PROBAST-2019: a 22 item tool to assess the risk of bias of studies on prediction models, and to assess their applicability for the targeted context and population.

Diagnostic & prognostic prediction models

A prediction model is defined as any form of mathematical equation that combines two or more predictors to estimate the probability that a certain outcome is currently present diagnostic prediction model – or will occur in some time period – prognostic prediction model.

Types of prediction model studies

PROBAST-2019 helps assessing studies on multivariable prediction models that are used to make predictions in individuals, that is individualised predictions, including studies on:

• development of new prediction models;
• development and validation of same prediction model(s);
• validation existing prediction models;
• development of new prediction model compared with validation of existing prediction models;
• updating or extension of existing prediction models;
• combination of any of the above.

Types of predictors, outcomes and modelling technique

PROBAST-2019 can be used to assess any type of diagnostic or prognostic prediction model used for individualized predictions, regardless of the type of:

• predictors; for example a demographic, a clinical, a biomarker, an imaging or omics;
• outcomes being predicted; for example binary, time-to-event, linear;
• statistical method used; for example logistic, survival, machine or deep learning techniques.