Primary Submission Category: Heterogeneous Treatment Effects
Developing and validating a treatment recommendation score with an application to the heterogeneous impact of transesophageal echocardiography on clinical outcomes post coronary artery bypass graft surgery
Authors: Charlotte Talham, Emily MacKay, Qijia He, Bo Zhang,
Presenting Author: Charlotte Talham*
While past observational research has shown transesophageal echocardiography (TEE) to improve patient outcomes post coronary artery bypass graft (CABG) surgery, resource constraints in clinical settings limit the use of TEE in practice. Our aim is thus to adopt a precision medicine approach to better assign TEE to patients who could most benefit from its use. Two practical challenges emerge. First, to be useful to practitioners, a summary of the heterogeneous treatment effect is needed to inform decisions about how to allocate limited resources. Second, validation of the proposed precision-TEE approach is critical to promoting its integration into medical practice. We present a framework that aims to address both challenges. We first perform an exploratory analysis on a random subset of data in which we take a machine learning approach to develop a treatment recommendation score (TRS) through solving a series of weighted classification problems, each subject to varying resource constraints. Next, we use the remainder of the data to validate the findings in a matched cohort study in which participants who received TEE are compared to those who did not receive TEE, but who were assigned the same TRS. Our matching procedure balances a large number of clinical and demographic covariates. The proposed framework is illustrated using synthetic data modeled after the Society of Thoracic Surgeons (STS) national registry. This method is also being used to analyze the STS registry data.