A flexible framework for recognizing events, temporal expressions, and temporal relations in clinical text

Chinese medical event detection based on event frequency distribution ratio and document consistency

Structured information especially medical events extracted from electronic medical records has extremely practical application value and play a basic role in various intelligent diagnosis and treatment systems. Fine-grained Chinese medical event detection is crucial in the process of structuring Chinese Electronic Medical Record (EMR). The current methods for detecting fine-grained Chinese medical events primarily rely on statistical machine learning and deep learning. However, they have two shortcomings: 1) they neglect to take into account the distribution characteristics of these fine-grained medical events. 2) they overlook the consistency in the distribution of medical events within each individual document. Therefore, this paper presents a fine-grained Chinese medical event detection method, which is based on event frequency distribution ratio and document consistency. To start with, a significant number of Chinese EMR texts are used to adapt the Chinese pre-training model BERT to the domain. Second, based on the fundamental features, the Event Frequency - Event Distribution Ratio (EF-DR) is devised to select distinct event information as supplementary features, taking into account the distribution of events within the EMR. Finally, using EMR document consistency within the model improves the outcome of event detection. Our experiments demonstrate that the proposed method significantly outperforms the baseline model.

Combining data augmentation and domain information with TENER model for Clinical Event Detection

Background

In recent years, with the development of artificial intelligence, the use of deep learning technology for clinical information extraction has become a new trend. Clinical Event Detection (CED) as its subtask has attracted the attention from academia and industry. However, directly applying the advancements in deep learning to CED task often yields unsatisfactory results. The main reasons are due to the following two points: (1) A great number of obscure professional terms in the electronic medical record leads to poor recognition performance of model. (2) The scarcity of datasets required for the task leads to poor model robustness. Therefore, it is urgent to solve these two problems to improve model performance.

Methods

This paper proposes a combining data augmentation and domain information with TENER Model for Clinical Event Detection.

Results

We use two evaluation metrics to compare the overall performance of the proposed model with the existing model on the 2012 i2b2 challenge dataset. Experimental results demonstrate that our proposed model achieves the best F1-score of 80.26%, type accuracy of 93% and Span F1-score of 90.33%, and outperforms the state-of-the-art approaches.

Conclusions

This paper proposes a multi-granularity information fusion encoder-decoder framework, which applies the TENER model to the CED task for the first time. It uses the pre-trained language model (BioBERT) to generate word-level features, solving the problem of a great number of obscure professional terms in the electronic medical record lead to poor recognition performance of model. In addition, this paper proposes a new data augmentation method for sequence labeling tasks, solving the problem of the scarcity of datasets required for the task leads to poor model robustness.

Extraction of temporal relations from clinical free text: A systematic review of current approaches

BACKGROUND

Temporal relations between clinical events play an important role in clinical assessment and decision making. Extracting such relations from free text data is a challenging task because it lies on between medical natural language processing, temporal representation and temporal reasoning.

OBJECTIVES

To survey existing methods for extracting temporal relations (TLINKs) between events from clinical free text in English; to establish the state-of-the-art in this field; and to identify outstanding methodological challenges.

METHODS

A systematic search in PubMed and the DBLP computer science bibliography was conducted for studies published between January 2006 and December 2018. The relevant studies were identified by examining the titles and abstracts. Then, the full text of selected studies was analyzed in depth and information were collected on TLINK tasks, TLINK types, data sources, features selection, methods used, and reported performance.

RESULTS

A total of 2834 publications were identified for title and abstract screening. Of these publications, 51 studies were selected. Thirty-two studies used machine learning approaches, 15 studies used a hybrid approaches, and only four studies used a rule-based approach. The majority of studies use publicly available corpora: THYME (28 studies) and the i2b2 corpus (17 studies).

CONCLUSION

The performance of TLINK extraction methods ranges widely depending on relation types and events (e.g. from 32% to 87% F-score for identifying relations between clinical events and document creation time). A small set of TLINKs (before, after, overlap and contains) has been widely studied with relatively good performance, whereas other types of TLINK (e.g., started by, finished by, precedes) are rarely studied and remain challenging. Machine learning classifiers (such as Support Vector Machine and Conditional Random Fields) and Deep Neural Networks were among the best performing methods for extracting TLINKs, but nearly all the work has been carried out and tested on two publicly available corpora only. The field would benefit from the availability of more publicly available, high-quality, annotated clinical text corpora.

Complexities, variations, and errors of numbering within clinical notes: the potential impact on information extraction and cohort-identification

BACKGROUND

Numbers and numerical concepts appear frequently in free text clinical notes from electronic health records. Knowledge of the frequent lexical variations of these numerical concepts, and their accurate identification, is important for many information extraction tasks. This paper describes an analysis of the variation in how numbers and numerical concepts are represented in clinical notes.

