A generalized synthetic control algorithm for sparse functional data

Area of Science:

• Causal Inference
• Biostatistics
• Longitudinal Data Analysis

Background:

• Synthetic Control Method (SCM) and Generalized Synthetic Control (GSC) are effective for panel data but struggle with irregular or sparse follow-up common in biomedical research.
• Biomedical cohort studies often feature unequally spaced measurements, posing challenges for traditional causal inference methods.
• Existing methods may not adequately capture complex outcome dynamics or handle limited observations per participant.

Purpose of the Study:

• To develop a Bayesian functional extension of Generalized Synthetic Control (GSC) that accommodates irregularly spaced and sparse measurements in biomedical cohort data.
• To enable robust causal inference by treating unit outcomes as smooth latent trajectories approximated by Functional Principal Components Analysis (FPCA).
• To estimate the causal effect of adolescent binge drinking on subsequent brain volumes using longitudinal neuroimaging data.

Main Methods:

• Developed a Bayesian functional extension of GSC, approximating unit outcome paths as smooth latent trajectories using FPCA.
• Learned unit and time latent factors jointly with FPCA scores from control data to construct counterfactual trajectories for treated units.
• Quantified uncertainty via posterior distributions and relied on latent-factor/weak-trend conditions and functional score space overlap for identification.

Main Results:

• Simulation studies demonstrated that the GSC-FPCA approach yields low bias and well-calibrated interval coverage even with irregular or sparse sampling.
• Application to the National Consortium on Alcohol and Neurodevelopment in Adolescence - Adulthood (NCANDA-A) study successfully estimated effects with 1 to 9 observations per participant.
• The method detected a negative impact of sustained high levels of adolescent binge drinking on gray-matter brain volumes.

Conclusions:

• Embedding Generalized Synthetic Control (GSC) within a functional framework (GSC-FPCA) enables robust causal inference in biomedical applications.
• The proposed method is suitable for studies with irregularly spaced visits, limited observations, and complex outcome dynamics.
• GSC-FPCA provides a powerful tool for analyzing longitudinal biomedical data and understanding developmental trajectories.