Faculty Bibliography

Dynamic survival analysis is a variant of traditional survival analysis where time-to-event predictions are updated as new information arrives about an individual over time. In this paper we propose a new approach to dynamic survival analysis based on learning a global parametric distribution, followed by individualization via truncating and renormalizing that distribution at different locations over time. We combine this approach with a likelihood-based loss that includes predictions at every time step within an individual's history, rather than just including one term per individual. The combination of this loss and model results in an interpretable approach to dynamic survival, requiring less fine tuning than existing methods, while still achieving good predictive performance. We evaluate the approach on the problem of predicting hospital mortality for a dataset with over 6900 COVID-19 patients.

BACKGROUND:Comorbid chronic conditions are common among people with type 2 diabetes. We developed an artificial intelligence algorithm, based on reinforcement learning (RL), for personalized diabetes and multimorbidity management, with strong potential to improve health outcomes relative to current clinical practice. METHODS:We modeled glycemia, blood pressure, and cardiovascular disease (CVD) risk as health outcomes, using a retrospective cohort of 16,665 patients with type 2 diabetes from New York University Langone Health ambulatory care electronic health records in 2009-2017. We trained an RL prescription algorithm that recommends a treatment regimen optimizing patients' cumulative health outcomes using their individual characteristics and medical history at each encounter. The RL recommendations were evaluated on an independent subset of patients. RESULTS:The single-outcome optimization RL algorithms, RL-glycemia, RL-blood pressure, and RL-CVD, recommended consistent prescriptions as that observed by clinicians in 86.1%, 82.9%, and 98.4% of the encounters, respectively. For patient encounters in which the RL recommendations differed from the clinician prescriptions, significantly fewer encounters showed uncontrolled glycemia (A1c > 8% in 35% of encounters), uncontrolled hypertension (blood pressure > 140 mmHg in 16% of encounters), and high CVD risk (risk > 20% in 25% of encounters) under RL algorithms compared with those observed under clinicians (43%, 27%, and 31% of encounters, respectively; all p < 0.001). CONCLUSIONS:A personalized RL prescriptive framework for type 2 diabetes yielded high concordance with clinicians' prescriptions, and substantial improvements in glycemia, blood pressure, and CVD risk outcomes.

Tumor-infiltrating lymphocytes (TIL) have potential prognostic value in melanoma and have been considered for inclusion in the American Joint Committee on Cancer (AJCC) staging criteria. However, interobserver discordance continues to prevent the adoption of TIL into clinical practice. Computational image analysis offers a solution to this obstacle, representing a methodological approach for reproducibly counting TIL. We sought to evaluate the ability of a TIL-quantifying machine learning algorithm to predict survival in primary melanoma. Digitized hematoxylin and eosin (H&E) slides from prospectively enrolled patients in the NYU melanoma database were scored for % TIL using machine learning and manually graded by pathologists using Clark's model. We evaluated the association of % TIL with recurrence-free survival (RFS) and overall survival (OS) using Cox proportional hazards modeling and concordance indices. Discordance between algorithmic and manual TIL quantification was assessed with McNemar's test and visually by an attending dermatopathologist. In total, 453 primary melanoma patients were scored using machine learning. Automated % TIL scoring significantly differentiated survival using an estimated cutoff of 16.6% TIL (log-rank P < 0.001 for RFS; P = 0.002 for OS). % TIL was associated with significantly longer RFS (adjusted HR = 0.92 [0.84-1.00] per 10% increase in % TIL) and OS (adjusted HR = 0.90 [0.83-0.99] per 10% increase in % TIL). In comparison, a subset of the cohort (n = 240) was graded for TIL by melanoma pathologists. However, TIL did not associate with RFS between groups (P > 0.05) when categorized as brisk, nonbrisk, or absent. A standardized and automated % TIL scoring algorithm can improve the prognostic impact of TIL. Incorporation of quantitative TIL scoring into the AJCC staging criteria should be considered.

