Peer-Reviewed Papers

Large Language Models (LLMs) have been widely used in real-world applications. However, as LLMs evolve and new datasets are released, it becomes crucial to build processes to evaluate and control the models’ performance. In this paper, we describe how to add Robustness, Accuracy, and Toxicity scores to model comparison tables, or leaderboards. We discuss the evaluation metrics, the approaches considered, and present the results of the first evaluation round for model Robustness, Accuracy, and Toxicity scores.

Our results show that GPT 4 achieves top performance on robustness and accuracy test, while Llama 2 achieves top performance on the toxicity test. We note that newer open-source models such as open chat 3.5 and neural chat 7B can perform well on these three test categories. Finally, domain-specific tests and models are also planned to be added to the leaderboard to allow for a more detailed evaluation of models in specific areas such as healthcare, legal, and finance.

Automatic generation of chemotherapy treatment timelines from electronic health records (EHRs) notes not only streamlines clinical workflows but also promotes better coordination and improvements in cancer treatment and quality of care. This paper describes the submission to the Chemotimelines 2024 shared task that aims to automatically build a chemotherapy treatment timeline for each patient using their complete set of EHR notes, spanning various sources such as primary care provider, oncology, discharge summaries, emergency department, pathology, radiology, and more. We report results from two large language models (LLMs), namely Llama 2 and Mistral 7B, applied to the shared task data using zero-shot prompting.

De-identification of medical images intended for research is a core requirement for data-sharing initiatives, particularly as the demand for data for artificial intelligence (AI) applications grows. The Center for Biomedical Informatics and Information Technology (CBIIT) of the US National Cancer Institute (NCI) convened a virtual workshop with the intent of summarizing the state of the art in de-identification technology and processes and exploring interesting aspects of the subject.

This paper summarizes the highlights of the first day of the workshop, the recordings, and presentations of which are publicly available for review. The topics covered included the report of the Medical Image De-Identification Initiative (MIDI) Task Group on best practices and recommendations, tools for conventional approaches to de-identification, international approaches to de-identification, and an industry panel.

Large language models (LLMs) are increasingly being applied to several tasks including text-to-SQL (the process of converting natural language to SQL queries). While most studies revolve around training LLMs on large SQL corpora for better generalization and then perform prompt engineering during inference, we investigate the notion of training LLMs for schema-less prompting. In particular, our approach uses simple natural language questions as input without any additional knowledge about the database schema. By doing so, we demonstrate that smaller models paired with simpler prompts result in considerable performance improvement while generating SQL queries.

Our model, based on the Flan-T5 architecture, achieves logical form accuracy (LFA) of 0.85 on the MIMICSQL dataset, significantly outperforming current state-of-the-art models such as Defog-SQL-Coder, GPT-3.5-Turbo, LLaMA-2-7B and GPT-4. This approach reduces the model size, lessening the amount of data and infrastructure cost required for training and serving, and improves the performance to enable the generation of much complex SQL queries.

Social determinants of health (SDoH) significantly influence health outcomes, accounting for nearly 40% of such outcomes globally. These determinants, pivotal in understanding health disparities, are insufficiently documented in clinical settings and academic clinical narratives. To address this gap, we examined clinical case reports from PubMed (1975–2022) to identify mentions of six specific SDoH, employing a pre-trained named-entity recognition (NER) model from John Snow Labs’ medical natural language processing (Healthcare NLP).

Precision and recall of John Snow Labs’ SDoH-NER model was compared with the zero-shot learning capabilities of GPT-3.5 and GPT-4. Regarding recall, SDoH-NER generally achieved better metrics for most types of SDoH compared to GPT-3.5 and GPT-4, although the differences were not statistically significant. In terms of precision, SDoH-NER outperformed GPT-4 in identifying marital status (p = 0.005) and homelessness/housing issues (p < 0.001). The SDoH-NER model is also far more efficient in inference, running on one AMD7 CPU with 32Gb of RAM.

The use of natural language processing (NLP) models, including the more recent large language models (LLM) in real-world applications obtained relevant success in the past years. To measure the performance of these systems, traditional performance metrics such as accuracy, precision, recall, and f1-score are used. Although it is important to measure the performance of the models in those terms, natural language often requires an holistic evaluation that consider other important aspects such as robustness, bias, accuracy, toxicity, fairness, safety, efficiency, clinical relevance, security, representation, disinformation, political orientation, sensitivity, factuality, legal concerns, and vulnerabilities.

