Communication About Childhood Obesity on Twitter

Abstract

Objectives. Little is known about the use of social media as a tool for health communication. We used a mixed-methods design to examine communication about childhood obesity on Twitter.

Methods. NodeXL was used to collect tweets sent in June 2013 containing the hashtag #childhoodobesity. Tweets were coded for content; tweeters were classified by sector and health focus. Data were also collected on the network of follower connections among the tweeters. We used descriptive statistics and exponential random graph modeling to examine tweet content, characteristics of tweeters, and the composition and structure of the network of connections facilitating communication among tweeters.

Results. We collected 1110 tweets originating from 576 unique Twitter users. More individuals (65.6%) than organizations (32.9%) tweeted. More tweets focused on individual behavior than environment or policy. Few government and educational tweeters were in the network, but they were more likely than private individuals to be followed by others.

Conclusions. There is an opportunity to better disseminate evidence-based information to a broad audience through Twitter by increasing the presence of credible sources in the #childhoodobesity conversation and focusing the content of tweets on scientific evidence.

The prevalence of obese and overweight youths in the United States has nearly doubled in the past 20 years,^1,2 with 32% of children and young people 2 to 19 years of age classified as overweight or obese as of 2012.³ Obesity contributes to poor health,^4–6 social problems,^6,7 and impaired school performance.^4,5,8 Being obese in childhood increases the risk of obesity in adulthood and the development of chronic diseases such as diabetes, hypertension, ischemic heart disease, and stroke.⁹ These conditions increase morbidity, reduce quality of life, and result in millions of dollars in health care–related costs.² Despite childhood obesity rates dropping by small but significant amounts among low-income children in 19 states between 2008 and 2011,¹⁰ evidence identifying effective interventions to combat childhood obesity remains limited.^11,12

As the use of social media has increased in recent years, its use as a way to find and share health information has also grown. Social media platforms are widely used by health care providers and public health practitioners^13–15 to share information with each other during training^16,17 and practice,¹⁸ reach the public with health information,^19–22 conduct surveillance,^23–25 and manage large-scale emergency situations.^26,27 As of 2013, 72% of online adults in the United States were using social media.²⁸ Of adult social media users, 23% follow their friends’ personal health experiences or updates, 17% use social media to remember or memorialize people with a specific health condition, and 15% obtain health information from social media sites.²⁹ Those using social media for purposes related to health tend to be consuming information³⁰; a few studies examining social media interventions for promoting healthy behavior have shown evidence of success in encouraging small behavior changes.^31,32

Twitter is one free social media application for microblogging, or brief, direct, one-to-many communications at little to no cost.^19,33 Twitter accounts can be followed by other Twitter users, allowing individuals or organizations to receive and spread information or “retweet” to others in their network. Twitter use among US adults more than doubled between 2010 and 2013, to 18% of adults overall.²⁸ In 2012 frequency of Twitter use, independent of gender, educational attainment, and income, was significantly higher among younger (vs older) age groups, among those residing in urban and suburban (vs rural) areas, and among Black non-Hispanics (vs White non-Hispanics and Hispanics).³⁴

By 2013, the frequency of Twitter use was significantly higher among Hispanics than among non-Hispanic Whites.²⁸ In addition, 24% of online teens between 12 and 17 years of age used Twitter in 2013, an increase from 16% in 2011.³⁵ Patterns of use suggest that Twitter may provide an important channel for reaching traditionally difficult-to-reach populations, including lower income, Hispanic, and non-Hispanic Black groups facing significantly higher rates of childhood obesity than their higher income and non-Hispanic White counterparts.³

Despite its widespread use by the public and public health professionals, there remains a shortage of evidence regarding the influence of social media on public health.^23,30,36 Examining social media can provide unique insights into the health information reaching, and possibly influencing, large segments of the general population.^37,38

To better understand Twitter use in communications among the public and public health professionals about childhood obesity, we collected and examined tweets incorporating the hashtag #childhoodobesity and the Twitter users who sent them. We organized our study around 5 inputs from McGuire’s communication–persuasion matrix known to influence communication effectiveness: source, message, channel, receiver, and destination.^39,40 These inputs correspond directly with Lasswell’s description of the communication process: who says what to whom in which channel with what effect?⁴¹ Specifically, we examined the characteristics of Twitter users tweeting about childhood obesity, the content of childhood obesity tweets, and the numbers and types of Twitter followers receiving tweets about childhood obesity.

