Anthony Matthews

The association between weight, and cardiovascular disease and mortality is well established, however, the causal effect of weight-loss in midlife on these outcomes is less clear. Bariatric surgery results in substantial weight-loss and is an ideal candidate to study the causal effects of weight-loss. We propose a project that willcausal inference and machine learning methods to answer two important questions: 2) Is bariatric surgery effective for reducing cardiovascular disease and mortality, and if so, for who? 3) Which type of bariatric surgery (gastric bypass or sleeve gastrectomy) is most effective, and for who?We will use data from various Swedish registers to identify individuals with obesity who are eligible for bariatric surgery. We will then compare cardiovascular and mortality outcomes among those undergoing different types of bariatric surgery with those receiving non-surgical obesity management using causal inference methods. We will use causal forests and expert knowledge to estimate indiviual treatment effects, and identify the groups of patients who benefit the most from these surgeries.This 3-year project will be undertaken by the CAUSALab, at the Unit of Epidemiology, Karolinska Institutet. The team of co-applicants, have extensive experience in using observational data to obtain causal inferences, particularly in the field of cardiovascular disease. A postdoctoral researcher will be hired to work full-time on this project.

Randomized trials do not estimate treatment effects in individuals that were eligible but did not enroll in the trial. However, guidelines established as a result of the trials apply to these populations in clinical practice. On average, around 40% of eligible individuals do not enroll in trials. A treatment that shows no benefit in trial participants may be effective in a subpopulation not likely to enroll, such as older individuals. Differences in effectiveness must be quantified to guide policy decisions. Extending randomized trial results to the treatment eligible population is often handled using real-world data alone, which are prone to bias.This project will propose a novel framework that uses transportability and benchmarking methods to simultaneously utilize data from the TASTE and VALIDATE randomized trials and real-world data from the SWEDEHEART registry to robustly fill the knowledge gaps left by the trials. Specifically, we will estimate the effect of thrombus aspiration and two anticoagulant medications after myocardial infarction in the whole trial-eligible population of Sweden

test the reliability of studies that use real-world data alone to estimate treatment effects

and create tutorials, teaching materials, and open-source code to facilitate the uptake of these methodologies. Our research will improve public health decisions in Sweden and serve as a model in the extension of findings from randomized trials to the population eligible for treatment.