HAI Book 2025 - Flipbook - Page 577
Wu, Chao-Yi
132
Twin-mapping techniques in clinical trials: a proof-of-concept
application to the Internet-Based Conversational Engagement Clinical Trial
(I-CONECT)
Chao-Yi Wu1, Liu Chen1, Steven Arnold1, Hiroko Dodge1
1
Neurology, Massachusetts General Hospital (MGH), Harvard Medical School, Charlestown, MA, US
Objectives: With U.S. Food and Drug Administration (FDA)-approved anti-amyloid, partially disease-modifying
treatments now available, the ethical justification for randomly assigning patients to placebo has become
controversial. The idea of twin-controls has gained significant attention in recent years, with the aim of creating
twin patient cohorts that can be used as a surrogate to predict the effects of treatment on a more personalized
level. While promising, the feasibility of twin-mapping techniques remains largely untested.
Methods: I-CONECT is a multi-site randomized controlled trial (RCT) examining the effects of conversational
interactions on cognition among socially-isolated subjects aged g75 years (normal cognition; mild cognitive
impairment). 186 participants were randomized into experimental or control groups. The experimental group
engaged in video-chats with study staff 4 times/week for 6 months, while the control group received weekly 10minute phone calls. The current analysis focused on the efficacy-shown Montreal Cognitive Assessment (MoCA;
global cognition) and category fluency animals (language-based executive function) at 6-month follow-up. Ten
digital twins were matched to each participant receiving the treatment using data from the National Alzheimer's
Coordinating Center Uniform Data Set (NACC-UDS) based on similarity distance metrics (Euclidean;Cosine).
Treatment effect sizes and p-values were compared between the original and the twin-control-based result.
Results: Approximately 10% of NACC-UDS participants (5,332 out of 50,259) were eligible for I-CONECT (Figure 1).
After twin-mapping, sample sizes for the original control vs. twin-control were 31 vs. 112 for MoCA and 86 vs. 330
for category fluency. Treatment effect sizes showed substantial agreement between original and twin-control
results (MoCA: 1.75±0.76 vs. 1.54±0.41; category fluency: 2.56±1.19 vs. 2.62±1.05) (Figure 1). P-values remained
significant and smaller when treating the twin samples as controls.
Conclusions: Digital twins from publicly available datasets enhance the rigor of RCTs by providing mapped twins
as controls for early-phase dementia trials. Figures 1. Proof-of-concept of twin-mapping technique.
Keywords: virtual twins; synthesized control; anti-amyloid drugs; disease modifying; natural history cohort study
HAI2025 - 577
