Prognostic models for cardiovascular disease risk in the general population: a systematic review

Tags: Oral
Damen J1, Hooft L1, Schuit E1, Debray T1, Collins GS2, Tzoulaki I3, Lassale CM4, Siontis G5, Black J6, Heus P1, van der Schouw YT7, Peelen L7, Moons K1
1Dutch Cochrane Centre, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, The Netherlands, 2Centre for Statistics in Medicine, NDORMS, Botnar Research Centre, University of Oxford, United Kingdom, 3Department of Epidemiology and Biostatistics, Imperial College of Medicine, London, United Kingdom, 4Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London, United Kingdom, 5Department of Cardiology, Bern University Hospital, Switzerland, 6MRC Epidemiology Unit, University of Cambridge School of Clinical Medicine, United Kingdom, 7Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, The Netherlands

Background: Cardiovascular disease (CVD) is a leading cause of morbidity and the leading cause of mortality worldwide. Many prediction models have been developed to assess individual CVD risk to allow targeting of preventive treatment.

Objectives: To give an overview of all prognostic models that predict future risk of CVD in the general population, and to describe predicted outcomes, study populations and included predictors.

Methods: In June 2013 a systematic search was performed in MEDLINE and Embase to identify studies that described the development, validation or incremental value of a multivariable prognostic model predicting CVD in the general population.

Results: Out of 9671 papers identified, 314 were included, describing the development of 373 prognostic models, 519 external validations and 278 incremental value assessments. In total 132 models (35%) were externally validated, of which 70 models (19%) were validated by independent researchers. Most models were developed in Europe (n = 258), and predicted risk of coronary heart disease (n = 115, Table 1) over a 10-year period (n = 211). Furthermore, most prevalent predictors were age and smoking (n = 323 and n = 332 respectively, Figure 1), frequently with separate models for males or females (n = 256). Discrimination and calibration were reported for only 62% and 52% of the external validations respectively. Crucial clinical and methodological information was often missing, and when reported, substantial heterogeneity in predictor and outcome definitions was seen between models. For 53 models the prediction time horizon was not reported and for 56 models the intercepts or baseline hazards were not reported, making it impossible to use for individual risk prediction.

Conclusion: In a similar manner to various other clinical domains, there is an excess of prognostic models predicting CVD in the general population. Additionally, the usefulness of most models remains unclear, due to incomplete presentation, lack of external validation, and heterogeneity in predicted outcomes and study populations. Future research should focus on validating, updating, meta-analyzing and studying impact of models.