RESUMEN

Despite the high creation cost, annotated corpora are indispensable for robust natural language processing systems. In the clinical field, in addition to annotating medical entities, corpus creators must also remove personally identifiable information (PII). This has become increasingly important in the era of large language models where unwanted memorization can occur. This paper presents a corpus annotated to anonymize personally identifiable information in 1,787 anamneses of work-related accidents and diseases in Spanish. Additionally, we applied a previously released model for Named Entity Recognition (NER) trained on referrals from primary care physicians to identify diseases, body parts, and medications in this work-related text. We analyzed the differences between the models and the gold standard curated by a physician in detail. Moreover, we compared the performance of the NER model on the original narratives, in narratives where personal information has been masked, and in texts where the personal data is replaced by another similar surrogate value (pseudonymization). Within this publication, we share the annotation guidelines and the annotated corpus.

Asunto(s)

RESUMEN

OBJECTIVE: To explore the feasibility of validating Dutch concept extraction tools using annotated corpora translated from English, focusing on preserving annotations during translation and addressing the scarcity of non-English annotated clinical corpora. MATERIALS AND METHODS: Three annotated corpora were standardized and translated from English to Dutch using 2 machine translation services, Google Translate and OpenAI GPT-4, with annotations preserved through a proposed method of embedding annotations in the text before translation. The performance of 2 concept extraction tools, MedSpaCy and MedCAT, was assessed across the corpora in both Dutch and English. RESULTS: The translation process effectively generated Dutch annotated corpora and the concept extraction tools performed similarly in both English and Dutch. Although there were some differences in how annotations were preserved across translations, these did not affect extraction accuracy. Supervised MedCAT models consistently outperformed unsupervised models, whereas MedSpaCy demonstrated high recall but lower precision. DISCUSSION: Our validation of Dutch concept extraction tools on corpora translated from English was successful, highlighting the efficacy of our annotation preservation method and the potential for efficiently creating multilingual corpora. Further improvements and comparisons of annotation preservation techniques and strategies for corpus synthesis could lead to more efficient development of multilingual corpora and accurate non-English concept extraction tools. CONCLUSION: This study has demonstrated that translated English corpora can be used to validate non-English concept extraction tools. The annotation preservation method used during translation proved effective, and future research can apply this corpus translation method to additional languages and clinical settings.

Asunto(s)

RESUMEN

Current animal protection laws require replacement of animal experiments with alternative methods, whenever such methods are suitable to reach the intended scientific objective. However, searching for alternative methods in the scientific literature is a time-consuming task that requires careful screening of an enormously large number of experimental biomedical publications. The identification of potentially relevant methods, e.g. organ or cell culture models, or computer simulations, can be supported with text mining tools specifically built for this purpose. Such tools are trained (or fine tuned) on relevant data sets labeled by human experts. We developed the GoldHamster corpus, composed of 1,600 PubMed (Medline) articles (titles and abstracts), in which we manually identified the used experimental model according to a set of eight labels, namely: "in vivo", "organs", "primary cells", "immortal cell lines", "invertebrates", "humans", "in silico" and "other" (models). We recruited 13 annotators with expertise in the biomedical domain and assigned each article to two individuals. Four additional rounds of annotation aimed at improving the quality of the annotations with disagreements in the first round. Furthermore, we conducted various machine learning experiments based on supervised learning to evaluate the corpus for our classification task. We obtained more than 7,000 document-level annotations for the above labels. After the first round of annotation, the inter-annotator agreement (kappa coefficient) varied among labels, and ranged from 0.42 (for "others") to 0.82 (for "invertebrates"), with an overall score of 0.62. All disagreements were resolved in the subsequent rounds of annotation. The best-performing machine learning experiment used the PubMedBERT pre-trained model with fine-tuning to our corpus, which gained an overall f-score of 0.83. We obtained a corpus with high agreement for all labels, and our evaluation demonstrated that our corpus is suitable for training reliable predictive models for automatic classification of biomedical literature according to the used experimental models. Our SMAFIRA - "Smart feature-based interactive" - search tool ( https://smafira.bf3r.de ) will employ this classifier for supporting the retrieval of alternative methods to animal experiments. The corpus is available for download ( https://doi.org/10.5281/zenodo.7152295 ), as well as the source code ( https://github.com/mariananeves/goldhamster ) and the model ( https://huggingface.co/SMAFIRA/goldhamster ).

