Evaluating Natural Language Processing Packages for Predicting Hospital-Acquired Pressure Injuries From Clinical Notes.

Incidence of hospital-acquired pressure injury, a key indicator of nursing quality, is directly proportional to adverse outcomes, increased hospital stays, and economic burdens on patients, caregivers, and society. Thus, predicting hospital-acquired pressure injury is important. Prediction models use structured data more often than unstructured notes, although the latter often contain useful patient information. We hypothesize that unstructured notes, such as nursing notes, can predict hospital-acquired pressure injury. We evaluate the impact of using various natural language processing packages to identify salient patient information from unstructured text. We use named entity recognition to identify keywords, which comprise the feature space of our classifier for hospital-acquired pressure injury prediction. We compare scispaCy and Stanza, two different named entity recognition models, using unstructured notes in Medical Information Mart for Intensive Care III, a publicly available ICU data set. To assess the impact of vocabulary size reduction, we compare the use of all clinical notes with only nursing notes. Our results suggest that named entity recognition extraction using nursing notes can yield accurate models. Moreover, the extracted keywords play a significant role in the prediction of hospital-acquired pressure injury.

Extracting medication information from unstructured public health data: a demonstration on data from population-based and tertiary-based samples.

BACKGROUND: Unstructured data from clinical epidemiological studies can be valuable and easy to obtain. However, it requires further extraction and processing for data analysis. Doing this manually is labor-intensive, slow and subject to error. In this study, we propose an automation framework for extracting and processing unstructured data. METHODS: The proposed automation framework consisted of two natural language processing (NLP) based tools for unstructured text data for medications and reasons for medication use. We first checked spelling using a spell-check program trained on publicly available knowledge sources and then applied NLP techniques. We mapped medication names into generic names using vocabulary from publicly available knowledge sources. We used WHO's Anatomical Therapeutic Chemical (ATC) classification system to map generic medication names to medication classes. We processed the reasons for medication with the Lancaster stemmer method and then grouped and mapped to disease classes based on organ systems. Finally, we demonstrated this automation framework on two data sources for Mylagic Encephalomyelitis/ Chronic Fatigue Syndrome (ME/CFS): tertiary-based (n = 378) and population-based (n = 664) samples. RESULTS: A total of 8681 raw medication records were used for this demonstration. The 1266 distinct medication names (omitting supplements) were condensed to 89 ATC classification system categories. The 1432 distinct raw reasons for medication use were condensed to 65 categories via NLP. Compared to completion of the entire process manually, our automation process reduced the number of the terms requiring manual labor for mapping by 84.4% for medications and 59.4% for reasons for medication use. Additionally, this process improved the precision of the mapped results. CONCLUSIONS: Our automation framework demonstrates the usefulness of NLP strategies even when there is no established mapping database. For a less established database (e.g., reasons for medication use), the method is easily modifiable as new knowledge sources for mapping are introduced. The capability to condense large features into interpretable ones will be valuable for subsequent analytical studies involving techniques such as machine learning and data mining.

Phenotype Instance Verification and Evaluation Tool (PIVET): A Scaled Phenotype Evidence Generation Framework Using Web-Based Medical Literature.

