Article type
Abstract
Background: Prediction models are often not validated nor developed with an a priori intended use. Despite this, prediction models are used in guidelines to inform prognostic or diagnostic recommendations. There might be discrepancies between how, and for whom, prediction models are developed and are used in clinical practice.
Objectives: To describe which prediction models are developed and (externally) validated, whether these prediction models are used in clinical guidelines, and how the prediction models are used in guideline recommendations.
Methods: Systematic reviews on prediction models and clinical guidelines in pulmonary embolism were searched. The corresponding guidelines for pulmonary embolism were selected from the following countries: USA, EU, Canada, UK, and the Netherlands. Data from the prediction models regarding development and validation was extracted, such as main prediction outcome, predictors, development setting, and patient characteristics, for both the development and internal/external validation phase. Description of prediction model use, and any use in guideline recommendations, was extracted from the guidelines. The development and validation between models, and development/intended use of models and their description in clinical guidelines were compared.
Results: The search and extraction for prediction models for pulmonary embolism included 16 models. Of these models, 4 were externally validated. Guidelines on pulmonary embolism used some combination of the Wells Score (4/5), Geneva Score (2/5), Pulmonary Embolism Rule-out Criteria (PERC) (3/5), with the Netherlands using the YEARS score. The Geneva score, Wells score, PERC, and YEARS rules were developed in hospital contexts, and only the Wells score has been further validated in other populations. The clinical guidelines typically do not specify in which setting they are intended for, and do not include extensive descriptions of the use and development of prediction models.
Conclusions: There are a lack of externally validated prediction models for pulmonary embolism, and a lack of (information) coordination between model development and implementation into guidelines regarding context. While the included guidelines do not have a set setting, commonly the lack external validation to other settings and of elaboration in the guidelines of context could have implications on the applicability of these scores to settings outside of hospitals.
Objectives: To describe which prediction models are developed and (externally) validated, whether these prediction models are used in clinical guidelines, and how the prediction models are used in guideline recommendations.
Methods: Systematic reviews on prediction models and clinical guidelines in pulmonary embolism were searched. The corresponding guidelines for pulmonary embolism were selected from the following countries: USA, EU, Canada, UK, and the Netherlands. Data from the prediction models regarding development and validation was extracted, such as main prediction outcome, predictors, development setting, and patient characteristics, for both the development and internal/external validation phase. Description of prediction model use, and any use in guideline recommendations, was extracted from the guidelines. The development and validation between models, and development/intended use of models and their description in clinical guidelines were compared.
Results: The search and extraction for prediction models for pulmonary embolism included 16 models. Of these models, 4 were externally validated. Guidelines on pulmonary embolism used some combination of the Wells Score (4/5), Geneva Score (2/5), Pulmonary Embolism Rule-out Criteria (PERC) (3/5), with the Netherlands using the YEARS score. The Geneva score, Wells score, PERC, and YEARS rules were developed in hospital contexts, and only the Wells score has been further validated in other populations. The clinical guidelines typically do not specify in which setting they are intended for, and do not include extensive descriptions of the use and development of prediction models.
Conclusions: There are a lack of externally validated prediction models for pulmonary embolism, and a lack of (information) coordination between model development and implementation into guidelines regarding context. While the included guidelines do not have a set setting, commonly the lack external validation to other settings and of elaboration in the guidelines of context could have implications on the applicability of these scores to settings outside of hospitals.