Article type
Year
Abstract
Background:The potential benefits of management for acute aortic dissection (AAD) depend on the accuracy of prognosis assessment. A variety of prediction models for in-hospital mortality in AAD have been reported in the past decades. These models distinctly composed of a single or combination of biomarkers, demographic information, and clinical presentations, and showed diverse performances.
Objectives: Previous studies identified several predictors and prognostic models for in-hospital mortality in AAD. Our objective was to identify studies evaluating these predictors and prediction models and to illustrate their performance in predicting in-hospital mortality in AAD.
Methods:We searched for studies in PubMed and EMBASE until July, 2019. Two reviewers independently screened records for inclusion, assessed risk of bias, and extracted data. We collected the following information from each eligible study: general study characteristics, predictors used, study population, performance of the model, and likelihood of use in practice.
Results:A total of 9526 reports were identified, of which 17 were included. Performance measures were poorly reported as only three studies reported both discrimination value and calibration value. For prediction model, the prediction model using International Registry of Acute Aortic Dissection (IRAD) from multinational data reported good calibration, while EuroSCORE II prediction models did not show good discriminative ability and good calibration. For biomarkers used in the prediction model, discriminatory power varied from 0.58 to 0.95; D-dimer, NLR, and CRP predictors were the most popular biomarkers for predicting in-hospital mortality in AAD. The risk of bias in the domains of participants, predictors, and outcome were rated as low for most studies, but the risk of bias in the domain of sample size and missing data and statistical analysis were rated as high or unclear for most studies.
Conclusions:In-hospital mortality risk prediction in AAD has been modelled, but many of these models are methodologically weak and biomarkers used in the prediction model have highly variable performance across different populations. A new model derived from IRAD from multinational data adding the relevant biomarkers may be required for improved prognostic performance.
Objectives: Previous studies identified several predictors and prognostic models for in-hospital mortality in AAD. Our objective was to identify studies evaluating these predictors and prediction models and to illustrate their performance in predicting in-hospital mortality in AAD.
Methods:We searched for studies in PubMed and EMBASE until July, 2019. Two reviewers independently screened records for inclusion, assessed risk of bias, and extracted data. We collected the following information from each eligible study: general study characteristics, predictors used, study population, performance of the model, and likelihood of use in practice.
Results:A total of 9526 reports were identified, of which 17 were included. Performance measures were poorly reported as only three studies reported both discrimination value and calibration value. For prediction model, the prediction model using International Registry of Acute Aortic Dissection (IRAD) from multinational data reported good calibration, while EuroSCORE II prediction models did not show good discriminative ability and good calibration. For biomarkers used in the prediction model, discriminatory power varied from 0.58 to 0.95; D-dimer, NLR, and CRP predictors were the most popular biomarkers for predicting in-hospital mortality in AAD. The risk of bias in the domains of participants, predictors, and outcome were rated as low for most studies, but the risk of bias in the domain of sample size and missing data and statistical analysis were rated as high or unclear for most studies.
Conclusions:In-hospital mortality risk prediction in AAD has been modelled, but many of these models are methodologically weak and biomarkers used in the prediction model have highly variable performance across different populations. A new model derived from IRAD from multinational data adding the relevant biomarkers may be required for improved prognostic performance.