METHODS

We used an inverted index of approximately 100 million notes to obtain the frequency of various permutations of numbers and numerical concepts, including the use of Roman numerals, numbers spelled as English words, and invalid dates, among others. Overall, twelve types of lexical variants were analyzed.

RESULTS

We found substantial variation in how these concepts were represented in the notes, including multiple data quality issues. We also demonstrate that not considering these variations could have substantial real-world implications for cohort identification tasks, with one case missing > 80% of potential patients.

CONCLUSIONS

Numbering within clinical notes can be variable, and not taking these variations into account could result in missing or inaccurate information for natural language processing and information retrieval tasks.

The Role of a Deep-Learning Method for Negation Detection in Patient Cohort Identification from Electroencephalography Reports

Detecting negation in biomedical texts entails the automatic identification of negation cues (e.g. "never", "not", "no longer") as well as the scope of these cues. When medical concepts or terms are identified within the scope of a negation cue, their polarity is inferred as "negative". All the other concepts or words receive a positive polarity. Correctly inferring the polarity is essential for patient cohort retrieval systems, as all inclusion criteria need to be automatically assigned positive polarity, whereas exclusion criteria should receive negative polarity. Motivated by the recent development of techniques using deep learning, we have experimented with a neural negation detection technique and compared it against an existing neural polarity recognition system, which were incorporated in a patient cohort system operating on clinical electroencephalography (EEG) reports. Our experiments indicate that the neural negation detection method produces better patient cohorts then the polarity recognition method.

Machine Learning Methods to Extract Documentation of Breast Cancer Symptoms From Electronic Health Records

CONTEXT

Clinicians document cancer patients' symptoms in free-text format within electronic health record visit notes. Although symptoms are critically important to quality of life and often herald clinical status changes, computational methods to assess the trajectory of symptoms over time are woefully underdeveloped.

OBJECTIVES

To create machine learning algorithms capable of extracting patient-reported symptoms from free-text electronic health record notes.

METHODS

The data set included 103,564 sentences obtained from the electronic clinical notes of 2695 breast cancer patients receiving paclitaxel-containing chemotherapy at two academic cancer centers between May 1996 and May 2015. We manually annotated 10,000 sentences and trained a conditional random field model to predict words indicating an active symptom (positive label), absence of a symptom (negative label), or no symptom at all (neutral label). Sentences labeled by human coder were divided into training, validation, and test data sets. Final model performance was determined on 20% test data unused in model development or tuning.

RESULTS

The final model achieved precision of 0.82, 0.86, and 0.99 and recall of 0.56, 0.69, and 1.00 for positive, negative, and neutral symptom labels, respectively. The most common positive symptoms were pain, fatigue, and nausea. Machine-based labeling of 103,564 sentences took two minutes.

CONCLUSION

We demonstrate the potential of machine learning to gather, track, and analyze symptoms experienced by cancer patients during chemotherapy. Although our initial model requires further optimization to improve the performance, further model building may yield machine learning methods suitable to be deployed in routine clinical care, quality improvement, and research applications.

Causality patterns and machine learning for the extraction of problem-action relations in discharge summaries

Clinical narrative text includes information related to a patient's medical history such as chronological progression of medical problems and clinical treatments. A chronological view of a patient's history makes clinical audits easier and improves quality of care. In this paper, we propose a clinical Problem-Action relation extraction method, based on clinical semantic units and event causality patterns, to present a chronological view of a patient's problem and a doctor's action. Based on our observation that a clinical text describes a patient's medical problems and a doctor's treatments in chronological order, a clinical semantic unit is defined as a problem and/or an action relation. Since a clinical event is a basic unit of the problem and action relation, events are extracted from narrative texts, based on the external knowledge resources context features of the conditional random fields. A clinical semantic unit is extracted from each sentence based on time expressions and context structures of events. Then, a clinical semantic unit is classified into a problem and/or action relation based on the event causality patterns of the support vector machines. Experimental results on Korean discharge summaries show 78.8% performance in the F1-measure. This result shows that the proposed method is effectively classifies clinical Problem-Action relations.

Temporal data representation, normalization, extraction, and reasoning: A review from clinical domain

BACKGROUND AND OBJECTIVE

We live our lives by the calendar and the clock, but time is also an abstraction, even an illusion. The sense of time can be both domain-specific and complex, and is often left implicit, requiring significant domain knowledge to accurately recognize and harness. In the clinical domain, the momentum gained from recent advances in infrastructure and governance practices has enabled the collection of tremendous amount of data at each moment in time. Electronic health records (EHRs) have paved the way to making these data available for practitioners and researchers. However, temporal data representation, normalization, extraction and reasoning are very important in order to mine such massive data and therefore for constructing the clinical timeline. The objective of this work is to provide an overview of the problem of constructing a timeline at the clinical point of care and to summarize the state-of-the-art in processing temporal information of clinical narratives.

