Isaac S. Kohane

Matthew B. A. McDermott, Bret Nestor, Peniel Argaw et al. · harvard

Generative, pre-trained transformers (GPTs, a.k.a. "Foundation Models") have reshaped natural language processing (NLP) through their versatility in diverse downstream tasks. However, their potential extends far beyond NLP. This paper provides a software utility to help realize this potential, extending the applicability of GPTs to continuous-time sequences of complex events with internal dependencies, such as medical record datasets. Despite their potential, the adoption of foundation models in these domains has been hampered by the lack of suitable tools for model construction and evaluation. To bridge this gap, we introduce Event Stream GPT (ESGPT), an open-source library designed to streamline the end-to-end process for building GPTs for continuous-time event sequences. ESGPT allows users to (1) build flexible, foundation-model scale input datasets by specifying only a minimal configuration file, (2) leverage a Hugging Face compatible modeling API for GPTs over this modality that incorporates intra-event causal dependency structures and autoregressive generation capabilities, and (3) evaluate models via standardized processes that can assess few and even zero-shot performance of pre-trained models on user-specified fine-tuning tasks.

Peniel N. Argaw, Elizabeth Healey, Isaac S. Kohane

Heterogeneous treatment effects (HTEs) are commonly identified during randomized controlled trials (RCTs). Identifying subgroups of patients with similar treatment effects is of high interest in clinical research to advance precision medicine. Often, multiple clinical outcomes are measured during an RCT, each having a potentially heterogeneous effect. Recently there has been high interest in identifying subgroups from HTEs, however, there has been less focus on developing tools in settings where there are multiple outcomes. In this work, we propose a framework for partitioning the covariate space to identify subgroups across multiple outcomes based on the joint CIs. We test our algorithm on synthetic and semi-synthetic data where there are two outcomes, and demonstrate that our algorithm is able to capture the HTE in both outcomes simultaneously.

Dianbo Liu, Karmel W. Choi, Paulo Lizano et al.

Importance: The prevalence of severe mental illnesses (SMIs) in the United States is approximately 3% of the whole population. The ability to conduct risk screening of SMIs at large scale could inform early prevention and treatment. Objective: A scalable machine learning based tool was developed to conduct population-level risk screening for SMIs, including schizophrenia, schizoaffective disorders, psychosis, and bipolar disorders,using 1) healthcare insurance claims and 2) electronic health records (EHRs). Design, setting and participants: Data from beneficiaries from a nationwide commercial healthcare insurer with 77.4 million members and data from patients from EHRs from eight academic hospitals based in the U.S. were used. First, the predictive models were constructed and tested using data in case-control cohorts from insurance claims or EHR data. Second, performance of the predictive models across data sources were analyzed. Third, as an illustrative application, the models were further trained to predict risks of SMIs among 18-year old young adults and individuals with substance associated conditions. Main outcomes and measures: Machine learning-based predictive models for SMIs in the general population were built based on insurance claims and EHR.

Stefan Feuerriegel, Dennis Frauen, Valentyn Melnychuk et al.

Causal machine learning (ML) offers flexible, data-driven methods for predicting treatment outcomes including efficacy and toxicity, thereby supporting the assessment and safety of drugs. A key benefit of causal ML is that it allows for estimating individualized treatment effects, so that clinical decision-making can be personalized to individual patient profiles. Causal ML can be used in combination with both clinical trial data and real-world data, such as clinical registries and electronic health records, but caution is needed to avoid biased or incorrect predictions. In this Perspective, we discuss the benefits of causal ML (relative to traditional statistical or ML approaches) and outline the key components and steps. Finally, we provide recommendations for the reliable use of causal ML and effective translation into the clinic.

Huaiyuan Ying, Hongyi Yuan, Jinsen Lu et al.