Asunto(s)

RESUMEN

Affective understanding of language is an important research focus in artificial intelligence. The large-scale annotated datasets of Chinese textual affective structure (CTAS) are the foundation for subsequent higher-level analysis of documents. However, there are very few published datasets for CTAS. This paper introduces a new benchmark dataset for the task of CTAS to promote development in this research direction. Specifically, our benchmark is a CTAS dataset with the following advantages: (a) it is Weibo-based, which is the most popular Chinese social media platform used by the public to express their opinions; (b) it includes the most comprehensive affective structure labels at present; and (c) we propose a maximum entropy Markov model that incorporates neural network features and experimentally demonstrate that it outperforms the two baseline models.

RESUMEN

OBJECTIVE: To develop a computable representation for medical evidence and to contribute a gold standard dataset of annotated randomized controlled trial (RCT) abstracts, along with a natural language processing (NLP) pipeline for transforming free-text RCT evidence in PubMed into the structured representation. MATERIALS AND METHODS: Our representation, EvidenceMap, consists of 3 levels of abstraction: Medical Evidence Entity, Proposition and Map, to represent the hierarchical structure of medical evidence composition. Randomly selected RCT abstracts were annotated following EvidenceMap based on the consensus of 2 independent annotators to train an NLP pipeline. Via a user study, we measured how the EvidenceMap improved evidence comprehension and analyzed its representative capacity by comparing the evidence annotation with EvidenceMap representation and without following any specific guidelines. RESULTS: Two corpora including 229 disease-agnostic and 80 COVID-19 RCT abstracts were annotated, yielding 12 725 entities and 1602 propositions. EvidenceMap saves users 51.9% of the time compared to reading raw-text abstracts. Most evidence elements identified during the freeform annotation were successfully represented by EvidenceMap, and users gave the enrollment, study design, and study Results sections mean 5-scale Likert ratings of 4.85, 4.70, and 4.20, respectively. The end-to-end evaluations of the pipeline show that the evidence proposition formulation achieves F1 scores of 0.84 and 0.86 in the adjusted random index score. CONCLUSIONS: EvidenceMap extends the participant, intervention, comparator, and outcome framework into 3 levels of abstraction for transforming free-text evidence from the clinical literature into a computable structure. It can be used as an interoperable format for better evidence retrieval and synthesis and an interpretable representation to efficiently comprehend RCT findings.

Asunto(s)

RESUMEN

Over the last few years, the number of corpora that can be used for language comparison has dramatically increased. The corpora are so diverse in their structure, size and annotation style, that a novice might not know where to start. The present paper charts this new and changing territory, providing a few landmarks, warning signs and safe paths. Although no corpus at present can replace the traditional type of typological data based on language description in reference grammars, corpora can help with diverse tasks, being particularly well suited for investigating probabilistic and gradient properties of languages and for discovering and interpreting cross-linguistic generalizations based on processing and communicative mechanisms. At the same time, the use of corpora for typological purposes has not only advantages and opportunities, but also numerous challenges. This paper also contains an empirical case study addressing two pertinent problems: the role of text types in language comparison and the problem of the word as a comparative concept.

