Poor reporting of multivariable prediction model studies: towards a targeted implementation strategy of the TRIPOD statement

Abstract

Background: As complete reporting is essential to judge the validity and applicability of multivariable prediction models, a guideline for the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) was introduced. We assessed the completeness of reporting of prediction model studies published just before the introduction of the TRIPOD statement, to refine and tailor its implementation strategy.

Methods: Within each of 37 clinical domains, 10 journals with the highest journal impact factor were selected. A PubMed search was performed to identify prediction model studies published before the launch of TRIPOD in these journals (May 2014). Eligible publications reported on the development or external validation of a multivariable prediction model (either diagnostic or prognostic) or on the incremental value of adding a predictor to an existing model.

Results: We included 146 publications (84% prognostic), from which we assessed 170 models: 73 (43%) on model development, 43 (25%) on external validation, 33 (19%) on incremental value, and 21 (12%) on combined development and external validation of the same model. Overall, publications adhered to a median of 44% (25th-75th percentile 35-52%) of TRIPOD items, with 44% (35-53%) for prognostic and 41% (34-48%) for diagnostic models. TRIPOD items that were completely reported for less than 25% of the models concerned abstract (2%), title (5%), blinding of predictor assessment (6%), comparison of development and validation data (11%), model updating (14%), model performance (14%), model specification (17%), characteristics of participants (21%), model performance measures (methods) (21%), and model-building procedures (24%). Most often reported were TRIPOD items regarding overall interpretation (96%), source of data (95%), and risk groups (90%).

Conclusions: More than half of the items considered essential for transparent reporting were not fully addressed in publications of multivariable prediction model studies. Essential information for using a model in individual risk prediction, i.e. model specifications and model performance, was incomplete for more than 80% of the models. Items that require improved reporting are title, abstract, and model-building procedures, as they are crucial for identification and external validation of prediction models.

Keywords: Development; Diagnosis; Incremental value; Prediction model; Prediction rule; Prognosis; Reporting guideline; Risk assessment; Risk score; TRIPOD; Validation.

Fig. 1

Flow diagram of selection procedure

Fig. 2

Reporting across publications: adherence to items of the TRIPOD statement

Fig. 3

Reporting of the items of the TRIPOD statement overall (a) and per type of prediction model study (b) (see Box 1 for list of items of the TRIPOD statement). NA not applicable (not all items of the TRIPOD statement are relevant to all types of prediction model studies). Percentages are based on number of models for which an item was applicable (and thus should have been reported). *Where this number deviates from the total number of models, this is indicated. This concerns the following items (N = number of models for which the item was applicable). Overall: 5c (N = 169), 10a (N = 127), 10b (N = 127), 10c (N = 84), 10e (N = 23), 11 (N = 70), 12 (N = 81), 13c (N = 97), 14a (N = 127), 14b (N = 94), 15a (N = 127), 15b (N = 127), 17 (N = 7), 19a (N = 92); Development: 5c (N = 72), 11 (N = 22), 14b (N = 55); External validation: 10e (N = 8), 11 (N = 15), 17 (N = 4); Incremental value: 10c (N = 20), 10e (N = 11), 11 (N = 20), 12 (N = 17), 14b (N = 25), 19a (N = 29); Development and external validation: 10e (N = 4), 11 (N = 13), 14b (N = 14), 17 (N = 3), 19a (N = 20). †Item 21 ’Provide information about the availability of supplementary resources, such as study protocol, Web calculator, and data sets’: the number of models for which this item was applicable is unknown. It probably was applicable to all models that reported this item. Instead of presenting a percentage of 100, we based the percentage on the total number of models.