Anna Goldenberg

Caitlin F. Harrigan, Gabriela Morgenshtern, Anna Goldenberg et al.

Clinician-facing predictive models are increasingly present in the healthcare setting. Regardless of their success with respect to performance metrics, all models have uncertainty. We investigate how to visually communicate uncertainty in this setting in an actionable, trustworthy way. To this end, we conduct a qualitative study with cardiac critical care clinicians. Our results reveal that clinician trust may be impacted most not by the degree of uncertainty, but rather by how transparent the visualization of what the sources of uncertainty are. Our results show a clear connection between feature interpretability and clinical actionability.

Kopal Garg, Jennifer Yu, Tina Behrouzi et al.

Identifying change points (CPs) in a time series is crucial to guide better decision making across various fields like finance and healthcare and facilitating timely responses to potential risks or opportunities. Existing Change Point Detection (CPD) methods have a limitation in tracking changes in the joint distribution of multidimensional features. In addition, they fail to generalize effectively within the same time series as different types of CPs may require different detection methods. As the volume of multidimensional time series continues to grow, capturing various types of complex CPs such as changes in the correlation structure of the time-series features has become essential. To overcome the limitations of existing methods, we propose TiVaCPD, an approach that uses a Time-Varying Graphical Lasso (TVGL) to identify changes in correlation patterns between multidimensional features over time, and combines that with an aggregate Kernel Maximum Mean Discrepancy (MMD) test to identify changes in the underlying statistical distributions of dynamic time windows with varying length. The MMD and TVGL scores are combined using a novel ensemble method based on similarity measures leveraging the power of both statistical tests. We evaluate the performance of TiVaCPD in identifying and characterizing various types of CPs and show that our method outperforms current state-of-the-art methods in real-world CPD datasets. We further demonstrate that TiVaCPD scores characterize the type of CPs and facilitate interpretation of change dynamics, offering insights into real-life applications.

Stanley Bryan Z. Hua, Mandy Rickard, John Weaver et al.

Previous work has shown the potential of deep learning to predict renal obstruction using kidney ultrasound images. However, these image-based classifiers have been trained with the goal of single-visit inference in mind. We compare methods from video action recognition (i.e. convolutional pooling, LSTM, TSM) to adapt single-visit convolutional models to handle multiple visit inference. We demonstrate that incorporating images from a patient's past hospital visits provides only a small benefit for the prediction of obstructive hydronephrosis. Therefore, inclusion of prior ultrasounds is beneficial, but prediction based on the latest ultrasound is sufficient for patient risk stratification.

Sujay Nagaraj, Andrew J. Goodwin, Dmytro Lopushanskyy et al.

Central Venous Lines (C-Lines) and Arterial Lines (A-Lines) are routinely used in the Critical Care Unit (CCU) for blood sampling, medication administration, and high-frequency blood pressure measurement. Judiciously accessing these lines is important, as over-utilization is associated with significant in-hospital morbidity and mortality. Documenting the frequency of line-access is an important step in reducing these adverse outcomes. Unfortunately, the current gold-standard for documentation is manual and subject to error, omission, and bias. The high-frequency blood pressure waveform data from sensors in these lines are often noisy and full of artifacts. Standard approaches in signal processing remove noise artifacts before meaningful analysis. However, from bedside observations, we characterized a distinct artifact that occurs during each instance of C-Line or A-Line use. These artifacts are buried amongst physiological waveform and extraneous noise. We focus on Machine Learning (ML) models that can detect these artifacts from waveform data in real-time - finding needles in needle stacks, in order to automate the documentation of line-access. We built and evaluated ML classifiers running in real-time at a major children's hospital to achieve this goal. We demonstrate the utility of these tools for reducing documentation burden, increasing available information for bedside clinicians, and informing unit-level initiatives to improve patient safety.

Abhishek Moturu, Anna Goldenberg, Babak Taati

Synthetic data is useful only when the added samples fill missing parts of the training distribution that matter for the downstream task. We introduce LiBaGS, a lightweight, generator-agnostic method for targeted synthetic training data selection. LiBaGS scores candidate synthetic samples by combining decision-boundary proximity, predictive uncertainty, real-data density, and support validity, so that selected samples are both informative and likely to remain on the real data manifold. We then use a boundary-gap allocation rule that targets sparse but realistic decision-boundary neighborhoods, rather than simply adding more data or selecting only the most uncertain candidates. LiBaGS also learns when enough synthetic samples have been added through a marginal-value stopping rule, assigns softer labels near ambiguous boundaries, and uses a diversity objective to avoid redundant near-duplicate selections. Experiments show that LiBaGS improves accuracy over classical oversampling, hard augmentation, uncertainty and density ablations, and targeted-generation selection criteria.

Zongliang Ji, Yifei Sun, Andre Amaral et al.