METHODS

Hashtags are metadata embedded in tweets that make it easier to find and share tweets. Hashtags facilitate the formation of ad hoc groups interested in specific topics or events⁴² and are positively associated with audience engagement (i.e., retweeting).^43,44 We used online tools (e.g., hashtagify.me) to identify hashtags specific to childhood obesity and found 2: #childhoodobesity and #childobesity. Because of the low frequency of use of #childobesity, we selected #childhoodobesity as our only search term. Using a search function in the NodeXL plug-in for Microsoft Excel,⁴⁵ we collected all tweets containing this hashtag in June 2013. As part of the search, we collected data on the number of followers for each Twitter user who tweeted with the hashtag and the other hashtags in each tweet. Finally, we collected data on the “who-follows-whom” network among Twitter users.

Coding

We coded each Twitter user (tweeter) with respect to type of user (individual, organization, unable to determine), health focus (yes, no, unable to determine, spam), and sector (private individual, education, government, nonprofit, for profit, media, unable to determine, spam). Two of the authors used a modified version of a Twitter user codebook previously tested for reliability⁴⁶ to code each tweeter, and they came to an agreement on final classifications. Tweets were reviewed for emerging themes to develop a codebook including 6 main categories: behavior, environment, policy, medical–heritable, consequences of obesity, childhood obesity as a problem, and something else. Tweets were assigned as many codes as relevant. We also coded whether each tweet appeared to be pro-health, anti-health (promoting unhealthy choices), or neither. Two of the authors classified each tweet independently and came to an agreement on final codes.

Finally, we collected information on whether each tweet was original or retweeted and whether any Twitter users were mentioned in the tweet by searching tweets for “RT” and “@.” Retweeting is forwarding a tweet sent by another user and adding “RT” to the tweet to show that it is not original. Mentions use the “@” symbol to include a specific Twitter user in a tweet to reach the specified user directly or to indicate that the user is relevant to the tweet.⁴⁷ RT and @ are often used together to demonstrate that a tweet has been retweeted from a specific user; including “RT @username” indicates that the tweet was retweeted from @username.

Data Analysis

We used 3 strategies to examine tweeters’ characteristics: descriptive statistics, network descriptive statistics and visualization (to determine which tweeters were common sources of information for others in the network), and network modeling (to examine characteristics associated with being followed by others, or being a source of information, within the network). We used IBM SPSS for descriptive statistics,⁴⁸ Pajek64 for network descriptive statistics and visualization,⁴⁹ and R-statnet for network modeling.^50,51

In a network of follower relationships on Twitter, links are directed, going from one Twitter user to another. For example, when B follows A, B receives tweets sent by A. This relationship is represented by A→B, where arrow direction represents information flow. The number of arrows coming from any network member is measured via outdegree centrality. For example, if A were followed by 10 people, 10 arrows would come from A, for an outdegree of 10. Outdegree was used to identify tweeters sending information to others in the network.

Exponential random graph modeling (ERGM) is a statistical technique similar to logistic regression wherein the outcome is a link between 2 network members.⁵¹ In this case, ERGM was used to estimate the probability of a tie between any 2 Twitter users based on their characteristics and overall network structures. Specifically, we tested whether Twitter users from different sectors were likely to be sources of information. In addition, past network research identified a tendency for homophily in observed networks, or for network members similar on certain characteristic to be connected.^52,53 We assessed homophily with respect to health focus and pro-health tweeting. To control for frequent tweeting, which is associated with having more followers,¹³ and to account for having many followers overall, we included these characteristics in the model. An edges term analogous to the constant in logistic regression is typically included in exponential random graph models to account for the number of ties in the network.

