CHARIOT: Development and Internal Validation of a Cardiovascular Health Assessment and Risk-based Intervention Optimisation Tool
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Objective
Clinical prediction models are used across the world to guide treatment for the primary prevention of cardiovascular disease. Such models are appropriate for estimating an individual’s risk of developing cardiovascular disease; however, they are sometimes used inappropriately to estimate risk under some intervention that results in changes to their risk factors, such as weight loss, or stopping smoking. The objective of this study is to develop a model that correctly predicts 10-year risk of cardiovascular disease under a wide range of interventions.
Design
Retrospective cohort study, prediction under intervention using causal inference.
Setting
English national primary care linked with secondary care, office for national statistics and index of multiple deprivation.
Participants
Adults (aged 18 – 86) free from cardiovascular disease between 2005 – 2020.
Main outcome measure
Incidence of cardiovascular disease
Results
The resulting model is designed to be used across multiple follow-up visits. We illustrate how a 70 year old woman with a 23.15% 10-year risk of cardiovascular disease could reduce this to 17.71% (statins), 18.82% (antihypertensives leading to 10mmHg reduction in systolic blood pressure), 19.42% (smoking cessation), 21.15% (weight loss of 5kg), 19.27 (weight loss of 10kg), or 14.78% through a combination of therapies. Under internal validation, the model has good calibration in the entire cohort and within subgroups defined by protected characteristics (sex, age and ethnicity) and 10 major English regions. The model had discrimination (c-statistic) of 0.874 (female) and 0.859 (male). Within subgroups defined by ethnic group, discrimination did not drop below 0.858 (Black Caribbean, Female) or 0.850 (White, male). Within subgroups defined by region, discrimination did not drop below 0.859 (West Midlands, Female) or 0.842 (North East, Male). There was a large drop in discrimination within subgroups defined by age given its strong predictive properties.
Conclusions
Alongside the novel prediction under intervention aspects, the model preserves good predictive performance with respect to traditional metrics - it is well calibrated and has discrimination meeting or exceeding the discrimination of models currently used to guide clinical care around cardiovascular disease. Further evaluation and piloting is required to support its clinical use.