Deploying clinical ML is slow and brittle: models that work at one hospital often degrade under distribution shifts at the next. In this work, we study a simple question -- can large language models (LLMs) create portable patient embeddings i.e. representations of patients enable a downstream predictor built on one hospital to be used elsewhere with minimal-to-no retraining and fine-tuning. To do so, we map from irregular ICU time series onto concise natural language summaries using a frozen LLM, then embed each summary with a frozen text embedding model to obtain a fixed length vector capable of serving as input to a variety of downstream predictors. Across three cohorts (MIMIC-IV, HIRID, PPICU), on multiple clinically grounded forecasting and classification tasks, we find that our approach is simple, easy to use and competitive with in-distribution with grid imputation, self-supervised representation learning, and time series foundation models, while exhibiting smaller relative performance drops when transferring to new hospitals. We study the variation in performance across prompt design, with structured prompts being crucial to reducing the variance of the predictive models without altering mean accuracy. We find that using these portable representations improves few-shot learning and does not increase demographic recoverability of age or sex relative to baselines, suggesting little additional privacy risk. Our work points to the potential that LLMs hold as tools to enable the scalable deployment of production grade predictive models by reducing the engineering overhead.

Zongliang Ji, Ziyang Zhang, Xincheng Tan et al.

Evidence-based medicine (EBM) is central to high-quality care, but remains difficult to implement in fast-paced primary care settings. Physicians face short consultations, increasing patient loads, and lengthy guideline documents that are impractical to consult in real time. To address this gap, we investigate the feasibility of using large language models (LLMs) as ambient assistants that surface targeted, evidence-based questions during physician-patient encounters. Our study focuses on question generation rather than question answering, with the aim of scaffolding physician reasoning and integrating guideline-based practice into brief consultations. We implemented two prompting strategies, a zero-shot baseline and a multi-stage reasoning variant, using Gemini 2.5 as the backbone model. We evaluated on a benchmark of 80 de-identified transcripts from real clinical encounters, with six experienced physicians contributing over 90 hours of structured review. Results indicate that while general-purpose LLMs are not yet fully reliable, they can produce clinically meaningful and guideline-relevant questions, suggesting significant potential to reduce cognitive burden and make EBM more actionable at the point of care.

Shihao Ma, Andy G. X. Zeng, Benjamin Haibe-Kains et al.

High-throughput omics profiling advancements have greatly enhanced cancer patient stratification. However, incomplete data in multi-omics integration presents a significant challenge, as traditional methods like sample exclusion or imputation often compromise biological diversity and dependencies. Furthermore, the critical task of accurately classifying new patients with partial omics data into existing subtypes is commonly overlooked. To address these issues, we introduce IntegrAO (Integrate Any Omics), an unsupervised framework for integrating incomplete multi-omics data and classifying new samples. IntegrAO first combines partially overlapping patient graphs from diverse omics sources and utilizes graph neural networks to produce unified patient embeddings. Our systematic evaluation across five cancer cohorts involving six omics modalities demonstrates IntegrAO's robustness to missing data and its accuracy in classifying new samples with partial profiles. An acute myeloid leukemia case study further validates its capability to uncover biological and clinical heterogeneity in incomplete datasets. IntegrAO's ability to handle heterogeneous and incomplete data makes it an essential tool for precision oncology, offering a holistic approach to patient characterization.

Zongliang Ji, Anna Goldenberg, Rahul G. Krishnan

Scheduling laboratory tests for ICU patients presents a significant challenge. Studies show that 20-40% of lab tests ordered in the ICU are redundant and could be eliminated without compromising patient safety. Prior work has leveraged offline reinforcement learning (Offline-RL) to find optimal policies for ordering lab tests based on patient information. However, new ICU patient datasets have since been released, and various advancements have been made in Offline-RL methods. In this study, we first introduce a preprocessing pipeline for the newly-released MIMIC-IV dataset geared toward time-series tasks. We then explore the efficacy of state-of-the-art Offline-RL methods in identifying better policies for ICU patient lab test scheduling. Besides assessing methodological performance, we also discuss the overall suitability and practicality of using Offline-RL frameworks for scheduling laboratory tests in ICU settings.

Tina Behrouzi, Sana Tonekaboni, Rahul G. Krishnan et al.

Real-world observational data often contain existing or emerging heterogeneous subpopulations that deviate from global patterns. The majority of models tend to overlook these underrepresented groups, leading to inaccurate or even harmful predictions. Existing solutions often rely on detecting these samples as Out-of-domain (OOD) rather than adapting the model to new emerging patterns. We introduce DynaSubVAE, a Dynamic Subgrouping Variational Autoencoder framework that jointly performs representation learning and adaptive OOD detection. Unlike conventional approaches, DynaSubVAE evolves with the data by dynamically updating its latent structure to capture new trends. It leverages a novel non-parametric clustering mechanism, inspired by Gaussian Mixture Models, to discover and model latent subgroups based on embedding similarity. Extensive experiments show that DynaSubVAE achieves competitive performance in both near-OOD and far-OOD detection, and excels in class-OOD scenarios where an entire class is missing during training. We further illustrate that our dynamic subgrouping mechanism outperforms standalone clustering methods such as GMM and KMeans++ in terms of both OOD accuracy and regret precision.