To address the gap, we introduce LangTest, an open source Python toolkit, aimed at reshaping the evaluation of LLMs and NLP models in real-world applications. The project aims to empower data scientists, enabling them to meet high standards in the ever-evolving landscape of AI model development. Specifically, it provides a comprehensive suite of more than 60 test types, ensuring a more comprehensive understanding of a model’s behavior and responsible AI use. In this experiment, a Named Entity Recognition (NER) clinical model showed significant improvement in its capabilities to identify clinical entities in text after applying data augmentation for robustness.

Recent research advances achieve human-level accuracy for de-identifying free-text clinical notes on research datasets, but gaps remain in reproducing this in large real-world settings. This paper summarizes lessons learned from building a system used to e-identify over one billion real clinical notes, in a fully automated way, that was independently certified by multiple organizations for production use.

A fully automated solution requires a very high level of accuracy that does not require manual review. A hybrid context-based model architecture is described, which outperforms a Named Entity Recogniton (NER)-only model by 10% on the i2b2-2014 benchmark. The proposed system makes 50%, 475%, and 575% fewer errors than the comparable AWS, Azure, and GCP services respectively while also outperforming ChatGPT by 33%. It exceeds 98% coverage of sensitive data across 7 European languages, without a need for fine tuning.

As Electronic Health Records (EHR) become ubiquitous in healthcare systems worldwide, including in Arabic-speaking countries, the dual imperative of safeguarding patient privacy and leveraging data for research and quality improvement grows. This paper presents a first-of-its-kind automated de-identification pipeline for medical text specifically tailored for the Arabic language.

This includes accurate medical Named Entity Recognition (NER) for identifying personal information; data obfuscation models to replace sensitive entities with fake entities; and an implementation that natively scales to large datasets on commodity clusters. This research makes two contributions. First, we adapt two existing NER architectures— BERT For Token Classification (BFTC) and BiLSTM-CNN-Char – to accommodate the unique syntactic and morphological characteristics of the Arabic language. Comparative analysis suggests that BFTC models outperform Bi-LSTM models, achieving higher F1 scores for both identifying and redacting personally identifiable information (PII) from Arabic medical texts.

Second, we augment the deep learning models with a contextual parser engine to handle commonly missed entities. Experiments show that the combined pipeline demonstrates superior performance with micro F1 scores ranging from 0.94 to 0.98 on the test dataset, which is a translated version of the i2b2 2014 de-identification challenge, across 17 sensitive entities. This level of accuracy is in line with that achieved with manual de-identification by domain experts, suggesting that a fully automated and scalable process is now viable.

Medical artificial intelligence (AI) has tremendous potential to advance healthcare by supporting and contributing to the evidence-based practice of medicine, personalizing patient treatment, reducing costs, and improving both healthcare provider and patient experience. Unlocking this potential requires systematic, quantitative evaluation of the performance of medical AI models on large-scale, heterogeneous data capturing diverse patient populations. Here, to meet this need, we introduce MedPerf, an open platform for benchmarking AI models in the medical domain. MedPerf focuses on enabling federated evaluation of AI models, by securely distributing them to different facilities, such as healthcare organizations.

This process of bringing the model to the data empowers each facility to assess and verify the performance of AI models in an efficient and human-supervised process, while prioritizing privacy. We describe the current challenges healthcare and AI communities face, the need for an open platform, the design philosophy of MedPerf, its current implementation status and real-world deployment, our roadmap and, importantly, the use of MedPerf with multiple international institutions within cloud-based technology and on-premises scenarios. Finally, we welcome new contributions by researchers and organizations to further strengthen MedPerf as an open benchmarking platform.

In healthcare product development and research, teams invest huge amounts of time to study publications and other relevant sources. There is a need for novel solutions to efficiently and reliably extract information from multiple clinical data sources, in addition to generating new insights which can only be achieved through structuring textual information and accessible intelligent synthesis across multiple relevant data sources. We are developing a cloud-based solution where data from heterogeneous sources is structured, integrated and harmonized, and users can easily leverage the combined database to answer domain-specific questions and generate insights efficiently.

Knowledge graphs provide us the advantage to leverage relationships in addition to concepts in the context of heterogeneous data. We leveraged graph and NLP (Natural Language Processing) methods to build a domain-specific knowledge graph. We extracted the biomedically-relevant subset of Wikidata and augmented it from the biomedical literature (PubMed), clinical trials (clinicaltrials.gov) and NIH grants.

Pharmacoepidemiology studies often have missing data that are nonmonotone (i.e., data missing in an arbitrary pattern). It is well known that complete case analysis (CC) reduces precision and may result in bias. Previously, inverse probability weighting (IPW) for nonmonotone missingness could not be easily implemented in standard software, but a new approach for estimating the missingness weights (unconstrained maximum likelihood estimator, UMLE) overcomes this difficulty.

