Introduction: Many risk prediction models are currently in use for predicting short-term mortality following coronary artery bypass graft (CABG) surgery. This review critically appraised the methods that were used for developing these models to assess their applicability in current practice setting as well as for the necessity of up-gradation.
Evidence acquisition: Medline via Ovid was searched for articles published between 1946 and 2016 and EMBASE via Ovid between 1974 and 2016 to identify risk prediction models for CABG. Article selection and data extraction was conducted using the CHARMS checklist for review of prediction model studies. Association between model development methods and model's discrimination was assessed using Kruskal-Wallis one-way analysis of variance and Mann-Whitney U-test.
Evidence synthesis: A total of 53 risk prediction models for short-term mortality following CABG were identified. The review found a wide variation in development methodology of risk prediction models in the field. Ambiguous predictor and outcome definition, sub-optimum sample size, inappropriate handling of missing data and inefficient predictor selection technique are major issues identified in the review. Quantitative synthesis in the review showed "missing value imputation" and "adopting machine learning algorithms" may result in better discrimination power of the models.
Conclusions: There are aspects in current risk modeling, where there is room for improvement to reflect current clinical practice. Future risk modelling needs to adopt a standardized approach to defining both outcome and predictor variables, rational treatment of missing data and robust statistical techniques to enhance performance of the mortality risk prediction.