Expert guided natural language processing using one-class classification

Introduction Automatically identifying specific phenotypes in free-text clinical notes is critically important for the reuse of clinical data. In this study, the authors combine expert-guided feature (text) selection with one-class classification for text processing.Objectives To compare the performance of one-class classification to traditional binary classification; to evaluate the utility of feature selection based on expert-selected salient text (snippets); and to determine the robustness of these models with respects to irrelevant surrounding text.Methods The authors trained one-class support vector machines (1C-SVMs) and two-class SVMs (2C-SVMs) to identify notes discussing breast cancer. Manually annotated visit summary notes (88 positive and 88 negative for breast cancer) were used to compare the performance of models trained on whole notes labeled as positive or negative to models trained on expert-selected text sections (snippets) relevant to breast cancer status. Model performance was evaluated using a 70:30 split for 20 iterations and on a realistic dataset of 10 000 records with a breast cancer prevalence of 1.4%.Results When tested on a balanced experimental dataset, 1C-SVMs trained on snippets had comparable results to 2C-SVMs trained on whole notes (F = 0.92 for both approaches). When evaluated on a realistic imbalanced dataset, 1C-SVMs had a considerably superior performance (F = 0.61 vs. F = 0.17 for the best performing model) attributable mainly to improved precision (p = .88 vs. p = .09 for the best performing model).Conclusions 1C-SVMs trained on expert-selected relevant text sections perform better than 2C-SVMs classifiers trained on either snippets or whole notes when applied to realistically imbalanced data with low prevalence of the positive class.     

Research and applications: A comprehensive study of named entity recognition in Chinese clinical text

Objective

Named entity recognition (NER) is one of the fundamental tasks in natural language processing. In the medical domain, there have been a number of studies on NER in English clinical notes; however, very limited NER research has been carried out on clinical notes written in Chinese. The goal of this study was to systematically investigate features and machine learning algorithms for NER in Chinese clinical text.

Materials and methods

We randomly selected 400 admission notes and 400 discharge summaries from Peking Union Medical College Hospital in China. For each note, four types of entity—clinical problems, procedures, laboratory test, and medications—were annotated according to a predefined guideline. Two-thirds of the 400 notes were used to train the NER systems and one-third for testing. We investigated the effects of different types of feature including bag-of-characters, word segmentation, part-of-speech, and section information, and different machine learning algorithms including conditional random fields (CRF), support vector machines (SVM), maximum entropy (ME), and structural SVM (SSVM) on the Chinese clinical NER task. All classifiers were trained on the training dataset and evaluated on the test set, and micro-averaged precision, recall, and F-measure were reported.

Results

Our evaluation on the independent test set showed that most types of feature were beneficial to Chinese NER systems, although the improvements were limited. The system achieved the highest performance by combining word segmentation and section information, indicating that these two types of feature complement each other. When the same types of optimized feature were used, CRF and SSVM outperformed SVM and ME. More specifically, SSVM achieved the highest performance of the four algorithms, with F-measures of 93.51% and 90.01% for admission notes and discharge summaries, respectively.     

CancerBERT: a cancer domain-specific language model for extracting breast cancer phenotypes from electronic health records

ObjectiveAccurate extraction of breast cancer patients’ phenotypes is important for clinical decision support and clinical research. This study developed and evaluated cancer domain pretrained CancerBERT models for extracting breast cancer phenotypes from clinical texts. We also investigated the effect of customized cancer-related vocabulary on the performance of CancerBERT models.Materials and MethodsA cancer-related corpus of breast cancer patients was extracted from the electronic health records of a local hospital. We annotated named entities in 200 pathology reports and 50 clinical notes for 8 cancer phenotypes for fine-tuning and evaluation. We kept pretraining the BlueBERT model on the cancer corpus with expanded vocabularies (using both term frequency-based and manually reviewed methods) to obtain CancerBERT models. The CancerBERT models were evaluated and compared with other baseline models on the cancer phenotype extraction task.ResultsAll CancerBERT models outperformed all other models on the cancer phenotyping NER task. Both CancerBERT models with customized vocabularies outperformed the CancerBERT with the original BERT vocabulary. The CancerBERT model with manually reviewed customized vocabulary achieved the best performance with macro F1 scores equal to 0.876 (95% CI, 0.873–0.879) and 0.904 (95% CI, 0.902–0.906) for exact match and lenient match, respectively.ConclusionsThe CancerBERT models were developed to extract the cancer phenotypes in clinical notes and pathology reports. The results validated that using customized vocabulary may further improve the performances of domain specific BERT models in clinical NLP tasks. The CancerBERT models developed in the study would further help clinical decision support.     

