Faculty Bibliography

With 15 drugs currently approved for the treatment of metastatic renal cell carcinoma (mRCC) and even more combination regimens with immunotherapy on the horizon, there remains a distinct lack of molecular biomarkers for therapeutic efficacy. Our study reports on real-world clinical outcomes of mRCC patients from a tertiary academic medical center treated with empirically selected standard-of-care therapy. We utilized the Stanford Renal Cell Carcinoma Database (RCCD) to report on various outcome measures, including overall survival (OS) and the median number of lines of targeted therapies received from the time of metastatic diagnosis. We found that most metastatic patients did not survive long enough to attempt even half of the available targeted therapies. We also noted that patients who failed to receive a clinical benefit within the first two lines of therapy could still go on to experience clinical benefit in later lines of therapy. The term, "clinical benefit" was assigned to a line of therapy if a patient remained on drug treatment for three months or longer. Moreover, patients with clinical benefit in at least one line of therapy experienced significantly longer OS compared to those who did not have clinical benefit in at least one line of therapy. Developing biomarkers that identify patients who will receive clinical benefit in individual lines of therapy is one potential strategy for achieving rational drug sequencing in mRCC.

Background/UNASSIGNED:Differentiating glioblastoma, brain metastasis, and central nervous system lymphoma (CNSL) on conventional magnetic resonance imaging (MRI) can present a diagnostic dilemma due to the potential for overlapping imaging features. We investigate whether machine learning evaluation of multimodal MRI can reliably differentiate these entities. Methods/UNASSIGNED:Preoperative brain MRI including diffusion weighted imaging (DWI), dynamic contrast enhanced (DCE), and dynamic susceptibility contrast (DSC) perfusion in patients with glioblastoma, lymphoma, or metastasis were retrospectively reviewed. Perfusion maps (rCBV, rCBF), permeability maps (K-trans, Kep, Vp, Ve), ADC, T1C+ and T2/FLAIR images were coregistered and two separate volumes of interest (VOIs) were obtained from the enhancing tumor and non-enhancing T2 hyperintense (NET2) regions. The tumor volumes obtained from these VOIs were utilized for supervised training of support vector classifier (SVC) and multilayer perceptron (MLP) models. Validation of the trained models was performed on unlabeled cases using the leave-one-subject-out method. Head-to-head and multiclass models were created. Accuracies of the multiclass models were compared against two human interpreters reviewing conventional and diffusion-weighted MR images. Results/UNASSIGNED:Twenty-six patients enrolled with histopathologically-proven glioblastoma (n=9), metastasis (n=9), and CNS lymphoma (n=8) were included. The trained multiclass ML models discriminated the three pathologic classes with a maximum accuracy of 69.2% accuracy (18 out of 26; kappa 0.540, P=0.01) using an MLP trained with the VpNET2 tumor volumes. Human readers achieved 65.4% (17 out of 26) and 80.8% (21 out of 26) accuracies, respectively. Using the MLP VpNET2 model as a computer-aided diagnosis (CADx) for cases in which the human reviewers disagreed with each other on the diagnosis resulted in correct diagnoses in 5 (19.2%) additional cases. Conclusions/UNASSIGNED:Our trained multiclass MLP using VpNET2 can differentiate glioblastoma, brain metastasis, and CNS lymphoma with modest diagnostic accuracy and provides approximately 19% increase in diagnostic yield when added to routine human interpretation.

Background/UNASSIGNED:Errors in grammar, spelling, and usage in radiology reports are common. To automatically detect inappropriate insertions, deletions, and substitutions of words in radiology reports, we proposed using a neural sequence-to-sequence (seq2seq) model. Methods/UNASSIGNED:Head CT and chest radiograph reports from Mount Sinai Hospital (MSH) (n=61,722 and 818,978, respectively), Mount Sinai Queens (MSQ) (n=30,145 and 194,309, respectively) and MIMIC-III (n=32,259 and 54,685) were converted into sentences. Insertions, substitutions, and deletions of words were randomly introduced. Seq2seq models were trained using corrupted sentences as input to predict original uncorrupted sentences. Three models were trained using head CTs from MSH, chest radiographs from MSH, and head CTs from all three collections. Model performance was assessed across different sites and modalities. A sample of original, uncorrupted sentences were manually reviewed for any error in syntax, usage, or spelling to estimate real-world proofreading performance of the algorithm. Results/UNASSIGNED:Seq2seq detected 90.3% and 88.2% of corrupted sentences with 97.7% and 98.8% specificity in same-site, same-modality test sets for head CTs and chest radiographs, respectively. Manual review of original, uncorrupted same-site same-modality head CT sentences demonstrated seq2seq positive predictive value (PPV) 0.393 (157/400; 95% CI, 0.346-0.441) and negative predictive value (NPV) 0.986 (789/800; 95% CI, 0.976-0.992) for detecting sentences containing real-world errors, with estimated sensitivity of 0.389 (95% CI, 0.267-0.542) and specificity 0.986 (95% CI, 0.985-0.987) over n=86,211 uncorrupted training examples. Conclusions/UNASSIGNED:Seq2seq models can be highly effective at detecting erroneous insertions, deletions, and substitutions of words in radiology reports. To achieve high performance, these models require site- and modality-specific training examples. Incorporating additional targeted training data could further improve performance in detecting real-world errors in reports.

