Metadata
Title
Identifying Medical Terms in Patient-Authored Text: A Crowdsourcing-based Approach
Category
general
UUID
a1232203c4a5445b8c002a08c6022a66
Source URL
https://idl.uw.edu/papers/adept
Parent URL
https://idl.uw.edu/papers
Crawl Time
2026-03-11T03:15:28+00:00
Rendered Raw Markdown

Identifying Medical Terms in Patient-Authored Text: A Crowdsourcing-based Approach

Source: https://idl.uw.edu/papers/adept Parent: https://idl.uw.edu/papers

Diana MacLean, Jeffrey Heer. J Am Med Inform Assoc, 2013

Diana MacLean, Jeffrey Heer

J Am Med Inform Assoc, 2013

A comparison of terms identified as medically-relevant (shown in black) by different models in five sample sentences. OBA and MetaMap are run using the SNOMED CT ontology.

Materials

PDF | Journal | Software

Abstract

Background and objective: As people increasingly engage in online health-seeking behavior and contribute to health-oriented websites, the volume of medical text authored by patients and other medical novices grows rapidly. However, we lack an effective method for automatically identifying medical terms in patient-authored text (PAT). We demonstrate that crowdsourcing PAT medical term identification tasks to non-experts is a viable method for creating large, accurately-labeled PAT datasets; moreover, such datasets can be used to train classifiers that outperform existing medical term identification tools. Materials and methods: To evaluate the viability of using non-expert crowds to label PAT, we compare expert (registered nurses) and non-expert (Amazon Mechanical Turk workers; Turkers) responses to a PAT medical term identification task. Next, we build a crowd-labeled dataset comprising 10 000 sentences from MedHelp. We train two models on this dataset and evaluate their performance, as well as that of MetaMap, Open Biomedical Annotator (OBA), and NaCTeM's TerMINE, against two gold standard datasets: one from MedHelp and the other from CureTogether. Results: When aggregated according to a corroborative voting policy, Turker responses predict expert responses with an F1 score of 84%. A conditional random field (CRF) trained on 10 000 crowd-labeled MedHelp sentences achieves an F1 score of 78% against the CureTogether gold standard, widely outperforming OBA (47%), TerMINE (43%), and MetaMap (39%). A failure analysis of the CRF suggests that misclassified terms are likely to be either generic or rare. Conclusions: Our results show that combining statistical models sensitive to sentence-level context with crowd-labeled data is a scalable and effective technique for automatically identifying medical terms in PAT.

BibTeX

@article{2013-adept,
  title = {Identifying Medical Terms in Patient-Authored Text: A Crowdsourcing-based Approach},
  author = {MacLean, Diana AND Heer, Jeffrey},
  journal = {J Am Med Inform Assoc},
  year = {2013},
  url = {https://idl.uw.edu/papers/adept},
  doi = {10.1136/amiajnl-2012-001110}
}

