Date of Award

Fall 10-1-2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Investigative Medicine

First Advisor

Krumholz, Harlan

Abstract

Understanding the patterns of postoperative recovery after cardiac surgery is important from several perspectives: to facilitate patient-centered treatment decision making, to inform health care policy targeted to improve postoperative recovery, and to guide patient care after cardiac surgery. Our works aimed to address the following: 1) to summarize existing approaches to measuring and reporting postoperative recovery after cardiac surgery, 2) to develop a framework to efficiently measure patient-reported outcome measures to understand longitudinal recovery process, and 3) to explore ways to summarize the longitudinal recovery data in an actionable way, and 4) to evaluate whether addition of patient information generated through different phases of care would improve the ability to predict patient’s outcome. We first conducted a systematic review of the studies reporting on postoperative recovery after cardiac surgery using patient-reported outcome measures. Our systematic review demonstrated that the current approaches to measuring and reporting recovery as a treatment outcome varied widely across studies. This made synthesis of collective knowledge challenging and highlighted key gaps in knowledge, which we sought to address in our prospective cohort study. We conducted a prospective single-center cohort study of patients after cardiac surgery to measure their recovery trajectory across multiple domains of recovery. Using a digital platform, we measured patient recovery in various domains over 30 days after surgery to visualize a granular evolution of patient recovery after cardiac surgery. We used a latent class analysis to facilitate identification of dominant trajectory patterns that had been obscured in a conventional way of reporting such time-series data using group-level means. For the pain domain, we identified 4 trajectory classes, one of which was a group of patients with persistently high pain trajectory that only became distinguishable from less concerning group after 10 days. Therefore, we obtained a potentially actionable insights to tailoring individualized follow-up timing after surgery to improve the pain control. The prospective study embodied several important features to successfully conducting such studies of patient-reported outcomes. This included the use of digital platform to facilitate efficient data collection extending after hospital discharge, iteratively improving the protocol to optimize patient engagement including evaluation of potential barriers to survey completion, and using latent class analysis to identify dominant patterns of recovery trajectories. We outlined these insights in the protocol manuscript to inform subsequent studies aiming to leverage such a digital platform to measure longitudinal patient-centered outcome. Finally, we evaluated the potential value of incorporating health care data generated in the different phases of patient care in improving the prediction of postoperative outcomes after cardiac surgery. The current standard of risk prediction in cardiac surgery is the Society of Thoracic Surgeons’ (STS) risk model, which only uses patient information available preoperatively. We demonstrated through prediction models fitted on the national STS risk model for coronary artery bypass graft surgery that the addition of intraoperative variables to the conventional preoperative variable set improved the performance of prediction models substantially. Using machine learning approach to such a high-dimensional dataset proved to be marginally important. This work demonstrated the potential value and importance of being able to leverage health care data to continuously update the prediction to inform patient outcomes and guide clinical care. Our work collectively advanced knowledge in several key aspects of postoperative recovery. First, we highlighted the knowledge gap in the existing literature through characterizing the variability in the ways such studies had been conducted. Second, we designed and described a framework to measure postoperative recovery and an analytical approach to informatively characterize longitudinal patient recovery. Third, we employed these designs in a prospective cohort study to measure and analyze recovery trajectories and described clinical insights obtained from the study. Finally, we demonstrated the potential value of a dynamic risk model to iteratively improve its predictive performance by incorporating new data generated as the patient progresses through the phase of care. Such a platform has the potential to individualize patient’s post-acute care in a data-driven manner.

COinS