Research and applications: Syntactic parsing of clinical text: guideline and corpus development with handling ill-formed sentences

Objective

To develop, evaluate, and share: (1) syntactic parsing guidelines for clinical text, with a new approach to handling ill-formed sentences; and (2) a clinical Treebank annotated according to the guidelines. To document the process and findings for readers with similar interest.

Methods

Using random samples from a shared natural language processing challenge dataset, we developed a handbook of domain-customized syntactic parsing guidelines based on iterative annotation and adjudication between two institutions. Special considerations were incorporated into the guidelines for handling ill-formed sentences, which are common in clinical text. Intra- and inter-annotator agreement rates were used to evaluate consistency in following the guidelines. Quantitative and qualitative properties of the annotated Treebank, as well as its use to retrain a statistical parser, were reported.

Results

A supplement to the Penn Treebank II guidelines was developed for annotating clinical sentences. After three iterations of annotation and adjudication on 450 sentences, the annotators reached an F-measure agreement rate of 0.930 (while intra-annotator rate was 0.948) on a final independent set. A total of 1100 sentences from progress notes were annotated that demonstrated domain-specific linguistic features. A statistical parser retrained with combined general English (mainly news text) annotations and our annotations achieved an accuracy of 0.811 (higher than models trained purely with either general or clinical sentences alone). Both the guidelines and syntactic annotations are made available at https://sourceforge.net/projects/medicaltreebank.

Conclusions

We developed guidelines for parsing clinical text and annotated a corpus accordingly. The high intra- and inter-annotator agreement rates showed decent consistency in following the guidelines. The corpus was shown to be useful in retraining a statistical parser that achieved moderate accuracy.     

Domain adaptation for semantic role labeling of clinical text

Objective Semantic role labeling (SRL), which extracts a shallow semantic relation representation from different surface textual forms of free text sentences, is important for understanding natural language. Few studies in SRL have been conducted in the medical domain, primarily due to lack of annotated clinical SRL corpora, which are time-consuming and costly to build. The goal of this study is to investigate domain adaptation techniques for clinical SRL leveraging resources built from newswire and biomedical literature to improve performance and save annotation costs.Materials and Methods Multisource Integrated Platform for Answering Clinical Questions (MiPACQ), a manually annotated SRL clinical corpus, was used as the target domain dataset. PropBank and NomBank from newswire and BioProp from biomedical literature were used as source domain datasets. Three state-of-the-art domain adaptation algorithms were employed: instance pruning, transfer self-training, and feature augmentation. The SRL performance using different domain adaptation algorithms was evaluated by using 10-fold cross-validation on the MiPACQ corpus. Learning curves for the different methods were generated to assess the effect of sample size.Results and Conclusion When all three source domain corpora were used, the feature augmentation algorithm achieved statistically significant higher F-measure (83.18%), compared to the baseline with MiPACQ dataset alone (F-measure, 81.53%), indicating that domain adaptation algorithms may improve SRL performance on clinical text. To achieve a comparable performance to the baseline method that used 90% of MiPACQ training samples, the feature augmentation algorithm required <50% of training samples in MiPACQ, demonstrating that annotation costs of clinical SRL can be reduced significantly by leveraging existing SRL resources from other domains.     

Biomedical negation scope detection with conditional random fields

Objective

Negation is a linguistic phenomenon that marks the absence of an entity or event. Negated events are frequently reported in both biological literature and clinical notes. Text mining applications benefit from the detection of negation and its scope. However, due to the complexity of language, identifying the scope of negation in a sentence is not a trivial task.

Design

Conditional random fields (CRF), a supervised machine-learning algorithm, were used to train models to detect negation cue phrases and their scope in both biological literature and clinical notes. The models were trained on the publicly available BioScope corpus.

