The TRIPOD extension for reporting of prediction model studies in large, clustered, datasets

The TRIPOD extension for reporting of prediction model studies in large, clustered, datasets
Speaker: Thomas Debray
  • 15th Jul 2019
  • 1:48 PM TO 2:06 PM
  • 40th Annual Conference of the International Society for Clinical Biostatistics
  • Leuven, Belgium

Context: The TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis) Statement is a guideline to improve the reporting of studies developing, validating, or updating a prediction model. The guideline has been adopted by numerous journals and is widely used by authors to prepare their manuscript for publication. With the increased availability of large datasets (or "big data") from electronic health care records and individual participant data meta-analysis, authors face novel challenges for conducting and reporting their prediction modeling research, as eminent by related reporting guidelines such as PRISMA-IPD, STROBE and RECORD.

Objective(s): To propose an extension for the TRIPOD statement to improve the conduct and reporting of prediction model studies using big and combined data sets.

Method(s): We formed a steering committee in 2016 to discuss which TRIPOD items needed revision, and whether additional items were needed. We subsequently developed a formal extension and evaluated the proposed modifications through a Delphi survey in February 2019. Hereto, we invited 77 experts in prediction model research from various countries including the Netherlands, the United Kingdom, and the United States.

Results: The original TRIPOD statement describes 37 reporting items, divided in 22 topics. Our extension modifies 18 of these items, and includes 9 entirely new items. For each item, we discuss good reporting practice and provide recent examples. In this talk, we will present the TRIPOD extension and the recommendations from the Delphi panel.

Conclusions: Our TRIPOD extension provides additional reporting guidance for the retrieval, evaluation, harmonization and analysis of multiple data sources when developing or validating a prediction model. We hope that it will provide a valuable reference when performing prediction model research using large, clustered, datasets, and when evaluating the resulting manuscripts.

#