RESUMEN

Texts are not monolithic entities but rather coherent collections of micro illocutionary acts which help to convey a unitary message of content and purpose. Identifying such text segments is challenging because they require a fine-grained level of analysis even within a single sentence. At the same time, accessing them facilitates the analysis of the communicative functions of a text as well as the identification of relevant information. We propose an empirical framework for modelling micro illocutionary acts at clause level, that we call content types, grounded on linguistic theories of text types, in particular on the framework proposed by Werlich in 1976. We make available a newly annotated corpus of 279 documents (for a total of more than 180,000 tokens) belonging to different genres and temporal periods, based on a dedicated annotation scheme. We obtain an average Cohen's kappa of 0.89 at token level. We achieve an average F1 score of 74.99% on the automatic classification of content types using a bi-LSTM model. Similar results are obtained on contemporary and historical documents, while performances on genres are more varied. This work promotes a discourse-oriented approach to information extraction and cross-fertilisation across disciplines through a computationally-aided linguistic analysis.

RESUMEN

BACKGROUND: The limited availability of clinical texts for Natural Language Processing purposes is hindering the progress of the field. This article investigates the use of synthetic data for the annotation and automated extraction of family history information from Norwegian clinical text. We make use of incrementally developed synthetic clinical text describing patients' family history relating to cases of cardiac disease and present a general methodology which integrates the synthetically produced clinical statements and annotation guideline development. The resulting synthetic corpus contains 477 sentences and 6030 tokens. In this work we experimentally assess the validity and applicability of the annotated synthetic corpus using machine learning techniques and furthermore evaluate the system trained on synthetic text on a corpus of real clinical text, consisting of de-identified records for patients with genetic heart disease. RESULTS: For entity recognition, an SVM trained on synthetic data had class weighted precision, recall and F1-scores of 0.83, 0.81 and 0.82, respectively. For relation extraction precision, recall and F1-scores were 0.74, 0.75 and 0.74. CONCLUSIONS: A system for extraction of family history information developed on synthetic data generalizes well to real, clinical notes with a small loss of accuracy. The methodology outlined in this paper may be useful in other situations where limited availability of clinical text hinders NLP tasks. Both the annotation guidelines and the annotated synthetic corpus are made freely available and as such constitutes the first publicly available resource of Norwegian clinical text.

Asunto(s)

RESUMEN

OBJECTIVE: We present the Berlin-Tübingen-Oncology corpus (BRONCO), a large and freely available corpus of shuffled sentences from German oncological discharge summaries annotated with diagnosis, treatments, medications, and further attributes including negation and speculation. The aim of BRONCO is to foster reproducible and openly available research on Information Extraction from German medical texts. MATERIALS AND METHODS: BRONCO consists of 200 manually deidentified discharge summaries of cancer patients. Annotation followed a structured and quality-controlled process involving 2 groups of medical experts to ensure consistency, comprehensiveness, and high quality of annotations. We present results of several state-of-the-art techniques for different IE tasks as baselines for subsequent research. RESULTS: The annotated corpus consists of 11 434 sentences and 89 942 tokens, annotated with 11 124 annotations for medical entities and 3118 annotations of related attributes. We publish 75% of the corpus as a set of shuffled sentences, and keep 25% as held-out data set for unbiased evaluation of future IE tools. On this held-out dataset, our baselines reach depending on the specific entity types F1-scores of 0.72-0.90 for named entity recognition, 0.10-0.68 for entity normalization, 0.55 for negation detection, and 0.33 for speculation detection. DISCUSSION: Medical corpus annotation is a complex and time-consuming task. This makes sharing of such resources even more important. CONCLUSION: To our knowledge, BRONCO is the first sizable and freely available German medical corpus. Our baseline results show that more research efforts are necessary to lift the quality of information extraction in German medical texts to the level already possible for English.