Measurement

The performance of the CRF models was evaluated on identifying the negation cue phrases and their scope by calculating recall, precision and F1-score. The models were compared with four competitive baseline systems.

Results

The best CRF-based model performed statistically better than all baseline systems and NegEx, achieving an F1-score of 98% and 95% on detecting negation cue phrases and their scope in clinical notes, and an F1-score of 97% and 85% on detecting negation cue phrases and their scope in biological literature.

Conclusions

This approach is robust, as it can identify negation scope in both biological and clinical text. To benefit text mining applications, the system is publicly available as a Java API and as an online application at http://negscope.askhermes.org.     

Optimizing the synthesis of clinical trial data using sequential trees

ObjectiveWith the growing demand for sharing clinical trial data, scalable methods to enable privacy protective access to high-utility data are needed. Data synthesis is one such method. Sequential trees are commonly used to synthesize health data. It is hypothesized that the utility of the generated data is dependent on the variable order. No assessments of the impact of variable order on synthesized clinical trial data have been performed thus far. Through simulation, we aim to evaluate the variability in the utility of synthetic clinical trial data as variable order is randomly shuffled and implement an optimization algorithm to find a good order if variability is too high.Materials and MethodsSix oncology clinical trial datasets were evaluated in a simulation. Three utility metrics were computed comparing real and synthetic data: univariate similarity, similarity in multivariate prediction accuracy, and a distinguishability metric. Particle swarm was implemented to optimize variable order, and was compared with a curriculum learning approach to ordering variables.ResultsAs the number of variables in a clinical trial dataset increases, there is a pattern of a marked increase in variability of data utility with order. Particle swarm with a distinguishability hinge loss ensured adequate utility across all 6 datasets. The hinge threshold was selected to avoid overfitting which can create a privacy problem. This was superior to curriculum learning in terms of utility.ConclusionsThe optimization approach presented in this study gives a reliable way to synthesize high-utility clinical trial datasets.     

Extracting seizure frequency from epilepsy clinic notes: a machine reading approach to natural language processing

ObjectiveSeizure frequency and seizure freedom are among the most important outcome measures for patients with epilepsy. In this study, we aimed to automatically extract this clinical information from unstructured text in clinical notes. If successful, this could improve clinical decision-making in epilepsy patients and allow for rapid, large-scale retrospective research.Materials and MethodsWe developed a finetuning pipeline for pretrained neural models to classify patients as being seizure-free and to extract text containing their seizure frequency and date of last seizure from clinical notes. We annotated 1000 notes for use as training and testing data and determined how well 3 pretrained neural models, BERT, RoBERTa, and Bio_ClinicalBERT, could identify and extract the desired information after finetuning.ResultsThe finetuned models (BERT_FT, Bio_ClinicalBERT_FT, and RoBERTa_FT) achieved near-human performance when classifying patients as seizure free, with BERT_FT and Bio_ClinicalBERT_FT achieving accuracy scores over 80%. All 3 models also achieved human performance when extracting seizure frequency and date of last seizure, with overall F₁ scores over 0.80. The best combination of models was Bio_ClinicalBERT_FT for classification, and RoBERTa_FT for text extraction. Most of the gains in performance due to finetuning required roughly 70 annotated notes.Discussion and ConclusionOur novel machine reading approach to extracting important clinical outcomes performed at or near human performance on several tasks. This approach opens new possibilities to support clinical practice and conduct large-scale retrospective clinical research. Future studies can use our finetuning pipeline with minimal training annotations to answer new clinical questions.     

