medicine in 2025.2,3
  4. 4A systematic review of accuracy studies. Use PRISMA-DTA for the review. The included primary studies are the ones that follow STARD.6
  5. 5Both a test and a model in one paper. Report each part against the guideline that matches it, and say so in the methods. This is Editverse practical advice, not a rule from either guideline.

What Changed: STARD 2003 to 2015, TRIPOD 2015 to TRIPOD+AI

STARD first appeared in 2003 in two dozen journals. The 2015 update grouped 30 items under the introduction, methods, results and discussion structure and made a flow diagram expected in all reports. New items include a structured abstract (item 2), intended use and clinical role of the test (item 3), study hypotheses (item 4), sample size (item 18), a structured discussion (items 26 and 27), registration (item 28), protocol (item 29) and funding (item 30).1

TRIPOD+AI covers prediction model studies that use any regression or machine learning method, harmonises the terms of the two communities and puts particular emphasis on fairness. The TRIPOD website announces TRIPOD+AI and TRIPOD-LLM as out (page read 31 October 2026).2,7

Stacked bar chart of checklist items per section. STARD 2015 has 30 items: title and abstract 2, introduction 2, methods 14, results 7, discussion 2, other 3. TRIPOD 2015 has 22: 2, 1, 9, 5, 3, 2. TRIPOD+AI has 27: 2, 2, 13, 5, 3, 2.
Items per section under each guideline’s own headings. Tallied from the published checklists and totalled in R by Editverse. In all three, methods is the largest section.

Real Case: A STARD Study Read Against the Checklist

Villalpando-Sánchez and colleagues tested whether advanced oxidative protein products (AOPPs), measured in blood, can identify advanced chronic kidney disease. It is a prospective diagnostic accuracy study from a family medicine clinic in Morelia, Mexico, run in 2021, and the authors state that it follows STARD. The paper is open access under CC BY 4.0, so we redrew its participant flow in R.8

The index test was AOPPs at or above 200 µmol/L. The reference was stage 4 to 5 kidney disease by glomerular filtration rate, estimated with three equations (Cockcroft-Gault, MDRD and CKD-EPI). A sample size of at least 296 was calculated. Of 392 adults selected, 90 were excluded (20 previous infection, 70 incomplete clinical studies), leaving 302; 152 were test negative and 150 test positive.8

Flow diagram: 392 adults selected, 90 excluded (20 previous infection, 70 incomplete clinical studies), 302 analysed, of whom 152 were test negative and 150 test positive at an AOPPs cut-off of 200 micromol per litre.
Participant flow redrawn by Editverse in R from the counts in the Results text of the case paper. Item 19 of STARD 2015 asks for such a diagram.

What the paper reports against STARD

Editverse read the Methods and Results text and the three tables of the paper against the STARD 2015 list. Items not listed were not checked.1,8

STARD itemWhat the paper showsStatus
18 Sample sizeA calculation giving at least 296 participantsReported
19 Flow diagramA STARD diagram with 392, 90 and 302 participantsReported
9 Type of seriesAdults were invited by a doctor and took part voluntarily; the text does not call the series consecutive, random or convenienceNot stated in these terms
12a Cut-offROC analysis estimated accuracy at the cut-off, and 200 µmol/L was chosen as the value closest to the median AOPPs; STARD asks authors to say whether a cut-off was pre-specified or exploratory, and the paper does not sayNot stated as pre-specified
13 BlindingThe Discussion says all tests were applied and interpreted blindly and independentlyReported
23 Cross-tabulationNo 2 x 2 table of test result by reference result in the text or tablesNot found
24 Accuracy with precisionAUC with 95% CI is reported; no CI is given in the text for sensitivity, specificity or predictive valuesPartly reported

Swipe sideways to see every column.

