Bitten by JAGuAR (Just Acronym Guidelines Annoying my Research)? — Blog and news on ethical AI and statistics in biomedicine.

Perhaps you are like me and feel like drowning in the acronym soup of statistics guidelines? CONSORT, STROBE, PRISMA, ARRIVE… each promises to rescue science from the flow of irreproducible studies. But sometimes, the more acronyms we have, the more tangled things seem to be. Yet, these guidelines deeply matter and deserve a post.

Today, I invite you to a quick tour through the main statistical recommendations that help keep life-science research credible, transparent, and, ideally, publishable.

The purpose of this post is threefold:

Create a short guide so that you may quickly spot the guidelines most relevant to your research
Help you find guidelines you may confidently rely on and refer to as citations in your scientific materials
Raise awareness about the omnipresence of statistics guidelines regardless of the life science sub-domain you work in

Some of the following guidelines, especially those linked to clinical research, are accessible through the Enhancing the QUAlity and Transparency Of health Research (EQUATOR) network database. Some were put together through a Delphi method, a procedure where a group of experts come to a consensus about a set of recommended items, others were initiatives from more isolated teams. In this latter case, the published guidelines are sometimes specific to certain publishers, but you may perfectly use (and cite) them to build your design and reporting. It is possible that some complementary guidelines, not cited here, may also be relevant to your specific field. Feel free to comment on your own experience and needs at the end of this article.

Opening disclaimer: reporting vs. methodology

Some guidelines (especially in clinical science) emphasize transparent reporting of statistical procedures to enhance reproducibility through clarity and openness. Others focus on promoting rigor in statistical design to strengthen reproducibility from methodology. Although their primary aims differ, these approaches often overlap. We will try to clarify which guideline belongs to each class.

In a nutshell, we can summarize from these guidelines a set of shared principles that, when consistently applied, promote both transparency and rigor in statistical practice. On one hand, transparency is fostered through exhaustive disclosure of all pieces of the study design, from randomization, blinding, and sample size justification to the precise definition of replicates, chosen significance levels, error used (eg.: SD, CI), analytical methods (tests), and software, complemented by open access to data and code. On the other hand, rigor stems from thoughtful planning during the design phase, ensuring that sample size, confounding factors, randomization, blinding, definitions of variables, and pre-specified future statistical analyses (with backup options that will be adjusted to the actual data) are all explicitly justified and subsequently implemented.

Clinical sciences: the big five

Credit where credit is due, we will start with clinical and epidemiological sciences, a field that boasts the largest number of guidelines owing to the ethical stakes tied to human rights and the demand for the highest research standards.

CONSORT and SPIRIT

Born in the mid-1990s and now updated several times (the latest version from 2025), the Consolidated Standards of Reporting Trials (CONSORT) statement was designed to improve transparency in reporting results from randomized controlled trials (RCTs). Its famous 25-item checklist and flow diagram help ensure that readers know who was enrolled in the trial, randomized, treated, and lost to follow-up. Without CONSORT, many RCTs reports would still read like irreproducible medical diaries. The question of reporting the RCT results being addressed, the issue of reporting protocols in a transparent and reproducible manner had then to be solved.

The Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT), was first published in 2013 with that exact objective. It is a mixed guideline in the sense that it aims to ensure both completeness and transparency of the protocol. These two sets of recommendations have very strong connections, and are now often referred to jointly as the combined SPIRIT/CONSORT guidelines.

Strengthening the Reporting of Observational Studies in Epidemiology (STROBE)

For observational studies, which collect data from individuals or groups of individuals and explore correlations between variables with no intervention, the STROBE guideline should be the guide. Therefore, if you use cohorts, case-control, or cross-sectional data, make sure you are familiar with the STROBE checklist.

Standards for Reporting Diagnostic Accuracy Studies (STARD)

Diagnostics research presents specificities such as the quantification of sensitivity, specificity, AUROC curves, etc, which are tackled by STARD. It focuses on reporting transparency: how were reference standards defined, who interpreted what, and what cut-offs were used.

Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD)

TRIPOD is the new-generation tool for predictive models relevant for multivariable models, those algorithms that aggregate many variables to predict relapse, recovery, or response to treatments. The statement includes a checklist with 22 items, which aims to improve the reporting of studies about the development, validation, or update of prediction models. The original 2015 TRIPOD statement has been updated and superseded by more comprehensive checklists that encompass AI algorithms (see below, be patient!).

Amended guidelines for AI-driven and AI-powered clinical research

AI has entered the life sciences. The consequence is a series of extensions of the above-introduced clinical guidelines that incorporate the use of AI, most of which were released very recently in 2025.

For example, in 2020, the CONSORT-AI for published trials and SPIRIT-AI for trial protocols came as a pair to ensure that AI-based interventions are described with enough detail for others to reproduce (or at least understand) what the model actually does. Similarly, the STARD-AI extends the STARD checklist to clarify model training, validation, and generalization.

