Presented at the 2019 AANS/CNS Section on Disorders of the Spine and Peripheral Nerves
Anshit Goyal, Che Ngufor, Panagiotis Kerezoudis, Brandon McCutcheon, Curtis Storlie and Mohamad Bydon
Nonhome discharge and unplanned readmissions represent important cost drivers following spinal fusion. The authors sought to utilize different machine learning algorithms to predict discharge to rehabilitation and unplanned readmissions in patients receiving spinal fusion.
The authors queried the 2012–2013 American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) for patients undergoing cervical or lumbar spinal fusion. Outcomes assessed included discharge to nonhome facility and unplanned readmissions within 30 days after surgery. A total of 7 machine learning algorithms were evaluated. Predictive hierarchical clustering of procedure codes was used to increase model performance. Model performance was evaluated using overall accuracy and area under the receiver operating characteristic curve (AUC), as well as sensitivity, specificity, and positive and negative predictive values. These performance metrics were computed for both the imputed and unimputed (missing values dropped) datasets.
A total of 59,145 spinal fusion cases were analyzed. The incidence rates of discharge to nonhome facility and 30-day unplanned readmission were 12.6% and 4.5%, respectively. All classification algorithms showed excellent discrimination (AUC > 0.80, range 0.85–0.87) for predicting nonhome discharge. The generalized linear model showed comparable performance to other machine learning algorithms. By comparison, all models showed poorer predictive performance for unplanned readmission, with AUC ranging between 0.63 and 0.66. Better predictive performance was noted with models using imputed data.
In an analysis of patients undergoing spinal fusion, multiple machine learning algorithms were found to reliably predict nonhome discharge with modest performance noted for unplanned readmissions. These results provide early evidence regarding the feasibility of modern machine learning classifiers in predicting these outcomes and serve as possible clinical decision support tools to facilitate shared decision making.
Bruce E. Pollock, Curtis B. Storlie, Michael J. Link, Scott L. Stafford, Yolanda I. Garces and Robert L. Foote
Successful stereotactic radiosurgery (SRS) for the treatment of arteriovenous malformations (AVMs) results in nidus obliteration without new neurological deficits related to either intracranial hemorrhage (ICH) or radiation-induced complications (RICs). In this study the authors compared 5 AVM grading scales (Spetzler-Martin grading scale, radiosurgery-based AVM score [RBAS], Heidelberg score, Virginia Radiosurgery AVM Scale [VRAS], and proton radiosurgery AVM scale [PRAS]) at predicting outcomes after SRS.
The study group consisted of 381 patients with sporadic AVMs who underwent Gamma Knife SRS between January 1990 and December 2009; none of the patients underwent prior radiation therapy. The primary end point was AVM obliteration without a decline in modified Rankin Scale (mRS) score (excellent outcome). Comparison of the area under the receiver operating characteristic curve (AUC) and accuracy was performed between the AVM grading scales and the best linear regression model (generalized linear model, elastic net [GLMnet]).
The median radiological follow-up after initial SRS was 77 months; the median clinical follow-up was 93 months. AVM obliteration was documented in 297 patients (78.0%). Obliteration was 59% at 4 years and 85% at 8 years. Fifty-five patients (14.4%) had a decline in mRS score secondary to RICs (n = 29, 7.6%) or ICH (n = 26, 6.8%). The mRS score declined by 10% at 4 years and 15% at 8 years. Overall, 274 patients (71.9%) had excellent outcomes. There was no difference between the AUC for the GLMnet (0.69 [95% CI 0.64–0.75]), RBAS (0.68 [95% CI 0.62–0.74]), or PRAS (0.69 [95% CI 0.62–0.74]). Pairwise comparison for accuracy showed no difference between the GLMnet and the RBAS (p = 0.08) or PRAS (p = 0.16), but it did show a significant difference between the GLMnet and the Spetzler-Martin grading system (p < 0.001), Heidelberg score (p < 0.001), and the VRAS (p < 0.001). The RBAS and the PRAS were more accurate when compared with the Spetzler-Martin grading scale (p = 0.03 and p = 0.01), Heidelberg score (p = 0.02 and p = 0.02), and VRAS (p = 0.03 and p = 0.02).
SRS provides AVM obliteration without functional decline in the majority of treated patients. AVM grading scales having continuous scores (RBAS and PRAS) outperformed integer-based grading systems in the prediction of AVM obliteration without mRS score decline after SRS.