What Editverse recomputed

Because item 23 is missing, we rebuilt the cross-tabulation in R from the printed numbers. The test result was positive for 150 and negative for 152 of the 302. For Cockcroft-Gault, exactly one table of whole numbers reproduces the reported sensitivity (96.58%) and specificity (80.00%): 113 true positive, 37 false positive, 4 false negative and 148 true negative. It also reproduces the reported positive likelihood ratio of 4.83. For CKD-EPI the matching table is 113, 37, 5 and 147, with likelihood ratio 4.76. For MDRD no table of whole numbers reproduces both reported values. Confidence intervals are Wilson intervals, computed by Editverse.8

Measure, percentCockcroft-Gault: reportedCockcroft-Gault: reconstructed (95% CI)CKD-EPI: reportedCKD-EPI: reconstructed (95% CI)
Sensitivity96.5896.58 (91.5 to 98.7)95.7695.76 (90.5 to 98.2)
Specificity80.0080.00 (73.7 to 85.1)79.8979.89 (73.5 to 85.0)
Positive predictive value96.3675.33 (67.9 to 81.5)95.7675.33 (67.9 to 81.5)
Negative predictive value80.8797.37 (93.4 to 99.0)79.8996.71 (92.5 to 98.6)

Swipe sideways to see every column.

Paired bars. Sensitivity 96.58 reported and recomputed; specificity 80.00 both; positive predictive value 96.36 reported versus 75.33 recomputed; negative predictive value 80.87 reported versus 97.37 recomputed.
Reported and recomputed accuracy for the Cockcroft-Gault reference. The reconstructed table is 113, 37, 4 and 148; all recomputed values were computed by Editverse in R.

What the case shows and what it does not

It shows why item 23 exists. With 150 test-positive participants, a positive predictive value of 96.36% would need 144.5 true positives, which is not a whole number, and the reported predictive values do not follow from the other printed numbers. The reported values may reflect a different calculation or a typing slip; only the authors’ data can settle it, and a printed 2 x 2 table would have let readers check in a minute. The case does not show that AOPPs are inaccurate, that the reported sensitivity or specificity is wrong, or anything about the MDRD result. The reconstruction assumes that the 150/152 split of test results applies to all three reference equations and that the published figures are rounded to two decimals.

Items Specific to Each Checklist

These are examples of items that matter in one checklist and have no close counterpart in the other. They are not complete lists.1,2

STARD 2015TRIPOD+AI
11 Rationale for the reference standard12c Model type, building steps and hyperparameter tuning
15 Handling of indeterminate results12e Measures and plots to evaluate performance, such as discrimination, calibration and clinical utility
22 Time interval between index test and reference standard14 Approaches used to address model fairness
23 Cross-tabulation of results22 Full model specification, such as formula or code, to allow predictions in new individuals
25 Adverse events from performing the tests18f Availability of the analytical code

Swipe sideways to see every column.

Use the Checklist on Your Manuscript

  1. 1Name what you evaluate. A test against a reference standard is a STARD paper; a model that outputs a risk is a TRIPOD+AI paper.
  2. 2Get the current checklist. Download STARD from the EQUATOR Network page and TRIPOD+AI from the TRIPOD website.9,7
  3. 3Add page numbers. STARD asks authors to state, at the start of each item row, the page where the item can be found.1
  4. 4Draw the flow diagram first. List who was selected, excluded, tested and analysed before you write the Results; the diagram then writes the first paragraph for you.
  5. 5Print the numbers readers need. For a test, print the 2 x 2 table and confidence intervals. For a model, report performance with confidence intervals and give the full model.1,2

Need the Analysis Run or Checked? Consult Editverse

Editverse subject-matter experts work to COPE, ICMJE and GPP 2022 guidance, with institutional and country-specific ethics followed.

Choose the stage you are at.

Editverse logo

Editverse

Statistical Analysis

Need the accuracy table or model performance run in SPSS, R or Stata?

You send: your dataset, the study protocol and the reporting guideline you are following.

You get: analysis in SPSS, R, STATA or SAS with annotated code, complete output files and an interpretation guide.

When: Your data are collected and the Results section is not written.

SPSS/R/STATA Statistical Analysis · $400

Get my analysis run →

Editverse logo

Editverse

Manuscript Writing

Want the paper written to STARD or TRIPOD from the first draft?

You send: your data, notes, checklist choice and target journal.

You get: a manuscript written and edited by PhD experts, with statistical review, journal selection and submission support.

When: Your study is finished and the paper is not.

Manuscript Writing Services · $1,500

Write my paper →

Editverse logo

Editverse

Journal Selection

Which journal asks for a STARD or TRIPOD checklist?

You send: your abstract and the kind of study you ran.

You get: 5-10 ranked journal suggestions with a rationale based on scope fit, impact factor and acceptance rates.

When: You have a manuscript and no shortlist.