Abhishek Moturu, Anna Goldenberg, Babak Taati

Training deep neural networks in the presence of noisy labels and data heterogeneity is a major challenge. We introduce Lightweight Learnable Adaptive Weighting (LiLAW), a novel method that dynamically adjusts the loss weight of each training sample based on its evolving difficulty level, categorized as easy, moderate, or hard. Using only three learnable parameters, LiLAW adaptively prioritizes informative samples throughout training by updating these weights using a single mini-batch gradient descent step on the validation set after each training mini-batch, without requiring excessive hyperparameter tuning or a clean validation set. Extensive experiments across multiple general and medical imaging datasets, noise levels and types, loss functions, and architectures with and without pretraining demonstrate that LiLAW consistently enhances performance, even in high-noise environments. It is effective without heavy reliance on data augmentation or advanced regularization, highlighting its practicality. It offers a computationally efficient solution to boost model generalization and robustness in any neural network training setup.

Addison Weatherhead, Anna Goldenberg

Time series data with missing values is common across many domains. Healthcare presents special challenges due to prolonged periods of sensor disconnection. In such cases, having a confidence measure for imputed values is critical. Most existing methods either overlook model uncertainty or lack mechanisms to estimate it. To address this gap, we introduce a general framework that quantifies and leverages uncertainty for selective imputation. By focusing on values the model is most confident in, highly unreliable imputations are avoided. Our experiments on multiple EHR datasets, covering diverse types of missingness, demonstrate that selectively imputing less-uncertain values not only reduces imputation errors but also improves downstream tasks. Specifically, we show performance gains in a 24-hour mortality prediction task, underscoring the practical benefit of incorporating uncertainty into time series imputation.

Zongliang Ji, Andre Carlos Kajdacsy-Balla Amaral, Anna Goldenberg et al.

Ordering a minimal subset of lab tests for patients in the intensive care unit (ICU) can be challenging. Care teams must balance between ensuring the availability of the right information and reducing the clinical burden and costs associated with each lab test order. Most in-patient settings experience frequent over-ordering of lab tests, but are now aiming to reduce this burden on both hospital resources and the environment. This paper develops a novel method that combines off-policy learning with privileged information to identify the optimal set of ICU lab tests to order. Our approach, EXplainable Off-policy learning with Side Information for ICU blood Test Orders (ExOSITO) creates an interpretable assistive tool for clinicians to order lab tests by considering both the observed and predicted future status of each patient. We pose this problem as a causal bandit trained using offline data and a reward function derived from clinically-approved rules; we introduce a novel learning framework that integrates clinical knowledge with observational data to bridge the gap between the optimal and logging policies. The learned policy function provides interpretable clinical information and reduces costs without omitting any vital lab orders, outperforming both a physician's policy and prior approaches to this practical problem.

Sujay Nagaraj, Walter Gerych, Sana Tonekaboni et al.

Many time series classification tasks, where labels vary over time, are affected by label noise that also varies over time. Such noise can cause label quality to improve, worsen, or periodically change over time. We first propose and formalize temporal label noise, an unstudied problem for sequential classification of time series. In this setting, multiple labels are recorded over time while being corrupted by a time-dependent noise function. We first demonstrate the importance of modeling the temporal nature of the label noise function and how existing methods will consistently underperform. We then propose methods to train noise-tolerant classifiers by estimating the temporal label noise function directly from data. We show that our methods lead to state-of-the-art performance under diverse types of temporal label noise on real-world datasets

Oleksii Tsepa, Bohdan Naida, Anna Goldenberg et al.

Drug synergy, characterized by the amplified combined effect of multiple drugs, is critically important for optimizing therapeutic outcomes. Limited data on drug synergy, arising from the vast number of possible drug combinations and testing costs, motivate the need for predictive methods. In this work, we introduce CongFu, a novel Conditional Graph Fusion Layer, designed to predict drug synergy. CongFu employs an attention mechanism and a bottleneck to extract local graph contexts and conditionally fuse graph data within a global context. Its modular architecture enables flexible replacement of layer modules, including readouts and graph encoders, facilitating customization for diverse applications. To evaluate the performance of CongFu, we conduct comprehensive experiments on four datasets, encompassing three distinct setups for drug synergy prediction. CongFu achieves state-of-the-art results on 11 out of 12 benchmark datasets, demonstrating its ability to capture intricate patterns of drug synergy. Through ablation studies, we validate the significance of individual layer components, affirming their contributions to overall predictive performance. Finally, we propose an explainability strategy for elucidating the effect of drugs on genes. By addressing the challenge of predicting drug synergy in untested drug pairs and utilizing our proposed explainability approach, CongFu opens new avenues for optimizing drug combinations and advancing personalized medicine.