{"status":200,"statusText":"","headers":{},"body":"[\n {\n \"fullName\": \"Proc. ACM Human Factors in Computing Systems (CHI)\",\n \"nickname\": \"CHI\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"IEEE Trans. Visualization & Comp. Graphics (Proc. VIS)\",\n \"nickname\": \"VIS\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Computer Graphics Forum (Proc. EuroVis)\",\n \"nickname\": \"EuroVis\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. EuroVis Short Papers\",\n \"nickname\": \"EuroVis-Short\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. IEEE VIS Short Papers\",\n \"nickname\": \"VIS-Short\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. ACM User Interface Software & Technology (UIST)\",\n \"nickname\": \"UIST\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. ACM Computer-Supported Cooperative Work (CSCW)\",\n \"nickname\": \"CSCW\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. ACM Intelligent User Interfaces\",\n \"nickname\": \"IUI\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"ACM Trans. on Computer-Human Interaction\",\n \"nickname\": \"ACM TOCHI\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. Advanced Visual Interfaces\",\n \"nickname\": \"AVI\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. Conference on Innovative Data Systems Research (CIDR)\",\n \"nickname\": \"CIDR\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. Very Large Database Endowment (PVLDB)\",\n \"nickname\": \"PVLDB\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. Empirical Methods in Natural Language Processing\",\n \"nickname\": \"EMNLP\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. NAACL-HLT\",\n \"nickname\": \"NAACL-HLT\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. International Conference on Weblogs and Social Media (ICWSM)\",\n \"nickname\": \"ICWSM\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"IEEE Trans. Visualization & Comp. Graphics (Proc. InfoVis)\",\n \"nickname\": \"InfoVis\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Beautiful Data\",\n \"nickname\": \"Beautiful Data\",\n \"venueType\": \"book\"\n },\n {\n \"fullName\": \"Information Visualization Journal\",\n \"nickname\": \"IV Journal\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. IEEE Visual Analytics Science & Technology (VAST)\",\n \"nickname\": \"VAST\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Cortex\",\n \"nickname\": \"Cortex\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. Hawaii International Conference on Systems Sciences (HICSS)\",\n \"nickname\": \"HICSS\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. IEEE Information Visualization (InfoVis)\",\n \"nickname\": \"InfoVis (Pre-TVCG)\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. Ubiquitous Computing\",\n \"nickname\": \"UbiComp\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. WEBKDD Workshop\",\n \"nickname\": \"WEBKDD\",\n \"venueType\": \"workshop\"\n },\n {\n \"fullName\": \"ACM Trans. on Information Systems\",\n \"nickname\": \"ACM TOIS\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Communications of the ACM\",\n \"nickname\": \"CACM\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. Workshop on Social Network Mining & Analysis, ACM KDD\",\n \"nickname\": \"SNAKDD\",\n \"venueType\": \"workshop\"\n },\n {\n \"fullName\": \"Proc. Social Visualization Workshop, ACM CHI\",\n \"nickname\": \"CHI Social Vis\",\n \"venueType\": \"workshop\"\n },\n {\n \"fullName\": \"Proc. AVI Workshop on Invisible & Transparent Interfaces\",\n \"nickname\": \"AVI ITI\",\n \"venueType\": \"workshop\"\n },\n {\n \"fullName\": \"Proc. Color Imaging Conference\",\n \"nickname\": \"Color Imaging Conf.\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. Workshop on Applications for Topic Models, NIPS\",\n \"nickname\": \"NIPS Topic Model Ws\",\n \"venueType\": \"workshop\"\n },\n {\n \"fullName\": \"Proc. Mining Software Repositories\",\n \"nickname\": \"MSR\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Journal of Animal Ecology\",\n \"nickname\": \"J Anim Eco\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"J Am Med Inform Assoc\",\n \"nickname\": \"JAMIA\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. International Conference on Machine Learning (ICML)\",\n \"nickname\": \"ICML\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Computer Graphics and Applications\",\n \"nickname\": \"CG&A\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. IEEE Biological Data Visualization (BioVis)\",\n \"nickname\": \"BioVis\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Poetics\",\n \"nickname\": \"Poetics\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. ACM Web Search and Data Mining (WSDM)\",\n \"nickname\": \"WSDM\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. User Modeling and User-Adapted Interaction (UMUAI)\",\n \"nickname\": \"UMUAI\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. Workshop on Eye Tracking and Visualization (ETVIS)\",\n \"nickname\": \"ETVIS\",\n \"venueType\": \"workshop\"\n },\n {\n \"fullName\": \"Trends in Ecology & Evolution\",\n \"nickname\": \"TREE\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"PLOS ONE\",\n \"nickname\": \"PLOS ONE\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. ACM SIGMOD Human-in-the-Loop Data Analysis (HILDA)\",\n \"nickname\": \"HILDA\",\n \"venueType\": \"workshop\"\n },\n {\n \"fullName\": \"IEEE Trans. Visualization & Comp. Graphics (Proc. VAST)\",\n \"nickname\": \"VAST-TVCG\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. Workshop