This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Infodemiology, is properly cited. The complete bibliographic information, a link to the original publication on https://infodemiology.jmir.org/, as well as this copyright and license information must be included.
COVID-19 severity is amplified among individuals with obesity, which may have influenced mainstream media coverage of the disease by both improving understanding of the condition and increasing weight-related stigma.
We aimed to measure obesity-related conversations on Facebook and Instagram around key dates during the first year of the COVID-19 pandemic.
Public Facebook and Instagram posts were extracted for 29-day windows in 2020 around January 28 (the first US COVID-19 case), March 11 (when COVID-19 was declared a global pandemic), May 19 (when obesity and COVID-19 were linked in mainstream media), and October 2 (when former US president Trump contracted COVID-19 and obesity was mentioned most frequently in the mainstream media). Trends in daily posts and corresponding interactions were evaluated using interrupted time series. The 10 most frequent obesity-related topics on each platform were also examined.
On Facebook, there was a temporary increase in 2020 in obesity-related posts and interactions on May 19 (posts +405, 95% CI 166 to 645; interactions +294,930, 95% CI 125,986 to 463,874) and October 2 (posts +639, 95% CI 359 to 883; interactions +182,814, 95% CI 160,524 to 205,105). On Instagram, there were temporary increases in 2020 only in interactions on May 19 (+226,017, 95% CI 107,323 to 344,708) and October 2 (+156,974, 95% CI 89,757 to 224,192). Similar trends were not observed in controls. Five of the most frequent topics overlapped (COVID-19, bariatric surgery, weight loss stories, pediatric obesity, and sleep); additional topics specific to each platform included diet fads, food groups, and clickbait.
Social media conversations surged in response to obesity-related public health news. Conversations contained both clinical and commercial content of possibly dubious accuracy. Our findings support the idea that major public health announcements may coincide with the spread of health-related content (truthful or otherwise) on social media.
The SARS-CoV-2 virus and COVID-19 pandemic have fundamentally changed society. The first US case was reported on January 20, 2020, and the World Health Organization (WHO) declared COVID-19 a global pandemic on March 11. COVID-19 is an infectious respiratory disease associated with a range of symptoms, and severity may be amplified in individuals with chronic, preexisting conditions such as obesity. This link was first reported in the spring of 2020, and studies have estimated that obesity may increase the risk of hospitalization due to COVID-19 between 7% and 33% and death by 8% to 61% [
US mainstream media coverage of the association between COVID-19 severity and obesity peaked in October 2020, when then US president Trump contracted COVID-19 [
Social media enables the documentation of heterogeneous opinions in near real time, making it an attractive avenue to assess shifts in opinion in response to major events. In particular, Facebook and Instagram are two popular social media platforms that were accessed by 70% and 59% of Americans on a daily basis in 2021, respectively [
This study included temporal and topical analyses to characterize how obesity-related content evolved on Facebook and Instagram surrounding 4 major events related to COVID-19 in 2020. Two dates were selected given their relevance to the broader pandemic: January 20, the date of the first US case, and March 11, the date when the WHO declared COVID-19 a global pandemic. Two other dates were directly related to obesity: May 19, the approximate date that obesity and COVID-19 were linked, and October 2, the date when then US President Trump contracted COVID-19 and obesity was most discussed in the news media, according to data from Media Cloud, an open-source content aggregation tool [
Facebook and Instagram posts were collected from CrowdTangle, a public insights tool owned and operated by Facebook [
Interrupted time series analysis was performed for each date using autoregressive integrated moving average (ARIMA) models. This method was chosen given its ability to control for highly cyclical and serially correlated data prior to each date and model complex postevent effects using one or a combination of transfer functions, including “pulse,” “step,” and “ramp” effects [
Each model was developed using the
Obesity-related posts were clustered into various topics using BERTopic, with a minimum topic size of 100 [
For all analyses, a Bonferroni-adjusted critical value of
Institutional review board approval was not required for this study given the public-facing nature of the social media data used [
Between January 6 and October 16, 2020, there were 175,242 posts across 66,497 public Facebook pages, groups, and pages in the CrowdTangle repository that contained the words “obesity” or “obese.” There was no significant change in posting behavior in the 14 days after January 20 and March 11 compared to the 14 days prior (
Autoregressive integrated moving average models for Facebook posts per day. Values in italics denote statistical significance at the Bonferroni-adjusted threshold of
Category/date (parameters) | Estimate (95% CI) | |||||
|
||||||
|
|
|||||
|
|
Ramp ( |
–3.00 (–12.0 to 6.03) | .515 | ||
|
|
|||||
|
|
Pulse ( |
–126 (–322 to 70.3) | .209 | ||
|
|
|||||
|
|
Pulse ( |
|
|
||
|
|
Step ( |
500 (60.0 to 941) | .026 | ||
|
|
|||||
|
|
Pulse ( |
|
|
||
|
|
Ramp ( |
9.06 (2.84 to 15.3) | .004 | ||
|
||||||
|
|
|||||
|
|
Ramp ( |
–0.67 (–29.8 to 28.5) | .964 | ||
|
|
|||||
|
|
Pulse ( |
–14.5 (–208 to 179) | .883 | ||
|
|
|||||
|
|
Pulse ( |
87.4 (–23.7 to 198) | .123 | ||
|
|
Step ( |
– |
|
||
|
|
|||||
|
|
Pulse ( |
196 (27.4 to 364) | .023 | ||
|
|
Ramp ( |
–6.08 (–10.6 to –1.56) | .008 | ||
|
||||||
|
|
|||||
|
|
Ramp ( |
0.20 (–2.00 to 2.40) | .859 | ||
|
|
|||||
|
|
Pulse ( |
–29.2 (–93.6 to 35.2) | .374 | ||
|
|
|||||
|
|
Pulse ( |
2.88 (–58.3 to 64.1) | .927 | ||
|
|
Step ( |
–17.0 (–54.0 to 20.0) | .368 | ||
|
|
|||||
|
|
Pulse ( |
–25.8 (–96.6 to 45.1) | .476 | ||
|
|
Ramp ( |
–0.95 (–3.51 to 1.60) | .464 |
a
bAICc: sample-corrected Akaike information criterion.
