New data-driven approaches change how researchers evaluate experimental therapies in clinical trials. Among these, synthetic control arms use existing datasets to provide a flexible framework for comparing treatment effects while reducing dependence on traditional placebo groups. As a result, this method has the potential to reshape how investigators assess efficacy, especially in complex diseases such as idiopathic pulmonary fibrosis (IPF).
In this Innovation Spotlight, Kiril Kirov, a senior data scientist at Qureight, describes how his team leverages external control arm generation and imaging biomarkers to support more efficient clinical trial design.
What challenges do researchers face when using traditional placebo-controlled clinical study designs?
Overall, the challenges are well understood. Clinical trials are costly and face problems with patient recruitment and withdrawal, as well as ethical dilemmas with treatment allocation. Global events like the recent COVID pandemic and armed conflicts also influence the availability of funding, patient participation, and trial timelines. Due to these challenges, clinical trials may not be sufficiently powered to detect treatment effects.
What are synthetic control arms?
As an umbrella term, synthetic control arms are trial groups generated using external data sources, including real-world hospital datasets and historical trial data. The approaches for generating synthetic control arms vary in complexity but can be grouped into methods that use real data and methods that deploy statistics or machine learning to produce computer-generated control data. There seems to be a shift in the terminology, where the terms synthetic or fully synthetic are used to denote the latter group, whereas the former group is referred to as external control arm generation.
Synthetic control arms allow existing data to be leveraged to reduce the number of patients who need to be recruited in clinical trials. As a result, this makes clinical trials in areas like rare diseases, for example, more feasible. They also help alleviate ethical concerns related to treatment versus placebo allocation.1
By leveraging external control arms generated from existing data, researchers can better evaluate treatment effects in clinical trials for idiopathic pulmonary fibrosis (IPF).
©iStock, magicmine
What approach does Qureight use to create synthetic control arms?
Our approach for synthetic control arm generation is conceptually straightforward and intuitive. At its core, it is based on repeatedly drawing a random selection of patients from an external dataset. The random samples are promoted to external control arms if they are similar to the trial group on several pre-selected baseline characteristics. This results in the generation of not one but several hundred external control arms, typically with the same size as the trial arm. Although the trial patients are fixed, the procedure is equivalent to performing the trial many times, each time with a different randomly generated control arm.
Since our approach does not involve learning any distribution parameters or training a model, it is more appropriate to refer to it as external control arm generation. The method uses tabular real-world data. Our work has so far focused on IPF, where we have used demographic and lung function data. In some experiments, we have also controlled for lung volume and fibrosis percentage, derived from high-resolution computed tomography (HRCT) scans. The use of HRCT biomarkers is not trivial; to our knowledge, we are the first to do this in IPF and potentially the first for synthetic control arm generation in general.
In Vicore Pharma’s single-arm AIR study in IPF, how did Qureight compare outcomes between the AIR study cohort and the external control arms, and what did this analysis show at 36 weeks?
For Vicore’s AIR trial in IPF, we matched post hoc the buloxibutid-treated and control groups on eight baseline characteristics in the external control arm generation stage. Then we compared the mean change in forced vital capacity (FVC) of the AIR cohort to the distribution of mean changes of the external control arms. FVC is the most common primary endpoint in respiratory drug trials. Only one of the 408 generated control arms had a better mean response than the AIR cohort. This shows that the probability of obtaining the trial result by chance is very small. Overall, our results demonstrate the potential efficacy of buloxibutid for treating IPF.
In Vicore Pharma’s ongoing ASPIRE study, how is Qureight using its Core Imaging Platform to collect and analyze HRCT data from patients with IPF?
To get a better understanding of how patients with different IPF phenotypes respond to orally administered buloxibutid, Vicore has included quantitative HRCT changes as a secondary trial endpoint in its ASPIRE study. Qureight receives the HRCT scans of ASPIRE patients on its platform, from 90 sites in 14 countries, in real time. The scans are verified using central reading, then processed and segmented into different lung tissues by our proprietary neural network segmentation algorithms. Finally, the demographic, lung function, and HRCT data of ASPIRE patients are automatically combined, analyzed, and compiled into an analytics report. These steps have been optimized for efficiency and speed to allow Vicore to continuously track the characteristics of the trial population as patients are enrolled and progress through the trial.
For the Avalyn study in IPF, how did Qureight’s external control arm support the assessment of treatment efficacy, and what insights have emerged?
Avalyn had previously reported a statistically significant difference in the mean change of FVC of patients in the ATLAS trial treated with two different doses of nebulized pirfenidone (AP01).2 To strengthen this finding further, we performed post hoc external control arm generation for both dose groups, using data for treatment-naive IPF patients. The experiments were done for different subsets of the trial data with and without imputation; some also included HRCT biomarkers in the external control generation. In all cases, we found a positive treatment effect for the high dose of AP01 and no treatment effect for the low dose. Our results are therefore in line with the reported finding that the high AP01 dose is effective in slowing down lung function decline associated with IPF.
How might synthetic control arms accelerate drug development and influence the design of future clinical trials?
Synthetic control arm generation is growing in acceptance in drug approval by regulatory authorities and could have a major impact on all clinical phases of drug development. Starting upstream, synthetic control arm generation could help provide a more reliable early indication of the efficacy of investigational therapies. By leveraging existing data to generate or complement existing cohorts in Phase 1 and 2 trials, synthetic control arm generation could dramatically decrease false discovery, prevent further waste of time and resources for larger clinical studies, and allow drug developers to focus on exploring other drug development strategies. Similarly, synthetic control arm generation can lead to trial size reduction for Phase 3 trials, faster patient recruitment, simplified logistics, and lower costs.
Synthetic control arm generation is not the only way to accelerate clinical trials. We are also working on developing tools and processes in our cloud data platform to allow efficient study management, fast site qualification, fast and accurate central read turnaround, and novel imaging endpoints and data analysis. All these measures taken together will enable us to establish a new standard for conducting clinical trials.
- Thorlund K, et al. Synthetic and external controls in clinical trials – A primer for researchers. Clin Epidemiol. 2020;12:457-467.
- West A, et al. Inhaled pirfenidone solution (AP01) for IPF: A randomised, open-label, dose–response trial. Thorax. 2023;78(9):882-889.