Scientific Journal Selection Services · $115

Find my journal →

Free Research Tools from Editverse

Frequently Asked Questions

What is the difference between STARD and TRIPOD?

STARD is for studies that measure how accurately a test identifies a target condition, using an index test and a reference standard. TRIPOD is for studies that develop or evaluate a multivariable prediction model for diagnosis or prognosis. The current TRIPOD checklist is TRIPOD+AI.

How many items do STARD and TRIPOD have?

STARD 2015 has 30 items. TRIPOD 2015 has 22 items in its own paper, which the TRIPOD+AI paper describes as 37 checklist items when sub-items are counted. TRIPOD+AI has 27 main items and 52 subitems.

Should I use TRIPOD or TRIPOD+AI?

Use TRIPOD+AI. Its authors state that it supersedes the TRIPOD 2015 checklist, which should no longer be used, and that it covers prediction models built with regression or machine learning.

Can I use STARD for a prediction model study?

STARD is written for test accuracy studies. The TRIPOD 2015 authors noted that STARD and other guidelines contain many items relevant to prediction models, but none is entirely appropriate for them. Use TRIPOD+AI for the model and STARD for a separate test accuracy question.

Is there a reporting guideline for AI diagnostic accuracy studies?

Yes. STARD-AI was published in Nature Medicine in 2025 and adds 18 new or modified items to the STARD 2015 checklist. An author correction was published in 2026, so download the current checklist.

Do I need a flow diagram?

STARD 2015 expects a diagram of participant flow in all reports (item 19). TRIPOD+AI item 20a asks you to describe the flow of participants and says a diagram may be helpful.

What should I use for a systematic review of diagnostic accuracy studies?

PRISMA-DTA, the reporting guideline for systematic reviews and meta-analyses of diagnostic test accuracy studies, published in JAMA in 2018. STARD is for the primary accuracy studies.

Why does STARD ask for a 2 x 2 table?

STARD item 23 asks for a cross-tabulation of index test results by reference standard results. With the counts, readers can recompute every accuracy measure and its confidence interval; without them they cannot, as the case in this post shows.

References

  1. Bossuyt PM, Reitsma JB, Bruns DE, et al. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. 2015;351:h5527. doi:10.1136/bmj.h5527. Open access, CC BY 4.0. pmc.ncbi.nlm.nih.gov/articles/PMC4623764/
  2. 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. doi:10.1136/bmj-2023-078378. Open access, CC BY 4.0. pmc.ncbi.nlm.nih.gov/articles/PMC11019967/
  3. Sounderajah V, Guni A, Liu X, et al. The STARD-AI reporting guideline for diagnostic accuracy studies using artificial intelligence. Nat Med. 2025;31(10):3283-3289. doi:10.1038/s41591-025-03953-8. An author correction was published in Nat Med. 2026;32(9):3493. doi.org/10.1038/s41591-025-03953-8
  4. Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMC Med. 2015;13:1. doi:10.1186/s12916-014-0241-z. Open access, CC BY 4.0. pmc.ncbi.nlm.nih.gov/articles/PMC4284921/
  5. Moons KGM, Altman DG, Reitsma JB, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): explanation and elaboration. Ann Intern Med. 2015;162(1):W1-W73. doi:10.7326/M14-0698. pubmed.ncbi.nlm.nih.gov/25560730/
  6. McInnes MDF, Moher D, Thombs BD, et al. Preferred reporting items for a systematic review and meta-analysis of diagnostic test accuracy studies: the PRISMA-DTA statement. JAMA. 2018;319(4):388-396. doi:10.1001/jama.2017.19163. pubmed.ncbi.nlm.nih.gov/29362800/
  7. TRIPOD statement website: TRIPOD+AI, a 27-item checklist (page read 31 October 2026). www.tripod-statement.org/
  8. Villalpando-Sánchez DC, Barajas-Medina CA, Alvarez-Aguilar C, et al. Advanced oxidative protein products had a diagnostic accuracy for identifying chronic kidney disease in adult population. Metabolites. 2024;14(1):37. doi:10.3390/metabo14010037. Open access, CC BY 4.0; flow figure and all derived numbers redrawn or computed by Editverse in R. pmc.ncbi.nlm.nih.gov/articles/PMC10821176/
  9. EQUATOR Network. STARD: reporting guideline page and checklist downloads. www.equator-network.org/reporting-guidelines/stard/

Editverse Research Team | Updated October 2026