Observed networks typically differ from randomly generated networks of the same size and density in 2 primary ways: distribution of degree and amount of transitivity. In a random network, links are randomly distributed, resulting in most network members having about the same number of links to others. In an observed network, links are often geometrically distributed, with a small number of well-connected network members and many members with few connections. In addition, observed networks often have more transitivity. That is, network members who are connected to each other have a tendency to have connections to the same other network members. Terms accounting for these features include a geometrically weighted term for outdegree distribution (geometrically weighted outdegree) and terms for 2 types of clustering associated with transitivity (geometrically weighted edge-wise shared partnerships and geometrically weighted dyad-wise shared partnerships).⁵⁴

Following Goodreau, we began with a null model, added main effects and homophily terms, and then added geometrically weighted terms.^51,54 To assess model fit, we simulated 100 networks from each model and assessed how well the outdegree distribution from the observed network was captured by the simulated networks. For example, the observed network included 216 network members with an outdegree value of 0; if more than 95% of the simulated networks included at least 216 network members with an outdegree of 0, the model accurately captured this characteristic. In addition, statistical measures of model fit (Akaike information criterion [AIC], Bayesian information criterion [BIC]) are reported. Although not considered the most appropriate means of assessing ERGM fit, given that the data do not meet standard assumptions, the AIC and BIC tend to correspond with simulation-based model fit measures and are often reported.⁵⁵

We used descriptive statistics to examine how many tweets fell into each category—behavior, environment, policy, medical–heritable, consequences of obesity, childhood obesity as a problem, and something else—and what proportion of tweets were sent most often by different types of tweeters.

Finally, we summed the number of followers for each Twitter user and each tweet sent to compute the number of impressions, or the number of times a #childhoodobesity tweet appeared in a Twitter feed. In addition, we visualized the network with members sized by indegree and calculated the average indegree value by sector to determine who was more likely to be following others in the network. Indegree measures how many incoming links a network member has; in this network, it shows how many sources of information a network member is receiving tweets from.

RESULTS

In June 2013, 576 unique tweeters sent 1110 tweets using the #childhoodobesity hashtag. The tweeters had a median of 322 followers (range = 0–80 925) and tweeted a median of 1520 times (range = 6–117 450). Most tweeters sent a single childhood obesity tweet, but 93 sent more than 1. One tweeter, a nonprofit organization focused on health, sent 69 #childhoodobesity tweets announcing a single event. We did not identify any unique coordinated social media efforts using the #childhoodobesity hashtag during the month.

Mentions and retweets are indicators of engagement,⁵⁶ and including a hashtag can prompt engagement (i.e., retweeting).⁴³ More than half (n = 618; 55.7%) of the tweets included a mention. About one quarter (25.1%) were retweeted from another source (n = 279). There were 438 unique hashtags aside from #childhoodobesity included in 697 of the tweets; the 10 most common were #obesity (n = 59), #health (n = 45), #nutrition (n = 45), #move (n = 37), #coc13 (n = 35), #healthykids (n = 31), #childobesity (n = 30), #playcityla (n = 28), #healthy (n = 23), and #physed (n = 22).

Source

More individuals (n = 378; 65.6%) than organizations (n = 185; 32.9%) tweeted using #childhoodobesity. Of the 378 individuals, 244 were private individuals not representing an organization or business with their Twitter account. More tweeters were health focused in their profiles (n = 309) than non–health focused (n = 267). However, more private individuals were non–health focused (n = 183) than health focused (n = 61). Figure 1 shows the distribution of health focus by sector. Private individuals had the fewest followers (median = 183.5), followed by tweeters in the education (median = 199), for-profit (median = 525), nonprofit (median = 556.5), government (median = 680), and media (median = 723.5) sectors. Table 1 presents examples of common categories of tweeters.

FIGURE 1—

Health focus and sector of #childhoodobesity tweeters, June 2013.

