The other pandemic: social media engagement around non-communicable disease preventive behaviours during Nigeria’s COVID-19 lockdowns

ABSTRACT Given the complexity of global health crises such as the COVID-19 pandemic, it is typical for crisis-focused interventions to have a multiplicity of impacts. Some of these impacts may yield positive or negative externalities for health priorities that do not have the same perceived urgency. The interplay between COVID-19 prevention (a high priority, high perceived urgency issue) and non-communicable disease (NCD) prevention (a high priority, low perceived urgency issue) provides a good case in point. By analysing tweets during Nigeria’s COVID-19 lockdowns, we identified avenues for social media to help adapt crisis responses to a wider range of wellbeing concerns.


Commentary
The coronavirus pandemic is unlikely to be the last major global disruption (Silva de Amorim and Baltazar 2020). Just like this pandemic, future disruptions such as climate change-related conflicts and natural disasters will also have repercussions beyond the traditionally assumed limits of public health (Mogo 2020). They will require public health actors to combine the traditional responsibility of scaling cost-effective top-down responses with addressing the diverse cultural, social, and economic contextual factors crucial for their success (Stake 2021). Interventions aimed at tackling such complex disruptions can have multiple and even conflicting impacts across time, health domain and place .
The interplay between COVID-19 prevention (a high priority, high perceived urgency issue) and noncommunicable disease (NCD) prevention (a high priority, low perceived urgency issue) provides a good case in point. NCDs such as diabetes and cardiovascular conditions are considered an emerging pandemic in Africa due to the rapid rise in disease burden (Allen 2017). Preventive behaviours to address this burgeoning epidemic are amenable to change through investments in inclusive access to healthy food, physical activity, and social participation which are significantly mediated through public spaces (Mogo et al. 2022). However, COVID-19 control measures such as lockdowns also curtail access to public spaces, which support behaviours that can prevent NCDs. Synergies may exist between crisis responses and non-crisis health priorities. For example, intersectoral collaborations in the context of COVID-19 prevention could yield co-benefits for NCD prevention (Oni et al. 2020). It is important to understand the impact of crisis responses on the broader wellbeing of populations, to build synergies and address possible negative externalities.
In 2019 (prior to the global outbreak of COVID-19), we began a study exploring NCD related risk factors and outcomes in the context of public space infrastructure development in Lagos, Nigeria and Yaounde, Cameroon (UrbanBetter 2022). With the governance of public spaces changing due to ensuing lockdowns, we expanded the focus of our research to consider how activities to address behaviours around the COVID-19 crisis may impact behaviour change around the (perceived to be) less urgent but similarly high priority issue of NCDs.
We identified social media as an avenue to gather insights into unfiltered and real-time engagement around NCD preventive behaviours, while bypassing the need for in-person data collection which was not feasible during lockdown. Social media analytics, a process of gathering and analysing data from social networks such as Facebook, Instagram, Twitter, and LinkedIn, is often used by brands for market research. This helps them track online conversations about their products, with emergent information used to guide corporate decision-making.
We had initially aimed to study social media engagement in both Nigeria and Cameroon given that these two countries were the focus areas for the ALPhA study. However, during a pilot search where we built a query on social media data originating from Nigeria and Cameroon, we noticed that social media content from Cameroon constituted only 2.4% of the total information analysed, with the rest being from Nigeria. All the top tweet authors recorded, that is, people whose tweets had the widest reach and mentions were from Nigeria. Relative population sizes may help to explain the comparative tweet volume. For example, Cameroon has a population of 27,744,989, about 13% of Nigeria's which is 214,028,302 (Worldometer 2021). About 53% of Cameroon's population, constituting 14,741,256 people, is urban, while about 44% of Nigeria's population, constituting about 102,806,948 people, is urban.
Another factor that may explain this difference is access to the internet. Nigeria had 61.2% of internet users in December 2019 compared to 23.1% in Cameroon in that same year (Internet World Stats 2021). Furthermore, political suppression has become a growing concern in many African countries. In 2016, the government of Cameroon released an official statement labelling social media as 'a new form of terrorism'. It pointed out that sites including Twitter and Facebook had created 'a social pandemic, perpetuated by amateurs, who do not have a sense of etiquette and decorum' (Ndongmo et al. 2021). Furthermore, while Nigeria implemented strict lockdowns with discernible phases of pre, peri and post lockdown, there was little by way of movement restrictions implemented and enforced in Cameroon. For these reasons, the analysis was limited to Nigeria, suggesting that social listening projects may be best suited to settings with high access to the internet, which tend to be urban areas.
