Rodney A Gabriel1,2,3, Beamy S Sharma1, Christina N Doan1, Xiaoqian Jiang2, Ulrich H Schmidt1, Florin Vaida4. 1. From the Department of Anesthesiology. 2. Health Sciences Department of Biomedical Informatics, University of California, San Diego, La Jolla, California. 3. Outcomes Research Consortium, Cleveland, Ohio. 4. Division of Biostatistics and Bioinformatics, University of California, San Diego, La Jolla, California.
Abstract
BACKGROUND: Hospital length of stay (LOS) is an important quality metric for total hip arthroplasty. Accurately predicting LOS is important to expectantly manage bed utilization and other hospital resources. We aimed to develop a predictive model for determining patients who do not require prolonged LOS. METHODS: This was a retrospective single-institution study analyzing patients undergoing elective unilateral primary total hip arthroplasty from 2014 to 2016. The primary outcome of interest was LOS less than or equal to the expected duration, defined as ≤3 days. Multivariable logistic regression was performed to generate a model for this outcome, and a point-based calculator was designed. The model was built on a training set, and performance was assessed on a validation set. The area under the receiver operating characteristic curve and the Hosmer-Lemeshow test were calculated to determine discriminatory ability and goodness-of-fit, respectively. Predictive models using other machine learning techniques (ridge regression, Lasso, and random forest) were created, and model performances were compared. RESULTS: The point-based score calculator included 9 variables: age, opioid use, metabolic equivalents score, sex, anemia, chronic obstructive pulmonary disease, hypertension, obesity, and primary anesthesia type. The area under the receiver operating characteristic curve of the calculator on the validation set was 0.735 (95% confidence interval, 0.675-0.787) and demonstrated adequate goodness-of-fit (Hosmer-Lemeshow test, P = .37). When using a score of 12 as a threshold for predicting outcome, the positive predictive value was 86.1%. CONCLUSIONS: A predictive model that can help identify patients at higher odds for not requiring a prolonged hospital LOS was developed and may aid hospital administrators in strategically planning bed availability to reduce both overcrowding and underutilization when coordinating with surgical volume.
BACKGROUND: Hospital length of stay (LOS) is an important quality metric for total hip arthroplasty. Accurately predicting LOS is important to expectantly manage bed utilization and other hospital resources. We aimed to develop a predictive model for determining patients who do not require prolonged LOS. METHODS: This was a retrospective single-institution study analyzing patients undergoing elective unilateral primary total hip arthroplasty from 2014 to 2016. The primary outcome of interest was LOS less than or equal to the expected duration, defined as ≤3 days. Multivariable logistic regression was performed to generate a model for this outcome, and a point-based calculator was designed. The model was built on a training set, and performance was assessed on a validation set. The area under the receiver operating characteristic curve and the Hosmer-Lemeshow test were calculated to determine discriminatory ability and goodness-of-fit, respectively. Predictive models using other machine learning techniques (ridge regression, Lasso, and random forest) were created, and model performances were compared. RESULTS: The point-based score calculator included 9 variables: age, opioid use, metabolic equivalents score, sex, anemia, chronic obstructive pulmonary disease, hypertension, obesity, and primary anesthesia type. The area under the receiver operating characteristic curve of the calculator on the validation set was 0.735 (95% confidence interval, 0.675-0.787) and demonstrated adequate goodness-of-fit (Hosmer-Lemeshow test, P = .37). When using a score of 12 as a threshold for predicting outcome, the positive predictive value was 86.1%. CONCLUSIONS: A predictive model that can help identify patients at higher odds for not requiring a prolonged hospital LOS was developed and may aid hospital administrators in strategically planning bed availability to reduce both overcrowding and underutilization when coordinating with surgical volume.
Authors: Olivier Q Groot; Michiel E R Bongers; Paul T Ogink; Joeky T Senders; Aditya V Karhade; Jos A M Bramer; Jorrit-Jan Verlaan; Joseph H Schwab Journal: Clin Orthop Relat Res Date: 2020-12 Impact factor: 4.755
Authors: Cesar D Lopez; Anastasia Gazgalis; Venkat Boddapati; Roshan P Shah; H John Cooper; Jeffrey A Geller Journal: Arthroplast Today Date: 2021-09-03