A Comprehensive Review of Predictive Risk Models for Cardiovascular Disease

Over recent years, multiple risk prediction models for cardiovascular disease (CVD) have been developed. These models use various patient characteristics to estimate the probability of outcomes over a certain period of time. A predictive model is defined as a model that provides a way to estimate a patient's individual risk for a cardiovascular (CV) outcome.1 With the development of so many predictive models, the question of when, which, and how to use these models arises.

A systematic review by Wessler et al. found that there is significant redundancy of many predictive models.1 In the past, there has been little effort to compare individual models to offer appropriate guidance on which models work best in specific situations.1 Previous reviews, however, were conducted decades ago and excluded models that were not internally or externally validated and excluded articles that only described external validation.1-3 Thus, it is important to compare these models in order to assess how they can best be applied to individualize patient care.

Damen et al. conducted a systematic review of various prediction models developed to assess the risk of CVD in the general population and published their findings in the BMJ.4 They also looked at the characteristics of the models' developments and whether they have undergone external validation.4 Eligible articles (a total of 212) were categorized into two groups: development articles (125 articles) and external validation articles (136 articles); often articles portrayed combinations of development and external validation.4

Large variations in predicted CV outcomes were observed, with the majority of models focusing on nonfatal or fatal CVD.4 The main categories of predictors were demographics (such as age, sex, race) and family history, lifestyle, comorbidities, blood pressure, physical examination, lipid levels, other blood variables, and genetics.4 It is important to note that none of the models included the use of lipid lowering agents, such as statins.

The prediction horizon ranged from 2 to 45 years, with 58% of studies predicting CVD outcomes for a 10-year period and 13% predicting CVD outcomes for a 5-year period.4 Despite some exceptions, including the Framingham and SCORE models, 64% of the models were not externally validated and only 19% of the models were validated by independent investigators.4

So what is there to be done to avoid confusion?

With the abundance of risk models, especially over the last two decades, that combine various predictors to estimate the risk of CVD, many clinicians are increasingly confused about which CVD prediction model to use. Instead of developing new models that further increase the confusion, studies should be aimed at comparing individual risk models against one another and should be intended to modify these models to be more applicable to certain settings and populations.

Furthermore, as the majority of prediction models were developed and validated in North American and European populations, a strong need remains for prediction models to incorporate various ethnicities (e.g., Asians) due to significant variations in CVD outcomes across different ethnicities.4,5 Damen et al. further explain the discrepancy in definitions of nonfatal or fatal CVD, thus making it more difficult for head-to-head comparisons.4

Although the authors mention that the prediction models do not incorporate various ethnicities, their search did specifically exclude non-English papers. These differences in definitions can confound the estimated predictor effects, alter the estimated probabilities for CVD outcomes, and subsequently provide inappropriate treatment strategies based on these results. Thus, clear outcome definitions can drastically improve the clinical utilization of these models in practice.

Comparing this review with previous studies shows similar results, as incomplete reporting of clinical and other information needed for validation was often present.4,6-8 The implication of risk models to guide informed health decision-making can be substantial. With the well-known CVD risk prediction models (Framingham, SCORE, and QRISK) lacking adequate head-to-head comparisons combined with the fact that the majority of models lack external validation by independent investigators, advocating for a specific model to use in specific populations or situations is very difficult to do at this time.4 Furthermore, these models likely have the same flaws as the atherosclerotic cardiovascular disease risk score (which was not studied in this paper), including the use of older data as the study reports a median publication date of 2002. Recent studies from some of the large cohorts (e.g., Framingham) show that the prevalence of traditional risk factors such as smoking, obesity, and diabetes has changed considerably over the past few decades.

The authors go on to state the importance of refining, updating, and even combining current CVD risk prediction models instead of repeating similar processes from the past and developing more models.4 Thus, it would seem imperative to devise clear definitions of appropriate CVD outcomes, to make better use of current evidence, and to compare individual risk models to assess how they can be applied to specific populations (including various ethnicities).

Combined with a clinician-patient risk discussion, the ultimate goal, perhaps using a range of various prediction modeling methods, is to develop a clinically pertinent CVD risk prediction model or models to guide clinicians on appropriate treatment decisions with regard to aspirin, statins, and possibly antihypertensive therapy for their patients.

References

  1. Wessler, BS, Lai Yh L, Kramer W, et al. Clinical prediction models for cardiovascular disease: Tufts Predictive Analytics and Comparative Effectiveness Clinical Prediction Model Database. Circ Cardiovasc Qual Outcomes 2015;8:368-75.
  2. Beswick AD, Brindle P, Fahey T, Ebrahim S. A Systematic Review of Risk Scoring Methods and Clinical Decision Aids Used in the Primary Prevention of Coronary Heart Disease (Supplement). London, England: Royal College of General Practitioners, 2008.
  3. Matheny M, McPheeters M, Glasser A, et al. Systematic Review of Cardiovascular Disease Risk Assessment Tools. Rockville, Maryland: Agency for Healthcare Research and Quality, 2011.
  4. Damen JA, Hooft L, Schuit E, et al. Prediction models for cardiovascular disease risk in the general population: systematic review. BMJ 2016;353:i2416.
  5. Gijsbetrts CM, Groenewegen KA, Hoefer IE, et al. Race/ethnic differences in the associations of the Framingham risk factors with carotid IMT and cardiovascular events. PLoS One 2015;10:e0132321.
  6. Bouwmeester W, Zuithoff NP, Mallett S, et al. Reporting and methods in clinical prediction research: a systematic review. PLoS Med 2012;9:1-12.
  7. Collins GS, Mallett S, Omar O, Yu LM. Developing risk prediction models for type 2 diabetes: a systematic review of methodology and reporting. BMC Med 2011;9:103.
  8. Collins GS, Omar O, Shanyinde M, Yu LM. A systematic review finds prediction models for chronic kidney disease were poorly reported and often developed using inappropriate methods. J Clin Epidemiol 2013;66:268-77.

Clinical Topics: Diabetes and Cardiometabolic Disease, Clinical Topic Collection: Dyslipidemia, Prevention, Lipid Metabolism, Smoking

Keywords: Antihypertensive Agents, Aspirin, Blood Pressure, Cardiovascular Diseases, Comorbidity, Diabetes Mellitus, Life Style, Lipids, Obesity, Risk Factors, Smoking, Dyslipidemias


< Back to Listings