TL;DR
This paper tackles the communication gap between healthcare professionals and consumers by creating a standardized Chinese consumer health vocabulary (CHT). They developed a novel approach to extract and map layperson health terms from online forums and patient education materials to professional medical terms, providing a valuable resource for improving health information access and NLP applications.
What Problem Does This Paper Address?
Healthcare professionals and consumers often use different language to describe medical conditions and concerns. This mismatch creates significant communication barriers, potentially leading to misdiagnoses, ineffective treatments, and difficulties for consumers in finding and understanding health information online. While consumer health vocabularies (CHVs) exist for English, there's a notable lack of such standardized resources for Chinese. This paper aims to bridge this gap by creating a comprehensive Chinese consumer health vocabulary.
Methods
The researchers employed a multi-stage approach combining automated and manual methods:
- Corpus Collection: They gathered data from two primary sources:
- Online Health Forums: Specifically, questions from the "Health and Medicine" section of Baidu Knows, a large Q&A platform.
- Patient Education Monographs: Articles written by experts to explain medical concepts in accessible language.
- Data Preprocessing: Raw data was filtered to remove common professional terms already present in the Chinese Medical Subject Headings (CMeSH) database. This step aimed to focus on consumer-specific language.
- Medical Term Extraction:
- Automated Extraction (Monographs): They used text-mining techniques to identify "AKA" (also known as) terms, which often represent alternative names or abbreviations for diseases within patient education materials.
- Manual Annotation (Both Sources): Trained annotators, familiar with CMeSH, manually extracted disease names and their alternative consumer-expressed synonyms. This involved double-annotation with a third annotator resolving discrepancies.
- Term Mapping: Extracted consumer health terms were mapped to corresponding professional medical terms using CMeSH. This process involved identifying the closest CMeSH term or its parent node when an exact match wasn't found.
- Standardization: The mapped terms were then linked to their respective MeSH IDs, creating a standardized vocabulary.
Datasets
The study utilized two main datasets:
- Online Forum Data: 8,436 questions from four specific departments (endocrinology, cardiology, gastroenterology, dermatology) were sampled from a larger corpus of over 2 million questions.
- Patient Education Monographs: 1,428 articles were collected.
The paper also introduced the Chinese Consumer Health Terms (CHT) dataset, which is publicly available. This dataset contains 1036 standardized Chinese consumer health terms mapped to 480 medical terms (MeSH IDs), along with their frequency and source information.
Key Insights and Ideas
- Consumer Language is Diverse and Informal: Consumers frequently use abbreviations, colloquialisms, and even typos to express health concerns, highlighting the need for a vocabulary that captures this informal language.
- Online Forums as a Rich Source: Despite the informality, online forums provide a valuable window into how the general public discusses health, offering a complementary perspective to expert-written materials.
- Monographs Offer Depth: Patient education materials, while more formal, provide a structured way to identify alternative names and synonyms for medical terms, aiding in comprehensive mapping.
- The "AKA" Pattern is Crucial: Identifying phrases like "is also called as" or "for short" within expert texts is a key strategy for uncovering consumer-friendly synonyms.
- High Inter-Annotator Agreement: The study achieved a high average inter-annotator agreement (93.91%), indicating the reliability and consistency of their manual annotation process.
Key Results
- The study successfully extracted 1036 Chinese consumer health terms and mapped them to 480 medical terms (MeSH IDs).
- The dataset includes terms from four medical departments: endocrinology, cardiology, gastroenterology, and dermatology.
- The online forum data showed higher familiarity with consumer terms compared to patient education monographs, suggesting consumers are more likely to use these informal terms when searching for information.
- The inter-annotator agreement (IAA) scores were high: 92.5% for forum data, 95.3% for monograph data, and 93.91% overall for mention annotation.
- The CHT dataset is publicly available, serving as a resource for future research.
Limitations and Future Work
- Limited Scope: The current study focused only on disease terms and was limited to four medical departments.
- Future Work: The authors plan to expand the extraction to include drug names and symptom names, as well as extend the scope to other medical departments. They also aim to investigate named entity recognition and explore automated annotation technologies based on the CHT dataset to further scale and improve the vocabulary. The goal is to create a comprehensive resource for Chinese consumer health vocabulary development and text mining.
Summary updated