To compare performance of IPW by UMLE and multiple imputation (MI) to address nonmonotone missingness when estimating the average treatment effect.

The surging amount of biomedical literature & digital clinical records presents a growing need for text mining techniques that can not only identify but also semantically relate entities in unstructured data. In this paper we propose a text mining framework comprising of Named Entity Recognition (NER) and Relation Extraction (RE) models, which expands on previous work in three main ways. First, we introduce two new RE model architectures — an accuracy-optimized one based on BioBERT and a speed-optimized one utilizing crafted features over a Fully Connected Neural Network (FCNN).

Second, we evaluate both models on public benchmark datasets and obtain new state-of-the-art F1 scores on the 2012 i2b2 Clinical Temporal Relations challenge (F1 of 73.6, +1.2% over the previous SOTA), the 2010 i2b2 Clinical Relations challenge (F1 of 69.1, +1.2%), the 2019 Phenotype-Gene Relations dataset (F1 of 87.9, +8.5%), the 2012 Adverse Drug Events Drug-Reaction dataset (F1 of 90.0, +6.3%), and the 2018 n2c2 Posology Relations dataset (F1 of 96.7, +0.6%). Third, we show two practical applications of this framework — for building a biomedical knowledge graph and for improving the accuracy of mapping entities to clinical codes. The system is built using the Spark NLP library which provides a production-grade, natively scalable, hardware-optimized, trainable & tunable NLP framework.

Adverse drug reactions / events (ADR/ADE) have a major impact on patient health and health care costs. Detecting ADR’s as early as possible and sharing them with regulators, pharma companies, and healthcare providers can prevent morbidity and save many lives. While most ADR’s are not reported via formal channels, they are often documented in a variety of unstructured conversations such as social media posts by patients, customer support call transcripts, or CRM notes of meetings between healthcare providers and pharma sales reps. In this paper, we propose a natural language processing (NLP) solution that detects ADR’s in such unstructured free-text conversations, which improves on previous work in three ways. First, a new Named Entity Recognition (NER) model obtains new state-of-the-art accuracy for ADR and Drug entity extraction on the ADE, CADEC, and SMM4H benchmark datasets (91.75%, 78.76%, and 83.41% F1 scores respectively).

Second, two new Relation Extraction (RE) models are introduced – one based on BioBERT while the other utilizing crafted features over a Fully Connected Neural Network (FCNN) – are shown to perform on par with existing state-of-the-art models, and outperform them when trained with a supplementary clinician-annotated RE dataset. Third, a new text classification model, for deciding if a conversation includes an ADR, obtains new state-of-the-art accuracy on the CADEC dataset (86.69% F1 score). The complete solution is implemented as a unified NLP pipeline in a production-grade library built on top of Apache Spark, making it natively scalable and able to process millions of batch or streaming records on commodity clusters.

Named entity recognition (NER) is a widely applicable natural language processing task and building block of question answering, topic modeling, information retrieval, etc. In the medical domain, NER plays a crucial role by extracting meaningful chunks from clinical notes and reports, which are then fed to downstream tasks like assertion status detection, entity resolution, relation extraction, and de-identification. Reimplementing a Bi-LSTM-CNN-Char deep learning architecture on top of Apache Spark, we present a single trainable NER model that obtains new state-of-the-art results on seven public biomedical benchmarks without using heavy contextual embeddings like BERT.

This includes improving BC4CHEMD to 93.72% (4.1% gain), Species800 to 80.91% (4.6% gain), and JNLPBA to 81.29% (5.2% gain). In addition, this model is freely available within a production-grade code base as part of the open-source Spark NLP library; can scale up for training and inference in any Spark cluster; has GPU support and libraries for popular programming languages such as Python, R, Scala and Java; and can be extended to support other human languages with no code changes.

Abstract: The clinical application of detecting COVID-19 factors is a challenging task. The existing named entity recognition models are usually trained on a limited set of named entities. Besides clinical, the non-clinical factors, such as social determinant of health (SDoH), are also important to study the infectious disease. In this paper, we propose a generalizable machine learning approach that improves on previous efforts by recognizing a large number of clinical risk factors and SDoH.

The novelty of the proposed method lies in the subtle combination of a number of deep neural networks, including the BiLSTM-CNN-CRF method and a transformer-based embedding layer. Experimental results on a cohort of COVID-19 data prepared from PubMed articles show the superiority of the proposed approach. When compared to other methods, the proposed approach achieves a performance gain of about 1–5% in terms of macro- and micro-average F1 scores. Clinical practitioners and researchers can use this approach to obtain accurate information regarding clinical risks and SDoH factors, and use this pipeline as a tool to end the pandemic or to prepare for future pandemics.