George Adam, Benjamin Haibe-Kains, Anna Goldenberg

Deployed machine learning models should be updated to take advantage of a larger sample size to improve performance, as more data is gathered over time. Unfortunately, even when model updates improve aggregate metrics such as accuracy, they can lead to errors on samples that were correctly predicted by the previous model causing per-sample regression in performance known as predictive churn. Such prediction flips erode user trust thereby reducing the effectiveness of the human-AI team as a whole. We propose a solution called Accumulated Model Combination (AMC) based keeping the previous and current model version, and generating a meta-output using the prediction of the two models. AMC is a general technique and we propose several instances of it, each having their own advantages depending on the model and data properties. AMC requires minimal additional computation and changes to training procedures. We motivate the need for AMC by showing the difficulty of making a single model consistent with its own predictions throughout training thereby revealing an implicit stability-plasticity tradeoff when training a single model. We demonstrate the effectiveness of AMC on a variety of modalities including computer vision, text, and tabular datasets comparing against state-of-the-art churn reduction methods, and showing superior churn reduction ability compared to all existing methods while being more efficient than ensembles.

Sana Tonekaboni, Chun-Liang Li, Sercan Arik et al.

Real-world time series data are often generated from several sources of variation. Learning representations that capture the factors contributing to this variability enables a better understanding of the data via its underlying generative process and improves performance on downstream machine learning tasks. This paper proposes a novel generative approach for learning representations for the global and local factors of variation in time series. The local representation of each sample models non-stationarity over time with a stochastic process prior, and the global representation of the sample encodes the time-independent characteristics. To encourage decoupling between the representations, we introduce counterfactual regularization that minimizes the mutual information between the two variables. In experiments, we demonstrate successful recovery of the true local and global variability factors on simulated data, and show that representations learned using our method yield superior performance on downstream tasks on real-world datasets. We believe that the proposed way of defining representations is beneficial for data modelling and yields better insights into the complexity of real-world data.

Chun-Hao Chang, George Alexandru Adam, Rich Caruana et al.

Although reinforcement learning (RL) has tremendous success in many fields, applying RL to real-world settings such as healthcare is challenging when the reward is hard to specify and no exploration is allowed. In this work, we focus on recovering clinicians' rewards in treating patients. We incorporate the what-if reasoning to explain the clinician's treatments based on their potential future outcomes. We use generalized additive models (GAMs) - a class of accurate, interpretable models - to recover the reward. In both simulation and a real-world hospital dataset, we show our model outperforms baselines. Finally, our model's explanations match several clinical guidelines when treating patients while we found the commonly-used linear model often contradicts them.

Chun-Hao Chang, Rich Caruana, Anna Goldenberg

Deployment of machine learning models in real high-risk settings (e.g. healthcare) often depends not only on the model's accuracy but also on its fairness, robustness, and interpretability. Generalized Additive Models (GAMs) are a class of interpretable models with a long history of use in these high-risk domains, but they lack desirable features of deep learning such as differentiability and scalability. In this work, we propose a neural GAM (NODE-GAM) and neural GA$^2$M (NODE-GA$^2$M) that scale well and perform better than other GAMs on large datasets, while remaining interpretable compared to other ensemble and deep learning models. We demonstrate that our models find interesting patterns in the data. Lastly, we show that we improve model accuracy via self-supervised pre-training, an improvement that is not possible for non-differentiable GAMs.

Chun-Hao Chang, George Alexandru Adam, Anna Goldenberg

Despite the success of machine learning applications in science, industry, and society in general, many approaches are known to be non-robust, often relying on spurious correlations to make predictions. Spuriousness occurs when some features correlate with labels but are not causal; relying on such features prevents models from generalizing to unseen environments where such correlations break. In this work, we focus on image classification and propose two data generation processes to reduce spuriousness. Given human annotations of the subset of the features responsible (causal) for the labels (e.g. bounding boxes), we modify this causal set to generate a surrogate image that no longer has the same label (i.e. a counterfactual image). We also alter non-causal features to generate images still recognized as the original labels, which helps to learn a model invariant to these features. In several challenging datasets, our data generations outperform state-of-the-art methods in accuracy when spurious correlations break, and increase the saliency focus on causal features providing better explanations.

Sana Tonekaboni, Danny Eytan, Anna Goldenberg

Time series are often complex and rich in information but sparsely labeled and therefore challenging to model. In this paper, we propose a self-supervised framework for learning generalizable representations for non-stationary time series. Our approach, called Temporal Neighborhood Coding (TNC), takes advantage of the local smoothness of a signal's generative process to define neighborhoods in time with stationary properties. Using a debiased contrastive objective, our framework learns time series representations by ensuring that in the encoding space, the distribution of signals from within a neighborhood is distinguishable from the distribution of non-neighboring signals. Our motivation stems from the medical field, where the ability to model the dynamic nature of time series data is especially valuable for identifying, tracking, and predicting the underlying patients' latent states in settings where labeling data is practically impossible. We compare our method to recently developed unsupervised representation learning approaches and demonstrate superior performance on clustering and classification tasks for multiple datasets.

Alex Chang, Vinith Suriyakumar, Abhishek Moturu et al.

