The full search protocol for this review is registered under PROSPERO (CRD42020148170). This review follows PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [16]. We searched PubMed, SocINDEX, Web of Science, and PsycINFO on November 6 and 7, 2019, using the search query ([“Social media” OR twitter OR tweet* OR facebook OR instagram OR youtube OR tumblr OR reddit OR “web 2.0” OR “public comments” OR hashtag*] and [“public health” OR “health research” OR “community health” OR “population health”]). We focused on YouTube only; findings for X (formerly Twitter) and Facebook are presented elsewhere [17,18].

English-language peer-reviewed journal articles concerning health and YouTube published between January 1, 2006 (the first full year after YouTube’s founding), and October 31, 2019 (when we completed our search), were eligible. This review assessed research ethics practices when using YouTube as a data source; studies using YouTube as a source for an intervention, health information dissemination, or health education and promotion were excluded (Textbox 1).

After removing duplicates, 2 blinded reviewers screened each title and abstract or full text, if not clear from the title and abstract, for eligibility. We further screened the eligible studies’ reference lists.

We extracted data using a standardized extraction form including fields for publication year, public health topic, outcome type, whether institutional review board (IRB) approval was sought, IRB determination, whether informed consent was sought, and whether identifying information was included (defined as any information that could help identify a YouTube user, content creator, or commenter). We categorized health conditions as noncommunicable and chronic diseases, infectious diseases, mental health and substance use, injury, or other. We categorized outcome types as descriptive (content analysis yielding video themes through quantitative or narrative comparisons), quality (measuring video quality), attitudes (measuring the video’s/commenters’ sentiment), utility (measuring a video’s usefulness to viewers), or other (behavioral analysis, accessibility, etc). JPT performed the initial data extraction, and HEJ reviewed all data for accuracy. As this review focuses on ethical research processes, we did not include an assessment of quality or risk of bias.

We report the frequency and percentage of research ethics practices overall and by whether they included user-identifying information. We provide references in the main text for results with ≤10 studies to facilitate readability. Those with >10 studies in a result can be verified in Table S1 in Multimedia Appendix 1.

The most common health topic was noncommunicable or chronic disease (44/119, 37%), followed by mental health and substance use (26/119, 21.8%) and infectious disease (20/119, 16.8%). Within chronic diseases, cancer was the most prevalent (10/119, 8.4%) [25,28,29,35,58-60,79,98,124], while vaccination was the most prevalent topic among articles on infectious disease (7/119, 5.9%) [21,34,52,57,62,85,136]. Most studies used content analysis to describe the videos’ themes (89/119, 74.8%), quality or utility (35/119, 29.4%), and public opinion or attitudes (31/119, 26.1%; Table S1 in Multimedia Appendix 1).

Most (82/119, 68.9%) articles did not mention ethical considerations in the study design or data collection, or the authors stated that their study did not meet the definition of human participant research (16/119, 13.4%). Of the 15 (12.6%) sent to the IRB for review, 12 (10.1%) were deemed to be nonhuman research by the IRB, and 3 (2.5%) [57,65,111] received IRB approval (Table 1). None of the 3 IRB-approved studies contained identifying information, one having been explicitly instructed not to include identifying information by their ethics committee [57].

In total, 33 (27.7%) studies contained identifying information, which included video titles (18/119, 15.1%), direct quotations of comments or videos (14/119, 11.8%), URLs for videos (12/119, 10.1%), names of video content creators (7/119, 5.9%) [29,53,84,103,126,127,131], and video screen captures (2/119, 1.7%) [29,127]. One study that included direct quotes excluded usernames in an attempt to anonymize the data, stating they did so because the YouTube users were unable to consent to participation in the study [108]. Two additional studies chose not to include videos that depicted perceived adolescents or children [84,102].

One study on diabetes burnout in people with type 1 diabetes with personal stories from individuals with type 1 diabetes in the videos sought informed consent from content creators through email or by commenting on the video itself [19]. The study used an opt-out solicitation and received no requests to be removed from the study.

We found some variation in research practice around ethical review for studies using YouTube data. Nearly three-quarters did not seek IRB review, considering YouTube publicly available data and thus not human participant research. Nevertheless, 12.6% (15/119) sought ethical review. Further, some researchers chose to anonymize their presentation of results or not to include identifiable content from children or adolescents; one study in Canada indicated that their ethical review board required content anonymization [57]. This variation in practice suggests ambiguity around privacy, personal identifiers, consent, and definitions of nonhuman participant research when using YouTube data.

This systematic review has several limitations. The review includes articles published from 2006 to 2019; as such, any changes to and evolution of the ethical and methodological use of social media data for research since 2019 are not captured. Additionally, given the breadth of public health research and the inclusion of only English-language articles, some studies may have been missed.

Given the continual debate about the definition and complexity of “public space” in social media and the shift away from a binary view of public versus private [138,139], and toward a consideration of intended use and “imagined audiences” [140], clearer guidelines are needed in social media research ethics. When studies include videos of real people sharing their personal health-related experiences, the notions of privacy and consent are blurred, with the possibility of social harm to nonconsented individuals.

Clearer ethical guidelines around the use of “publicly available” social media data in scientific research are needed.