Faculty Bibliography

BACKGROUND:Delays in postoperative referrals to rehabilitation or skilled nursing facilities contribute toward extended hospital stays. Facilitating more efficient referrals through accurate preoperative prediction algorithms has the potential to reduce unnecessary economic burden and minimize risk of hospital-acquired complications. We develop a robust machine learning algorithm to predict non-home discharge after thoracolumbar spine surgery that generalizes to unseen populations and identifies markers for prediction. METHODS:Retrospective electronic health records were obtained from the single-center data warehouse (SCDW) to identify patients undergoing thoracolumbar spine surgeries between 2008 and 2019 for algorithm development and internal validation. The National Inpatient Sample (NIS) database was queried to identify thoracolumbar surgeries between 2009 and 2017 for out-of-sample validation. Ensemble decision trees were constructed for prediction and area under the receiver operating characteristic curve (AUROC) was used to assess performance. SHAP values were derived to identify drivers of non-home discharge for interpretation of algorithm predictions. RESULTS:5,224 cases of thoracolumbar spine surgeries were isolated from the SCDW and 492,312 cases were identified from NIS. The model achieved an AUROCs of 0.81 (SD=0.01) on the SCDW test set and 0.77 (SD=0.01) on the nationwide NIS dataset, thereby demonstrating robust prediction of non-home discharge across all diverse patient cohorts. Age, total Elixhauser comorbidities, Medicare insurance, weighted Elixhauser score, and female gender were among the most important predictors of non-home discharge. CONCLUSIONS:Machine learning algorithms reliably predict non-home discharge after thoracolumbar spine surgery across single-center and national cohorts and identify preoperative features of importance that elucidate algorithm decision-making.

BACKGROUND:Extended postoperative hospital stays are associated with numerous clinical risks and increased economic cost. Accurate preoperative prediction of extended length of stay (LOS) can facilitate targeted interventions to mitigate clinical harm and resource utilization. OBJECTIVE:To develop a machine learning algorithm aimed at predicting extended LOS after cervical spine surgery on a national level and elucidate drivers of prediction. METHODS:Electronic medical records from a large, urban academic medical center were retrospectively examined to identify patients who underwent cervical spine fusion surgeries between 2008 and 2019 for machine learning algorithm development and in-sample validation. The National Inpatient Sample database was queried to identify cervical spine fusion surgeries between 2009 and 2017 for out-of-sample validation of algorithm performance. Gradient-boosted trees predicted LOS and efficacy was assessed using the area under the receiver operating characteristic curve (AUROC). Shapley values were calculated to characterize preoperative risk factors for extended LOS and explain algorithm predictions. RESULTS:Gradient-boosted trees accurately predicted extended LOS across cohorts, achieving an AUROC of 0.87 (SD = 0.01) on the single-center validation set and an AUROC of 0.84 (SD = 0.00) on the nationwide National Inpatient Sample data set. Anterior approach only, elective admission status, age, and total number of Elixhauser comorbidities were important predictors that affected the likelihood of prolonged LOS. CONCLUSION/CONCLUSIONS:Machine learning algorithms accurately predict extended LOS across single-center and national patient cohorts and characterize key preoperative drivers of increased LOS after cervical spine surgery.

BACKGROUND:The merit-based incentive payment system (MIPS) program was implemented to tie Medicare reimbursements to value-based care measures. Neurosurgical performance in MIPS has not yet been described. OBJECTIVE:To characterize neurosurgical performance in the first 2 years of MIPS. METHODS:Publicly available data regarding MIPS performance for neurosurgeons in 2017 and 2018 were queried. Descriptive statistics about physician characteristics, MIPS performance, and ensuing payment adjustments were performed, and predictors of bonus payments were identified. RESULTS:There were 2811 physicians included in 2017 and 3147 in 2018. Median total MIPS scores (99.1 vs 90.4, P < .001) and quality scores (97.9 vs 88.5, P < .001) were higher in 2018 than in 2017. More neurosurgeons (2758, 87.6%) received bonus payments in 2018 than in 2017 (2013, 71.6%). Of the 2232 neurosurgeons with scores in both years, 1347 (60.4%) improved their score. Reporting through an alternative payment model (odds ratio [OR]: 32.3, 95% CI: 16.0-65.4; P < .001) and any practice size larger than 10 (ORs ranging from 2.37 to 10.2, all P < .001) were associated with receiving bonus payments. Increasing years in practice (OR: 0.99; 95% CI: 0.982-0.998, P = .011) and having 25% to 49% (OR: 0.72; 95% CI: 0.53-0.97; P = .029) or ≥50% (OR: 0.48; 95% CI: 0.28-0.82; P = .007) of a physician's patients eligible for Medicaid were associated with lower rates of bonus payments. CONCLUSION/CONCLUSIONS:Neurosurgeons performed well in MIPS in 2017 and 2018, although the program may be biased against surgeons who practice in small groups or take care of socially disadvantaged patients.