SynTEG: a framework for temporal structured electronic health data simulation

ObjectiveSimulating electronic health record data offers an opportunity to resolve the tension between data sharing and patient privacy. Recent techniques based on generative adversarial networks have shown promise but neglect the temporal aspect of healthcare. We introduce a generative framework for simulating the trajectory of patients’ diagnoses and measures to evaluate utility and privacy.Materials and MethodsThe framework simulates date-stamped diagnosis sequences based on a 2-stage process that 1) sequentially extracts temporal patterns from clinical visits and 2) generates synthetic data conditioned on the learned patterns. We designed 3 utility measures to characterize the extent to which the framework maintains feature correlations and temporal patterns in clinical events. We evaluated the framework with billing codes, represented as phenome-wide association study codes (phecodes), from over 500 000 Vanderbilt University Medical Center electronic health records. We further assessed the privacy risks based on membership inference and attribute disclosure attacks.ResultsThe simulated temporal sequences exhibited similar characteristics to real sequences on the utility measures. Notably, diagnosis prediction models based on real versus synthetic temporal data exhibited an average relative difference in area under the ROC curve of 1.6% with standard deviation of 3.8% for 1276 phecodes. Additionally, the relative difference in the mean occurrence age and time between visits were 4.9% and 4.2%, respectively. The privacy risks in synthetic data, with respect to the membership and attribute inference were negligible.ConclusionThis investigation indicates that temporal diagnosis code sequences can be simulated in a manner that provides utility and respects privacy.     

UMLS-based data augmentation for natural language processing of clinical research literature

ObjectiveThe study sought to develop and evaluate a knowledge-based data augmentation method to improve the performance of deep learning models for biomedical natural language processing by overcoming training data scarcity.Materials and MethodsWe extended the easy data augmentation (EDA) method for biomedical named entity recognition (NER) by incorporating the Unified Medical Language System (UMLS) knowledge and called this method UMLS-EDA. We designed experiments to systematically evaluate the effect of UMLS-EDA on popular deep learning architectures for both NER and classification. We also compared UMLS-EDA to BERT.ResultsUMLS-EDA enables substantial improvement for NER tasks from the original long short-term memory conditional random fields (LSTM-CRF) model (micro-F1 score: +5%, + 17%, and +15%), helps the LSTM-CRF model (micro-F1 score: 0.66) outperform LSTM-CRF with transfer learning by BERT (0.63), and improves the performance of the state-of-the-art sentence classification model. The largest gain on micro-F1 score is 9%, from 0.75 to 0.84, better than classifiers with BERT pretraining (0.82).ConclusionsThis study presents a UMLS-based data augmentation method, UMLS-EDA. It is effective at improving deep learning models for both NER and sentence classification, and contributes original insights for designing new, superior deep learning approaches for low-resource biomedical domains.     

Clinical concept extraction using transformers

ObjectiveThe goal of this study is to explore transformer-based models (eg, Bidirectional Encoder Representations from Transformers [BERT]) for clinical concept extraction and develop an open-source package with pretrained clinical models to facilitate concept extraction and other downstream natural language processing (NLP) tasks in the medical domain.MethodsWe systematically explored 4 widely used transformer-based architectures, including BERT, RoBERTa, ALBERT, and ELECTRA, for extracting various types of clinical concepts using 3 public datasets from the 2010 and 2012 i2b2 challenges and the 2018 n2c2 challenge. We examined general transformer models pretrained using general English corpora as well as clinical transformer models pretrained using a clinical corpus and compared them with a long short-term memory conditional random fields (LSTM-CRFs) mode as a baseline. Furthermore, we integrated the 4 clinical transformer-based models into an open-source package.Results and ConclusionThe RoBERTa-MIMIC model achieved state-of-the-art performance on 3 public clinical concept extraction datasets with F1-scores of 0.8994, 0.8053, and 0.8907, respectively. Compared to the baseline LSTM-CRFs model, RoBERTa-MIMIC remarkably improved the F1-score by approximately 4% and 6% on the 2010 and 2012 i2b2 datasets. This study demonstrated the efficiency of transformer-based models for clinical concept extraction. Our methods and systems can be applied to other clinical tasks. The clinical transformer package with 4 pretrained clinical models is publicly available at https://github.com/uf-hobi-informatics-lab/ClinicalTransformerNER. We believe this package will improve current practice on clinical concept extraction and other tasks in the medical domain.     

