26 May Too Many Cardiovascular Disease Prediction Models Lack Clear Validation
MedicalResearch.com Interview with:
Johanna Damen, MSc
Julius Center for Health Sciences and
Primary Care Cochrane Netherlands
University Medical Center Utrecht,
MedicalResearch.com:What is the background for this study?
Response: Prediction models for cardiovascular disease (CVD) estimate the probability that an individual will develop a certain cardiovascular condition in the future. For instance, prognostic models for CVD are typically used to decide which patients need treatment (e.g. antihypertensive or lipid lowering drugs, or life-style interventions) to reduce their 10 year risk. Previous reviews have shown there are a lot of CVD prediction models, but no systematic review has given an overview of which models exist, which predictors they use, which outcome they predict and which models have been externally validated.
MedicalResearch.com: What are the main findings?
Response: We found that there is an excess of prediction models for CVD (363 models in total) with extreme variation in predicted outcomes and included predictors. For instance, the majority of cardiovascular models predicted the risk of coronary heart disease of cardiovascular disease, although more than 70 different definitions were reported for these outcomes. Most models included predictors like age, smoking status, blood pressure, and cholesterol levels, but over 100 other predictors were identified. Remarkably, the performance of these models was almost never validated in new patients or settings, indicating that much more emphasis is placed on repeating the process of identifying new predictors rather than validating existing models. Quality of reporting was often insufficient to actually use the model for individual risk predictions, and model performance was infrequently reported.
MedicalResearch.com: What should readers take away from your report?
Response: Despite a clear excess of CVD prediction models, there is a lack of studies in which their predictive performance and usefulness in clinical practice is assessed. Furthermore, there is a lack of studies in which existing models are compared and the performance in local settings or populations is assessed. Although there are many differences between the available CVD models, most of them predict similar outcomes and use similar predictors.
MedicalResearch.com: What recommendations do you have for future research as a result of this study?
Response: Future research should focus on validating existing prediction models, rather than developing new models. Ideally, models are compared head-to-head to identify the best model in a certain setting.
Furthermore, these models can be combined and tailored to specific settings, and other predictors can be added to improve their predictive performance.
Finally, impact studies are urgently needed to investigate the effect of using CVD prediction models on doctor’s prescription behavior and patient outcomes.
MedicalResearch.com: Thank you for your contribution to the MedicalResearch.com community.
Citation: Damen Johanna A A G, Hooft Lotty, Schuit Ewoud,Debray Thomas P A, Collins Gary S, Tzoulaki Ioanna et al. Prediction models for cardiovascular disease risk in the general population: systematic review BMJ2016; 353 :i2416
Note: Content is Not intended as medical advice.
Please consult your health care provider regarding your specific medical condition and questions.
More Medical Research Interviews on MedicalResearch.com.