cRamp (ωr) functions are 0 before the intervention and (t–T+1) after the intervention (where
dPulse (ωp) functions are 1 if it is the day of the intervention and 0 all other days.
eStep (ωs) functions are 0 before the intervention and 1 the day of and after the intervention.
Changes in interactions on obesity posts for the 14 days following January 20 and March 11 were not significant (
Autoregressive integrated moving average models for Facebook interactions. Values in italics denote statistical significance at the Bonferroni-adjusted threshold of
Category/date (parameters) | Estimate (95% CI) | |||||
|
||||||
|
|
|||||
|
|
Step (ωs,0)c | –61,169 (–139,294 to 16,957) | .125 | ||
|
|
|||||
|
|
Pulse (ωp,1)d | –3298 (–60,578 to 53,981) | .910 | ||
|
|
|||||
|
|
Pulse (ωp,2) |
|
|
||
|
|
Step (ωs,2) |
|
|
||
|
|
Ramp (ωr,2)e | –38,596 (–64,268 to –12,923) | .003 | ||
|
|
|||||
|
|
Pulse (ωp,3) |
|
|
||
|
|
Step (ωs,3) | 5791 (1449 to 10,134) | .009 | ||
|
||||||
|
|
|||||
|
|
Step (ωs,0) | 37,827 (–9152 to 84,807) | .115 | ||
|
|
|||||
|
|
Pulse (ωp,1) |
|
|
||
|
|
|||||
|
|
Pulse (ωp,2) | –10,492 (–86,771 to 65,786) | .787 | ||
|
|
Step (ωs,2) | 1826 (–43,247 to 46,890) | .937 | ||
|
|
Ramp (ωr,2) | –1139 (–5542 to 3265) | .612 | ||
|
|
|||||
|
|
Pulse (ωp,3) |
|
|
||
|
|
Step (ωs,3) | –49.6 (–29,591 to 29,492) | .997 | ||
|
||||||
|
|
|||||
|
|
Step (ωs,0) | –107 (–1835 to 1621) | .903 | ||
|
|
|||||
|
|
Pulse (ωp,1) | –2189 (–4738 to 359) | .092 | ||
|
|
|||||
|
|
Pulse (ωp,2) | –331 (–3632 to 2969) | .844 | ||
|
|
Step (ωs,2) | 377 (–5442 to 6196) | .899 | ||
|
|
Ramp (ωr,2) | –589 (–4759 to 3582) | .782 | ||
|
|
|||||
|
|
Pulse (ωp,3) | –138 (–1901 to 1626) | .878 | ||
|
|
Step (ωs,3) | 66.1 (–578 to 710) | .840 |
a
bAICc: sample-corrected Akaike information criterion.
cStep (ωs) functions are 0 before the intervention and 1 the day of and after the intervention.
dPulse (ωp) functions are 1 if it is the day of the intervention and 0 all other days.
eRamp (ωr) functions are 0 before the intervention and (t–T+1) after the intervention (where
Between January 6 and October 16, 2020, there were 18,129 posts across 3170 unique usernames in the CrowdTangle repository containing “obese” or “obesity.” Of the 4 dates, only a ramp effect after January 20 (
Of 175,242 obesity-related posts, 87,470 (49.9%) could not be classified into a topic; a random sample of these can be found in
The distribution of most frequent topics changed around each date (
Each topic also had distinct daily posting behavior (
Distribution of the top 10 most frequent topics for Facebook around each date of interest.