METHODS

This review surveys the methods used in three important area: modeling and representing of time, medical NLP methods for extracting time, and methods of time reasoning and processing. The review emphasis on the current existing gap between present methods and the semantic web technologies and catch up with the possible combinations.

RESULTS

The main findings of this review are revealing the importance of time processing not only in constructing timelines and clinical decision support systems but also as a vital component of EHR data models and operations.

CONCLUSIONS

Extracting temporal information in clinical narratives is a challenging task. The inclusion of ontologies and semantic web will lead to better assessment of the annotation task and, together with medical NLP techniques, will help resolving granularity and co-reference resolution problems.

Towards a Generalizable Time Expression Model for Temporal Reasoning in Clinical Notes

Accurate temporal identification and normalization is imperative for many biomedical and clinical tasks such as generating timelines and identifying phenotypes. A major natural language processing challenge is developing and evaluating a generalizable temporal modeling approach that performs well across corpora and institutions. Our long-term goal is to create such a model. We initiate our work on reaching this goal by focusing on temporal expression (TIMEX3) identification. We present a systematic approach to 1) generalize existing solutions for automated TIMEX3 span detection, and 2) assess similarities and differences by various instantiations of TIMEX3 models applied on separate clinical corpora. When evaluated on the 2012 i2b2 and the 2015 Clinical TempEval challenge corpora, our conclusion is that our approach is successful - we achieve competitive results for automated classification, and we identify similarities and differences in TIMEX3 modeling that will be informative in the development of a simplified, general temporal model.

Normalization of relative and incomplete temporal expressions in clinical narratives

OBJECTIVE

To improve the normalization of relative and incomplete temporal expressions (RI-TIMEXes) in clinical narratives.

METHODS

We analyzed the RI-TIMEXes in temporally annotated corpora and propose two hypotheses regarding the normalization of RI-TIMEXes in the clinical narrative domain: the anchor point hypothesis and the anchor relation hypothesis. We annotated the RI-TIMEXes in three corpora to study the characteristics of RI-TMEXes in different domains. This informed the design of our RI-TIMEX normalization system for the clinical domain, which consists of an anchor point classifier, an anchor relation classifier, and a rule-based RI-TIMEX text span parser. We experimented with different feature sets and performed an error analysis for each system component.

RESULTS

The annotation confirmed the hypotheses that we can simplify the RI-TIMEXes normalization task using two multi-label classifiers. Our system achieves anchor point classification, anchor relation classification, and rule-based parsing accuracy of 74.68%, 87.71%, and 57.2% (82.09% under relaxed matching criteria), respectively, on the held-out test set of the 2012 i2b2 temporal relation challenge.

DISCUSSION

Experiments with feature sets reveal some interesting findings, such as: the verbal tense feature does not inform the anchor relation classification in clinical narratives as much as the tokens near the RI-TIMEX. Error analysis showed that underrepresented anchor point and anchor relation classes are difficult to detect.

CONCLUSIONS

We formulate the RI-TIMEX normalization problem as a pair of multi-label classification problems. Considering only RI-TIMEX extraction and normalization, the system achieves statistically significant improvement over the RI-TIMEX results of the best systems in the 2012 i2b2 challenge.

Automatic identification of methotrexate-induced liver toxicity in patients with rheumatoid arthritis from the electronic medical record

OBJECTIVES

To improve the accuracy of mining structured and unstructured components of the electronic medical record (EMR) by adding temporal features to automatically identify patients with rheumatoid arthritis (RA) with methotrexate-induced liver transaminase abnormalities.

MATERIALS AND METHODS

Codified information and a string-matching algorithm were applied to a RA cohort of 5903 patients from Partners HealthCare to select 1130 patients with potential liver toxicity. Supervised machine learning was applied as our key method. For features, Apache clinical Text Analysis and Knowledge Extraction System (cTAKES) was used to extract standard vocabulary from relevant sections of the unstructured clinical narrative. Temporal features were further extracted to assess the temporal relevance of event mentions with regard to the date of transaminase abnormality. All features were encapsulated in a 3-month-long episode for classification. Results were summarized at patient level in a training set (N=480 patients) and evaluated against a test set (N=120 patients).

RESULTS

The system achieved positive predictive value (PPV) 0.756, sensitivity 0.919, F1 score 0.829 on the test set, which was significantly better than the best baseline system (PPV 0.590, sensitivity 0.703, F1 score 0.642). Our innovations, which included framing the phenotype problem as an episode-level classification task, and adding temporal information, all proved highly effective.

CONCLUSIONS

Automated methotrexate-induced liver toxicity phenotype discovery for patients with RA based on structured and unstructured information in the EMR shows accurate results. Our work demonstrates that adding temporal features significantly improved classification results.