A significant amount of clinical research is observational by nature and derived from medical records, clinical trials, and large-scale registries. While there is no substitute for randomized, controlled experimentation, such experiments or trials are often costly, time consuming, and even ethically or practically impossible to execute. Combining classical regression and structural equation modeling with matching techniques can leverage the value of observational data. Nevertheless, identifying variables of greatest interest in high-dimensional data is frequently challenging, even with application of classical dimensionality reduction and/or propensity scoring techniques. Here, we demonstrate that projecting high-dimensional medical data onto a lower-dimensional manifold using deep autoencoders and post-hoc generation of treatment/control cohorts based on proximity in the lower-dimensional space results in better matching of confounding variables compared to classical propensity score matching (PSM) in the original high-dimensional space (P<0.0001) and performs similarly to PSM models constructed by experts with prior knowledge of the underlying pathology when evaluated on predicting risk ratios from real-world clinical data. Thus, in cases when the underlying problem is poorly understood and the data is high-dimensional in nature, matching in the autoencoder latent space might be of particular benefit.

PURPOSE/UNASSIGNED:Vision loss following surgery for pituitary adenoma is poorly described in the literature and cannot be reliably predicted with current prognostic models. Detailed characterization of this population is warranted to further understand the factors that predispose a minority of patients to post-operative vision loss. MATERIALS AND METHODS/UNASSIGNED:The medical records of 587 patients who underwent endoscopic transsphenoidal surgery at the Mount Sinai Medical Centre between January 2013 and August 2018 were reviewed. Patients who experienced post-operative vision deterioration, defined by reduced visual acuity, worsened VFDs, or new onset of blurry vision, were identified and analysed. RESULTS/UNASSIGNED:Eleven out of 587 patients who received endoscopic surgery for pituitary adenoma exhibited post-operative vision deterioration. All eleven patients presented with preoperative visual impairment (average duration of 13.1 months) and pre-operative optic chiasm compression. Seven patients experienced visual deterioration within 24 h of surgery. The remaining four patients experienced delayed vision loss within one month of surgery. Six patients had complete blindness in at least one eye, one patient had complete bilateral blindness. Four patients had reduced visual acuity compared with preoperative testing, and four patients reported new-onset blurriness that was not present before surgery. High rates of graft placement (10/11 patients) and opening of the diaphragma sellae (9/11 patients) were found in this series. Four patients had hematomas and four patients had another significant post-operative complication. CONCLUSIONS/UNASSIGNED:While most patients with pituitary adenoma experience favourable ophthalmological outcomes following endoscopic transsphenoidal surgery, a subset of patients exhibit post-operative vision deterioration. The present study reports surgical and disease features of this population to further our understanding of factors that may underlie vision loss following pituitary adenoma surgery. Graft placement and opening of the diaphragma sellae may be important risk factors in vision loss following ETS and should be an area of future investigation.

BACKGROUND AND PURPOSE/OBJECTIVE:Quantitative volumetric segmentation of gliomas has important implications for diagnosis, treatment, and prognosis. We present a deep-learning model that accommodates automated preoperative and postoperative glioma segmentation with a pipeline for clinical implementation. Developed and engineered in concert, the work seeks to accelerate clinical realization of such tools. MATERIALS AND METHODS/METHODS:A deep learning model, autoencoder regularization-cascaded anisotropic, was developed, trained, and tested fusing key elements of autoencoder regularization with a cascaded anisotropic convolutional neural network. We constructed a dataset consisting of 437 cases with 40 cases reserved as a held-out test and the remainder split 80:20 for training and validation. We performed data augmentation and hyperparameter optimization and used a mean Dice score to evaluate against baseline models. To facilitate clinical adoption, we developed the model with an end-to-end pipeline including routing, preprocessing, and end-user interaction. RESULTS:The autoencoder regularization-cascaded anisotropic model achieved median and mean Dice scores of 0.88/0.83 (SD, 0.09), 0.89/0.84 (SD, 0.08), and 0.81/0.72 (SD, 0.1) for whole-tumor, tumor core/resection cavity, and enhancing tumor subregions, respectively, including both preoperative and postoperative follow-up cases. The overall total processing time per case was ∼10 minutes, including data routing (∼1 minute), preprocessing (∼6 minute), segmentation (∼1-2 minute), and postprocessing (∼1 minute). Implementation challenges were discussed. CONCLUSIONS:We show the feasibility and advantages of building a coordinated model with a clinical pipeline for the rapid and accurate deep learning segmentation of both preoperative and postoperative gliomas. The ability of the model to accommodate cases of postoperative glioma is clinically important for follow-up. An end-to-end approach, such as used here, may lead us toward successful clinical translation of tools for quantitative volume measures for glioma.

