OBJECTIVES: Large language models (LLMs) are increasingly utilized in healthcare, transforming medical practice through advanced language processing capabilities. However, the evaluation of LLMs predominantly relies on human qualitative assessment, which is time-consuming, resource-intensive, and may be subject to variability and bias. There is a pressing need for quantitative metrics to enable scalable, objective, and efficient evaluation. MATERIALS AND METHODS: We propose a unified evaluation framework that bridges qualitative and quantitative methods to assess LLM performance in healthcare settings. This framework maps evaluation aspects-such as linguistic quality, efficiency, content integrity, trustworthiness, and usefulness-to both qualitative assessments and quantitative metrics. We apply our approach to empirically evaluate the Epic In-Basket feature, which uses LLM to generate patient message replies. RESULTS: The empirical evaluation demonstrates that while Artificial Intelligence (AI)-generated replies exhibit high fluency, clarity, and minimal toxicity, they face challenges with coherence and completeness. Clinicians' manual decision to use AI-generated drafts correlates strongly with quantitative metrics, suggesting that quantitative metrics have the potential to reduce human effort in the evaluation process and make it more scalable. DISCUSSION: Our study highlights the potential of a unified evaluation framework that integrates qualitative and quantitative methods, enabling scalable and systematic assessments of LLMs in healthcare. Automated metrics streamline evaluation and monitoring processes, but their effective use depends on alignment with human judgment, particularly for aspects requiring contextual interpretation. As LLM applications expand, refining evaluation strategies and fostering interdisciplinary collaboration will be critical to maintaining high standards of accuracy, ethics, and regulatory compliance. CONCLUSION: Our unified evaluation framework bridges the gap between qualitative human assessments and automated quantitative metrics, enhancing the reliability and scalability of LLM evaluations in healthcare. While automated quantitative evaluations are not ready to fully replace qualitative human evaluations, they can be used to enhance the process and, with relevant benchmarks derived from the unified framework proposed here, they can be applied to LLM monitoring and evaluation of updated versions of the original technology evaluated using qualitative human standards.

OBJECTIVE: Leverage electronic health record (EHR) audit logs to develop a machine learning (ML) model that predicts which notes a clinician wants to review when seeing oncology patients. MATERIALS AND METHODS: We trained logistic regression models using note metadata and a Term Frequency Inverse Document Frequency (TF-IDF) text representation. We evaluated performance with precision, recall, F1, AUC, and a clinical qualitative assessment. RESULTS: The metadata only model achieved an AUC 0.930 and the metadata and TF-IDF model an AUC 0.937. Qualitative assessment revealed a need for better text representation and to further customize predictions for the user. DISCUSSION: Our model effectively surfaces the top 10 notes a clinician wants to review when seeing an oncology patient. Further studies can characterize different types of clinician users and better tailor the task for different care settings. CONCLUSION: EHR audit logs can provide important relevance data for training ML models that assist with note-writing in the oncology setting.

We study the application of large language models to zero-shot and few-shot classification of tabular data. We prompt the large language model with a serialization of the tabular data to a natural-language string, together with a short description of the classification problem. In the few-shot setting, we fine-tune the large language model using some labeled examples. We evaluate several serialization methods including templates, table-to-text models, and large language models. Despite its simplicity, we find that this technique outperforms prior deep-learning-based tabular classification methods on several benchmark datasets. In most cases, even zero-shot classification obtains non-trivial performance, illustrating the method’s ability to exploit prior knowledge encoded in large language models. Unlike many deep learning methods for tabular datasets, this approach is also competitive with strong traditional baselines like gradient-boosted trees, especially in the very-few-shot setting.

Label-scarce, high-dimensional domains such as healthcare present a challenge for modern machine learning techniques. To overcome the difficulties posed by a lack of labeled data, we explore an "order-contrastive" method for self-supervised pre-training on longitudinal data. We sample pairs of time segments, switch the order for half of them, and train a model to predict whether a given pair is in the correct order. Intuitively, the ordering task allows the model to attend to the least time-reversible features (for example, features that indicate progression of a chronic disease). The same features are often useful for downstream tasks of interest. To quantify this, we study a simple theoretical setting where we prove a finite-sample guarantee for the downstream error of a representation learned with order-contrastive pre-training. Empirically, in synthetic and longitudinal healthcare settings, we demonstrate the effectiveness of order-contrastive pre-training in the small-data regime over supervised learning and other self-supervised pre-training baselines. Our results indicate that pre-training methods designed for particular classes of distributions and downstream tasks can improve the performance of self-supervised learning.

A long-running goal of the clinical NLP community is the extraction of important variables trapped in clinical notes.However, roadblocks have included dataset shift from the general domain and a lack of public clinical corpora and annotations.In this work, we show that large language models, such as InstructGPT (Ouyang et al., 2022), perform well at zero-and few-shot information extraction from clinical text despite not being trained specifically for the clinical domain.Whereas text classification and generation performance have already been studied extensively in such models, here we additionally demonstrate how to leverage them to tackle a diverse set of NLP tasks which require more structured outputs, including span identification, token-level sequence classification, and relation extraction.Further, due to the dearth of available data to evaluate these systems, we introduce new datasets for benchmarking fewshot clinical information extraction based on a manual re-annotation of the CASI dataset (Moon et al., 2014) for new tasks 1 .On the clinical extraction tasks we studied, the GPT-3 systems significantly outperform existing zeroand few-shot baselines.