BACKGROUND: Researchers are developing methods to automatically extract clinically relevant and useful patient characteristics from raw healthcare datasets. These characteristics, often capturing essential properties of patients with common medical conditions, are called computational phenotypes. Being generated by automated or semiautomated, data-driven methods, such potential phenotypes need to be validated as clinically meaningful (or not) before they are acceptable for use in decision making. OBJECTIVE: The objective of this study was to present Phenotype Instance Verification and Evaluation Tool (PIVET), a framework that uses co-occurrence analysis on an online corpus of publically available medical journal articles to build clinical relevance evidence sets for user-supplied phenotypes. PIVET adopts a conceptual framework similar to the pioneering prototype tool PheKnow-Cloud that was developed for the phenotype validation task. PIVET completely refactors each part of the PheKnow-Cloud pipeline to deliver vast improvements in speed without sacrificing the quality of the insights PheKnow-Cloud achieved. METHODS: PIVET leverages indexing in NoSQL databases to efficiently generate evidence sets. Specifically, PIVET uses a succinct representation of the phenotypes that corresponds to the index on the corpus database and an optimized co-occurrence algorithm inspired by the Aho-Corasick algorithm. We compare PIVET's phenotype representation with PheKnow-Cloud's by using PheKnow-Cloud's experimental setup. In PIVET's framework, we also introduce a statistical model trained on domain expert-verified phenotypes to automatically classify phenotypes as clinically relevant or not. Additionally, we show how the classification model can be used to examine user-supplied phenotypes in an online, rather than batch, manner. RESULTS: PIVET maintains the discriminative power of PheKnow-Cloud in terms of identifying clinically relevant phenotypes for the same corpus with which PheKnow-Cloud was originally developed, but PIVET's analysis is an order of magnitude faster than that of PheKnow-Cloud. Not only is PIVET much faster, it can be scaled to a larger corpus and still retain speed. We evaluated multiple classification models on top of the PIVET framework and found ridge regression to perform best, realizing an average F1 score of 0.91 when predicting clinically relevant phenotypes. CONCLUSIONS: Our study shows that PIVET improves on the most notable existing computational tool for phenotype validation in terms of speed and automation and is comparable in terms of accuracy.

Limestone: high-throughput candidate phenotype generation via tensor factorization.

The rapidly increasing availability of electronic health records (EHRs) from multiple heterogeneous sources has spearheaded the adoption of data-driven approaches for improved clinical research, decision making, prognosis, and patient management. Unfortunately, EHR data do not always directly and reliably map to medical concepts that clinical researchers need or use. Some recent studies have focused on EHR-derived phenotyping, which aims at mapping the EHR data to specific medical concepts; however, most of these approaches require labor intensive supervision from experienced clinical professionals. Furthermore, existing approaches are often disease-centric and specialized to the idiosyncrasies of the information technology and/or business practices of a single healthcare organization. In this paper, we propose Limestone, a nonnegative tensor factorization method to derive phenotype candidates with virtually no human supervision. Limestone represents the data source interactions naturally using tensors (a generalization of matrices). In particular, we investigate the interaction of diagnoses and medications among patients. The resulting tensor factors are reported as phenotype candidates that automatically reveal patient clusters on specific diagnoses and medications. Using the proposed method, multiple phenotypes can be identified simultaneously from data. We demonstrate the capability of Limestone on a cohort of 31,815 patient records from the Geisinger Health System. The dataset spans 7years of longitudinal patient records and was initially constructed for a heart failure onset prediction study. Our experiments demonstrate the robustness, stability, and the conciseness of Limestone-derived phenotypes. Our results show that using only 40 phenotypes, we can outperform the original 640 features (169 diagnosis categories and 471 medication types) to achieve an area under the receiver operator characteristic curve (AUC) of 0.720 (95% CI 0.715 to 0.725). Moreover, in consultation with a medical expert, we confirmed 82% of the top 50 candidates automatically extracted by Limestone are clinically meaningful.

An AdaBoost-based algorithm to detect hospital-acquired pressure injury in the presence of conflicting annotations.

Hospital-acquired pressure injury is one of the most harmful events in clinical settings. Patients who do not receive early prevention and treatment can experience a significant financial burden and physical trauma. Several hospital-acquired pressure injury prediction algorithms have been developed to tackle this problem, but these models assume a consensus, gold-standard label (i.e., presence of pressure injury or not) is present for all training data. Existing definitions for identifying hospital-acquired pressure injuries are inconsistent due to the lack of high-quality documentation surrounding pressure injuries. To address this issue, we propose in this paper an ensemble-based algorithm that leverages truth inference methods to resolve label inconsistencies between various case definitions and the level of disagreements in annotations. Application of our method to MIMIC-III, a publicly available intensive care unit dataset, gives empirical results that illustrate the promise of learning a prediction model using truth inference-based labels and observed conflict among annotators.

