Akash A Shah1, Sai K Devana2, Changhee Lee3, Amador Bugarin2, Elizabeth L Lord2, Arya N Shamie2, Don Y Park2, Mihaela van der Schaar4, Nelson F SooHoo2. 1. Department of Orthopaedic Surgery, David Geffen School of Medicine at UCLA, Los Angeles, California, USA. Electronic address: aashah@mednet.ucla.edu. 2. Department of Orthopaedic Surgery, David Geffen School of Medicine at UCLA, Los Angeles, California, USA. 3. Department of Electrical and Computer Engineering, University of California, Los Angeles, California, USA. 4. Department of Electrical and Computer Engineering, University of California, Los Angeles, California, USA; Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, United Kingdom.
Abstract
BACKGROUND: Given the significant cost and morbidity of patients undergoing lumbar fusion, accurate preoperative risk-stratification would be of great utility. We aim to develop a machine learning model for prediction of major complications and readmission after lumbar fusion. We also aim to identify the factors most important to performance of each tested model. METHODS: We identified 38,788 adult patients who underwent lumbar fusion at any California hospital between 2015 and 2017. The primary outcome was major perioperative complication or readmission within 30 days. We build logistic regression and advanced machine learning models: XGBoost, AdaBoost, Gradient Boosting, and Random Forest. Discrimination and calibration were assessed using area under the receiver operating characteristic curve and Brier score, respectively. RESULTS: There were 4470 major complications (11.5%). The XGBoost algorithm demonstrates the highest discrimination of the machine learning models, outperforming regression. The variables most important to XGBoost performance include angina pectoris, metastatic cancer, teaching hospital status, history of concussion, comorbidity burden, and workers' compensation insurance. Teaching hospital status and concussion history were not found to be important for regression. CONCLUSIONS: We report a machine learning algorithm for prediction of major complications and readmission after lumbar fusion that outperforms logistic regression. Notably, the predictors most important for XGBoost differed from those for regression. The superior performance of XGBoost may be due to the ability of advanced machine learning methods to capture relationships between variables that regression is unable to detect. This tool may identify and address potentially modifiable risk factors, helping risk-stratify patients and decrease complication rates.
BACKGROUND: Given the significant cost and morbidity of patients undergoing lumbar fusion, accurate preoperative risk-stratification would be of great utility. We aim to develop a machine learning model for prediction of major complications and readmission after lumbar fusion. We also aim to identify the factors most important to performance of each tested model. METHODS: We identified 38,788 adult patients who underwent lumbar fusion at any California hospital between 2015 and 2017. The primary outcome was major perioperative complication or readmission within 30 days. We build logistic regression and advanced machine learning models: XGBoost, AdaBoost, Gradient Boosting, and Random Forest. Discrimination and calibration were assessed using area under the receiver operating characteristic curve and Brier score, respectively. RESULTS: There were 4470 major complications (11.5%). The XGBoost algorithm demonstrates the highest discrimination of the machine learning models, outperforming regression. The variables most important to XGBoost performance include angina pectoris, metastatic cancer, teaching hospital status, history of concussion, comorbidity burden, and workers' compensation insurance. Teaching hospital status and concussion history were not found to be important for regression. CONCLUSIONS: We report a machine learning algorithm for prediction of major complications and readmission after lumbar fusion that outperforms logistic regression. Notably, the predictors most important for XGBoost differed from those for regression. The superior performance of XGBoost may be due to the ability of advanced machine learning methods to capture relationships between variables that regression is unable to detect. This tool may identify and address potentially modifiable risk factors, helping risk-stratify patients and decrease complication rates.
Authors: Andre Esteva; Brett Kuprel; Roberto A Novoa; Justin Ko; Susan M Swetter; Helen M Blau; Sebastian Thrun Journal: Nature Date: 2017-06-28 Impact factor: 49.962
Authors: José H Jiménez-Almonte; Gavin S Hautala; Eric J Abbenhaus; Jonathan D Grabau; Ifeanyi N Nzegwu; Syed K Mehdi; Zeeshan M Akhtar; Boshen Liu; Cale A Jacobs; R Carter Cassidy Journal: Spine J Date: 2020-06-02 Impact factor: 4.166
Authors: Brook I Martin; Richard A Deyo; Sohail K Mirza; Judith A Turner; Bryan A Comstock; William Hollingworth; Sean D Sullivan Journal: JAMA Date: 2008-02-13 Impact factor: 56.272
Authors: Vadim Goz; Jeffrey H Weinreb; Ian McCarthy; Frank Schwab; Virginie Lafage; Thomas J Errico Journal: Spine (Phila Pa 1976) Date: 2013-10-15 Impact factor: 3.468
Authors: Nathan J Lee; Parth Kothari; Kevin Phan; John I Shin; Holt S Cutler; Nikita Lakomkin; Dante M Leven; Javier Z Guzman; Samuel K Cho Journal: Spine (Phila Pa 1976) Date: 2018-01-01 Impact factor: 3.468
Authors: Daniel D Bohl; Benjamin C Mayo; Dustin H Massel; Stephanie E Iantorno; Junyoung Ahn; Bryce A Basques; Jonathan N Grauer; Kern Singh Journal: Spine (Phila Pa 1976) Date: 2016-06 Impact factor: 3.468