We demonstrate that co-training (Blum & Mitchell, 1998) can improve the performance of prompt-based learning by using unlabeled data. While prompting has emerged as a promising paradigm for few-shot and zero-shot learning, it is often brittle and requires much larger models compared to the standard supervised setup. We find that co-training makes it possible to improve the original prompt model and at the same time learn a smaller, downstream task-specific model. In the case where we only have partial access to a prompt model (e.g., output probabilities from GPT-3 (Brown et al., 2020)) we learn a calibration model over the prompt outputs. When we have full access to the prompt model’s gradients but full finetuning remains prohibitively expensive (e.g., T0 (Sanh et al., 2021)), we learn a set of soft prompt continuous vectors to iteratively update the prompt model. We find that models trained in this manner can significantly improve performance on challenging datasets where there is currently a large gap between prompt-based learning and fully-supervised models.

Data cleaning is naturally framed as probabilistic inference in a generative model of ground-truth data and likely errors, but the diversity of real-world error patterns and the hardness of inference make Bayesian approaches difficult to automate. We present PClean, a probabilistic programming language (PPL) for leveraging dataset-specific knowledge to automate Bayesian cleaning. Compared to general-purpose PPLs, PClean tackles a restricted problem domain, enabling three modeling and inference innovations: (1) a non-parametric model of relational database instances, which users’ programs customize; (2) a novel sequential Monte Carlo inference algorithm that exploits the structure of PClean’s model class; and (3) a compiler that generates near-optimal SMC proposals and blocked-Gibbs rejuvenation kernels based on the user’s model and data. We show empirically that short (< 50-line) PClean programs can: be faster and more accurate than generic PPL inference on data-cleaning benchmarks; match state-of-the-art data-cleaning systems in terms of accuracy and runtime (unlike generic PPL inference in the same runtime); and scale to real-world datasets with millions of records.

Automated decision support can accelerate tedious tasks as users can focus their attention where it is needed most. However, a key concern is whether users overly trust or cede agency to automation. In this paper, we investigate the effects of introducing automation to annotating clinical texts — a multi-step, error-prone task of identifying clinical concepts (e.g., procedures) in medical notes, and mapping them to labels in a large ontology. We consider two forms of decision aid: recommending which labels to map concepts to, and pre-populating annotation suggestions. Through laboratory studies, we find that 18 clinicians generally build intuition of when to rely on automation and when to exercise their own judgement. However, when presented with fully pre-populated suggestions, these expert users exhibit less agency: accepting improper mentions, and taking less initiative in creating additional annotations. Our findings inform how systems and algorithms should be designed to mitigate the observed issues.

Clinical documentation can be transformed by Electronic Health Records, yet the documentation process is still a tedious, time-consuming, and error-prone process. Clinicians are faced with multi-faceted requirements and fragmented interfaces for information exploration and documentation. These challenges are only exacerbated in the Emergency Department—clinicians often see 35 patients in one shift, during which they have to synthesize an often previously unknown patient’s medical records in order to reach a tailored diagnosis and treatment plan. To better support this information synthesis, clinical documentation tools must enable rapid contextual access to the patient’s medical record. MedKnowts is an integrated note-taking editor and information retrieval system which unifies the documentation and search process and provides concise synthesized concept-oriented slices of the patient’s medical record. MedKnowts automatically captures structured data while still allowing users the flexibility of natural language. MedKnowts leverages this structure to enable easier parsing of long notes, auto-populated text, and proactive information retrieval, easing the documentation burden.

Motivation: The use of drug combinations, termed polypharmacy, is common to treat patients with complex diseases or co-existing conditions. However, a major consequence of polypharmacy is a much higher risk of adverse side effects for the patient. Polypharmacy side effects emerge because of drug-drug interactions, in which activity of one drug may change, favorably or unfavorably, if taken with another drug. The knowledge of drug interactions is often limited because these complex relationships are rare, and are usually not observed in relatively small clinical testing. Discovering polypharmacy side effects thus remains an important challenge with significant implications for patient mortality and morbidity. Results: Here, we present Decagon, an approach for modeling polypharmacy side effects. The approach constructs a multimodal graph of protein-protein interactions, drug-protein target interactions and the polypharmacy side effects, which are represented as drug-drug interactions, where each side effect is an edge of a different type. Decagon is developed specifically to handle such multimodal graphs with a large number of edge types. Our approach develops a new graph convolutional neural network for multirelational link prediction in multimodal networks. Unlike approaches limited to predicting simple drug-drug interaction values, Decagon can predict the exact side effect, if any, through which a given drug combination manifests clinically. Decagon accurately predicts polypharmacy side effects, outperforming baselines by up to 69%. We find that it automatically learns representations of side effects indicative of co-occurrence of polypharmacy in patients. Furthermore, Decagon models particularly well polypharmacy side effects that have a strong molecular basis, while on predominantly non-molecular side effects, it achieves good performance because of effective sharing of model parameters across edge types. Decagon opens up opportunities to use large pharmacogenomic and patient population data to flag and prioritize polypharmacy side effects for follow-up analysis via formal pharmacological studies. Availability and implementation: Source code and preprocessed datasets are at: http://snap.stanford.edu/decagon.