Using nursing notes to improve clinical outcome prediction in intensive care patients: A retrospective cohort study

ObjectiveElectronic health record documentation by intensive care unit (ICU) clinicians may predict patient outcomes. However, it is unclear whether physician and nursing notes differ in their ability to predict short-term ICU prognosis. We aimed to investigate and compare the ability of physician and nursing notes, written in the first 48 hours of admission, to predict ICU length of stay and mortality using 3 analytical methods.Materials and MethodsThis was a retrospective cohort study with split sampling for model training and testing. We included patients ≥18 years of age admitted to the ICU at Beth Israel Deaconess Medical Center in Boston, Massachusetts, from 2008 to 2012. Physician or nursing notes generated within the first 48 hours of admission were used with standard machine learning methods to predict outcomes.ResultsFor the primary outcome of composite score of ICU length of stay ≥7 days or in-hospital mortality, the gradient boosting model had better performance than the logistic regression and random forest models. Nursing and physician notes achieved area under the curves (AUCs) of 0.826 and 0.796, respectively, with even better predictive power when combined (AUC, 0.839).DiscussionModels using only nursing notes more accurately predicted short-term prognosis than did models using only physician notes, but in combination, the models achieved the greatest accuracy in prediction. ConclusionsOur findings demonstrate that statistical models derived from text analysis in the first 48 hours of ICU admission can predict patient outcomes. Physicians’ and nurses’ notes are both uniquely important in mortality prediction and combining these notes can produce a better predictive model.     

A sequence labeling approach to link medications and their attributes in clinical notes and clinical trial announcements for information extraction

Objective

The goal of this work was to evaluate machine learning methods, binary classification and sequence labeling, for medication–attribute linkage detection in two clinical corpora.

Data and methods

We double annotated 3000 clinical trial announcements (CTA) and 1655 clinical notes (CN) for medication named entities and their attributes. A binary support vector machine (SVM) classification method with parsimonious feature sets, and a conditional random fields (CRF)-based multi-layered sequence labeling (MLSL) model were proposed to identify the linkages between the entities and their corresponding attributes. We evaluated the system''s performance against the human-generated gold standard.

Results

The experiments showed that the two machine learning approaches performed statistically significantly better than the baseline rule-based approach. The binary SVM classification achieved 0.94 F-measure with individual tokens as features. The SVM model trained on a parsimonious feature set achieved 0.81 F-measure for CN and 0.87 for CTA. The CRF MLSL method achieved 0.80 F-measure on both corpora.

Discussion and conclusions

We compared the novel MLSL method with a binary classification and a rule-based method. The MLSL method performed statistically significantly better than the rule-based method. However, the SVM-based binary classification method was statistically significantly better than the MLSL method for both the CTA and CN corpora. Using parsimonious feature sets both the SVM-based binary classification and CRF-based MLSL methods achieved high performance in detecting medication name and attribute linkages in CTA and CN.     

Effects of personal identifier resynthesis on clinical text de-identification

Objective

De-identified medical records are critical to biomedical research. Text de-identification software exists, including “resynthesis” components that replace real identifiers with synthetic identifiers. The goal of this research is to evaluate the effectiveness and examine possible bias introduced by resynthesis on de-identification software.

Design

We evaluated the open-source MITRE Identification Scrubber Toolkit, which includes a resynthesis capability, with clinical text from Vanderbilt University Medical Center patient records. We investigated four record classes from over 500 patients'' files, including laboratory reports, medication orders, discharge summaries and clinical notes. We trained and tested the de-identification tool on real and resynthesized records.

Measurements

We measured performance in terms of precision, recall, F-measure and accuracy for the detection of protected health identifiers as designated by the HIPAA Safe Harbor Rule.

Results

The de-identification tool was trained and tested on a collection of real and resynthesized Vanderbilt records. Results for training and testing on the real records were 0.990 accuracy and 0.960 F-measure. The results improved when trained and tested on resynthesized records with 0.998 accuracy and 0.980 F-measure but deteriorated moderately when trained on real records and tested on resynthesized records with 0.989 accuracy 0.862 F-measure. Moreover, the results declined significantly when trained on resynthesized records and tested on real records with 0.942 accuracy and 0.728 F-measure.

Conclusion

The de-identification tool achieves high accuracy when training and test sets are homogeneous (ie, both real or resynthesized records). The resynthesis component regularizes the data to make them less “realistic,” resulting in loss of performance particularly when training on resynthesized data and testing on real data.     