Modern deep unsupervised learning methods have shown great promise for detecting diseases across a variety of medical imaging modalities. While previous generative modeling approaches successfully perform anomaly detection by learning the distribution of healthy 2D image slices, they process such slices independently and ignore the fact that they are correlated, all being sampled from a 3D volume. We show that incorporating the 3D context and processing whole-body MRI volumes is beneficial to distinguishing anomalies from their benign counterparts. In our work, we introduce a multi-channel sliding window generative model to perform lesion detection in whole-body MRI (wbMRI). Our experiments demonstrate that our proposed method significantly outperforms processing individual images in isolation and our ablations clearly show the importance of 3D reasoning. Moreover, our work also shows that it is beneficial to include additional patient-specific features to further improve anomaly detection in pediatric scans.

Erik Drysdale, Devin Singh, Anna Goldenberg

Predicting patient volumes in a hospital setting is a well-studied application of time series forecasting. Existing tools usually make forecasts at the daily or weekly level to assist in planning for staffing requirements. Prompted by new COVID-related capacity constraints placed on our pediatric hospital's emergency department, we developed an hourly forecasting tool to make predictions over a 24 hour window. These forecasts would give our hospital sufficient time to be able to martial resources towards expanding capacity and augmenting staff (e.g. transforming wards or bringing in physicians on call). Using Gaussian Process Regressions (GPRs), we obtain strong performance for both point predictions (average R-squared: 82%) as well as classification accuracy when predicting the ordinal tiers of our hospital's capacity (average precision/recall: 82%/74%). Compared to traditional regression approaches, GPRs not only obtain consistently higher performance, but are also robust to the dataset shifts that have occurred throughout 2020. Hospital stakeholders are encouraged by the strength of our results, and we are currently working on moving our tool to a real-time setting with the goal of augmenting the capabilities of our healthcare workers.

Vinith M. Suriyakumar, Nicolas Papernot, Anna Goldenberg et al.

Machine learning models in health care are often deployed in settings where it is important to protect patient privacy. In such settings, methods for differentially private (DP) learning provide a general-purpose approach to learn models with privacy guarantees. Modern methods for DP learning ensure privacy through mechanisms that censor information judged as too unique. The resulting privacy-preserving models, therefore, neglect information from the tails of a data distribution, resulting in a loss of accuracy that can disproportionately affect small groups. In this paper, we study the effects of DP learning in health care. We use state-of-the-art methods for DP learning to train privacy-preserving models in clinical prediction tasks, including x-ray classification of images and mortality prediction in time series data. We use these models to perform a comprehensive empirical investigation of the tradeoffs between privacy, utility, robustness to dataset shift, and fairness. Our results highlight lesser-known limitations of methods for DP learning in health care, models that exhibit steep tradeoffs between privacy and utility, and models whose predictions are disproportionately influenced by large demographic groups in the training data. We discuss the costs and benefits of differentially private learning in health care.

Matthew B. A. McDermott, Bret Nestor, Evan Kim et al.

Multi-task learning (MTL) is a machine learning technique aiming to improve model performance by leveraging information across many tasks. It has been used extensively on various data modalities, including electronic health record (EHR) data. However, despite significant use on EHR data, there has been little systematic investigation of the utility of MTL across the diverse set of possible tasks and training schemes of interest in healthcare. In this work, we examine MTL across a battery of tasks on EHR time-series data. We find that while MTL does suffer from common negative transfer, we can realize significant gains via MTL pre-training combined with single-task fine-tuning. We demonstrate that these gains can be achieved in a task-independent manner and offer not only minor improvements under traditional learning, but also notable gains in a few-shot learning context, thereby suggesting this could be a scalable vehicle to offer improved performance in important healthcare contexts.

Chun-Hao Chang, Sarah Tan, Ben Lengerich et al.

Generalized additive models (GAMs) have become a leading modelclass for interpretable machine learning. However, there are many algorithms for training GAMs, and these can learn different or even contradictory models, while being equally accurate. Which GAM should we trust? In this paper, we quantitatively and qualitatively investigate a variety of GAM algorithms on real and simulated datasets. We find that GAMs with high feature sparsity (only using afew variables to make predictions) can miss patterns in the data and be unfair to rare subpopulations. Our results suggest that inductive bias plays a crucial role in what interpretable models learn and that tree-based GAMs represent the best balance of sparsity, fidelity and accuracy and thus appear to be the most trustworthy GAM.

Alex Chang, Vinith M. Suriyakumar, Abhishek Moturu et al.

Early detection of cancer is key to a good prognosis and requires frequent testing, especially in pediatrics. Whole-body magnetic resonance imaging (wbMRI) is an essential part of several well-established screening protocols, with screening starting in early childhood. To date, machine learning (ML) has been used on wbMRI images to stage adult cancer patients. It is not possible to use such tools in pediatrics due to the changing bone signal throughout growth, the difficulty of obtaining these images in young children due to movement and limited compliance, and the rarity of positive cases. We evaluate the quality of wbMRI images generated using generative adversarial networks (GANs) trained on wbMRI data from The Hospital for Sick Children in Toronto. We use the Frchet Inception Distance (FID) metric, Domain Frchet Distance (DFD), and blind tests with a radiology fellow for evaluation. We demonstrate that StyleGAN2 provides the best performance in generating wbMRI images with respect to all three metrics.