Hypergraph Transformers for EHR-based Clinical Predictions.

Electronic health records (EHR) data contain rich information about patients' health conditions including diagnosis, procedures, medications and etc., which have been widely used to facilitate digital medicine. Despite its importance, it is often non-trivial to learn useful representations for patients' visits that support downstream clinical predictions, as each visit contains massive and diverse medical codes. As a result, the complex interactions among medical codes are often not captured, which leads to substandard predictions. To better model these complex relations, we leverage hypergraphs, which go beyond pairwise relations to jointly learn the representations for visits and medical codes. We also propose to use the self-attention mechanism to automatically identify the most relevant medical codes for each visit based on the downstream clinical predictions with better generalization power. Experiments on two EHR datasets show that our proposed method not only yields superior performance, but also provides reasonable insights towards the target tasks.

Hepatitis C care cascade in a large academic healthcare system, 2012 to 2018.

To determine the hepatitis C virus (HCV) care cascade among persons who were born during 1945 to 1965 and received outpatient care on or after January 2014 at a large academic healthcare system. Deidentified electronic health record data in an existing research database were analyzed for this study. Laboratory test results for HCV antibody and HCV ribonucleic acid (RNA) indicated seropositivity and confirmatory testing. HCV genotyping was used as a proxy for linkage to care. A direct-acting antiviral (DAA) prescription indicated treatment initiation, an undetectable HCV RNA at least 20 weeks after initiation of antiviral treatment indicated a sustained virologic response. Of the 121,807 patients in the 1945 to 1965 birth cohort who received outpatient care between January 1, 2014 and June 30, 2017, 3399 (3%) patients were screened for HCV; 540 (16%) were seropositive. Among the seropositive, 442 (82%) had detectable HCV RNA, 68 (13%) had undetectable HCV RNA, and 30 (6%) lacked HCV RNA testing. Of the 442 viremic patients, 237 (54%) were linked to care, 65 (15%) initiated DAA treatment, and 32 (7%) achieved sustained virologic response. While only 3% were screened for HCV, the seroprevalence was high in the screened sample. Despite the established safety and efficacy of DAAs, only 15% initiated treatment during the study period. To achieve HCV elimination, improved HCV screening and linkage to HCV care and DAA treatment are needed.

A Comprehensive and Improved Definition for Hospital-Acquired Pressure Injury Classification Based on Electronic Health Records: Comparative Study.

BACKGROUND: Patients develop pressure injuries (PIs) in the hospital owing to low mobility, exposure to localized pressure, circulatory conditions, and other predisposing factors. Over 2.5 million Americans develop PIs annually. The Center for Medicare and Medicaid considers hospital-acquired PIs (HAPIs) as the most frequent preventable event, and they are the second most common claim in lawsuits. With the growing use of electronic health records (EHRs) in hospitals, an opportunity exists to build machine learning models to identify and predict HAPI rather than relying on occasional manual assessments by human experts. However, accurate computational models rely on high-quality HAPI data labels. Unfortunately, the different data sources within EHRs can provide conflicting information on HAPI occurrence in the same patient. Furthermore, the existing definitions of HAPI disagree with each other, even within the same patient population. The inconsistent criteria make it impossible to benchmark machine learning methods to predict HAPI. OBJECTIVE: The objective of this project was threefold. We aimed to identify discrepancies in HAPI sources within EHRs, to develop a comprehensive definition for HAPI classification using data from all EHR sources, and to illustrate the importance of an improved HAPI definition. METHODS: We assessed the congruence among HAPI occurrences documented in clinical notes, diagnosis codes, procedure codes, and chart events from the Medical Information Mart for Intensive Care III database. We analyzed the criteria used for the 3 existing HAPI definitions and their adherence to the regulatory guidelines. We proposed the Emory HAPI (EHAPI), which is an improved and more comprehensive HAPI definition. We then evaluated the importance of the labels in training a HAPI classification model using tree-based and sequential neural network classifiers. RESULTS: We illustrate the complexity of defining HAPI, with <13% of hospital stays having at least 3 PI indications documented across 4 data sources. Although chart events were the most common indicator, it was the only PI documentation for >49% of the stays. We demonstrate a lack of congruence across existing HAPI definitions and EHAPI, with only 219 stays having a consensus positive label. Our analysis highlights the importance of our improved HAPI definition, with classifiers trained using our labels outperforming others on a small manually labeled set from nurse annotators and a consensus set in which all definitions agreed on the label. CONCLUSIONS: Standardized HAPI definitions are important for accurately assessing HAPI nursing quality metric and determining HAPI incidence for preventive measures. We demonstrate the complexity of defining an occurrence of HAPI, given the conflicting and incomplete EHR data. Our EHAPI definition has favorable properties, making it a suitable candidate for HAPI classification tasks.