Incremental cost-effectiveness of dobutamine stress cardiac magnetic resonance imaging in patients at intermediate risk for coronary artery disease

Aims

The effectiveness of stress cardiac magnetic resonance (CMR) as a gatekeeper for coronary angiography (CA) has been established. Level five HTA studies according to the hierarchical model of diagnostic test evaluation are not available.

Methods

This cohort study included 1,158 consecutive patients (mean age 63 ± 11 years, 42 % women) presenting at our institution between January 1, 2003 and December 31, 2004 with suspected coronary artery disease (CAD) for an elective CA. The patients were assessed for eligibility and propensity score matching was applied to address selection bias regarding the patients’ allocation to CMR or direct CA. Median patient follow-up was 7.9 years (95 % CI 7.8–8.0 years). The primary effect was calculated as relative survival difference. The cost unit calculation (per patient) at our institute was the source of costs.

Results

Survival was similar in CMR and CA (p = 0.139). Catheterizations ruling out CAD were significantly reduced by the CMR gate-keeper strategy. Patients with prior CMR had significantly lower costs at the initial hospital stay and at follow-up (CMR vs. CA, initial: 2,904€ vs. 3,421€, p = 0.018; follow-up: 2,045€ vs. 3,318€, p = 0.037). CMR was cost-effective in terms of a contribution of 12,466€ per life year to cover a part of the CMR costs.

Conclusion

Stress CMR prior to CA was saving 12,466€ of hospital costs per life year. Lower costs at follow-up suggest sustained cost-effectiveness of the CMR-guided strategy.

Electronic supplementary material

The online version of this article (doi:10.1007/s00392-014-0793-0) contains supplementary material, which is available to authorized users.

Extraction of events and temporal expressions from clinical narratives

This paper addresses an important task of event and timex extraction from clinical narratives in context of the i2b2 2012 challenge. State-of-the-art approaches for event extraction use a multi-class classifier for finding the event types. However, such approaches consider each event in isolation. In this paper, we present a sentence-level inference strategy which enforces consistency constraints on attributes of those events which appear close to one another. Our approach is general and can be used for other tasks as well. We also design novel features like clinical descriptors (from medical ontologies) which encode a lot of useful information about the concepts. For timex extraction, we adapt a state-of-the-art system, HeidelTime, for use in clinical narratives and also develop several rules which complement HeidelTime. We also give a robust algorithm for date extraction. For the event extraction task, we achieved an overall F1 score of 0.71 for determining span of the events along with their attributes. For the timex extraction task, we achieved an F1 score of 0.79 for determining span of the temporal expressions. We present detailed error analysis of our system and also point out some factors which can help to improve its accuracy.

Temporal reasoning over clinical text: the state of the art

OBJECTIVES

To provide an overview of the problem of temporal reasoning over clinical text and to summarize the state of the art in clinical natural language processing for this task.

TARGET AUDIENCE

This overview targets medical informatics researchers who are unfamiliar with the problems and applications of temporal reasoning over clinical text.

SCOPE

We review the major applications of text-based temporal reasoning, describe the challenges for software systems handling temporal information in clinical text, and give an overview of the state of the art. Finally, we present some perspectives on future research directions that emerged during the recent community-wide challenge on text-based temporal reasoning in the clinical domain.

Evaluating temporal relations in clinical text: 2012 i2b2 Challenge

BACKGROUND

The Sixth Informatics for Integrating Biology and the Bedside (i2b2) Natural Language Processing Challenge for Clinical Records focused on the temporal relations in clinical narratives. The organizers provided the research community with a corpus of discharge summaries annotated with temporal information, to be used for the development and evaluation of temporal reasoning systems. 18 teams from around the world participated in the challenge. During the workshop, participating teams presented comprehensive reviews and analysis of their systems, and outlined future research directions suggested by the challenge contributions.

METHODS

The challenge evaluated systems on the information extraction tasks that targeted: (1) clinically significant events, including both clinical concepts such as problems, tests, treatments, and clinical departments, and events relevant to the patient's clinical timeline, such as admissions, transfers between departments, etc; (2) temporal expressions, referring to the dates, times, durations, or frequencies phrases in the clinical text. The values of the extracted temporal expressions had to be normalized to an ISO specification standard; and (3) temporal relations, between the clinical events and temporal expressions. Participants determined pairs of events and temporal expressions that exhibited a temporal relation, and identified the temporal relation between them.

RESULTS

For event detection, statistical machine learning (ML) methods consistently showed superior performance. While ML and rule based methods seemed to detect temporal expressions equally well, the best systems overwhelmingly adopted a rule based approach for value normalization. For temporal relation classification, the systems using hybrid approaches that combined ML and heuristics based methods produced the best results.