RESUMEN

OBJECTIVE: To annotate a corpus of randomized controlled trial (RCT) publications with the checklist items of CONSORT reporting guidelines and using the corpus to develop text mining methods for RCT appraisal. METHODS: We annotated a corpus of 50 RCT articles at the sentence level using 37 fine-grained CONSORT checklist items. A subset (31 articles) was double-annotated and adjudicated, while 19 were annotated by a single annotator and reconciled by another. We calculated inter-annotator agreement at the article and section level using MASI (Measuring Agreement on Set-Valued Items) and at the CONSORT item level using Krippendorff's α. We experimented with two rule-based methods (phrase-based and section header-based) and two supervised learning approaches (support vector machine and BioBERT-based neural network classifiers), for recognizing 17 methodology-related items in the RCT Methods sections. RESULTS: We created CONSORT-TM consisting of 10,709 sentences, 4,845 (45%) of which were annotated with 5,246 labels. A median of 28 CONSORT items (out of possible 37) were annotated per article. Agreement was moderate at the article and section levels (average MASI: 0.60 and 0.64, respectively). Agreement varied considerably among individual checklist items (Krippendorff's α= 0.06-0.96). The model based on BioBERT performed best overall for recognizing methodology-related items (micro-precision: 0.82, micro-recall: 0.63, micro-F1: 0.71). Combining models using majority vote and label aggregation further improved precision and recall, respectively. CONCLUSION: Our annotated corpus, CONSORT-TM, contains more fine-grained information than earlier RCT corpora. Low frequency of some CONSORT items made it difficult to train effective text mining models to recognize them. For the items commonly reported, CONSORT-TM can serve as a testbed for text mining methods that assess RCT transparency, rigor, and reliability, and support methods for peer review and authoring assistance. Minor modifications to the annotation scheme and a larger corpus could facilitate improved text mining models. CONSORT-TM is publicly available at https://github.com/kilicogluh/CONSORT-TM.

Asunto(s)

RESUMEN

BACKGROUND: Chronic pain affects more than 20% of adults in the United States and is associated with substantial physical, mental, and social burden. Clinical text contains rich information about chronic pain, but no systematic appraisal has been performed to assess the electronic health record (EHR) narratives for these patients. A formal content analysis of the unstructured EHR data can inform clinical practice and research in chronic pain. OBJECTIVE: We characterized individual episodes of chronic pain by annotating and analyzing EHR notes for a stratified cohort of adults with known chronic pain. METHODS: We used the Rochester Epidemiology Project infrastructure to screen all residents of Olmsted County, Minnesota, for evidence of chronic pain, between January 1, 2005, and September 30, 2015. Diagnosis codes were used to assemble a cohort of 6586 chronic pain patients; people with cancer were excluded. The records of an age- and sex-stratified random sample of 62 patients from the cohort were annotated using an iteratively developed guideline. The annotated concepts included date, location, severity, causes, effects on quality of life, diagnostic procedures, medications, and other treatment modalities. RESULTS: A total of 94 chronic pain episodes from 62 distinct patients were identified by reviewing 3272 clinical notes. Documentation was written by clinicians across a wide spectrum of specialties. Most patients (40/62, 65%) had 1 pain episode during the study period. Interannotator agreement ranged from 0.78 to 1.00 across the annotated concepts. Some pain-related concepts (eg, body location) had 100% (94/94) coverage among all the episodes, while others had moderate coverage (eg, effects on quality of life) (55/94, 59%). Back pain and leg pain were the most common types of chronic pain in the annotated cohort. Musculoskeletal issues like arthritis were annotated as the most common causes. Opioids were the most commonly captured medication, while physical and occupational therapies were the most common nonpharmacological treatments. CONCLUSIONS: We systematically annotated chronic pain episodes in clinical text. The rich content analysis results revealed complexity of the chronic pain episodes and of their management, as well as the challenges in extracting pertinent information, even for humans. Despite the pilot study nature of the work, the annotation guideline and corpus should be able to serve as informative references for other institutions with shared interest in chronic pain research using EHRs.