on Dealing with Cognitive Biases in Visualisations (DECISIVe), IEEE VIS\",\n \"nickname\": \"DECISIVe\",\n \"venueType\": \"workshop\"\n },\n {\n \"fullName\": \"arXiv\",\n \"nickname\": \"arXiv\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"The Journal of Open Source Software\",\n \"nickname\": \"JOSS\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proceedings of the National Academy of Sciences\",\n \"nickname\": \"PNAS\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. Association for Computational Linguistics (ACL)\",\n \"nickname\": \"ACL\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Distill\",\n \"nickname\": \"Distill\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Harvard Data Science Review\",\n \"nickname\": \"HDSR\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Organizational Behavior and Human Decision Processes\",\n \"nickname\": \"OBHDP\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"EPJ Data Science\",\n \"nickname\": \"EPJ-DS\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. IEEE Symposium on Visual Languages and Human Centric Computing (VL/HCC)\",\n \"nickname\": \"VL/HCC\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Proc. ACM Management of Data (SIGMOD)\",\n \"nickname\": \"SIGMOD\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Companion of ACM Management of Data (SIGMOD)\",\n \"nickname\": \"SIGMOD-Demo\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"IEEE Trans. Visualization & Comp. Graphics\",\n \"nickname\": \"TVCG\",\n \"venueType\": \"journal\"\n },\n {\n \"fullName\": \"Proc. ACM Creativity & Cognition\",\n \"nickname\": \"C&C\",\n \"venueType\": \"conference\"\n },\n {\n \"fullName\": \"Workshop on Intelligent and Interactive Writing Assistants (In2Writing)\",\n \"nickname\": \"In2Writing\",\n \"venueType\": \"workshop\"\n }\n]\n"} {"status":200,"statusText":"","headers":{},"body":"{\n \"title\": \"Identifying Medical Terms in Patient-Authored Text: A Crowdsourcing-based Approach\",\n \"year\": 2013,\n \"start_page\": null,\n \"end_page\": null,\n \"volume\": null,\n \"issue\": null,\n \"editors\": \"\",\n \"publisher\": \"\",\n \"location\": \"\",\n \"pdf\": \"https://idl.cs.washington.edu/files/2013-ADEPT-JAMIA.pdf\",\n \"abstract\": \"Background and objective: As people increasingly engage in online health-seeking behavior and contribute to health-oriented websites, the volume of medical text authored by patients and other medical novices grows rapidly. However, we lack an effective method for automatically identifying medical terms in patient-authored text (PAT). We demonstrate that crowdsourcing PAT medical term identification tasks to non-experts is a viable method for creating large, accurately-labeled PAT datasets; moreover, such datasets can be used to train classifiers that outperform existing medical term identification tools.\r\n\r\nMaterials and methods: To evaluate the viability of using non-expert crowds to label PAT, we compare expert (registered nurses) and non-expert (Amazon Mechanical Turk workers; Turkers) responses to a PAT medical term identification task. Next, we build a crowd-labeled dataset comprising 10 000 sentences from MedHelp. We train two models on this dataset and evaluate their performance, as well as that of MetaMap, Open Biomedical Annotator (OBA), and NaCTeM's TerMINE, against two gold standard datasets: one from MedHelp and the other from CureTogether.\r\n\r\nResults: When aggregated according to a corroborative voting policy, Turker responses predict expert responses with an F1 score of 84%. A conditional random field (CRF) trained on 10 000 crowd-labeled MedHelp sentences achieves an F1 score of 78% against the CureTogether gold standard, widely outperforming OBA (47%), TerMINE (43%), and MetaMap (39%). A failure analysis of the CRF suggests that misclassified terms are likely to be either generic or rare.\r\n\r\nConclusions: Our results show that combining statistical models sensitive to sentence-level context with crowd-labeled data is a scalable and effective technique for automatically identifying medical terms in PAT.\",\n \"thumbnail\": \"images/thumbs/adept.png\",\n \"figure\": \"images/figures/adept.png\",\n \"caption\": \"A comparison of terms identified as medically-relevant (shown in black) by different models in five sample sentences. OBA and MetaMap are run using the SNOMED CT ontology.\",\n \"web_name\": \"adept\",\n \"visible\": true,\n \"mod_date\": \"2013-05-07\",\n \"note\": \"\",\n \"pub_date\": \"2013-05-06\",\n \"venue\": \"JAMIA\",\n \"authors\": [\n {\n \"first_name\": \"Diana\",\n \"last_name\": \"MacLean\",\n \"url\": \"http://www.stanford.edu/~malcdi/\"\n },\n {\n \"first_name\": \"Jeffrey\",\n \"last_name\": \"Heer\",\n \"url\": \"http://homes.cs.washington.edu/~jheer/\"\n }\n ],\n \"materials\": [\n {\n \"name\": \"Journal\",\n \"link\": \"http://jamia.bmj.com/content/early/2013/05/04/amiajnl-2012-001110.full\"\n },\n {\n \"name\": \"Software\",\n \"link\": \"http://vis.stanford.edu/projects/adept\"\n }\n ],\n \"tags\": [],\n \"doi\": \"10.1136/amiajnl-2012-001110\"\n}"}

{ __sveltekit_17copn9 = { base: new URL("..", location).pathname.slice(0, -1), assets: "/uwdata.github.io" }; const element = document.currentScript.parentElement; const data = [null,null]; Promise.all([ import("../_app/immutable/entry/start.CZdZnu7S.js"), import("../_app/immutable/entry/app.qRA-U4ZQ.js") ]).then(([kit, app]) => { kit.start(app, element, { node_ids: [0, 7], data, form: null, error: null }); }); }