Instrumental variable approach to estimating individual causal effects in N-of-1 trials: application to ISTOP study

Area of Science:

• Biostatistics
• Personalized Medicine
• Wearable Technology

Background:

• N-of-1 trials are crucial for personalized treatment decisions, especially with heterogeneous responses.
• Wearable devices enhance the feasibility of N-of-1 trials for identifying optimal individual treatment plans.
• The I-STOP-AFib study investigated triggers for atrial fibrillation (AF).

Purpose of the Study:

• To develop a causal framework for N-of-1 trials.
• To define and estimate individual causal effects considering continuous and observed behaviors.
• To address challenges like imperfect compliance, binary outcomes, and serial correlation.

Main Methods:

• Utilized a causal framework with potential treatment and outcome paths.
• Employed a Bayesian instrumental variable (IV) approach, using randomization as the IV.
• Developed a system of two parametric Bayesian models to estimate individual causal effects, addressing non-collapsibility and serial correlation.

Main Results:

• The proposed Bayesian IV method demonstrated reduced bias and improved coverage for estimated causal effects in simulations.
• The method effectively handled imperfect compliance, binary outcomes (non-collapsibility), and longitudinal data autocorrelation.
• Applied to the I-STOP-AFib study, it estimated the individual effect of alcohol on AF occurrence.

Conclusions:

• The Bayesian IV approach provides a robust framework for N-of-1 trials, enhancing personalized medicine.
• This method accurately estimates individual causal effects in complex scenarios, outperforming existing methods.
• It offers a valuable tool for analyzing data from wearable devices in clinical studies like I-STOP-AFib.