RESUMEN

Large-scale neuroscience literature call for effective methods to mine the knowledge from species perspective to link the brain and neuroscience communities, neurorobotics, computing devices, and AI research communities. Structured knowledge can motivate researchers to better understand the functionality and structure of the brain and link the related resources and components. However, the abstracts of massive scientific works do not explicitly mention the species. Therefore, in addition to dictionary-based methods, we need to mine species using cognitive computing models that are more like the human reading process, and these methods can take advantage of the rich information in the literature. We also enable the model to automatically distinguish whether the mentioned species is the main research subject. Distinguishing the two situations can generate value at different levels of knowledge management. We propose SpecExplorer project which is used to explore the knowledge associations of different species for brain and neuroscience. This project frees humans from the tedious task of classifying neuroscience literature by species. Species classification task belongs to the multi-label classification which is more complex than the single-label classification due to the correlation between labels. To resolve this problem, we present the sequence-to-sequence classification framework to adaptively assign multiple species to the literature. To model the structure information of documents, we propose the hierarchical attentive decoding (HAD) to extract span of interest (SOI) for predicting each species. We create three datasets from PubMed and PMC corpora. We present two versions of annotation criteria (mention-based annotation and semantic-based annotation) for species research. Experiments demonstrate that our approach achieves improvements in the final results. Finally, we perform species-based analysis of brain diseases, brain cognitive functions, and proteins related to the hippocampus and provide potential research directions for certain species.

RESUMEN

BACKGROUND: To robustly identify synergistic combinations of drugs, high-throughput screenings are desirable. It will be of great help to automatically identify the relations in the published papers with machine learning based tools. To support the chemical disease semantic relation extraction especially for chronic diseases, a chronic disease specific corpus for combination therapy discovery in Chinese (RCorp) is manually annotated. METHODS: In this study, we extracted abstracts from a Chinese medical literature server and followed the annotation framework of the BioCreative CDR corpus, with the guidelines modified to make the combination therapy related relations available. An annotation tool was incorporated to the standard annotation process. RESULTS: The resulting RCorp consists of 339 Chinese biomedical articles with 2367 annotated chemicals, 2113 diseases, 237 symptoms, 164 chemical-induce-disease relations, 163 chemical-induce-symptom relations, and 805 chemical-treat-disease relations. Each annotation includes both the mention text spans and normalized concept identifiers. The corpus gets an inter-annotator agreement score of 0.883 for chemical entities, 0.791 for disease entities which are measured by F score. And the F score for chemical-treat-disease relations gets 0.788 after unifying the entity mentions. CONCLUSIONS: We extracted and manually annotated a chronic disease specific corpus for combination therapy discovery in Chinese. The result analysis of the corpus proves its quality for the combination therapy related knowledge discovery task. Our annotated corpus would be a useful resource for the modelling of entity recognition and relation extraction tools. In the future, an evaluation based on the corpus will be held.

Asunto(s)

RESUMEN

Pharmacovigilance (PV) databases record the benefits and risks of different drugs, as a means to ensure their safe and effective use. Creating and maintaining such resources can be complex, since a particular medication may have divergent effects in different individuals, due to specific patient characteristics and/or interactions with other drugs being administered. Textual information from various sources can provide important evidence to curators of PV databases about the usage and effects of drug targets in different medical subjects. However, the efficient identification of relevant evidence can be challenging, due to the increasing volume of textual data. Text mining (TM) techniques can support curators by automatically detecting complex information, such as interactions between drugs, diseases and adverse effects. This semantic information supports the quick identification of documents containing information of interest (e.g., the different types of patients in which a given adverse drug reaction has been observed to occur). TM tools are typically adapted to different domains by applying machine learning methods to corpora that are manually labelled by domain experts using annotation guidelines to ensure consistency. We present a semantically annotated corpus of 597 MEDLINE abstracts, PHAEDRA, encoding rich information on drug effects and their interactions, whose quality is assured through the use of detailed annotation guidelines and the demonstration of high levels of inter-annotator agreement (e.g., 92.6% F-Score for identifying named entities and 78.4% F-Score for identifying complex events, when relaxed matching criteria are applied). To our knowledge, the corpus is unique in the domain of PV, according to the level of detail of its annotations. To illustrate the utility of the corpus, we have trained TM tools based on its rich labels to recognise drug effects in text automatically. The corpus and annotation guidelines are available at: http://www.nactem.ac.uk/PHAEDRA/ .