TABLE 1—

Examples of Common Categories of #childhoodobesity Tweeters: June 2013

Type of Follower/Health Focus	Twitter User’s Self-Description
Private person
Yes	Love to read, travel and eat/Public Health Nutritionist, MPH
No	Live your life your way, have fun, take it as it comes and enjoy it while it lasts:) I love cats:) &music:) I post videos on youtube & make a fool of myself
Nonprofit organization
Yes	Working together to prevent obesity among children younger than 5, through healthy changes in our homes and our communities
No	We provide free creative writing workshops throughout NYC for formerly voiceless members of society. http://t.co/rYWnhi8Ewo #nywriteathon
Media member
Yes	Investigative reporter, Caller-Times. Diabetes, health care, education in Corpus Christi and Coastal Bend
No	I cover K-12 and #highered for the Corpus Christi Caller-Times. Follow me for the latest #education information concerning #SouthTexas

Many of the 576 tweeters were following each other on Twitter, forming a large network. From the time of tweet collection in June 2013 until collection of data on network ties in July 2013, 12 tweeters had changed their account name or left Twitter and were therefore not included in the network. Figure 2a shows the entire network, with node shading representing sector and node size representing outdegree centrality; the larger the node, the more followers it has. Consistent with the figure showing several large nonprofit nodes in the center of the network, the median number of followers in the network was highest for the nonprofit sector (median = 4), followed by government (median = 3.5), education (median = 2.5), for profit (median = 2), media (median = 1), and private individuals (median = 0). Note that 152 of the 244 private individuals in the network had no followers in the network; 127 were isolates, with no connections in the network.

FIGURE 2—

Largest connected network of #childhoodobesity tweeters (n = 358), with shading showing health focus by (a) the number of followers within the network and (b) the number following within the network: June 2013.

Model 3 had the best fit (Table 2), with the highest percentage of the outdegree distribution captured by simulated networks and the lowest AIC and BIC values; we therefore adopted model 3 as the final model. After control for other network structures and network member characteristics, network members from the government sector were 1.83 times more likely to be followed (95% confidence interval [CI] = 1.35, 2.49) than private individuals, and those from the media sector were 1.53 times more likely to be followed (95% CI = 1.16, 2.01).

TABLE 2—

Predictors of Being Followed Among Twitter Users Tweeting About #childhoodobesity: June 2013

Variable	Null Model, OR (95% CI)	Model 1, OR (95% CI)	Model 2, OR (95% CI)	Model 3, OR (95% CI)
Constant	0.009 (0.009, 0.010)	0.003 (0.003, 0.003)	0.0003 (0.0002, 0.0004)	0.001 (0.00001, 0.480)
Main effects
General predictors of being followed
No. of followers		1.04 (1.04, 1.05)	1.04 (1.04, 1.04)	1.03 (1.03, 1.04)
No. of tweets		0.99 (0.99, 0.99)	1.00 (1.00, 1.00)	0.99 (0.99, 1.00)
Influence on being followed, by sector membership
Private person (Ref)		1.00	1.00	1.00
Education		2.79 (2.05, 3.81)	1.12 (0.82, 1.53)	1.09 (0.89, 1.33)
Government		8.90 (7.42, 10.67)	4.44 (3.69, 5.34)	1.83 (1.35, 2.49)
Nonprofit		6.95 (6.18, 7.81)	3.42 (3.03, 3.86)	1.41 (0.76, 2.62)
For profit		2.56 (2.25, 2.91)	1.22 (1.07, 1.39)	1.06 (0.78, 1.43)
Media		4.44 (3.87, 5.09)	2.34 (2.03, 2.70)	1.53 (1.16, 2.01)
Other		1.24 (0.31, 5.00)	0.47 (0.12, 1.90)	1.21 (0.04, 36.79)
Homophily
Tweeter has health focus in profile
No			0.94 (0.77, 1.14)	1.31 (1.11, 1.54)
Yes			6.28 (5.68, 6.95)	1.81 (1.70, 1.93)
Tweeter sends pro-health #childhoodobesity tweets
No			8.72 (4.51, 16.88)	6.52 (0.02, 2669)
Yes			6.40 (4.62, 8.87)	2.29 (0.13, 38.97)
Structural terms
Geometrically weighted outdegree				0.58 (0.000004, 76 230)
Geometrically weighted edge-wise shared partnerships				3.93 (1.11, 13.84)
Geometrically weighted dyad-wise shared partnerships				1.00 (0.97, 1.03)
Fit
Akaike information criterion	33 554	31 008	28 395	23 779
Bayesian information criterion	33 565	31 104	28 534	23 950
Outdegree captured in model simulations, %	7.6	70.3	75.4	96.6

Note. CI = confidence interval; OR = odds ratio.