Evaluation of available risk scores to predict multiple cardiovascular complications for patients with type 2 diabetes mellitus using electronic health records.

Aims: Various cardiovascular risk prediction models have been developed for patients with type 2 diabetes mellitus. Yet few models have been validated externally. We perform a comprehensive validation of existing risk models on a heterogeneous population of patients with type 2 diabetes using secondary analysis of electronic health record data. Methods: Electronic health records of 47,988 patients with type 2 diabetes between 2013 and 2017 were used to validate 16 cardiovascular risk models, including 5 that had not been compared previously, to estimate the 1-year risk of various cardiovascular outcomes. Discrimination and calibration were assessed by the c-statistic and the Hosmer-Lemeshow goodness-of-fit statistic, respectively. Each model was also evaluated based on the missing measurement rate. Sub-analysis was performed to determine the impact of race on discrimination performance. Results: There was limited discrimination (c-statistics ranged from 0.51 to 0.67) across the cardiovascular risk models. Discrimination generally improved when the model was tailored towards the individual outcome. After recalibration of the models, the Hosmer-Lemeshow statistic yielded p-values above 0.05. However, several of the models with the best discrimination relied on measurements that were often imputed (up to 39% missing). Conclusion: No single prediction model achieved the best performance on a full range of cardiovascular endpoints. Moreover, several of the highest-scoring models relied on variables with high missingness frequencies such as HbA1c and cholesterol that necessitated data imputation and may not be as useful in practice. An open-source version of our developed Python package, cvdm, is available for comparisons using other data sources.

PubMed Author-assigned Keyword Extraction (PubMedAKE) Benchmark.

With the ever-increasing abundance of biomedical articles, improving the accuracy of keyword search results becomes crucial for ensuring reproducible research. However, keyword extraction for biomedical articles is hard due to the existence of obscure keywords and the lack of a comprehensive benchmark. PubMedAKE is an author-assigned keyword extraction dataset that contains the title, abstract, and keywords of over 843,269 articles from the PubMed open access subset database. This dataset, publicly available on Zenodo, is the largest keyword extraction benchmark with sufficient samples to train neural networks. Experimental results using state-of-the-art baseline methods illustrate the need for developing automatic keyword extraction methods for biomedical literature.

Counterfactual and Factual Reasoning over Hypergraphs for Interpretable Clinical Predictions on EHR.

Electronic Health Record modeling is crucial for digital medicine. However, existing models ignore higher-order interactions among medical codes and their causal relations towards downstream clinical predictions. To address such limitations, we propose a novel framework CACHE, to provide effective and insightful clinical predictions based on hypergraph representation learning and counterfactual and factual reasoning techniques. Experiments on two real EHR datasets show the superior performance of CACHE. Case studies with a domain expert illustrate a preferred capability of CACHE in generating clinically meaningful interpretations towards the correct predictions.

Word centrality constrained representation for keyphrase extraction.