Use of unstructured text in prognostic clinical prediction models: a systematic review

ObjectiveThis systematic review aims to assess how information from unstructured text is used to develop and validate clinical prognostic prediction models. We summarize the prediction problems and methodological landscape and determine whether using text data in addition to more commonly used structured data improves the prediction performance.Materials and MethodsWe searched Embase, MEDLINE, Web of Science, and Google Scholar to identify studies that developed prognostic prediction models using information extracted from unstructured text in a data-driven manner, published in the period from January 2005 to March 2021. Data items were extracted, analyzed, and a meta-analysis of the model performance was carried out to assess the added value of text to structured-data models.ResultsWe identified 126 studies that described 145 clinical prediction problems. Combining text and structured data improved model performance, compared with using only text or only structured data. In these studies, a wide variety of dense and sparse numeric text representations were combined with both deep learning and more traditional machine learning methods. External validation, public availability, and attention for the explainability of the developed models were limited.ConclusionThe use of unstructured text in the development of prognostic prediction models has been found beneficial in addition to structured data in most studies. The text data are source of valuable information for prediction model development and should not be neglected. We suggest a future focus on explainability and external validation of the developed models, promoting robust and trustworthy prediction models in clinical practice.     

BoB,a best-of-breed automated text de-identification system for VHA clinical documents

Objective

De-identification allows faster and more collaborative clinical research while protecting patient confidentiality. Clinical narrative de-identification is a tedious process that can be alleviated by automated natural language processing methods. The goal of this research is the development of an automated text de-identification system for Veterans Health Administration (VHA) clinical documents.

Materials and methods

We devised a novel stepwise hybrid approach designed to improve the current strategies used for text de-identification. The proposed system is based on a previous study on the best de-identification methods for VHA documents. This best-of-breed automated clinical text de-identification system (aka BoB) tackles the problem as two separate tasks: (1) maximize patient confidentiality by redacting as much protected health information (PHI) as possible; and (2) leave de-identified documents in a usable state preserving as much clinical information as possible.

Results

We evaluated BoB with a manually annotated corpus of a variety of VHA clinical notes, as well as with the 2006 i2b2 de-identification challenge corpus. We present evaluations at the instance- and token-level, with detailed results for BoB''s main components. Moreover, an existing text de-identification system was also included in our evaluation.

Discussion

BoB''s design efficiently takes advantage of the methods implemented in its pipeline, resulting in high sensitivity values (especially for sensitive PHI categories) and a limited number of false positives.

Conclusions

Our system successfully addressed VHA clinical document de-identification, and its hybrid stepwise design demonstrates robustness and efficiency, prioritizing patient confidentiality while leaving most clinical information intact.     

MT-clinical BERT: scaling clinical information extraction with multitask learning

ObjectiveClinical notes contain an abundance of important, but not-readily accessible, information about patients. Systems that automatically extract this information rely on large amounts of training data of which there exists limited resources to create. Furthermore, they are developed disjointly, meaning that no information can be shared among task-specific systems. This bottleneck unnecessarily complicates practical application, reduces the performance capabilities of each individual solution, and associates the engineering debt of managing multiple information extraction systems.Materials and MethodsWe address these challenges by developing Multitask-Clinical BERT: a single deep learning model that simultaneously performs 8 clinical tasks spanning entity extraction, personal health information identification, language entailment, and similarity by sharing representations among tasks.ResultsWe compare the performance of our multitasking information extraction system to state-of-the-art BERT sequential fine-tuning baselines. We observe a slight but consistent performance degradation in MT-Clinical BERT relative to sequential fine-tuning.DiscussionThese results intuitively suggest that learning a general clinical text representation capable of supporting multiple tasks has the downside of losing the ability to exploit dataset or clinical note-specific properties when compared to a single, task-specific model.ConclusionsWe find our single system performs competitively with all state-the-art task-specific systems while also benefiting from massive computational benefits at inference.     

