Transforming Clinical Trials with Intelligent Control Arms

Summary:

Because clinical trials are inefficient, expensive, and time-consuming, clinical trial sponsors are seeking ways to improve the process, in order to increase the likelihood of finding an effective treatment. One way to improve this process is to use digital twins populated in intelligent control arms, a type of external control arm. Using digital twins in intelligent control arms increases the efficiency of clinical trials and confidence in trial results by: 

  1. reducing the time needed for patient recruitment with fewer subjects and a higher treated-to-placebo ratio
  2. increasing study power by reducing variability between control and treated subjects
  3. providing individualized patient information about response to treatment

Read further: 

Clinical trials are often inefficient, expensive, and time-consuming. This is especially true for Alzheimer's clinical trials, which take years to complete [1] and can cost millions of dollars [2]. 

Why is that? 

A significant cause is the inability to recruit a sufficient number of patients [3]. In many cases, patients don't meet eligibility criteria [4] or don't enroll because they don't want to receive a placebo [5] -- thus preventing trials from reaching the number of patients they need. 

To address this problem, we can use historical trial data and machine learning to create digital twins that can act as actual subjects in a trial [6]. Digital twins are comprehensive, longitudinal, and computationally generated clinical records that describe what would likely have happened to a specific subject if they had received a placebo.

So how can we use these digital twins in clinical trials?

Digital twins can be used to populate an intelligent control arm, a type of external control arm with a variety of applications: 

Application 1: Supplement a single-arm trial

If we plan to run a single-arm, open-label trial, an intelligent control arm can be used as an external control arm for such a trial. This means that every subject in the treatment arm will have a digital twin that is a perfectly matched control counterpart (figure 1). Because the digital twin is perfectly matched to each subject, we reduce variability between patient populations at baseline when using digital twins*, compared to using traditional historical controls. 

Application 2: Augment a randomized controlled trial (RCT)

If we plan to run an RCT with a concurrent control, we can add an intelligent control arm to the trial to reduce the number of subjects in the control arm. The randomization ratio of an Alzheimer's trial has recently been 1:1, meaning that 50% of subjects are in the control arm, and 50% are in the treatment arm. By adding an intelligent control arm, we can change the ratio to e.g. 3:1** with 75% of subjects in the treatment arm and 25% of subjects in the concurrent control arm. 

As shown in figure 2, each subject in the treatment arm and the concurrent control arm will have a digital twin. Creating digital twins for all subjects prior to treatment assignment preserves blinding and randomization of the trial. 


The above applications represent two major use cases of digital twins in intelligent control arms. To summarize, we can see three benefits of using intelligent control arms: 

  1. They reduce the time needed for patient recruitment by requiring fewer subjects and increasing the treated-to-placebo ratio. 
  2. They increase study power by reducing variability. Because digital twins are perfectly matched to actual subjects in the treatment arm at baseline, clinical trial sponsors can assess the effectiveness of a specific treatment with greater certainty.
  3. They provide individualized patient information about response to treatment. When we have digital twins matched to each subject of the trial, each twin acts as a counterfactual for the paired subject. We can get an estimate of how effective the treatment was for every individual subject and then identify subgroups of patients who respond best to the treatment. 

Footnotes

[1]http://phrma-docs.phrma.org/files/dmfile/TEConomy_PhRMA-Clinical-Trials-Impacts.pdf?origin_team=T4QH08C78

[2] Phase III clinical trials for Alzheimer's disease, for example, cost ~$40,000/subject. If 2000 subjects are recruited, the cost would amount to $80 million.(http://phrma-docs.phrma.org/files/dmfile/TEConomy_PhRMA-Clinical-Trials-Impacts.pdf?origin_team=T4QH08C78, https://www.nejm.org/doi/full/10.1056/NEJMoa1705971)

[3] https://www.ascopost.com/News/60106

[4] https://forteresearch.com/news/recruitment-efforts-fail-enroll-enough-patients/

[5] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5327530/

[6] https://www.unlearn.health/blog/digital-twins-for-matching-third-times-a-charm

*If we assume the model we use for creating digital twins is perfect, then we will reduce variability when including digital twins, compared to using traditional historical controls. However, if the model has some error, then we will add error to the estimate for how effective the treatment is. 

**The 3:1 randomization ratio is recommended as an example. What constitutes the best randomization ratio for reducing the right number of actual subjects will be partly based on what the FDA can accept. 

Enter your email address to download paper.

Click the link to begin download.
Oops! Something went wrong while submitting the form.
Blog

CTAD 2020: Unlearn will present abstracts to demonstrate how Digital Twins enable smaller, more efficient trials

Blog

Announcing The Unlearn Opportunities Internship Program

Podcasts

UCSF Rosenman Institute - The Health Technology Podcast #59: Digital Twins for Clinical Trials

Charles Fisher: Digital Twins for Clinical Trials
Creating Opportunities for Students of Data Science and Business from Underrepresented Groups in STEM
Unlearn will present data from two separate abstracts demonstrating how Digital Twins can power novel trial designs and accelerate timelines.