Mert Karabacak, Konstantinos Margetis
Abstract
By predicting short-term postoperative outcomes before surgery, patients who undergo posterior cervical fusion (PCF) surgery may benefit from more precise patient care plans that reduce the likelihood of unfavorable outcomes. We developed machine learning models for predicting short-term postoperative outcomes and incorporate these models into an open-source web application in this study.
Introduction
Given the importance of preoperative risk identification and facilitating shared decision-making, the adoption of machine learning (ML) classifiers in clinical prediction models offers a substantial edge over conventional methods. This advantage stems from their capacity to handle complex, high-dimensional, and non-linear relationships among variables [5]. Conventional methods like logistic regression are restricted by their linear nature and require assumptions of variable independence [6].
Materials and methods
This study utilized data from the American College of Surgeons (ACS) National Surgical Quality Improvement Program (NSQIP) database to identify patients who underwent PCF between 2014 and 2020. Excluding trauma and transplant cases, the ACS-NSQIP database serves as a national surgical registry for adult patients who have had major surgical procedures in various subspecialties at over 700 participating medical centers throughout the United States [10, 11]. Comprehensive information regarding the database and its data collection techniques can be found in other sources [12].
Results
The most accurately predicted outcome in terms of AUROC was the non-home discharges, with a mean AUROC of 0.812 and an accuracy of 81.5%. The most accurately predicting algorithm in terms of AUROC was LightGBM, with a mean AUROC of 0.766, followed by CatBoost, with a mean AUROC of 0.764. The mean AUROCs for Random Forest and XGBoost were 0.763 and 0.752, respectively. Detailed metrics regarding the algorithms’ performances are presented in Table 1. AUROC and AUPRC curves for the three outcomes are shown in Figs 2 and 3.
Discussion
A group of ML models that can predict prolonged LOS, non-home discharges, and readmissions for patients who undergo PCF surgery is presented in this research. Our study suggests that the use of ML models could help with the risk stratification process for PCF surgery, and there is a substantial potential for predicting surgical outcomes. By providing patients with better information about the risks of surgery, clinicians may be able to customize patient care plans for those who are at risk of experiencing adverse outcomes following PCF surgery. This study contributes to the existing knowledge by describing the advantages and effectiveness of incorporating ML into patient care to anticipate outcomes after spine surgery [24].
Conclusion
ML models hold significant potential in predicting postoperative outcomes after PCF surgery. These algorithms can be integrated into decision-making tools that have clinical relevance. The development and utilization of predictive models as easily accessible tools could substantially enhance prognosis and risk management. In this study, we introduce a predictive algorithm for PCF surgery with the goal of fulfilling the aforementioned objectives, which is available to the general public.
Citation: Karabacak M, Margetis K (2023) Interpretable machine learning models to predict short-term postoperative outcomes following posterior cervical fusion. PLoS ONE 18(7): e0288939. https://doi.org/10.1371/journal.pone.0288939
Editor: Mohamed El-Sayed Abdel-Wanis, Sohag University Faculty of Medicine, EGYPT
Received: March 30, 2023; Accepted: July 6, 2023; Published: July 21, 2023
Copyright: © 2023 Karabacak, Margetis. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: "Restrictions apply to the availability of these data, as the data is shared solely with fellows of American College of Surgeons. Data was obtained from American College of Surgeons National Surgical Quality Improvement Program and are available (https://www.facs.org/quality-programs/data-and-registries/acs-nsqip/) with the permission of American College of Surgeons."
Funding: The authors received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0288939#abstract0