Building longitudinal medication dose data using medication information extracted from clinical notes in electronic health records

ObjectiveTo develop an algorithm for building longitudinal medication dose datasets using information extracted from clinical notes in electronic health records (EHRs).Materials and MethodsWe developed an algorithm that converts medication information extracted using natural language processing (NLP) into a usable format and builds longitudinal medication dose datasets. We evaluated the algorithm on 2 medications extracted from clinical notes of Vanderbilt’s EHR and externally validated the algorithm using clinical notes from the MIMIC-III clinical care database.ResultsFor the evaluation using Vanderbilt’s EHR data, the performance of our algorithm was excellent; F1-measures were ≥0.98 for both dose intake and daily dose. For the external validation using MIMIC-III, the algorithm achieved F1-measures ≥0.85 for dose intake and ≥0.82 for daily dose.DiscussionOur algorithm addresses the challenge of building longitudinal medication dose data using information extracted from clinical notes. Overall performance was excellent, but the algorithm can perform poorly when incorrect information is extracted by NLP systems. Although it performed reasonably well when applied to the external data source, its performance was worse due to differences in the way the drug information was written. The algorithm is implemented in the R package, “EHR,” and the extracted data from Vanderbilt’s EHRs along with the gold standards are provided so that users can reproduce the results and help improve the algorithm.ConclusionOur algorithm for building longitudinal dose data provides a straightforward way to use EHR data for medication-based studies. The external validation results suggest its potential for applicability to other systems.     

Extracting social determinants of health from electronic health records using natural language processing: a systematic review

ObjectiveSocial determinants of health (SDoH) are nonclinical dispositions that impact patient health risks and clinical outcomes. Leveraging SDoH in clinical decision-making can potentially improve diagnosis, treatment planning, and patient outcomes. Despite increased interest in capturing SDoH in electronic health records (EHRs), such information is typically locked in unstructured clinical notes. Natural language processing (NLP) is the key technology to extract SDoH information from clinical text and expand its utility in patient care and research. This article presents a systematic review of the state-of-the-art NLP approaches and tools that focus on identifying and extracting SDoH data from unstructured clinical text in EHRs.Materials and MethodsA broad literature search was conducted in February 2021 using 3 scholarly databases (ACL Anthology, PubMed, and Scopus) following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A total of 6402 publications were initially identified, and after applying the study inclusion criteria, 82 publications were selected for the final review.ResultsSmoking status (n = 27), substance use (n = 21), homelessness (n = 20), and alcohol use (n = 15) are the most frequently studied SDoH categories. Homelessness (n = 7) and other less-studied SDoH (eg, education, financial problems, social isolation and support, family problems) are mostly identified using rule-based approaches. In contrast, machine learning approaches are popular for identifying smoking status (n = 13), substance use (n = 9), and alcohol use (n = 9).ConclusionNLP offers significant potential to extract SDoH data from narrative clinical notes, which in turn can aid in the development of screening tools, risk prediction models, and clinical decision support systems.     

Development of a predictive model for retention in HIV care using natural language processing of clinical notes

ObjectiveAdherence to a treatment plan from HIV-positive patients is necessary to decrease their mortality and improve their quality of life, however some patients display poor appointment adherence and become lost to follow-up (LTFU). We applied natural language processing (NLP) to analyze indications towards or against LTFU in HIV-positive patients’ notes.Materials and MethodsUnstructured lemmatized notes were labeled with an LTFU or Retained status using a 183-day threshold. An NLP and supervised machine learning system with a linear model and elastic net regularization was trained to predict this status. Prevalence of characteristics domains in the learned model weights were evaluated.ResultsWe analyzed 838 LTFU vs 2964 Retained notes and obtained a weighted F1 mean of 0.912 via nested cross-validation; another experiment with notes from the same patients in both classes showed substantially lower metrics. “Comorbidities” were associated with LTFU through, for instance, “HCV” (hepatitis C virus) and likewise “Good adherence” with Retained, represented with “Well on ART” (antiretroviral therapy).DiscussionMentions of mental health disorders and substance use were associated with disparate retention outcomes, however history vs active use was not investigated. There remains further need to model transitions between LTFU and being retained in care over time.ConclusionWe provided an important step for the future development of a model that could eventually help to identify patients who are at risk for falling out of care and to analyze which characteristics could be factors for this. Further research is needed to enhance this method with structured electronic medical record fields.