Date of Award

January 2025

Document Type

Thesis

Degree Name

Medical Doctor (MD)

Department

Medicine

First Advisor

Jinlei Li

Abstract

In elective spinal fusion surgeries, effective postoperative control is integral to early and expedite recovery. Machine learning (ML) has been used in various fields of medicine to assist in clinical decision making and pattern recognition. In this thesis, we hypothesize that we will be able to develop AI models from ML algorithms that can reasonably predict postoperative pain and opioid usage. We also hypothesize that using erector spinae plane (ESP) block in elective spinal fusion surgeries will lead to decreased postoperative reported pain and opioid usage. Retrospective chart reviews of 3071 cases were conducted at Hartford Healthcare. These cases were divided into control group which did not receive the ESP blocks and the treatment group which received the blocks. The clinical and demographic data obtained from each study were then assessed and analyzed using appropriate statistical analytic methods based on the normality of each dataset. Python was used as the programming language and appropriate libraries such as Pandas and SciKit-Learn were used for each statistical analysis and for the machine learning portion of the study. Gradient boosting ensemble tree methodology was used to create the AI predictive models. Results suggested that ESP block may be effective in reducing total opioid usage. However, it was associated with higher postoperative pain. The four AI models developed from the study showed promising results in predicting postsurgical pain and opioid usage, with the best model showing mean absolute error of 13.3%. Overall, ESP block may provide clinical benefit in reducing opioid usage in perioperative periods for elective spine surgery. The increased postoperative pain associated with ESP may be due to the implicit bias of the anesthesiologist knowing the patients who received preoperative ESP. This knowledge may have led to them providing less than adequate pain control perioperatively, leading to increased pain reported. Pursuing this study further with a prospective, double-blinded RCT design would allow controlling for such biases and help clarify the potential clinical benefit of ESP blocks in elective posterior fusion spine surgery.

Comments

This thesis is restricted to Yale network users only. It will be made publicly available on 05/14/2027

Download

Share

COinS