Using AI To Create Digital Twin And Reduce Clinical Trial Cycle Times

A digital twin is a longitudinal clinical record built from a patient's baseline data—before they receive their first therapy—that forecasts how that patient would likely evolve if given a placebo.

FREMONT, CA: Most professionals in drug research are familiar with the nerve-wracking wait for the results of a large trial. Is the experimental therapy useless if the impact is negative? Or is the failure due to an unanticipated defect in the protocol's design or implementation rather than a lack of efficacy? The researchers could spend weeks examining data, but a definite answer may be elusive due to limited statistical power for such analyses in the already completed experiment. These issues are exacerbated if the trial has a lower enrollment or higher dropout rate than anticipated due to an unforeseen occurrence such as Covid-19. And if an attempt is unsuccessful, the next one will likely be larger and more expensive, if it occurs at all.

Sponsors typically employ external control arms in proof-of-concept trials to save schedules and costs. Unfortunately, this opens the door to confounders or discrepancies between patients in the external control arm and those in the treatment arm that prevent attributing differences in outcomes to the treatment's effect. Thus, trials employing outer control arms are effective yet unreliable.

With the availability of massive historical datasets of longitudinal patient data and the rapid development of artificial intelligence (AI) technology, clinical trials are no longer required to adhere to the status quo. It is possible to combine the dependability of a randomized controlled trial (RCT) with the effectiveness of an external control arm. The technological advancement that makes this feasible is known as a digital twin.

A digital twin is a detailed, longitudinal clinical record built from baseline data taken from a patient—before they get their first therapy—that forecasts how that patient would likely evolve throughout the study if a placebo were administered. In other words, a digital twin is analogous to a simulated control group for a certain patient.

Digital twins are viewed as factors that are optimized to explain outcome variability. By employing pre-specified covariate adjustment—essentially comparing anticipated placebo outcomes to actual placebo results and removing any bias—digital twins can be introduced into randomized controlled trials without introducing bias to increase power and efficiency.

Digital twins do not constitute Synthetic Controls. Synthetic Control Arms (SCAs) include patients not included in the trial's initial patient sample. Because SCAs can raise Type I error, regulatory guidelines limit their applications5. Digital twins provide additional information on study participants. Because projected outcomes are considered covariates, they do not create bias and can be applied to all phases of clinical development.