RESUMEN

BACKGROUND: Consumers increasingly use online resources for their health information needs. While current search engines can address these needs to some extent, they generally do not take into account that most health information needs are complex and can only fully be expressed in natural language. Consumer health question answering (QA) systems aim to fill this gap. A major challenge in developing consumer health QA systems is extracting relevant semantic content from the natural language questions (question understanding). To develop effective question understanding tools, question corpora semantically annotated for relevant question elements are needed. In this paper, we present a two-part consumer health question corpus annotated with several semantic categories: named entities, question triggers/types, question frames, and question topic. The first part (CHQA-email) consists of relatively long email requests received by the U.S. National Library of Medicine (NLM) customer service, while the second part (CHQA-web) consists of shorter questions posed to MedlinePlus search engine as queries. Each question has been annotated by two annotators. The annotation methodology is largely the same between the two parts of the corpus; however, we also explain and justify the differences between them. Additionally, we provide information about corpus characteristics, inter-annotator agreement, and our attempts to measure annotation confidence in the absence of adjudication of annotations. RESULTS: The resulting corpus consists of 2614 questions (CHQA-email: 1740, CHQA-web: 874). Problems are the most frequent named entities, while treatment and general information questions are the most common question types. Inter-annotator agreement was generally modest: question types and topics yielded highest agreement, while the agreement for more complex frame annotations was lower. Agreement in CHQA-web was consistently higher than that in CHQA-email. Pairwise inter-annotator agreement proved most useful in estimating annotation confidence. CONCLUSIONS: To our knowledge, our corpus is the first focusing on annotation of uncurated consumer health questions. It is currently used to develop machine learning-based methods for question understanding. We make the corpus publicly available to stimulate further research on consumer health QA.

Asunto(s)

RESUMEN

BACKGROUND: One important type of information contained in biomedical research literature is the newly discovered relationships between phenotypes and genotypes. Because of the large quantity of literature, a reliable automatic system to identify this information for future curation is essential. Such a system provides important and up to date data for database construction and updating, and even text summarization. In this paper we present a machine learning method to identify these genotype-phenotype relationships. No large human-annotated corpus of genotype-phenotype relationships currently exists. So, a semi-automatic approach has been used to annotate a small labelled training set and a self-training method is proposed to annotate more sentences and enlarge the training set. RESULTS: The resulting machine-learned model was evaluated using a separate test set annotated by an expert. The results show that using only the small training set in a supervised learning method achieves good results (precision: 76.47, recall: 77.61, F-measure: 77.03) which are improved by applying a self-training method (precision: 77.70, recall: 77.84, F-measure: 77.77). CONCLUSIONS: Relationships between genotypes and phenotypes is biomedical information pivotal to the understanding of a patient's situation. Our proposed method is the first attempt to make a specialized system to identify genotype-phenotype relationships in biomedical literature. We achieve good results using a small training set. To improve the results other linguistic contexts need to be explored and an appropriately enlarged training set is required.

Asunto(s)

RESUMEN

The free text notes typed by physicians during patient consultations contain valuable information for the study of disease and treatment. These notes are difficult to process by existing natural language analysis tools since they are highly telegraphic (omitting many words), and contain many spelling mistakes, inconsistencies in punctuation, and non-standard word order. To support information extraction and classification tasks over such text, we describe a de-identified corpus of free text notes, a shallow syntactic and named entity annotation scheme for this kind of text, and an approach to training domain specialists with no linguistic background to annotate the text. Finally, we present a statistical chunking system for such clinical text with a stable learning rate and good accuracy, indicating that the manual annotation is consistent and that the annotation scheme is tractable for machine learning.

