Tuesday, October 13 @ 11:00 - 12:30 PM (ONLINE)
A resampling approach for causal inference on two-point time-series with application to identify risk factors for type-2 diabetes and cardiovascular disease
Xiaowu Dai & Saad Mouti, UC Berkeley
ABSTRACT: Two-point time-series data, characterized by baseline and follow-up observations, are frequently encountered in health research. In analyzing such time-series data, the two-point pattern must be adequately accounted to balance the trade-off between efficient inference and estimation bias. Although many methods have been proposed in the literature to analyze two-point time-series data, little attention has been given to exploit the two-point pattern to identify important risk factors that could indicate specific health outcomes or intervention effects. We propose a resampling approach called `i-randomization' for independently subsampling one of the two time points for each individual and making inference on the estimated causal effects based on matching methods. The proposed method is illustrated with data from a dietary intervention trial to promote a low-carbohydrate diet (LCD), and to identify risk factors for type-2 diabetes (T2D) and cardiovascular disease (CVD). Baseline data in the trial contain a preintervention health record of study participants, and the health data after LCD intervention are recorded at the follow-up visit, yielding two-point time-series pattern. We find that obesity is a significant risk factor of T2D and CVD, and LCD would significantly mitigate the risks of T2D and CVD. A software package is provided that implements our method.
Please register to attend this seminar.