Predictive Modeling of Length of Hospital Stay Following Adult Spinal Deformity Correction: Analysis of 653 Patients with an Accuracy of 75% within 2 Days.
BACKGROUND: Length of stay (LOS) after surgery for adult spinal deformity (ASD) is a critical period that allows for optimal recovery. Predictive models that estimate LOS allow for stratification of high-risk patients. METHODS: A prospectively acquired multicenter database of patients with ASD was used. Patients with staged surgery or LOS >30 days were excluded. Univariable predictor importance ≥0.90, redundancy, and collinearity testing were used to identify variables for model building. A generalized linear model was constructed using a training dataset developed from a bootstrap sample; patients not randomly selected for the bootstrap sample were selected to the training dataset. LOS predictions were compared with actual LOS to calculate an accuracy percentage. RESULTS: Inclusion criteria were met by 653 patients. The mean LOS was 7.9 ± 4.1 days (median 7 days; range, 1-28 days). Following bootstrapping, 893 patients were modeled (653 in the training model and 240 in the testing model). Linear correlations for the training and testing datasets were 0.632 and 0.507, respectively. The prediction accuracy within 2 days of actual LOS was 75.4%. CONCLUSIONS: Our model successfully predicted LOS after ASD surgery with an accuracy of 75% within 2 days. Factors relating to actual LOS, such as rehabilitation bed availability and social support resources, are not captured in large prospective datasets. Predictive analytics will play an increasing role in the future of ASD surgery, and future models will seek to improve the accuracy of these tools.
Safaee, MM; Scheer, JK; Ailon, T; Smith, JS; Hart, RA; Burton, DC; Bess, S; Neuman, BJ; Passias, PG; Miller, E; Shaffrey, CI; Schwab, F; Lafage, V; Klineberg, EO; Ames, CP; International Spine Study Group,
Volume / Issue
Start / End Page
Pubmed Central ID
Electronic International Standard Serial Number (EISSN)
Digital Object Identifier (DOI)