Following the global COVID-19 pandemic, the number of scientific papers studying the virus has grown massively, leading to increased interest in automated literate review. We present a clinical text mining system that improves on previous efforts in three ways. First, it can recognize over 100 different entity types including social determinants of health, anatomy, risk factors, and adverse events in addition to other commonly used clinical and biomedical entities.

Second, the text processing pipeline includes assertion status detection, to distinguish between clinical facts that are present, absent, conditional, or about someone other than the patient. Third, the deep learning models used are more accurate than previously available, leveraging an integrated pipeline of state-of-the-art pretrained named entity recognition models, and improving on the previous best performing benchmarks for assertion status detection.

We illustrate extracting trends and insights, e.g. most frequent disorders and symptoms, and most common vital signs and EKG findings, from the COVID-19 Open Research Dataset (CORD-19). The system is built using the Spark NLP library which natively supports scaling to use distributed clusters, leveraging GPUs, configurable and reusable NLP pipelines, healthcare specific embeddings, and the ability to train models to support new entity types or human languages with no code changes.

Named entity recognition (NER) is one of the most important building blocks of NLP tasks in the medical domain by extracting meaningful chunks from clinical notes and reports, which are then fed to downstream tasks like assertion status detection, entity resolution, relation extraction, and de-identification. Due to the growing volume of healthcare data in unstructured format, an increasingly important challenge is providing high accuracy implementations of state-of-the-art deep learning (DL) algorithms at scale.

While recent advances in NLP like Transformers and BERT have pushed the boundaries for accuracy, these methods are significantly slow and difficult to scale on millions of records.

In this study, we introduce an agile, production-grade clinical and biomedical NER algorithm based on a modified BiLSTM-CNN-Char DL architecture built on top of Apache Spark.

Our NER implementation establishes new state-of-the-art accuracy on 7 of 8 well-known biomedical NER benchmarks and 3 clinical concept extraction challenges: 2010 i2b2/VA clinical concept extraction, 2014 n2c2 de-identification, and 2018 n2c2 medication extraction. Moreover, clinical NER models trained using this implementation outperform the accuracy of commercial entity extraction solutions such AWS Medical Comprehend and Google Cloud Healthcare API by a large margin (8.9% and 6.7% respectively), without using memory-intensive language models.

The proposed model requires no handcrafted features or task-specific resources, requires minimal hyperparameter tuning for a given dataset from any domain, can be trained with any embeddings including BERT, and can be trained to support more human languages with no code changes. It is available within a production-grade code base as part of the Spark NLP library, the only open-source NLP library that can scale to make use of a Spark cluster for training and inference, has GPU support, and provides libraries for Python, R, Scala and Java.

Developing prediction models for emerging infectious diseases from relatively small numbers of cases is a critical need for improving pandemic preparedness. Using COVID-19 as an exemplar, we propose a transfer learning methodology for developing predictive models from multi-modal electronic healthcare records by leveraging information from more prevalent diseases with shared clinical characteristics. Our novel hierarchical, multi-modal model (TRANSMED) integrates baseline risk factors from the natural language processing of clinical notes at admission, time-series measurements of biomarkers obtained from laboratory tests, and discrete diagnostic, procedure and drug codes.

We demonstrate the alignment of TRANSMED’s predictions with well-established clinical knowledge about COVID-19 through univariate and multivariate risk factor driven sub-cohort analysis. TRANSMED’s superior performance over state-of-the-art methods shows that leveraging patient data across modalities and transferring prior knowledge from similar disorders is critical for accurate prediction of patient outcomes, and this approach may serve as an important tool in the early response to future pandemics.

Artificial intelligence (AI) can be applied in the practice of gastroenterology to acquire and analyze information. Besides speed and duplicability, AI has the potential of also offering insight with results that surpass medical specialists. Natural language processing (NLP) is being used to extract information from text, organize and categorize documents. Processing unstructured data with NLP will result in structured data and medical codes can be extracted more easily (ICD10, medical procedure codes, etc) for reimbursement purposes among others. Recent research is studying the use of AI for automated interpretation of text from endoscopy and medical documents for better quality and patient phenotyping as well as enhanced detection and descriptions of endoscopic lesions such as colon polyps. In this paper, we present a method of extracting medical data using Spark NLP (John Snow Labs, DE, USA), by annotating endoscopy reports and training a model to automatically extract labels in order to obtain
structured medical data. This can be used in combination with other forms of structured data for an optimal and novel patient profiling.