In addition, there was homophily in connections between those with as well as without a health focus in their profile; however, homophily was not observed for those sending pro-health or anti-health #childhoodobesity tweets. Two network members who both had a health focus were more likely to be connected to each other (odds ratio [OR] = 1.81; 95% CI = 1.70, 1.93) than 2 network members not both having a health focus. Likewise, those without a health focus were more likely to be connected to one another (OR = 1.31; 95% CI = 1.11, 1.54) than were 2 network members who did not both have such a focus. The significant odds ratio of 3.93 (95% CI = 1.11, 13.84) for geometrically weighted edge-wise shared partnerships indicates that the more shared partners 2 network members had, the more likely they were to be connected, all else held constant.

Message

Of the tweets, 1037 (93.4%) were pro-health, 34 (3.1%) were anti-health (promoting unhealthy choices), and 39 (3.5%) were neither. More than half were coded as relating to behavior; 24.8% were coded as environment, 4.9% as policy, 5.2% as medical–heritable, 5.4% as consequences of obesity, 20.1% as childhood obesity as a problem, and 36.7% as something else. Retweeting was least common for tweets about the consequences of obesity and most common for medical–heritable causes of obesity. “Something else” tweets included those related to event promotion, community programs, and eating disorders. Tweeters from all sectors sent more behavioral tweets than any other type. Table 3 shows tweet topic frequencies; each tweet was coded for all relevant topics.

TABLE 3—

Tweet Topic Codes and Frequencies of Appearance: June 2013

Category	Tweets, No. (%)	Retweets, No. (%)
Behavior	563 (50.7)	165 (29.3)
Food as cause of obesity	198 (17.8)	40 (20.2)
Food as strategy to combat obesity	227 (20.5)	60 (26.4)
Physical inactivity as cause of obesity	46 (4.1)	16 (34.8)
Physical activity as strategy to combat obesity	199 (17.9)	77 (38.7)
Environment	275 (24.8)	83 (30.2)
Home/family as cause of obesity	35 (3.2)	10 (28.6)
Home/family as strategy to combat obesity	95 (8.6)	16 (16.8)
School as cause of obesity	10 (0.9)	1 (10.0)
School as strategy to combat obesity	120 (10.8)	52 (43.3)
Built environment as cause of obesity	18 (1.6)	6 (33.3)
Built environment as strategy to combat obesity	21 (1.9)	6 (28.6)
Policy	54 (4.9)	14 (25.9)
Food policy as cause of obesity	7 (0.6)	3 (42.9)
Food policy as strategy to combat obesity	36 (3.2)	5 (13.9)
Physical activity policy as cause of obesity	1 (0.1)	1 (100.0)
Physical activity policy as strategy to combat obesity	16 (1.4)	7 (43.8)
Medical/heritable	58 (5.2)	19 (32.8)
Medical/heritable factors as cause of obesity	54 (4.9)	17 (31.5)
Medical/heritable factors as strategy to combat obesity	4 (0.4)	2 (50.0)
Consequences of obesity	60 (5.4)	7 (11.7)
Individual	18 (1.6)	0 (0.0)
Group	43 (3.9)	7 (16.3)
Childhood obesity as problem	223 (20.1)	58 (26.0)
Magnitude	110 (9.9)	29 (26.4)
Understanding/researching the problem	126 (11.4)	34 (27.0)
Something else	407 (36.7)	89 (21.9)

Note. Tweets were coded to all categories that apply; the percentages listed are for the number in each group from the 1110 tweets overall.

