In order to bring you the best possible user experience, this site uses Javascript. If you are seeing this message, it is likely that the Javascript option in your browser is disabled. For optimal viewing of this site, please ensure that Javascript is enabled for your browser.

Abstract of the day - A novel digital twin-based approach to translate RCT findings to different populations

31 Aug 2024
Abstract of the Day

The concept of digital twins – sophisticated virtual representations built using machine learning and multiscale modelling – is gaining increasing attention with the shift towards precision medicine. Today, a digital-twin approach developed to evaluate the generalisability of randomised controlled trials (RCTs) will be presented.

RCTs remain the gold standard for the generation of data to optimise clinical practice, but questions remain regarding how representative their findings are when applied to real-world populations. Moreover, there are situations where RCTs testing similar interventions differ in their findings, which raises other potential issues about generalisability. An example is the discrepancy between the improved cardiovascular outcomes observed with intensive blood pressure control in the SPRINT trial (hazard ratio [HR] 0.75; 95% confidence interval [CI] 0.64 to 0.89)1 and the lack of improvement seen in the ACCORD trial (HR 0.88; 95% CI 0.73 to 1.06).2

Digital twin synthesis using models such as generative adversarial networks (GANs) can integrate multiple patient population characteristics by constructing a synthetic cohort that accurately represents their covariate distributions. While GANs have been utilised to estimate individual treatment effects, their potential for evidence translation across patient populations has not been explored.

Today, Doctor Phyllis Thangaraj (Cardiovascular Data Science [CarDS] Lab, Yale University - New Haven, USA) presents the creation of digital RCT-Twin GAN models, conditioned on the characteristics of another patient population, to assess replication of treatment effects using data from the SPRINT and ACCORD trials. An RCT-Twin of SPRINT was generated conditioned on the most disparate covariates drawn from ACCORD, and an RCT-Twin of ACCORD was generated conditioned on the SPRINT cohort.

The key finding was that the conditioned RCT-Twins reproduced the treatment effects of the other RCT on 5-year major adverse cardiovascular events. A SPRINT-conditioned ACCORD twin reproduced the SPRINT treatment effect (median HR 0.79; 95% CI 0.72 to 0.86), while the ACCORD-conditioned SPRINT twin reproduced the ACCORD treatment effect (median HR 0.87; 95% CI 0.68 to 1.13).

In this way, the application of digital twins provides a novel approach for evidence translation across populations. The results indicate that it is possible to create an RCT-Twin GAN model to generalise an RCT across a second patient population, in this case the population of another RCT. It may also be possible to evaluate treatment effects by conditioning RCT data on covariate distributions from real-world populations, thereby enhancing inference for real-world patients.

References

  1. SPRINT Research Group. N Engl J Med. 2015;373:2103–2116.
  2. ACCORD Study Group. N Engl J Med. 2010;362:1575–1585.
Data Privacy: During the congress your online activity is tracked and collected by the ESC. We use this to award your CME credits; for reporting and statistical purposes; and to provide you with the best experience.
Please note your details will be shared with the sponsor of any Industry Session, Tutorial or Exhibition you choose to view. More information is available in the ESC Congress 2024 London Registration terms and conditions