The fundamental challenge in machine learning is ensuring that trained models generalize well to unseen data. We developed a general technique for ameliorating the effect of dataset shift using generative adversarial networks (GANs) on a dataset of 149,298 handwritten digits and dataset of 868,549 chest radiographs obtained from four academic medical centers. Efficacy was assessed by comparing area under the curve (AUC) pre- and post-adaptation. On the digit recognition task, the baseline CNN achieved an average internal test AUC of 99.87% (95% CI, 99.87-99.87%), which decreased to an average external test AUC of 91.85% (95% CI, 91.82-91.88%), with an average salvage of 35% from baseline upon adaptation. On the lung pathology classification task, the baseline CNN achieved an average internal test AUC of 78.07% (95% CI, 77.97-78.17%) and an average external test AUC of 71.43% (95% CI, 71.32-71.60%), with a salvage of 25% from baseline upon adaptation. Adversarial domain adaptation leads to improved model performance on radiographic data derived from multiple out-of-sample healthcare populations. This work can be applied to other medical imaging domains to help shape the deployment toolkit of machine learning in medicine.

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.

Real-time seizure detection is a resource intensive process as it requires continuous monitoring of patients on stereoelectroencephalography. This study improves real-time seizure detection in drug resistant epilepsy (DRE) patients by developing patient-specific deep learning models that utilize a novel self-supervised dynamic thresholding approach. Deep neural networks were constructed on over 2000 h of high-resolution, multichannel SEEG and video recordings from 14 DRE patients. Consensus labels from a panel of epileptologists were used to evaluate model efficacy. Self-supervised dynamic thresholding exhibited improvements in positive predictive value (PPV; difference: 39.0%; 95% CI 4.5-73.5%; Wilcoxon-Mann-Whitney test; N = 14; p = 0.03) with similar sensitivity (difference: 14.3%; 95% CI - 21.7 to 50.3%; Wilcoxon-Mann-Whitney test; N = 14; p = 0.42) compared to static thresholds. In some models, training on as little as 10 min of SEEG data yielded robust detection. Cross-testing experiments reduced PPV (difference: 56.5%; 95% CI 25.8-87.3%; Wilcoxon-Mann-Whitney test; N = 14; p = 0.002), while multimodal detection significantly improved sensitivity (difference: 25.0%; 95% CI 0.2-49.9%; Wilcoxon-Mann-Whitney test; N = 14; p < 0.05). Self-supervised dynamic thresholding improved the efficacy of real-time seizure predictions. Multimodal models demonstrated potential to improve detection. These findings are promising for future deployment in epilepsy monitoring units to enable real-time seizure detection without annotated data and only minimal training time in individual patients.

Purpose/UNASSIGNED:To train a deep learning classification algorithm to predict chest radiograph severity scores and clinical outcomes in patients with coronavirus disease 2019 (COVID-19). Materials and Methods/UNASSIGNED:= 110) populations. Bootstrapping was used to compute CIs. Results/UNASSIGNED:The model trained on the chest radiograph severity score produced the following areas under the receiver operating characteristic curves (AUCs): 0.80 (95% CI: 0.73, 0.88) for the chest radiograph severity score, 0.76 (95% CI: 0.68, 0.84) for admission, 0.66 (95% CI: 0.56, 0.75) for intubation, and 0.59 (95% CI: 0.49, 0.69) for death. The model trained on clinical variables produced an AUC of 0.64 (95% CI: 0.55, 0.73) for intubation and an AUC of 0.59 (95% CI: 0.50, 0.68) for death. Combining chest radiography and clinical variables increased the AUC of intubation and death to 0.88 (95% CI: 0.79, 0.96) and 0.82 (95% CI: 0.72, 0.91), respectively. Conclusion/UNASSIGNED:© RSNA, 2020.