TL;DR
This paper explores the 2019 vaping-associated lung injury (EVALI) outbreak by analyzing data from four diverse online sources: a participatory surveillance platform (YouVape), Reddit, Google Trends, and Bing. By combining these complementary datasets, the researchers gained a more holistic understanding of vaping behaviors, product associations, and adverse reactions, demonstrating the power of multi-source digital surveillance for public health emergencies.
What Problem Does This Paper Address?
In September 2019, the CDC issued an alert about a concerning outbreak of lung illness linked to e-cigarette use. At that time, critical information about the causes, at-risk populations, and specific contributing factors was scarce. Traditional public health data collection methods, while valuable, can be time-consuming and costly, especially in the early stages of an emerging outbreak. This paper addresses the challenge of rapidly gathering and analyzing information to understand such a public health crisis by leveraging the vast and diverse data available online. The goal was to provide near real-time insights into the EVALI outbreak by integrating data from various digital sources that capture different facets of user behavior and reported experiences.
Methods
The study employed a multi-pronged approach to data collection and analysis, integrating four distinct online data sources:
- Participatory Surveillance (YouVape): This platform, developed by Boston Children's Hospital and Harvard Medical School, allowed self-reporting of sociodemographic information, vaping behaviors, and medical symptoms related to EVALI. Users were recruited via press releases, leading to a self-selected, potentially more engaged cohort.
- Web Search Data (Google Trends & Bing): Google Trends provided aggregate, anonymized data on search query volumes, offering insights into geographic spread and product popularity over time. Bing search data, on the other hand, allowed for the analysis of individual, anonymous user queries, which proved useful for identifying potential adverse reactions, similar to previous work on drug recalls.
- Social Media Data (Reddit): Data from the "Vaping101" subreddit was analyzed to understand user experiences, particularly focusing on first-time users to distinguish ongoing health issues from those related to the onset of vaping.
The analysis involved correlating data across these sources, identifying associations between specific vaping products/ingredients and adverse reactions using statistical methods like chi-squared tests and logistic regression. The researchers also developed a methodology to identify adverse reactions by comparing users who searched for vaping products with a control group who searched for symptoms but not vaping products.
Datasets
The study utilized data from four distinct online sources:
- YouVape: A participatory surveillance platform with 720 users, providing data from October 29, 2019, to January 25, 2020.
- Google Trends: Provided aggregate web search data, with the study focusing on the period from January 1, 2018, to January 1, 2020.
- Bing: Provided anonymous individual search data, with a focus on the nine months prior to the outbreak (October 2018 to June 2019). This included data from approximately 1.03 million users in a "vaping group" and 3.2 million in a control group.
- Reddit: Data was extracted from the "Vaping101" subreddit and other related subreddits, including 4331 users, covering the period from January 1, 2015, to December 31, 2019.
Key Insights and Ideas
- Complementary Nature of Digital Data: The study strongly emphasizes that no single online data source is sufficient. Each platform (participatory, search, social media) offers a unique perspective, and combining them provides a more comprehensive and robust understanding of complex public health issues like the EVALI outbreak.
- Geographic Variation in Vaping Products: Significant differences in the popularity of vaping products were observed across different geographic regions in the US, as indicated by Google Trends data.
- Ingredient-Symptom Associations: The analysis revealed specific correlations between certain vaping ingredients and adverse reactions. Notably, cannabidiol (CBD) was associated with fever, and tetrahydrocannabinol (THC) was linked to diarrhea.
- Product-Symptom Associations: Beyond ingredients, certain vaping products were also associated with specific symptoms. For instance, Juul was linked to cough, Dank Vape to nausea, and TKO to stomach pain.
- Demographic Trends: The study confirmed that males are more likely to vape, aligning with previous research. While YouVape users tended to be older than Reddit users, both platforms showed age distributions broadly consistent with national surveys.
- Purchase Location Matters: For "other" (non-branded) vaping products, the location of purchase was significantly associated with the likelihood of experiencing adverse reactions, with online purchases showing the lowest risk.
Key Results
- High Adverse Reaction Reporting: 60% of YouVape users reported experiencing adverse reactions.
- Correlation Across Data Sources: Significant correlations were found between the reported adverse reactions across the different data sources, validating the findings.
- Specific Ingredient/Product Associations:
- CBD was significantly associated with fever.
- THC was significantly associated with diarrhea.
- Juul was associated with cough.
- Dank Vape was associated with nausea.
- TKO was associated with stomach pain.
- Geographic Patterns: Google Trends data revealed distinct geographic patterns in the popularity of various vaping brands across the US.
- Demographic Patterns: YouVape data indicated that brand use was strongly associated with age but not gender. Conversely, vaping ingredients were associated with gender.
Limitations and Future Work
The authors acknowledge several limitations:
- Selection Bias: Participatory surveillance platforms like YouVape and community forums like Reddit may attract users who are more engaged or symptomatic, potentially leading to a selection bias that affects the generalizability of findings to the broader population.
- Self-Identification: User data from these platforms is based on self-identification, which is not independently verified.
- Limited Sample Size: While YouVape provided detailed information, its user base was relatively small, limiting the volume of data.
- Known Problems Only: Participatory surveillance mechanisms can only be initiated once a problem is recognized.
- Control Symptom Limitations: While control symptoms were used to estimate baseline reporting, the difference in reporting rates between known and control symptoms was not statistically significant, suggesting a need for more nuanced control symptom selection in future studies.
Future work could involve:
- Further Investigation of Specific Symptoms: Investigating the newly identified adverse symptoms (e.g., abdomen pain, shivering, hemoptysis) and their potential links to specific brands or ingredients.
- Replication with Similar Data Sources: Encouraging the use of similar private datasets to replicate findings and expand the scope of research.
- Longitudinal Studies: Developing methods for continuous digital cohort monitoring that can be augmented with active inquiries during emerging health crises.
- Exploring Additional Data Sources: Investigating other online platforms and data streams to further enrich the understanding of vaping-related health issues.
Summary updated