To keep pace with the increased generation and digitization of documents, automated methods that can improve search, discovery and mining of the vast body of literature are essential. Keyphrases provide a concise representation by identifying salient concepts in a document. Various supervised approaches model keyphrase extraction using local context to predict the label for each token and perform much better than the unsupervised counterparts. Unfortunately, this method fails for short documents where the context is unclear. Moreover, keyphrases, which are usually the gist of a document, need to be the central theme. We propose a new extraction model that introduces a centrality constraint to enrich the word representation of a Bidirectional long short-term memory. Performance evaluation on two publicly available datasets demonstrate our model outperforms existing state-of-the art approaches. Our model is publicly available at https://github.com/ZHgero/keyphrases_centrality.git.

Uncertainty-based Self-training for Biomedical Keyphrase Extraction.

To keep pace with the increased generation and digitization of documents, automated methods that can improve search, discovery and mining of the vast body of literature are essential. Keyphrases provide a concise representation by identifying salient concepts in a document. Various supervised approaches model keyphrase extraction using local context to predict the label for each token and perform much better than the unsupervised counterparts. However, existing supervised datasets have limited annotated examples to train better deep learning models. In contrast, many domains have large amount of un-annotated data that can be leveraged to improve model performance in keyphrase extraction. We introduce a self-learning based model that incorporates uncertainty estimates to select instances from large-scale unlabeled data to augment the small labeled training set. Performance evaluation on a publicly available biomedical dataset demonstrates that our method improves performance of keyphrase extraction over state of the art models.

CATAN: Chart-aware temporal attention network for adverse outcome prediction.

There is an increased adoption of electronic health record systems by a variety of hospitals and medical centers. This provides an opportunity to leverage automated computer systems in assisting healthcare workers. One of the least utilized but rich source of patient information is the unstructured clinical text. In this work, we develop CATAN, a chart-aware temporal attention network for learning patient representations from clinical notes. We introduce a novel representation where each note is considered a single unit, like a sentence, and composed of attention-weighted words. The notes in turn are aggregated into a patient representation using a second weighting unit, note attention. Unlike standard attention computations which focus only on the content of the note, we incorporate the chart-time for each note as a constraint for attention calculation. This allows our model to focus on notes closer to the prediction time. Using the MIMIC-III dataset, we empirically show that our patient representation and attention calculation achieves the best performance in comparison with various state-of-the-art baselines for one-year mortality prediction and 30-day hospital readmission. Moreover, the attention weights can be used to offer transparency into our model's predictions.

SMAT: An attention-based deep learning solution to the automation of schema matching.

Schema matching aims to identify the correspondences among attributes of database schemas. It is frequently considered as the most challenging and decisive stage existing in many contemporary web semantics and database systems. Low-quality algorithmic matchers fail to provide improvement while manually annotation consumes extensive human efforts. Further complications arise from data privacy in certain domains such as healthcare, where only schema-level matching should be used to prevent data leakage. For this problem, we propose SMAT, a new deep learning model based on state-of-the-art natural language processing techniques to obtain semantic mappings between source and target schemas using only the attribute name and description. SMAT avoids directly encoding domain knowledge about the source and target systems, which allows it to be more easily deployed across different sites. We also introduce a new benchmark dataset, OMAP, based on real-world schema-level mappings from the healthcare domain. Our extensive evaluation of various benchmark datasets demonstrates the potential of SMAT to help automate schema-level matching tasks.

Controlled Molecule Generator for Optimizing Multiple Chemical Properties.

Generating a novel and optimized molecule with desired chemical properties is an essential part of the drug discovery process. Failure to meet one of the required properties can frequently lead to failure in a clinical test which is costly. In addition, optimizing these multiple properties is a challenging task because the optimization of one property is prone to changing other properties. In this paper, we pose this multi-property optimization problem as a sequence translation process and propose a new optimized molecule generator model based on the Transformer with two constraint networks: property prediction and similarity prediction. We further improve the model by incorporating score predictions from these constraint networks in a modified beam search algorithm. The experiments demonstrate that our proposed model, Controlled Molecule Generator (CMG), outperforms state-of-the-art models by a significant margin for optimizing multiple properties simultaneously.