RESUMEN

Ethical issues reported with paid crowdsourcing include unfairly low wages. It is assumed that such issues are under the control of the task requester. Can one control the amount that a worker earns by controlling the amount that one pays? 412 linguistic data development tasks were submitted to Amazon Mechanical Turk. The pay per HIT was manipulated through a range of values. We examined the relationship between the pay that is offered per HIT and the effective pay rate. There is no such relationship. Paying more per HIT does not cause workers to earn more: the higher the pay per HIT, the more time workers spend on them (R = 0.92). So, the effective hourly rate stays roughly the same. The finding has clear implications for language resource builders who want to behave ethically: other means must be found in order to compensate workers fairly. The findings of this paper should not be taken as an endorsement of unfairly low pay rates for crowdsourcing workers. Rather, the intention is to point out that additional measures, such as pre-calculating and communicating to the workers an average hourly, rather than per-task, rate must be found in order to ensure an ethical rate of pay.

RESUMEN

For the purpose of post-marketing drug safety surveillance, which has traditionally relied on the voluntary reporting of individual cases of adverse drug events (ADEs), other sources of information are now being explored, including electronic health records (EHRs), which give us access to enormous amounts of longitudinal observations of the treatment of patients and their drug use. Adverse drug events, which can be encoded in EHRs with certain diagnosis codes, are, however, heavily underreported. It is therefore important to develop capabilities to process, by means of computational methods, the more unstructured EHR data in the form of clinical notes, where clinicians may describe and reason around suspected ADEs. In this study, we report on the creation of an annotated corpus of Swedish health records for the purpose of learning to identify information pertaining to ADEs present in clinical notes. To this end, three key tasks are tackled: recognizing relevant named entities (disorders, symptoms, drugs), labeling attributes of the recognized entities (negation, speculation, temporality), and relationships between them (indication, adverse drug event). For each of the three tasks, leveraging models of distributional semantics - i.e., unsupervised methods that exploit co-occurrence information to model, typically in vector space, the meaning of words - and, in particular, combinations of such models, is shown to improve the predictive performance. The ability to make use of such unsupervised methods is critical when faced with large amounts of sparse and high-dimensional data, especially in domains where annotated resources are scarce.

Asunto(s)

RESUMEN

BACKGROUND: Chronic obstructive pulmonary disease (COPD) is a life-threatening lung disorder whose recent prevalence has led to an increasing burden on public healthcare. Phenotypic information in electronic clinical records is essential in providing suitable personalised treatment to patients with COPD. However, as phenotypes are often "hidden" within free text in clinical records, clinicians could benefit from text mining systems that facilitate their prompt recognition. This paper reports on a semi-automatic methodology for producing a corpus that can ultimately support the development of text mining tools that, in turn, will expedite the process of identifying groups of COPD patients. METHODS: A corpus of 30 full-text papers was formed based on selection criteria informed by the expertise of COPD specialists. We developed an annotation scheme that is aimed at producing fine-grained, expressive and computable COPD annotations without burdening our curators with a highly complicated task. This was implemented in the Argo platform by means of a semi-automatic annotation workflow that integrates several text mining tools, including a graphical user interface for marking up documents. RESULTS: When evaluated using gold standard (i.e., manually validated) annotations, the semi-automatic workflow was shown to obtain a micro-averaged F-score of 45.70% (with relaxed matching). Utilising the gold standard data to train new concept recognisers, we demonstrated that our corpus, although still a work in progress, can foster the development of significantly better performing COPD phenotype extractors. CONCLUSIONS: We describe in this work the means by which we aim to eventually support the process of COPD phenotype curation, i.e., by the application of various text mining tools integrated into an annotation workflow. Although the corpus being described is still under development, our results thus far are encouraging and show great potential in stimulating the development of further automatic COPD phenotype extractors.