Background: Myocardial infarction with non-obstructive coronary arteries (MINOCA) occurs in 6-15% of MI and disproportionately affects women. Scientific statements recommend multi-modality imaging in MINOCA to define the underlying cause. We performed coronary optical coherence tomography (OCT) and cardiac magnetic resonance imaging (CMR) to assess mechanisms of MINOCA. Methods: In this prospective, multicenter, international, observational study, we enrolled women with a clinical diagnosis of MI. If invasive coronary angiography revealed <50% stenosis in all major arteries, multi-vessel OCT was performed, followed by CMR (cine imaging, late gadolinium enhancement, and T2-weighted imaging and/or T1 mapping). Angiography, OCT, and CMR were evaluated at blinded, independent core laboratories. Culprit lesions identified by OCT were classified as definite or possible. The CMR core laboratory identified ischemia-related and non-ischemic myocardial injury. Imaging results were combined to determine the mechanism of MINOCA, when possible. Results: Among 301 women enrolled at 16 sites, 170 were diagnosed with MINOCA, of whom 145 had adequate OCT image quality for analysis; 116 of these underwent CMR. A definite or possible culprit lesion was identified by OCT in 46.2% (67/145) of participants, most commonly plaque rupture, intra-plaque cavity or layered plaque. CMR was abnormal in 74.1% (86/116) of participants. An ischemic pattern of CMR abnormalities (infarction or myocardial edema in a coronary territory) was present in 53.4% of participants undergoing CMR (62/116). A non-ischemic pattern of CMR abnormalities (myocarditis, takotsubo syndrome or non-ischemic cardiomyopathy) was present in 20.7% (24/116). A cause of MINOCA was identified in 84.5% of the women with multi-modality imaging (98/116), higher than with OCT alone (p<0.001) or CMR alone (p=0.001). An ischemic etiology was identified in 63.8% of women with MINOCA (74/116), a non-ischemic etiology was identified in 20.7% (24/116), and no mechanism was identified in 15.5% (18/116). Conclusions: Multi-modality imaging with coronary OCT and CMR identified potential mechanisms in 84.5% of women with a diagnosis of MINOCA, three-quarters of which were ischemic and one-quarter of which were non-ischemic, alternate diagnoses to MI. Identification of the etiology of MINOCA is feasible and has the potential to guide medical therapy for secondary prevention. Clinical Trial Registration: URL: https://clinicaltrials.gov Unique Identifier: NCT02905357.

BACKGROUND:Recent preclinical data suggest that there may be therapeutic synergy between immune checkpoint blockade and inhibition of the coagulation cascade. Here, we investigate whether patients who received immune checkpoint inhibitors (ICI) and were on concomitant anticoagulation (AC) experienced better treatment outcomes than individuals not on AC.Affiliation: Kindly confirm if corresponding authors affiliation is identified correctly.The corresponding author's affiliation is correct. METHODS:We studied a cohort of 728 advanced cancer patients who received 948 lines of ICI at NYU (2010-2020). Patients were classified based on whether they did (n = 120) or did not (n = 828) receive therapeutic AC at any point during their treatment with ICI. We investigated the relationship between AC status and multiple clinical endpoints including best overall response (BOR), objective response rate (ORR), disease control rate (DCR), progression free survival (PFS), overall survival (OS), and the incidence of bleeding complications.Affiliations: Journal instruction requires a country for affiliations; however, this is missing in affiliations 1 to 5. Please verify if the provided country is correct and amend if necessary.The country is correct for all affiliations (1 - 5). RESULTS:Treatment with AC was not associated with significantly different BOR (P = 0.80), ORR (P =0.60), DCR (P =0.77), PFS (P = 0.59), or OS (P =0.64). Patients who received AC were significantly more likely to suffer a major or clinically relevant minor bleed (P = 0.05). CONCLUSION/CONCLUSIONS:AC does not appear to impact the activity or efficacy of ICI in advanced cancer patients. On the basis of our findings, we caution that there is insufficient evidence to support prospectively evaluating the combination of AC and immunotherapy.

PURPOSE/OBJECTIVE:Several biomarkers of response to immune checkpoint inhibitors (ICI) show potential but are not yet scalable to the clinic. We developed a pipeline that integrates deep learning on histology specimens with clinical data to predict ICI response in advanced melanoma. EXPERIMENTAL DESIGN/METHODS:We used a training cohort from New York University (New York, NY) and a validation cohort from Vanderbilt University (Nashville, TN). We built a multivariable classifier that integrates neural network predictions with clinical data. A ROC curve was generated and the optimal threshold was used to stratify patients as high versus low risk for progression. Kaplan-Meier curves compared progression-free survival (PFS) between the groups. The classifier was validated on two slide scanners (Aperio AT2 and Leica SCN400). RESULTS:= 0.03 for the Leica SCN400). CONCLUSIONS:Histology slides and patients' clinicodemographic characteristics are readily available through standard of care and have the potential to predict ICI treatment outcomes. With prospective validation, we believe our approach has potential for integration into clinical practice.

The associations between covariates and the outcomes often vary over time, regardless of whether the covariate is time-varying or time-invariant. For example, we hypothesize that the impact of chronic diseases, such as diabetes and heart disease, on people's physical functions differ with aging. However, the age-varying effect would be missed if one models the covariate simply as a time-invariant covariate (yes/no) with a time-constant coefficient. We propose a fused lasso-based time-varying linear mixed effect (FTLME) model and an efficient two-stage parameter estimation algorithm to estimate the longitudinal trajectories of fixed-effect coefficients. Simulation studies are presented to demonstrate the efficacy of the method and its computational efficiency in estimating smooth time-varying effects in high dimensional settings. A real data example on the Health and Retirement Study (HRS) analysis is used to demonstrate the practical usage of our method to infer age-varying impact of chronic disease on older people's physical functions.