Evaluating diagnostic tests and prediction models

Aim and background of TG6

TG6 aims to produce guidance on methods for the development and evaluation of diagnostic tests and prediction models.

Recent years have seen a spectacular rise in the number of assays, imaging modalities, models and other tests to improve medical care. This is likely related to an increase in knowledge about biology and the recent popularity of flexible algorithms, including machine learning and artificial intelligence. Reviews have indicated that many model development studies use suboptimal methods, that many lab tests and models are not properly validated, and that validation studies often do not address all relevant issues. In addition, development and validation studies for prediction models are poorly reported, which limits their usability.

Some of the members of TG6 have been active in related initiatives such as the EQUATOR network. They have published reporting guidelines for diagnosis (STARD: Standards for reporting diagnostic accuracy studies) and prognosis (TRIPOD: Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis).

Members of TG6 and researchers with related interests have regularly met over recent years to collaborate. TG6 has links with all other topic groups in the STRATOS initiative, and specifically with TG2 (Selection of Variables and Functional Forms in Multivariable Analysis) and TG8 (Survival Analysis).