Introduction
China currently faces the world’s largest diabetes burden, with over 114 million people affected. As the internet has become a primary tool for patients to seek advice and share experiences, it offers a rich repository of data on consumer health information needs. However, healthcare providers often lack a systematic understanding of these needs. This study leverages modern deep learning to analyze a massive, 15-year longitudinal dataset from a major Chinese health portal, providing a data-driven roadmap for improving public health education and patient self-management.
Methods
Researchers developed a custom web crawler to aggregate 151,589 diabetes-related questions from 39 Health (39.net) spanning 2003 to 2017. To process this "noisy" data—characterized by medical misspellings and informal language—the team employed a two-step Natural Language Processing (NLP) pipeline:
- Word2vec: A word embedding model was trained on a large medical corpus to map words into 300-dimensional vectors, allowing the system to recognize semantically identical terms despite misspellings.
- Multi-channel CNN: A Convolutional Neural Network was used to capture semantic patterns across phrases, rather than looking at words in isolation.
- Active Learning: The model was refined through iterative annotation, using a manually labeled subset of 3,000 questions for training and validation.
Results
The CNN model achieved an impressive 96.7% accuracy in categorizing patient queries. The analysis revealed a clear hierarchy in what patients seek online.
Categorical Distribution
As shown in the figure above, the primary concerns were Diagnosis (34.95%), Treatment (25.17%), and Lifestyle (21.09%). Notably, despite the rising prevalence of the disease, only 1.23% of questions focused on Prevention, indicating a significant "prevention gap" where users primarily seek information only after a suspected or confirmed diagnosis.
Demographic and Longitudinal Trends
- Gender and Age: Women were more active seekers, accounting for 61.3% of queries. Information needs shifted dramatically with age: younger users (20–40) focused on maternity-related issues and prognosis, while older users (>60) prioritized treatment regimens and complications like cardiovascular disease and diabetic eye disease.
- The Lifestyle Pivot: Interest in diet and exercise nearly doubled after 2010, reflecting increased socioeconomic development and awareness of chronic disease management.
- Medication Trends: Insulin and Metformin were the most discussed drugs. However, significant interest in "Xiaoke pills" (a Chinese patent medicine) highlights a need for better education regarding the risks of hypoglycemia associated with certain traditional treatments.
Discussion and Conclusion
The study demonstrates that internet-based analysis can effectively map the landscape of patient concerns. The findings suggest that diabetes health education in China should be tailored to different age groups and must urgently address the lack of focus on prevention. Furthermore, as the volume of online health queries continues to grow—increasing tenfold between 2007 and 2017—healthcare providers should utilize these insights to improve patient follow-up and self-management support.
Summary updated