Electronic Health Records (EHRs) hold immense potential for advancing healthcare, offering rich, longitudinal data that combines structured information with valuable insights from unstructured clinical notes. However, the unstructured nature of clinical text poses significant challenges for secondary applications. Traditional methods for structuring EHR free-text data, such as rule-based systems and multi-stage pipelines, are often limited by their time-consuming configurations and inability to adapt across clinical notes from diverse healthcare settings. Few systems provide a comprehensive attribute extraction for terminologies. While giant large language models (LLMs) like GPT-4 and LLaMA 405B excel at structuring tasks, they are slow, costly, and impractical for large-scale use. To overcome these limitations, we introduce GENIE, a Generative Note Information Extraction system that leverages LLMs to streamline the structuring of unstructured clinical text into usable data with standardized format. GENIE processes entire paragraphs in a single pass, extracting entities, assertion statuses, locations, modifiers, values, and purposes with high accuracy. Its unified, end-to-end approach simplifies workflows, reduces errors, and eliminates the need for extensive manual intervention. Using a robust data preparation pipeline and fine-tuned small scale LLMs, GENIE achieves competitive performance across multiple information extraction tasks, outperforming traditional tools like cTAKES and MetaMap and can handle extra attributes to be extracted. GENIE strongly enhances real-world applicability and scalability in healthcare systems. By open-sourcing the model and test data, we aim to encourage collaboration and drive further advancements in EHR structurization.

Ayush Noori, Adam Rodman, Alan Karthikesalingam et al.

Modern computer systems often rely on syslog, a simple, universal protocol that records every critical event across heterogeneous infrastructure. However, healthcare's rapidly growing clinical AI stack has no equivalent. As hospitals rush to pilot large language models and other AI-based clinical decision support tools, we still lack a standard way to record how, when, by whom, and for whom these AI models are used. Without that transparency and visibility, it is challenging to measure real-world performance and outcomes, detect adverse events, or correct bias or dataset drift. In the spirit of syslog, we introduce MedLog, a protocol for event-level logging of clinical AI. Any time an AI model is invoked to interact with a human, interface with another algorithm, or act independently, a MedLog record is created. This record consists of nine core fields: header, model, user, target, inputs, artifacts, outputs, outcomes, and feedback, providing a structured and consistent record of model activity. To encourage early adoption, especially in low-resource settings, and minimize the data footprint, MedLog supports risk-based sampling, lifecycle-aware retention policies, and write-behind caching; detailed traces for complex, agentic, or multi-stage workflows can also be captured under MedLog. MedLog can catalyze the development of new databases and software to store and analyze MedLog records. Realizing this vision would enable continuous surveillance, auditing, and iterative improvement of medical AI, laying the foundation for a new form of digital epidemiology.

Michelle M. Li, Ben Y. Reis, Adam Rodman et al.

Medical AI, including clinical language models, vision-language models, and multimodal health record models, already summarizes notes, answers questions, and supports decisions. Their adaptation to new populations, specialties, or care settings often relies on fine-tuning, prompting, or retrieval from external knowledge bases. These strategies can scale poorly and risk contextual errors: outputs that appear plausible but miss critical patient or situational information. We envision context switching as a solution. Context switching adjusts model reasoning at inference without retraining. Generative models can tailor outputs to patient biology, care setting, or disease. Multimodal models can reason on notes, laboratory results, imaging, and genomics, even when some data are missing or delayed. Agent models can coordinate tools and roles based on tasks and users. In each case, context switching enables medical AI to adapt across specialties, populations, and geographies. It requires advances in data design, model architectures, and evaluation frameworks, and establishes a foundation for medical AI that scales to infinitely many contexts while remaining reliable and suited to real-world care.

Isaac Kohane

As large language models (LLMs) are deployed in high-stakes domains like healthcare, understanding how well their decision-making aligns with human preferences and values becomes crucial, especially when we recognize that there is no single gold standard for these preferences. This paper applies a systematic methodology for evaluating preference alignment in LLMs on categorical decision-making with medical triage as a domain-specific use case. It also measures how effectively an alignment procedure will change the alignment of a specific model. Key to this methodology is a novel simple measure, the Alignment Compliance Index (ACI), that quantifies how effectively a LLM can be aligned to a given preference function or gold standard. Since the ACI measures the effect rather than the process of alignment, it is applicable to alignment methods beyond the in-context learning used in this study. Using a dataset of simulated patient pairs, three frontier LLMs (GPT4o, Claude 3.5 Sonnet, and Gemini Advanced) were assessed on their ability to make triage decisions consistent with an expert clinician's preferences. The models' performance before and after alignment attempts was evaluated using various prompting strategies. The results reveal significant variability in alignment effectiveness across models and alignment approaches. Notably, models that performed well, as measured by ACI, pre-alignment sometimes degraded post-alignment, and small changes in the target preference function led to large shifts in model rankings. The implicit ethical principles, as understood by humans, underlying the LLMs' decisions were also explored through targeted questioning. This study motivates the use of a practical set of methods and the ACI, in the near term, to understand the correspondence between the variety of human and LLM decision-making values in categorical decision-making such as triage.