Tweets about causes of childhood obesity.

Of the tweets, 253 (22.8%) identified at least one behavioral, environmental, or policy cause of childhood obesity. Of these tweets, 23.7% were retweets, a figure slightly lower than the overall percentage of retweets (25.1%). Behavior was the most commonly identified cause, with 198 tweets involving food behavior (17.8%) and 46 involving physical inactivity (4.1%). Examples of tweets in each topic category are included in the sections to follow.

Food behavior as a cause: I’m in the Park & amount of kids drinking sugary ‘fruit juice’ drinks is rather alarming. 1 in 4 peeps...That's a scary no. #childhoodobesity

Physical inactivity behavior as a cause: I’m seriously so outta shape right now. I can’t go for a 30 second run before I’m out of breath #ChildhoodObesity

Few tweets implicated a policy, environment, or medical–heritable cause. The least common causes identified were food policies and physical activity policies. Slightly more tweets included home–family environment, built environment, and school environment as a cause.

Home–family environment as a cause: I fear too many parents are scared to say no when their kids demand unhealthy foods #childhoodobesity #diabetes #fitness #personaltrainer

School environment as a cause: Study finds that most public elementary schools don’t regulate access to #junkfoods http://t.co/rjXbkaDEYb #childhoodobesity #schoolfood

Tweets about strategies to combat obesity.

A behavioral, policy, or environmental strategy to combat obesity was included in 473 tweets (42.6%); 32.8% of these tweets were retweets. A total of 227 tweets (20.5%) involved food behavior to combat obesity, whereas 199 (17.9%) focused on physical activity.

Food behavior as a strategy: Or deep-fried?—Children eat more veggies when flavored dips are offered http://t.co/klRTeBEB2b #childhoodobesity #vegetables

Physical activity behavior as a strategy: Get kids moving more during school day. Plan for next year to help fight against #childhoodobesity. http://t.co/L39xMI6l32

Food policies and physical activity policies were named as strategies in 36 and 16 tweets (3.2% and 1.4%), respectively.

Food policy as a strategy: Bye bye junk food, new #USDA rules require healthier snacks for kids #healthyschools #childhoodobesity #schoolfood http://t.co/NxfxxsFGy8

Physical activity policy as a strategy: RT @CassClayalive: Why physical education classes should be adopted as a core subject. RT if you agree! #PE #childhoodobesity http://t.co/DwtucQ93nk

School environment was the most commonly mentioned environmental strategy to combat obesity, with 120 tweets (10.8%); home–family strategies were included in 95 tweets (8.6%).

School environment as a strategy: So why such limited resources given for #PhysEd? “Americans Expect Schools to Lead in Preventing #ChildhoodObesity” http://t.co/tcaxZFjprT

Home–family environment as a strategy: #ExerciseTip: America’s fight against #childhoodobesity starts with parents. Try these tips! http://t.co/pMAgvfxBhE

The built environment as a strategy to combat childhood obesity was mentioned in 21 tweets (1.9%), and medical strategies were included in 4 tweets (0.4%).

Tweets about the problem of childhood obesity.

Childhood obesity as a societal problem was mentioned in 110 tweets (9.9%). According to one: “It’s a situation so dire it’s being called an EPIDEMIC. #ChildhoodObesity.” Childhood obesity research was mentioned in 126 tweets (11.4%).

Anti-health tweets.

Few tweets were anti-health (n = 34; 3.1%); although it is not possible to know tweeters’ intent, many appeared to use sarcasm or humor.

Can we all just marvel at how Oreos has remade Dunkaroos? #yes #childhoodobesity #diabetes #idontcare… http://t.co/g4t9MtSx00

GOTS TO HAVE DAT DR P #whatstheworsthatcouldhappen #childhoodobesity #ohwell

Receiver

In June 2013, the 1110 #childhoodobesity tweets made 3 570 886 impressions, or appearances in Twitter feeds. Within the network of #childhoodobesity tweeters, nonprofit organizations followed the most network members on average (median = 7; range = 0–32) and therefore received #childhoodobesity tweets from the most sources. Network members from the government and education sectors followed a median of 5 and 4.5 sources, respectively; those affiliated with media followed a median of 2.5 sources. More than half of private individuals (n = 130) followed none of the other network members and so were not receiving #childhoodobesity tweets. Figure 2b shows the network, with nodes shaded by sector and sized according to how many network members tweeters followed.