Sana Tonekaboni, Shalmali Joshi, Kieran Campbell et al.

Explanations of time series models are useful for high stakes applications like healthcare but have received little attention in machine learning literature. We propose FIT, a framework that evaluates the importance of observations for a multivariate time-series black-box model by quantifying the shift in the predictive distribution over time. FIT defines the importance of an observation based on its contribution to the distributional shift under a KL-divergence that contrasts the predictive distribution against a counterfactual where the rest of the features are unobserved. We also demonstrate the need to control for time-dependent distribution shifts. We compare with state-of-the-art baselines on simulated and real-world clinical data and demonstrate that our approach is superior in identifying important time points and observations throughout the time series.

Bret Nestor, Matthew B. A. McDermott, Willie Boag et al.

When training clinical prediction models from electronic health records (EHRs), a key concern should be a model's ability to sustain performance over time when deployed, even as care practices, database systems, and population demographics evolve. Due to de-identification requirements, however, current experimental practices for public EHR benchmarks (such as the MIMIC-III critical care dataset) are time agnostic, assigning care records to train or test sets without regard for the actual dates of care. As a result, current benchmarks cannot assess how well models trained on one year generalise to another. In this work, we obtain a Limited Data Use Agreement to access year of care for each record in MIMIC and show that all tested state-of-the-art models decay in prediction quality when trained on historical data and tested on future data, particularly in response to a system-wide record-keeping change in 2008 (0.29 drop in AUROC for mortality prediction, 0.10 drop in AUROC for length-of-stay prediction with a random forest classifier). We further develop a simple yet effective mitigation strategy: by aggregating raw features into expert-defined clinical concepts, we see only a 0.06 drop in AUROC for mortality prediction and a 0.03 drop in AUROC for length-of-stay prediction. We demonstrate that this aggregation strategy outperforms other automatic feature preprocessing techniques aimed at increasing robustness to data drift. We release our aggregated representations and code to encourage more deployable clinical prediction models.

Sana Tonekaboni, Shalmali Joshi, Melissa D McCradden et al.

Translating machine learning (ML) models effectively to clinical practice requires establishing clinicians' trust. Explainability, or the ability of an ML model to justify its outcomes and assist clinicians in rationalizing the model prediction, has been generally understood to be critical to establishing trust. However, the field suffers from the lack of concrete definitions for usable explanations in different settings. To identify specific aspects of explainability that may catalyze building trust in ML models, we surveyed clinicians from two distinct acute care specialties (Intenstive Care Unit and Emergency Department). We use their feedback to characterize when explainability helps to improve clinicians' trust in ML models. We further identify the classes of explanations that clinicians identified as most relevant and crucial for effective translation to clinical practice. Finally, we discern concrete metrics for rigorous evaluation of clinical explainability methods. By integrating perceptions of explainability between clinicians and ML researchers we hope to facilitate the endorsement and broader adoption and sustained use of ML systems in healthcare.

George Adam, Petr Smirnov, Benjamin Haibe-Kains et al.

Deep learning algorithms have increasingly been shown to lack robustness to simple adversarial examples (AdvX). An equally troubling observation is that these adversarial examples transfer between different architectures trained on different datasets. We investigate the transferability of adversarial examples between models using the angle between the input-output Jacobians of different models. To demonstrate the relevance of this approach, we perform case studies that involve jointly training pairs of models. These case studies empirically justify the theoretical intuitions for why the angle between gradients is a fundamental quantity in AdvX transferability. Furthermore, we consider the asymmetry of AdvX transferability between two models of the same architecture and explain it in terms of differences in gradient norms between the models. Lastly, we provide a simple modification to existing training setups that reduces transferability of adversarial examples between pairs of models.

Erik Drysdale, Yingwei Peng, Timothy P. Hanna et al.

In high dimensional settings where a small number of regressors are expected to be important, the Lasso estimator can be used to obtain a sparse solution vector with the expectation that most of the non-zero coefficients are associated with true signals. While several approaches have been developed to control the inclusion of false predictors with the Lasso, these approaches are limited by relying on asymptotic theory, having to empirically estimate terms based on theoretical quantities, assuming a continuous response class with Gaussian noise and design matrices, or high computation costs. In this paper we show how: (1) an existing model (the SQRT-Lasso) can be recast as a method of controlling the number of expected false positives, (2) how a similar estimator can used for all other generalized linear model classes, and (3) this approach can be fit with existing fast Lasso optimization solvers. Our justification for false positive control using randomly weighted self-normalized sum theory is to our knowledge novel. Moreover, our estimator's properties hold in finite samples up to some approximation error which we find in practical settings to be negligible under a strict mutual incoherence condition.

Chun-Hao Chang, Mingjie Mai, Anna Goldenberg

Imagine a patient in critical condition. What and when should be measured to forecast detrimental events, especially under the budget constraints? We answer this question by deep reinforcement learning (RL) that jointly minimizes the measurement cost and maximizes predictive gain, by scheduling strategically-timed measurements. We learn our policy to be dynamically dependent on the patient's health history. To scale our framework to exponentially large action space, we distribute our reward in a sequential setting that makes the learning easier. In our simulation, our policy outperforms heuristic-based scheduling with higher predictive gain and lower cost. In a real-world ICU mortality prediction task (MIMIC3), our policies reduce the total number of measurements by $31\%$ or improve predictive gain by a factor of $3$ as compared to physicians, under the off-policy policy evaluation.