Samuel G. Finlayson, Matthew B. A. McDermott, Alex V. Pickering et al.

Modeling the relationship between chemical structure and molecular activity is a key goal in drug development. Many benchmark tasks have been proposed for molecular property prediction, but these tasks are generally aimed at specific, isolated biomedical properties. In this work, we propose a new cross-modal small molecule retrieval task, designed to force a model to learn to associate the structure of a small molecule with the transcriptional change it induces. We develop this task formally as multi-view alignment problem, and present a coordinated deep learning approach that jointly optimizes representations of both chemical structure and perturbational gene expression profiles. We benchmark our results against oracle models and principled baselines, and find that cell line variability markedly influences performance in this domain. Our work establishes the feasibility of this new task, elucidates the limitations of current data and systems, and may serve to catalyze future research in small molecule representation learning.

Samuel G. Finlayson, Hyunkwang Lee, Isaac S. Kohane et al.

Medical students and radiology trainees typically view thousands of images in order to "train their eye" to detect the subtle visual patterns necessary for diagnosis. Nevertheless, infrastructural and legal constraints often make it difficult to access and quickly query an abundance of images with a user-specified feature set. In this paper, we use a conditional generative adversarial network (GAN) to synthesize $1024\times1024$ pixel pelvic radiographs that can be queried with conditioning on fracture status. We demonstrate that the conditional GAN learns features that distinguish fractures from non-fractures by training a convolutional neural network exclusively on images sampled from the GAN and achieving an AUC of $>0.95$ on a held-out set of real images. We conduct additional analysis of the images sampled from the GAN and describe ongoing work to validate educational efficacy.

Brett K. Beaulieu-Jones, Isaac S. Kohane, Andrew L. Beam

Biomedical association studies are increasingly done using clinical concepts, and in particular diagnostic codes from clinical data repositories as phenotypes. Clinical concepts can be represented in a meaningful, vector space using word embedding models. These embeddings allow for comparison between clinical concepts or for straightforward input to machine learning models. Using traditional approaches, good representations require high dimensionality, making downstream tasks such as visualization more difficult. We applied Poincaré embeddings in a 2-dimensional hyperbolic space to a large-scale administrative claims database and show performance comparable to 100-dimensional embeddings in a euclidean space. We then examine disease relationships under different disease contexts to better understand potential phenotypes.

Samuel G. Finlayson, Hyung Won Chung, Isaac S. Kohane et al.

The discovery of adversarial examples has raised concerns about the practical deployment of deep learning systems. In this paper, we demonstrate that adversarial examples are capable of manipulating deep learning systems across three clinical domains. For each of our representative medical deep learning classifiers, both white and black box attacks were highly successful. Our models are representative of the current state of the art in medical computer vision and, in some cases, directly reflect architectures already seeing deployment in real world clinical settings. In addition to the technical contribution of our paper, we synthesize a large body of knowledge about the healthcare system to argue that medicine may be uniquely susceptible to adversarial attacks, both in terms of monetary incentives and technical vulnerability. To this end, we outline the healthcare economy and the incentives it creates for fraud and provide concrete examples of how and why such attacks could be realistically carried out. We urge practitioners to be aware of current vulnerabilities when deploying deep learning systems in clinical settings, and encourage the machine learning community to further investigate the domain-specific characteristics of medical learning systems.

Andrew L. Beam, Benjamin Kompa, Allen Schmaltz et al.

Word embeddings are a popular approach to unsupervised learning of word relationships that are widely used in natural language processing. In this article, we present a new set of embeddings for medical concepts learned using an extremely large collection of multimodal medical data. Leaning on recent theoretical insights, we demonstrate how an insurance claims database of 60 million members, a collection of 20 million clinical notes, and 1.7 million full text biomedical journal articles can be combined to embed concepts into a common space, resulting in the largest ever set of embeddings for 108,477 medical concepts. To evaluate our approach, we present a new benchmark methodology based on statistical power specifically designed to test embeddings of medical concepts. Our approach, called cui2vec, attains state-of-the-art performance relative to previous methods in most instances. Finally, we provide a downloadable set of pre-trained embeddings for other researchers to use, as well as an online tool for interactive exploration of the cui2vec embeddings