TRIPOD+AI has been coined for multivariable prediction models using AI, which adds reporting elements for machine-learning–based approaches (e.g.: details of hyperparameter tuning). Importantly, TRIPOD+AI provides guidance for reporting prediction studies, irrespective of whether regression or machine learning have been used, meaning that It now supplanted the obsolete 2015 TRIPOD checklist, which should not be used anymore. Last mutated form of TRIPOD: TRIPOD-LLM, an extension of 19 main items and some subitems, which particularly insist on the context of Large Language Models (LLMs) in healthcare. Because if you use an LLM to help interpret data or write text, readers need to know where the “AI hand” starts and ends (if you are based in Switzerland, I recommend you read my recent blog post about LLM in medicine).

Finally, let’s mention the Chatbot Assessment Reporting Tool (CHART) that offers reporting guidelines for studies evaluating the performance of chatbots using generative artificial intelligence when summarizing clinical evidence and providing health advice.

Non-clinical guidelines

Of course, life sciences extend far broader than clinical research. The worlds of cell lines, molecular biology, transgenic plants, flies, worms, or rodent models, all equally deserve rigorous statistics. I am, myself, well placed to comment on that considering the countless hours I spent in tissue culture rooms or doing preclinical molecular biology.

Animal Research: Reporting of In Vivo Experiments (ARRIVE)

If you work with animal models, the ARRIVE guidelines are non-negotiable. First published in 2010, the checklist was revised in 2020 (ARRIVE 2.0) as two separate checklists of 10 essential items (e.g.: type of design, sample size, randomization and blinding methods, statistical tests used) and 11 recommended items. ARRIVE aims at transparent reporting of research involving animals, with the ultimate objective to prevent waste of laboratory animals in full alignment with the guiding concept of reduction in the 3R (Reduction, Replacement, Refinement) ethical principles of animal research.

Cell biology and physiology

In cell biology and physiology, Pollard and colleagues (2019) and Curran-Everett and Benos (2004) provided complete guidance beyond reporting, which you may refer to. For recommendations more specifically targeting data display and result reporting, you may explore the guidelines from Michel and colleagues (2020). These three articles were released as journal guidelines, but you may use them as inspirations and references for your research.

Statistical standards in plant sciences

If you are a plant biologist searching for official statistical guidelines, you may find yourself a little left behind. It is rather ironic, and somewhat disappointing, that so little material is available, given that Ronald Fisher, one of the founding fathers of statistics, made most of his monumental contributions in agricultural research. Researchers in plant biology may not be under the same ethical pressure as clinical or preclinical researchers, but their data deserve no less rigor to ensure reproducibility in biotechnology and prevent potential wastes of financial resources. We may mention the journal guidelines by Jeon (2022) who proposed statistical guidance tailored for the Plant pathology journal, as well as the article by Massart (2022) addressed to scientists involved in the validation of plant diagnostic tests.

Reporting in antibody research and quantitative RT-PCR

Biochemistry or molecular biology too are concerned with statistical considerations. Rigor in experimental design, data analysis, and transparent reporting is essential in biochemistry and molecular biology, to ensure reproducibility, but also because these methods often form part of larger studies with significant ethical implications, such as clinical or preclinical research.

Brooks and Lindley (2018) have released a general guideline for research involving antibodies, updated by a more comprehensive guideline in 2024. They make it clear that in western-blots, ELISA, FACS or cell labelling, statistically-relevant items should be included in reporting, such as statistical methods or whether blinding is used. I am particularly sympathetic to this section, owing to a decade I dedicated performing immunoblots and immunolabelings where I recurrently witnessed the lack of statistical culture, or even interest, in the field.

Similarly, the Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE) guidelines, first published in 2009 and updated in 2025 (MIQE 2.0), establish standards for the design, execution, and reporting of quantitative PCR (qPCR). For example, the MIQE 2.0 checklist includes items on the transparency of statistical procedures, the choice of significance level, and calculation of statistical power.

The case of evidence synthesis: the PRISMA checklist

Evidence syntheses (i.e.: systematic reviews and meta-analyses) are at the top of evidence-based research, their transparency is guaranteed by the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) checklist. Its structured approach makes sure reviews and meta-analyses can be replicated and updated. The PRISMA guidelines were revised in 2020. The PRISMA guideline has spawned multiple extensions specific to certain research fields, with some still in development. Although the world of systematic reviews and meta-analysis is very much clinical, honorable exceptions must be made to the PRISMA-EcoEvo that is relevant to primary research in ecology and evolutionary biology, and to the prospective and awaited PRISMA Extension of Preclinical In Vivo Animal Experiments whose protocol was registered in 2020.

We may also mention TRIPOD-SRMA that provides guidance for systematic reviews and meta-analyses (hence the SRMA) in the field of multivariate prediction models.

Conclusion: relying on one reporting guideline is usually not enough

Guidelines are sometimes overlapping in the context of a single research project. For example two separate research projects on human clinical sciences and molecular plant biology may refer to distinct guidelines on clinical trials and plant pathology, but they might also both need to abide by guidelines on antibody research if they use western-blots or ELISA. The good news is that these guidelines often have quite overlapping recommendations bearing the mark of statistical common sense such as the full disclosure of the statistical methods used or details about randomization and blinding procedures. Furthermore, interdisciplinary projects can perfectly refer to multiple guidelines in the sections dedicated to specific techniques or fields. The more cautious in terms of planning and reporting, the better. I will conclude by quoting Curran-Everett and Benos (2004) in their opening recommendation: “If in doubt, consult a statistician when you plan your study”.