MOTIVATION:Radiologists have used algorithms for Computer-Aided Diagnosis (CAD) for decades. These algorithms use machine learning with engineered features, and there have been mixed findings on whether they improve radiologists' interpretations. Deep learning offers superior performance but requires more training data and has not been evaluated in joint algorithm-radiologist decision systems. RESULTS:We developed the Computer-Aided Note and Diagnosis Interface (CANDI) for collaboratively annotating radiographs and evaluating how algorithms alter human interpretation. The annotation app collects classification, segmentation, and image captioning training data, and the evaluation app randomizes the availability of CAD tools to facilitate clinical trials on radiologist enhancement. AVAILABILITY AND IMPLEMENTATION:Demonstrations and source code are hosted at (https://candi.nextgenhealthcare.org), and (https://github.com/mbadge/candi), respectively, under GPL-3 license. SUPPLEMENTARY INFORMATION:Supplementary material is available at Bioinformatics online.

STUDY DESIGN/METHODS:A retrospective analysis of prospectively collected data. OBJECTIVE:The aim of this study was to determine the ability of Revised Cardiac Risk Index (RCRI) to predict adverse cardiac events following posterior lumbar decompression (PLD). SUMMARY OF BACKGROUND DATA/BACKGROUND:PLD is an increasingly common procedure used to treat a variety of degenerative spinal conditions. The RCRI is used to predict risk for cardiac events following noncardiac surgery. There is a paucity of literature that directly addresses the relationship between RCRI and outcomes following PLD, specifically, the discriminative ability of the RCRI to predict adverse postoperative cardiac events. METHODS:ACS-NSQIP was utilized to identify patients undergoing PLD from 2006 to 2014. Fifty-two thousand sixty-six patients met inclusion criteria. Multivariate and ROC analysis was utilized to identify associations between RCRI and postoperative complications. RESULTS:Membership in the RCRI=1 cohort was a predictor for myocardial infarction (MI) [odds ratio (OR) = 3.3, P = 0.002] and cardiac arrest requiring cardiopulmonary resuscitation (CPR) (OR = 3.4, P = 0.013). Membership in the RCRI = 2 cohort was a predictor for MI (OR = 5.9, P = 0.001) and cardiac arrest requiring CPR (OR = 12.5), Membership in the RCRI = 3 cohort was a predictor for MI (OR = 24.9) and cardiac arrest requiring CPR (OR = 26.9, P = 0.006). RCRI had a good discriminative ability to predict both MI [area under the curve (AUC) = 0.876] and cardiac arrest requiring CPR (AUC = 0.855). The RCRI had a better discriminative ability to predict these outcomes that did ASA status, which had discriminative abilities of "fair" (AUC = 0.799) and "poor" (AUC = 0.674), respectively. P < 0.001 unless otherwise specified. CONCLUSION/CONCLUSIONS:RCRI was predictive of cardiac events following PLD, and RCRI had a better discriminative ability to predict MI and cardiac arrest requiring CPR than did ASA status. Consideration of the RCRI as a component of preoperative surgical risk stratification can minimize patient morbidity and mortality. Studies such as this can allow for implementation of guidelines that better estimate the preoperative risk profile of surgical patients. LEVEL OF EVIDENCE/METHODS:3.

