Clinical Prediction Models for Cardiovascular Disease: Tufts Predictive Analytics and Comparative Effectiveness
Benjamin S Wessler1, Lana Lai Yh1, Whitney Kramer1
1From the Division of Cardiology, Tufts Medical Center, Boston, MA (B.S.W.); Predictive Analytics and Comparative Effectiveness (PACE) Center, Institute for Clinical Research and Health Policy Studies (ICRHPS), Tufts Medical Center/Tufts University School of Medicine, Boston, MA (B.S.W., L.L.Y., J.S.L., D.M.K.); Business Intelligence and Analytics, Iora Health, Cambridge, MA (W.K.); Health Economics and Reimbursement, Boston Scientific, Marlborough, MA (M.C.); and Center for Clinical Evidence Synthesis, ICRHPS, Medical Center/Tufts University School of Medicine, Boston, MA (G.R.).
Insights
Clinical prediction models (CPMs) for cardiovascular disease are abundant, with over 796 models identified. However, their comparative performance and clinical impact remain poorly understood, indicating significant literature redundancy.
Area of Science:
- Cardiovascular Medicine
- Medical Informatics
Background:
- Clinical prediction models (CPMs) aid decision-making by estimating clinical outcome probabilities.
- Numerous CPMs exist for cardiovascular disease (CVD), but the scope of this literature is not well-defined.
Purpose of the Study:
- To systematically review and characterize the landscape of CPMs for cardiovascular disease.
- To quantify the extent and types of CPMs developed for various CVD conditions.
Main Methods:
- Conducted a systematic review of CPMs for CVD published between January 1990 and May 2012.
- Created a novel database to characterize 796 identified CPMs by development stage, population, performance, covariates, and outcomes.
- Included conditions such as coronary heart disease, heart failure, arrhythmias, stroke, venous thromboembolism, and peripheral vascular disease.
Main Results:
- Identified 796 CPMs for 31 index conditions, with a steady increase in publications over time.
- The majority (90%) were de novo models; 63% reported c-statistics and 36% provided calibration data.
- Specific models included 215 for coronary artery disease and 79 for heart failure, with 77 distinct index/outcome pairings.
Conclusions:
- A substantial and redundant body of CPMs exists for cardiovascular diseases.
- The comparative performance, consistency of risk estimates, and clinical utility of these numerous models are not well understood.
Background:
Clinical prediction models (CPMs) estimate the probability of clinical outcomes and hold the potential to improve decision making and individualize care. For patients with cardiovascular disease, there are numerous CPMs available although the extent of this literature is not well described.
Methods And Results:
We conducted a systematic review for articles containing CPMs for cardiovascular disease published between January 1990 and May 2012. Cardiovascular disease includes coronary heart disease, heart failure, arrhythmias, stroke, venous thromboembolism, and peripheral vascular disease. We created a novel database and characterized CPMs based on the stage of development, population under study, performance, covariates, and predicted outcomes. There are 796 models included in this database. The number of CPMs published each year is increasing steadily over time. Seven hundred seventeen (90%) are de novo CPMs, 21 (3%) are CPM recalibrations, and 58 (7%) are CPM adaptations. This database contains CPMs for 31 index conditions, including 215 CPMs for patients with coronary artery disease, 168 CPMs for population samples, and 79 models for patients with heart failure. There are 77 distinct index/outcome pairings. Of the de novo models in this database, 450 (63%) report a c-statistic and 259 (36%) report some information on calibration.
Conclusions:
There is an abundance of CPMs available for a wide assortment of cardiovascular disease conditions, with substantial redundancy in the literature. The comparative performance of these models, the consistency of effects and risk estimates across models and the actual and potential clinical impact of this body of literature is poorly understood.