Angeline Yasodhara, Mamatha Bhat, Anna Goldenberg

25% of people who received a liver transplant will go on to develop diabetes within the next 5 years. These thousands of individuals are at 2-fold higher risk of cardiovascular events, graft loss, infections, as well as lower long-term survival. This is partly due to the medication used during and/or after transplant that significantly impacts metabolic balance. To assess which medication best suits the patient's condition, clinicians need an accurate estimate of diabetes risk. Both patient's historical data and observations at the current visit are informative in predicting whether the patient will develop diabetes within the following year. In this work we compared a variety of time-to-event prediction models as well as classifiers predicting the likelihood of the event within a year from the current checkup. We are particularly interested in comparing two types of models: 1) standard time-to-event predictors where the historical measurements are merely concatenated, 2) incorporating Deep Markov Model to first obtain low-dimensional embedding of historical data and then using this embedding as an additional input into the model. We compared a variety of algorithms including standard and regularized Cox proportional-hazards model (CPH), mixed effect random forests, survival-forests and Weibull Time-To-Event Recurrent Neural Network (WTTE-RNN). The results show that although all methods' performances varied from year to year and there was no clear winner across all the time points, regularized CPH model that used 1 to 3 years of historical visits data on average achieved a high, clinically relevant Concordance Index of .863. We thus recommend this model for further prospective clinical validation and hopefully, an eventual use in the clinic to improve clinicians' ability to personalize post-operative care and reduce the incidence of new-onset diabetes post liver transplant.

Chun-Hao Chang, Mingjie Mai, Anna Goldenberg

Current clinical practice to monitor patients' health follows either regular or heuristic-based lab test (e.g. blood test) scheduling. Such practice not only gives rise to redundant measurements accruing cost, but may even lead to unnecessary patient discomfort. From the computational perspective, heuristic-based test scheduling might lead to reduced accuracy of clinical forecasting models. Computationally learning an optimal clinical test scheduling and measurement collection, is likely to lead to both, better predictive models and patient outcome improvement. We address the scheduling problem using deep reinforcement learning (RL) to achieve high predictive gain and low measurement cost, by scheduling fewer, but strategically timed tests. We first show that in the simulation our policy outperforms heuristic-based measurement scheduling with higher predictive gain or lower cost measured by accumulated reward. We then learn a scheduling policy for mortality forecasting in the real-world clinical dataset (MIMIC3), our learned policy is able to provide useful clinical insights. To our knowledge, this is the first RL application on multi-measurement scheduling problem in the clinical setting.

Bret Nestor, Matthew B. A. McDermott, Geeticka Chauhan et al.

Machine learning for healthcare often trains models on de-identified datasets with randomly-shifted calendar dates, ignoring the fact that data were generated under hospital operation practices that change over time. These changing practices induce definitive changes in observed data which confound evaluations which do not account for dates and limit the generalisability of date-agnostic models. In this work, we establish the magnitude of this problem on MIMIC, a public hospital dataset, and showcase a simple solution. We augment MIMIC with the year in which care was provided and show that a model trained using standard feature representations will significantly degrade in quality over time. We find a deterioration of 0.3 AUC when evaluating mortality prediction on data from 10 years later. We find a similar deterioration of 0.15 AUC for length-of-stay. In contrast, we demonstrate that clinically-oriented aggregates of raw features significantly mitigate future deterioration. Our suggested aggregated representations, when retrained yearly, have prediction quality comparable to year-agnostic models.

George A. Adam, Petr Smirnov, David Duvenaud et al.

Many deep learning algorithms can be easily fooled with simple adversarial examples. To address the limitations of existing defenses, we devised a probabilistic framework that can generate an exponentially large ensemble of models from a single model with just a linear cost. This framework takes advantage of neural network depth and stochastically decides whether or not to insert noise removal operators such as VAEs between layers. We show empirically the important role that model gradients have when it comes to determining transferability of adversarial examples, and take advantage of this result to demonstrate that it is possible to train models with limited adversarial attack transferability. Additionally, we propose a detection method based on metric learning in order to detect adversarial examples that have no hope of being cleaned of maliciously engineered noise.

Chun-Hao Chang, Elliot Creager, Anna Goldenberg et al.

When an image classifier makes a prediction, which parts of the image are relevant and why? We can rephrase this question to ask: which parts of the image, if they were not seen by the classifier, would most change its decision? Producing an answer requires marginalizing over images that could have been seen but weren't. We can sample plausible image in-fills by conditioning a generative model on the rest of the image. We then optimize to find the image regions that most change the classifier's decision after in-fill. Our approach contrasts with ad-hoc in-filling approaches, such as blurring or injecting noise, which generate inputs far from the data distribution, and ignore informative relationships between different parts of the image. Our method produces more compact and relevant saliency maps, with fewer artifacts compared to previous methods.