Longitudinal variations in top 10 most frequent topics about obesity on Facebook. The dashed lines indicate the 4 key dates of interest (January 20, March 11, May 19, and October 2, 2020).
Of the 18,129 obesity-related Instagram posts, 6856 (37.8%) could not be classified into a topic; a random sample of these can be found in
The ranking of topic frequency was consistent around each date (
Two Instagram topics changed significantly surrounding the 4 dates (
Distribution of the top 10 most frequent topics for Instagram around each date of interest.
Longitudinal variations in top 10 most frequent topics about obesity on Instagram. The dashed lines indicate the 4 key dates of interest (January 20, March 11, May 19, and October 2, 2020).
This study is the first to comprehensively evaluate obesity-related content throughout the pandemic on Facebook and Instagram. On Facebook, obesity-related content surged around the dates of 4 key news stories related to obesity or COVID-19. Posting behavior of obesity-related content on Instagram was not affected, although changes in interactions occurred. Frequent content on each platform had some overlapping themes (ie, COVID-19, bariatric surgery, childhood obesity, weight loss stories, and sleep), while other topics varied in popularity. These findings demonstrate how social media conversations regarding prevalent health conditions may be influenced by news media and global events.
On Facebook, there were immediate changes in posting and interaction behavior for obesity-related content on both May 19 and October 2 that were not sustained in the following 14 days. A similar effect was observed for interactions on Instagram. The lack of statistical significance in the control data for any of the May 19 outcomes provides evidence that change in online discussion about obesity was specific to the reported association between COVID-19 and obesity that was shared in the mainstream media on that day. For October 2, a significant pulse effect was observed for interactions on Facebook posts in the health control data, while the nonhealth control data remained insignificant. Since the keyword for the health control data (ie, “headache”) is also a symptom of COVID-19, this may suggest that the surge in interactions occurred on posts that discussed the same topic as the obesity data (ie, the prognosis of then US president Trump, who had contracted COVID-19).
When broken down by topic, 5 frequent topics overlapped. Three (ie, bariatric surgery, pediatric obesity, and sleep) were clinical in nature and received the fewest interactions from users, suggesting that social media may not be an ideal platform to communicate this kind of content. In contrast, weight loss stories were present on both platforms and received a high number of interactions. This consistency may suggest that individuals feel comfortable sharing personal stories on these platforms as a show of support for others, and frequent mentions of obesity online or in mainstream media may empower individuals to communicate their own experiences with obesity or weight loss [
Prior work has pointed to a limited amount of healthy dietary advice on Instagram among posts with the hashtags #weightloss or #quarantine15 [
The practical implications and importance of these finding are 3-fold. Broadly, the ability to isolate the impact that media mentions of public health topics have on social media discussion contributes to the growing body of literature that demonstrates how social media can help gauge public opinions during times of crisis [
Strengths of this study include its expansion on prior work to understand obesity discourse more broadly, inclusion of multiple social media platforms, and evaluation of both temporal and topical patterns. However, there are several limitations that are important to note. First, there were no demographic data for users who created and viewed the content, which is a common challenge of epidemiologic research on social media. This study attempted to address this in part by using multiple platforms, which broadened the possible generalizability of the study. For instance, while approximately 71% of US adults aged between 18 and 29 years report ever using each platform, only 13% of US adults over the age of 65 years report using Instagram, compared to 50% for Facebook. Differences exist in other demographic groups as well, including race, income, and education [
Overall, these findings suggest that the pandemic had distinct impacts on the frequency of and attention to obesity-related conversations on 2 popular social media platforms. Posts about obesity and corresponding interactions did not shift after two COVID-19–specific dates (ie, January 28 and March 11), suggesting that events of public health significance that do not relate to obesity do not dramatically alter conversations about the disease on Facebook and Instagram. In contrast, posts and interactions about obesity increased after 2 dates of importance to both COVID-19 and obesity (ie, May 19 and October 2). This pattern was not observed in health and nonhealth control data for the same time period, demonstrating how the relationship between COVID-19 and obesity amplified discussions about obesity. Clinical topics were similar between the platforms, as were weight loss stories. Dietary topics were more prevalent on Instagram, while “clickbait” was more prevalent on Facebook. Taken together, these results suggest that the impact of major public health events (including mainstream media attention and government campaigns) on social media discourse can be successfully isolated and monitored. Public health officials should consider leveraging social media campaigns to prevent the spread of misleading, deleterious content, such as misinformation that may spike around such events.
Models with matched p and q parameters.
Representative documents from Facebook topics.
Model selection and equations for Facebook topics.
Representative documents from Instagram topics.
Model selection and equations for Instagram topics.
sample-corrected Akaike information criterion
autoregressive integrated moving average
bidirectional encoder representations from transformers
latent Dirichlet allocation
World Health Organization
None declared.