JMIR Publications

ABSTRACT

Background:

Online wellness influencers (individuals dispensing unregulated health and wellness advice over social media) may have incentives to oppose traditional medical authorities. Their messaging may decrease the overall effectiveness of public health campaigns during global health crises like the COVID-19 pandemic.

Objective:

As a probe for how wellness influencers respond to a public health campaign, we examined how a sample of wellness influencers on Twitter identified before the COVID-19 pandemic on Twitter took stances on the COVID-19 vaccine during 2020-2022. We evaluated the prevalence of pro-vaccination messaging among wellness influencers compared to a control group, as well as the rhetorical strategies these influencers used when supporting or opposing vaccination.

Methods:

Following a longitudinal design, wellness influencer accounts were identified on Twitter from a random sample of tweets posted in 2019. Accounts were identified using a combination of topic modeling and hand-annotation for adherence to influencer criteria. Their tweets from 2020-2022 containing vaccine keywords were collected and labeled as pro/anti-vaccination stances using a language model. We compared their stances to a Control group of non-influencer accounts that discussed similar health topics before the pandemic using a generalized linear model with mixed effects and a nearest-neighbors classifier. We also used topic modeling to locate key themes in influencer’s pro- and anti- vaccine messages.

Results:

Wellness influencers (n=161) had lower rates of pro-vaccination stances in their on-topic tweets (20%; 614/3,045) compared to controls (n=242 accounts, with 42% or 3,201/7,584 pro-vaccination tweets). Using generalized linear model of tweet stance with mixed effects to model tweets from the same account, the main effect of group was significant (β1=−2.2668, SE = 0.2940, P < .001). Covariate analysis suggests an association between anti-vaccination tweets and accounts representing individuals (β =−0.9591, SE = 0.2917, P = .001), but not social network position. A complementary modeling exercise of stance within user accounts showed a significant different in the proportion of anti-vaccination users by group χ2(1, N =321) =36.125, P < .001). While nearly half of influencer accounts were labeled by a K-nearest neighbor classifier as predominantly anti-vaccination (48%; 58/120), only 16% of control accounts were labeled this way (33/201). Topic modeling of influencer tweets showed that the most prevalent anti-vaccination themes were protecting children, guarding against government overreach, and the corruption of the pharmaceutical industry. Pro-vaccination messaging tended to encourage followers to take action or emphasize the efficacy of the vaccine.

Conclusions:

Wellness influencers showed higher rates of vaccine opposition compared to other accounts that participated in health discourse before the pandemic. This pattern supports the theory that unregulated wellness influencers have incentives to resist messaging from establishment authorities like public health agencies.