Marinka Zitnik, Francis Nguyen, Bo Wang et al.

New technologies have enabled the investigation of biology and human health at an unprecedented scale and in multiple dimensions. These dimensions include a myriad of properties describing genome, epigenome, transcriptome, microbiome, phenotype, and lifestyle. No single data type, however, can capture the complexity of all the factors relevant to understanding a phenomenon such as a disease. Integrative methods that combine data from multiple technologies have thus emerged as critical statistical and computational approaches. The key challenge in developing such approaches is the identification of effective models to provide a comprehensive and relevant systems view. An ideal method can answer a biological or medical question, identifying important features and predicting outcomes, by harnessing heterogeneous data across several dimensions of biological variation. In this Review, we describe the principles of data integration and discuss current methods and available implementations. We provide examples of successful data integration in biology and medicine. Finally, we discuss current challenges in biomedical integrative methods and our perspective on the future development of the field.

Chun-Hao Chang, Ladislav Rampasek, Anna Goldenberg

Deep neural networks (DNNs) achieve state-of-the-art results in a variety of domains. Unfortunately, DNNs are notorious for their non-interpretability, and thus limit their applicability in hypothesis-driven domains such as biology and healthcare. Moreover, in the resource-constraint setting, it is critical to design tests relying on fewer more informative features leading to high accuracy performance within reasonable budget. We aim to close this gap by proposing a new general feature ranking method for deep learning. We show that our simple yet effective method performs on par or compares favorably to eight strawman, classical and deep-learning feature ranking methods in two simulations and five very different datasets on tasks ranging from classification to regression, in both static and time series scenarios. We also illustrate the use of our method on a drug response dataset and show that it identifies genes relevant to the drug-response.

Ladislav Rampasek, Daniel Hidru, Petr Smirnov et al.

We present two deep generative models based on Variational Autoencoders to improve the accuracy of drug response prediction. Our models, Perturbation Variational Autoencoder and its semi-supervised extension, Drug Response Variational Autoencoder (Dr.VAE), learn latent representation of the underlying gene states before and after drug application that depend on: (i) drug-induced biological change of each gene and (ii) overall treatment response outcome. Our VAE-based models outperform the current published benchmarks in the field by anywhere from 3 to 11% AUROC and 2 to 30% AUPR. In addition, we found that better reconstruction accuracy does not necessarily lead to improvement in classification accuracy and that jointly trained models perform better than models that minimize reconstruction error independently.

Lauren Erdman, Ekansh Sharma, Eva Unternahrer et al.

More than two thirds of mental health problems have their onset during childhood or adolescence. Identifying children at risk for mental illness later in life and predicting the type of illness is not easy. We set out to develop a platform to define subtypes of childhood social-emotional development using longitudinal, multifactorial trait-based measures. Subtypes discovered through this study could ultimately advance psychiatric knowledge of the early behavioural signs of mental illness. To this extent we have examined two types of models: latent class mixture models and GP-based models. Our findings indicate that while GP models come close in accuracy of predicting future trajectories, LCMMs predict the trajectories as well in a fraction of the time. Unfortunately, neither of the models are currently accurate enough to lead to immediate clinical impact. The available data related to the development of childhood mental health is often sparse with only a few time points measured and require novel methods with improved efficiency and accuracy.

Daniel Hidru, Anna Goldenberg

Nonnegative matrix factorization (NMF) methods have proved to be powerful across a wide range of real-world clustering applications. Integrating multiple types of measurements for the same objects/subjects allows us to gain a deeper understanding of the data and refine the clustering. We have developed a novel Graph-reguarized multiview NMF-based method for data integration called EquiNMF. The parameters for our method are set in a completely automated data-specific unsupervised fashion, a highly desirable property in real-world applications. We performed extensive and comprehensive experiments on multiview imaging data. We show that EquiNMF consistently outperforms other single-view NMF methods used on concatenated data and multi-view NMF methods with different types of regularizations.

Bo Wang, Anna Goldenberg

Analysis of high dimensional noisy data is of essence across a variety of research fields. Feature selection techniques are designed to find the relevant feature subset that can facilitate classification or pattern detection. Traditional (supervised) feature selection methods utilize label information to guide the identification of relevant feature subsets. In this paper, however, we consider the unsupervised feature selection problem. Without the label information, it is particularly difficult to identify a small set of relevant features due to the noisy nature of real-world data which corrupts the intrinsic structure of the data. Our Gradient-based Laplacian Feature Selection (GLFS) selects important features by minimizing the variance of the Laplacian regularized least squares regression model. With $\ell_1$ relaxation, GLFS can find a sparse subset of features that is relevant to the Laplacian manifolds. Extensive experiments on simulated, three real-world object recognition and two computational biology datasets, have illustrated the power and superior performance of our approach over multiple state-of-the-art unsupervised feature selection methods. Additionally, we show that GLFS selects a sparser set of more relevant features in a supervised setting outperforming the popular elastic net methodology.