Communication Efficient Federated Generalized Tensor Factorization for Collaborative Health Data Analytics.

Modern healthcare systems knitted by a web of entities (e.g., hospitals, clinics, pharmacy companies) are collecting a huge volume of healthcare data from a large number of individuals with various medical procedures, medications, diagnosis, and lab tests. To extract meaningful medical concepts (i.e., phenotypes) from such higher-arity relational healthcare data, tensor factorization has been proven to be an effective approach and received increasing research attention, due to their intrinsic capability to represent the high-dimensional data. Recently, federated learning offers a privacy-preserving paradigm for collaborative learning among different entities, which seemingly provides an ideal potential to further enhance the tensor factorization-based collaborative phenotyping to handle sensitive personal health data. However, existing attempts to federated tensor factorization come with various limitations, including restrictions to the classic tensor factorization, high communication cost and reduced accuracy. We propose a communication efficient federated generalized tensor factorization, which is flexible enough to choose from a variate of losses to best suit different types of data in practice. We design a three-level communication reduction strategy tailored to the generalized tensor factorization, which is able to reduce the uplink communication cost up to 99.90%. In addition, we theoretically prove that our algorithm does not compromise convergence speed despite the aggressive communication compression. Extensive experiments on two real-world electronics health record datasets demonstrate the efficiency improvements in terms of computation and communication cost.

Temporal Network Embedding via Tensor Factorization.

Representation learning on static graph-structured data has shown a significant impact on many real-world applications. However, less attention has been paid to the evolving nature of temporal networks, in which the edges are often changing over time. The embeddings of such temporal networks should encode both graph-structured information and the temporally evolving pattern. Existing approaches in learning temporally evolving network representations fail to capture the temporal interdependence. In this paper, we propose Toffee, a novel approach for temporal network representation learning based on tensor decomposition. Our method exploits the tensor-tensor product operator to encode the cross-time information, so that the periodic changes in the evolving networks can be captured. Experimental results demonstrate that Toffee outperforms existing methods on multiple real-world temporal networks in generating effective embeddings for the link prediction tasks.

GASP: Graph-based Approximate Sequential Pattern Mining for Electronic Health Records.

Sequential pattern mining can be used to extract meaningful sequences from electronic health records. However, conventional sequential pattern mining algorithms that discover all frequent sequential patterns can incur a high computational and be susceptible to noise in the observations. Approximate sequential pattern mining techniques have been introduced to address these shortcomings yet, existing approximate methods fail to reflect the true frequent sequential patterns or only target single-item event sequences. Multi-item event sequences are prominent in healthcare as a patient can have multiple interventions for a single visit. To alleviate these issues, we propose GASP, a graph-based approximate sequential pattern mining, that discovers frequent patterns for multi-item event sequences. Our approach compresses the sequential information into a concise graph structure which has computational benefits. The empirical results on two healthcare datasets suggest that GASP outperforms existing approximate models by improving recoverability and extracts better predictive patterns.

Privacy-preserving Sequential Pattern Mining in distributed EHRs for Predicting Cardiovascular Disease.

From electronic health records (EHRs), the relationship between patients' conditions, treatments, and outcomes can be discovered and used in various healthcare research tasks such as risk prediction. In practice, EHRs can be stored in one or more data warehouses, and mining from distributed data sources becomes challenging. Another challenge arises from privacy laws because patient data cannot be used without some patient privacy guarantees. Thus, in this paper, we propose a privacy-preserving framework using sequential pattern mining in distributed data sources. Our framework extracts patterns from each source and shares patterns with other sources to discover discriminative and representative patterns that can be used for risk prediction while preserving privacy. We demonstrate our framework using a case study of predicting Cardiovascular Disease in patients with type 2 diabetes and show the effectiveness of our framework with several sources and by applying differential privacy mechanisms.