Why do authors publish radiological multivariable models: Questionnaire survey of motivations, evaluation, and usage...

Multivariable models currently comprise a large proportion of indexed radiological literature but authors' motivation to develop and publish models, available resource, plans for external evaluation, and whether models are used, is generally unknown. We aimed to answer these questions via a questionnaire survey.

We administered an online questionnaire to authors reporting multivariable prognostic or diagnostic models developed using regression, published in the top 3 indexed general radiological journals 2013 to 2015 and December 2022 to May 2023 inclusive.

34 authors responded. Regarding external evaluation, the most frequent response (49%, 17/35 responses) stated no intention to do this. Regarding barriers to evaluation, respondents cited lack of time (27%, 18/67 responses; multiple responses were possible), funding (22%, n = 15), and interest in development alone (19%, n = 13). 67% (n = 23) declared no or little statistical support. Most (82%, n = 28) did not use their own model in clinical practice and most (56%, n = 19) also believed other clinicians would not use the model either. Most (79%, n = 27) did not use models in general. A substantial proportion (29%, n = 10) reported model development to progress their career via publication.

Most respondents had neither evaluated their model nor intended to do so. Most did not use their own model in clinical practice, nor did they believe others would. Statistical assistance was largely absent or minimal.

Our data suggest that the large volume of multivariable models currently available in the indexed medical literature have little clinical utility.