Back in the 1990s, the idea of running computer-simulated clinical trials would have seemed far-fetched. The internet had only just emerged, personal computers were the size of suitcases, and...
Control arms are key to reliably determining whether a new treatment works better than what already exists. The data they generate – from trial patients who receive a placebo, the standard of care (SOC), a comparator therapy or no treatment at all – are key to the ‘fair test’ that kids learn about in primary school science.
Yet industry – with regulators’ blessing – has begun to tamper with control arms in some circumstances. Historical data from similar, past trials, or from real world data sources like medical records and claims, may now replace traditional control data. So-called ‘external control arms’ are justified in trials of rare diseases where there aren’t many patients or for medicines treating serious, end-stage unmet needs, where it’s ethically questionable to deprive any patient of a novel therapy, even if unproven. Pulling together control arm data from external sources can also help address the challenge of rapidly changing SOC regimens in some areas of cancer.
Historical control data rarely matches up with data in a current study – patient selection criteria, endpoints and other variables may differ considerably. Real world data is also tricky to align with trial data, not least because much of what patients do in a non-experimental setting goes unmeasured. (RWD is, after all, defined by its differentiation from controlled trial data.)
However, statistical tools for reconciling historical data (or any parallel data that is not collected in the same trial) are evolving fast. Historical controls can be ‘matched’ or adjusted to mimic active arm patient data as closely as possible, and techniques to reduce (or at least account for) uncertainties are becoming more sophisticated .
But they’re not a complete solution. The quality, consistency and completeness of health data remain less-than-perfect. Bias is still a problem. Sponsors may select data, or data borrowing methods, that make their drug appear most effective.
Fully synthetic control arms offer all the benefits (and more) of external control arms without the data-matching limitations. They are made up of virtual patients which share baseline characteristics with their real-life counterparts. These virtual patients – built and calibrated using actual clinical and non-clinical data – are effectively ‘in-person’ controls, or digital twins, that evolve contemporaneously. They are unconstrained by the limits of real data sets or recruitment challenges, yet they capture both the accuracy of clinical data and the variability of real world data.
These digital twins provide a rigorous baseline – perhaps even more rigorous than that of traditional control groups, which necessarily comprise different individuals to those in the treatment arm. Clusters of similar control patients can be created to evolve in parallel, strengthening the predictive power of the virtual arm further still. And since the mechanistic disease model and virtual patients are created, calibrated and validated up front, the approach is less vulnerable to bias than RWD-based controls.
Synthetic control arms can do more than reduce the time, costs, and practical and ethical hurdles of clinical trials. They can also help quantify a drug’s benefit, post-approval, by predicting the outcomes of patients who do not receive a given treatment. How worse off would they likely be, over given time frames, and at what cost to the system? (There is precedent for external controls here: Roche accelerated coverage of ALK-positive lung cancer drug Alecensa (alectinib) using a control group built from patient records at Flatiron, the health data company it acquired in 2018.)
Virtual populations can also be used to capture the impact of changing treatment standards. They can model today’s SOC, and anticipate the impact of tomorrow’s, based on the mechanism of action of drug candidates in late-stage development or registration. Ultimately, this approach could help determine which treatment regimen – SOC or otherwise – might work best for any given individual, taking us a step closer towards truly personalised medicine.
NOVA’s clinical trial simulation platform, JINKŌ®, allows drug developers to create virtual patients and populations to help elucidate the effects of a given drug regimen or combination, and to compare this to an alternative, or no, treatment.