New TRIPOD-Code Protocol Aims to Standardize Prediction Model Reporting

Researchers have published a detailed protocol outlining the development of TRIPOD-Code, a new reporting guideline designed to improve the transparency and reproducibility of code repositories associated with diagnostic and prognostic prediction model studies. The protocol, published in the journal Diagnostic and Prognostic Research in February 2026, describes the specific methods and processes that will be used to create this extension to the existing Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement. The primary focus of the new guideline is the transparent reporting of analytical code, including the software used for data preprocessing, model development, and model evaluation, which is essential for verifying the results of prediction model studies.

The need for such a guideline arises from the increasing complexity of prediction models in the biomedical literature. While early models typically relied on regression methods, recent years have seen a rapid increase in the use of machine learning and other forms of artificial intelligence. In response to this shift, the TRIPOD+AI guidelines were published in 2024 to provide unified reporting recommendations for studies using either traditional statistical models or machine learning approaches. A key addition in the TRIPOD+AI guidelines was a new “Open Science” section, which includes an item encouraging researchers to report whether the analytical code is available, yet sharing this code in a standardised way remains a significant challenge.

Current practices for sharing code vary considerably, often leading to issues with usability and completeness. Important details are frequently missing from code repositories, such as how the code is structured, what dependencies it requires, any license restrictions, testing details, or how it will be archived for long-term access. Without this information, it becomes difficult for other researchers to reproduce results, assess the quality of a study, or use and extend the code for future research. TRIPOD-Code aims to address this gap by providing a checklist of minimum reporting requirements for code quality and availability in prediction model studies.

The development of TRIPOD-Code will follow published guidance from the EQUATOR Network and will comprise five distinct stages. Stage 1 involves a methodological review, specifically a meta review, to evaluate how code availability is currently reported in published prediction model studies. This review will assess adherence to the TRIPOD Statement on code sharing and identify specific issues related to code transparency and reproducibility that are not currently addressed by existing guidelines. The findings from this stage will help identify candidate reporting items for consideration in the new extension.

Stage 2 of the protocol involves a Delphi exercise, a structured process for achieving consensus among experts. A diverse group of key stakeholders, including researchers, journal editors, funders, and healthcare providers, will participate in this international survey. The Delphi process will comprise a series of rounds, with a maximum of three, in which panelists independently and anonymously evaluate proposed reporting items. Participants will rate items on a 5-point Likert scale regarding their agreement with inclusion in the TRIPOD-Code extension, and items achieving a high level of agreement, defined as at least 70 percent support, will be taken forward to the next stage.

Stage 3 consists of a virtual consensus meeting where the results from the Delphi exercise will be discussed and finalised. This meeting will include members of the executive committee and a purposively selected subset of Delphi participants, aiming to ensure representation across key stakeholder groups. The objective is to finalize the items to be included in the TRIPOD-Code reporting guideline. An item will be retained for inclusion in the final checklist if it achieves at least 70 percent support from the consensus meeting participants, while items falling just below this threshold may be flagged for further consideration or merged with overlapping concepts.

Stage 4 involves the actual development of the TRIPOD-Code statement. The executive committee will lead this process, based on the list of items agreed during the consensus meeting. The committee will reserve the right to revise checklist items during development, potentially as a result of feedback from pilot testing. A full draft manuscript, including the finalized checklist and a description of its development process, will be circulated to all consensus meeting participants for their comments, further refinement, and approval before publication.

The final stage, Stage 5, focuses on the dissemination of the guideline. The strategy will aim for simultaneous publication in key journals to reach diverse readerships. The TRIPOD-Code checklist will be published open access and made available on the TRIPOD website alongside other extensions, as well as indexed on the EQUATOR Network website. Social media channels and conferences will also be utilized to promote the uptake of the new guideline, with the executive committee and consensus meeting participants encouraged to publicize the statement widely.

The TRIPOD-Code working group includes an executive committee, an advisory and working group, and a large international Delphi panel. The executive committee comprises lead authors of the original TRIPOD reporting guidelines, along with prediction model experts and researchers with a focus on code availability and reproducibility in machine learning. The study has been determined to be exempt from ongoing Institutional Review Board oversight by the Massachusetts Institute of Technology Committee on the Use of Humans as Experimental Subjects, under Exempt ID E-6675.

TRIPOD-Code is intended to complement the existing TRIPOD guidance by addressing reporting considerations specific to analytical code and computational reproducibility. It is not intended to replace the extensive guidance developed in the software engineering field on good coding practice, documentation, and testing, but rather to draw on and contextualise the most relevant principles for prediction model studies. The checklist will cover key elements such as documentation of software dependencies, specification of license terms, and whether the code is modular, tested, or reproducible. By standardising expectations around code availability and quality, the guideline aims to support reproducibility, facilitate independent validation and implementation, and ultimately increase the impact of prediction models in healthcare.

The protocol highlights the distinction between analytical code and model implementation code. TRIPOD-Code primarily targets the transparent reporting of analytical and evaluation code that underpins published results. However, it recognises that model implementation code, which is required for applying a trained model to new data, may be shared separately as part of model specification. The growing use of complex AI models has introduced new challenges, including the need to manage model artifacts, such as trained weights or checkpoint files, separately from the codebase. TRIPOD-Code will treat model sharing as distinct from code sharing and may refer to complementary practices or emerging standards specific to model specification and availability.

The study also notes that while analytical code is central to computational reproducibility, full reproducibility typically depends on access to the underlying data. TRIPOD-Code focuses specifically on reporting the availability and structure of code and aims to enhance transparency even in settings where data sharing is not feasible due to privacy, governance, or licensing considerations. In such cases, providing demo or synthetic data can offer substantial benefits by enabling others to understand, test, and reuse the code. Although designed with clinical prediction models in mind, the researchers anticipate that many of the recommendations will be applicable to any field where scientific code underpins research findings, as the principles of transparency, reproducibility, and accessibility are broadly applicable.

The publication of this protocol marks a significant step towards improving the quality and reliability of prediction model studies in the biomedical literature. By providing a clear framework for reporting code, TRIPOD-Code aims to assist reviewers, editors, policymakers, and end users in understanding the methodologies and findings of these studies. The expected outcomes include a study protocol, a review paper, and the final TRIPOD-Code statement. All participants of the online consensus will be offered authorship on the final statement, provided they meet the requirements for authorship, ensuring that the guideline reflects the collective expertise and agreement of the international research community.

Subject of Research: Development protocol for the TRIPOD-Code reporting guideline for prediction model code repositories

Article Title: Protocol for development of a reporting guideline (TRIPOD-Code) for code repositories associated with diagnostic and prognostic prediction model studies

Article References: Pollard, T., Sounack, T., Gao, C. A., Celi, L. A., Lindvall, C., Lee, H., Lee, H.-C., Moons, K. G. M., & Collins, G. S. (2026). Protocol for development of a reporting guideline (TRIPOD-Code) for code repositories associated with diagnostic and prognostic prediction model studies. Diagnostic and Prognostic Research, 10(1), Article 4. https://doi.org/10.1186/s41512-025-00217-4

Image Credits: AI Generated

DOI: 10.1186/s41512-025-00217-4

Keywords: TRIPOD-Code, prediction models, reproducibility, code sharing, reporting guidelines, machine learning, EQUATOR Network, Delphi method, computational reproducibility, biomedical research, Protocol, development