DISCUSSION

An examination of 1 month of childhood obesity tweets suggests opportunities to increase exposure of the large and diverse Twitter audience to evidence-based public health information by influencing who says what and to whom about childhood obesity. Specifically, there are opportunities to increase the presence of credible sources from the government, media, and education sectors disseminating public health information consistent with current evidence and to increase the number of Twitter users exposed to this information.

Health information sources thought to be credible can be persuasive.³⁹ Evidence on the credibility of Twitter users is beginning to emerge, suggesting that Twitter users with more influence (i.e., more followers, retweets, and mentions), more topical expertise (topical Twitter account, history of on-topic tweeting), and a favorable reputation (tweet author is Twitter verified, followed by the tweet recipient, or familiar to the tweet recipient) are deemed more credible.⁵⁷

In addition, organizations on Twitter are perceived as more authoritative than individuals.⁵⁸ Government and media sources in our study had the most followers overall and were significantly more likely to be followed within the network of those tweeting about childhood obesity. Twitter tends to verify the accounts of popular users, which typically include organizations and celebrities; government and media Twitter accounts are among the types of accounts often verified. Thus, government and media sources are likely to be influential and reputable, and government and educational sources are likely to demonstrate public health expertise. The overall lack of government, media, and educational sources of childhood obesity tweets in this study suggests a limited presence of credible sources of childhood obesity information on Twitter. Increasing the presence of government, media, and educational sources tweeting about childhood obesity and other public health topics is a promising strategy for increasing the credibility of public health information on Twitter.

Emerging evidence suggests that effective strategies for reducing childhood obesity rates include reducing television viewing¹¹; implementing programs that combine physical activity, health education, and nutrition¹¹; and employing school-based interventions alone or in combination with activities in other settings.⁵⁹ Although we did not code specifically for television viewing, television viewing primarily occurs in the home environment. The small proportions of tweets including home- or school-based strategies for reducing childhood obesity demonstrate the limited focus on these 2 evidence-based strategies.

One of the greatest challenges facing public health is the lack of dissemination and implementation of scientific research.^60–62 Efforts to increase the use of evidence-based prevention information and strategies have too often been “unsystematic, uncoordinated, and insufficiently capitalized.”^62(p443) Exposure to information is the first necessary step in the dissemination process eventually leading to adoption and action.⁶³ Increasing the number of tweets incorporating public health evidence to increase exposure to such evidence among the large and diverse Twitter audience is a promising strategy to aid in reaching the goal of moving evidence into practice. Because government and educational sources are likely the producers of much of the public health evidence, increasing the presence of these credible sources may increase the number of tweets with content that is consistent with current evidence.

There are several limitations to this study, including the use of a hashtag for data collection. Tweets about childhood obesity may not contain #childhoodobesity, so the number of tweets on the topic is probably higher than that reported here. There is also little research on patterns of hashtag use on Twitter,⁶⁴ so it is unclear whether certain types of users would be more likely to use hashtags or whether certain topics are more likely to be hashtagged than others. The use of hashtags in sarcastic ways, changing the intention of a tweet, is also common on Twitter but difficult to discern.⁶⁵ Many anti-health tweets in this study appeared to use the childhood obesity hashtag sarcastically.

Finally, retweeting is one of the primary modes of dissemination on Twitter and is related to characteristics of the tweet and its sender⁴³; one important direction for future research is to provide a better understanding of retweeting behavior and its consequences for the dissemination of evidence-based public health information. Despite its limitations, this study of a growing communication channel suggests opportunities to expose the large and diverse Twitter audience to evidence-based public health information.