PURPOSE/OBJECTIVE:To determine if weakly supervised learning with surrogate metrics and active transfer learning can hasten clinical deployment of deep learning models. MATERIALS AND METHODS/METHODS:test, Bland-Altman analysis, and intraclass correlation; survival analysis was performed with the Kaplan-Meier method and a Mantel-Cox test. RESULTS:= .0005). In overall survival analysis, predicted liver volumes using the best active learning-trained model (LiTS-OU) were at least comparable with liver volumes extracted from radiology reports and MELD-Na scores in predicting survival. CONCLUSION/CONCLUSIONS:

This study trained long short-term memory (LSTM) recurrent neural networks (RNNs) incorporating an attention mechanism to predict daily sepsis, myocardial infarction (MI), and vancomycin antibiotic administration over two week patient ICU courses in the MIMIC-III dataset. These models achieved next-day predictive AUC of 0.876 for sepsis, 0.823 for MI, and 0.833 for vancomycin administration. Attention maps built from these models highlighted those times when input variables most influenced predictions and could provide a degree of interpretability to clinicians. These models appeared to attend to variables that were proxies for clinician decision-making, demonstrating a challenge of using flexible deep learning approaches trained with EHR data to build clinical decision support. While continued development and refinement is needed, we believe that such models could one day prove useful in reducing information overload for ICU physicians by providing needed clinical decision support for a variety of clinically important tasks.

BACKGROUND:The Revised Cardiac Risk Index (RCRI) was designed to predict risk for cardiac events after noncardiac surgery. However, there is a paucity of literature that directly addresses the relationship between RCRI and noncardiac outcomes after posterior lumbar decompression (PLD). The objective of this study is to determine the ability of RCRI to predict noncardiac adverse events after PLD. METHODS:The American College of Surgeons National Surgical Quality Improvement Program was used to identify patients undergoing PLD from 2006 to 2014. Multivariate and receiver operating characteristic analysis was used to identify associations between RCRI and postoperative complications. RESULTS:A total of 52,066 patients met the inclusion criteria. Membership in the RCRI=1 cohort independently predicted unplanned intubation, ventilation >48 hours, progressive renal insufficiency, acute renal failure, urinary tract infection (UTI), sepsis, septic shock, and readmission. Membership in the RCRI=2 cohort independently predicted for superficial surgical site infection, pneumonia, unplanned intubation, ventilation >48 hours, bleeding transfusion, progressive renal insufficiency, acute renal failure, UTI, sepsis, septic shock, and readmission. Membership in the RCRI=3 cohort independently predicted unplanned intubation (odds ratio [OR], 11.8), ventilation >48 hours (OR, 23.0), acute renal failure (OR, 84.5), and UTI (OR, 3.6). RCRI had a poor discriminative ability (DA) (area under the curve = 0.623), and American Society of Anesthesiologists status had a fair DA (area under the curve = 0.770) to predict a composite of noncardiac complications. CONCLUSIONS:RCRI was predictive of a wide range of noncardiac complications after PLD but had a diminished DA to predict a composite of any noncardiac complication than did American Society of Anesthesiologists score. Consideration of the RCRI as a component of preoperative surgical risk stratification can minimize patient morbidity and mortality after lumbar decompression.

OBJECTIVE:Machine learning algorithms excel at leveraging big data to identify complex patterns that can be used to aid in clinical decision-making. The objective of this study is to demonstrate the performance of machine learning models in predicting postoperative complications following anterior cervical discectomy and fusion (ACDF). METHODS:Artificial neural network (ANN), logistic regression (LR), support vector machine (SVM), and random forest decision tree (RF) models were trained on a multicenter data set of patients undergoing ACDF to predict surgical complications based on readily available patient data. Following training, these models were compared to the predictive capability of American Society of Anesthesiologists (ASA) physical status classification. RESULTS:A total of 20,879 patients were identified as having undergone ACDF. Following exclusion criteria, patients were divided into 14,615 patients for training and 6,264 for testing data sets. ANN and LR consistently outperformed ASA physical status classification in predicting every complication (p < 0.05). The ANN outperformed LR in predicting venous thromboembolism, wound complication, and mortality (p < 0.05). The SVM and RF models were no better than random chance at predicting any of the postoperative complications (p < 0.05). CONCLUSION/CONCLUSIONS:ANN and LR algorithms outperform ASA physical status classification for predicting individual postoperative complications. Additionally, neural networks have greater sensitivity than LR when predicting mortality and wound complications. With the growing size of medical data, the training of machine learning on these large datasets promises to improve risk prognostication, with the ability of continuously learning making them excellent tools in complex clinical scenarios.