We aimed to understand social media engagement on NCD risk reduction e.g. physical activity, obesity, mental wellbeing, in the context of a crisis. Previous research findings have demonstrated the potential for social media to identify health service needs (Saha et al. 2022) and perceptions around preventive healthcare seeking behaviours (Ryu and Pratt 2022), as well as the value of social media in facilitating the spread of information and mobilisation efforts during crises (Osoro 2017). We considered the Helsinki ethical principles in the conduct of our study (World Medical Association 2013). Our project satisfied these principles by engaging in a research project that could be beneficial to the population studied, only making use of data that was already in the public domain, and not creating further risks of transmission of COVID-19 through in-person data collection.
We used Brandwatch (2022), a tool that tracks publicly available social media data from various sites to aggregate data from these sources. Brandwatch captures public reactions to marketing campaigns, classifies related negative and positive sentiments, and monitors the geographical spread of these sentiments. Dimensions of publicly available data collected by Brandwatch include shares, likes, conversations, comments, mentions, impressions, sentiments, location, and industries of sharers. Sentiment analysis on Brandwatch adopts three main steps, which include the use of knowledge-based rules, machine learning, and the development of customisable rules (Brandwatch 2022). Information shared online is categorised as positive, negative or neutral based on generic language specific characteristics of the text. Information that is categorised as neutral or unclassified based on the knowledge-based system is passed through language and industry specific classified information which has been trained to identify sentiments. Users are also able to set up specific rules around sentiment classification that are peculiar to unusual events or to their brands to refine their results. This process has recorded between 30-80% agreement with human classifiers in the past (Brandwatch 2022). However, it is useful for identifying key trends which can then be triangulated against other sources of information for validation.
Other social listening platforms available include Digimind, Sprout Social, Falcon.io, Sprinklr, to mention but a few. However, we decided to use Brandwatch based on the expert recommendation of our collaborators at the Centerfor Analytics and Behavioural Change in South Africa (CABC 2022). At the CABC, output from Brandwatch had been previously triangulated against expert input, government reports and their in-house social listening platform, Anansi, and found reliable for purposes similar to those of our study. The team also used Brandwatch in several African countries including South Africa, Ghana, Senegal, Nigeria and Tanzania on social listening projects that were used to identify and understand social unrest, to track social cohesion and division, and to explore vaccine hesitancy (CABC 2022). For this reason, we considered it most appropriate to our study aims.
Information we captured through our query ( Figure 1) included: i) which individuals and organisations were generating the most discussions on NCD preventive behaviours and overall wellbeing during the pandemic; ii) where people were sharing from; and iii) how the content of discussions relating to NCD preventive behaviours evolved before, during, and after the lockdown period.
A key component of the query development was identifying a near exhaustive list of synonyms for words related to NCD behavioural risk factors e.g. mental health, fitness, walking. We also compiled a list of iterations of commonly misspelt words so as not to leave out any relevant posts. Given the vastness of social media conversations, a 'NOT' parameter was added to the Boolean query, to remove extraneous information that constituted noise. In the field of social media listening, noise is a term that is used to describe tweets that appeared in our dataset because they met some parameters within our query but were unrelated to our query. Trending issues present an opportunity for hashtag hijacking, when a trending hashtag is used to draw attention to an unrelated topic. A large volume of mentions was still returned in the final cleaned dataset, therefore approximately 93% of the data was retained for the analysis.
The period of analysis covered an eight-month period from January to August 2020. This allowed us to capture how conversations and public perceptions around behavioural NCD risk factors changed in response to the spread of COVID-19 and the enforcement of mitigation measures. Data were analysed using simple descriptive statistics such as tweet mentions, the highest number of mentions concerning a particular query, the nature of people and organisations tweeting, and the underlying sentiments the tweets expressed. Charts, frequencies, percentages, and sample tweet images were used to display the findings. Below, we reflect on the insights gathered from this approach.

Identifying distinct changes in public concerns in the context of crisis responses
We designed our query to focus on Nigeria's lockdown period, which can be broken down into four stages between January and August 2020 (Ibrahim et al. 2020). These periods are the pre-lockdown period from 2 February to 29 March 2020; the total lockdown during which major public space restrictions occurred from 30 March − 3 May 2020; the easing up of lockdown from 05 May − 15 July 2020, and the post-lockdown period from 16 July and beyond. We did not restrict our search by age and gender. Our approach allowed us to explore how social media engagement changed during these periods, both in terms of the conversation topics and the conversation volume. Our query output suggests that while there was a fair amount of conversation online in Nigeria around health, fitness in the context of COVID-19 during the pre-lockdown phase (Figure 2), total mentions increased by 46% as the total lockdown was put in place ( Figure 3) and the access of residents to public spaces was further curtailed.
In our data set, conversation spikes related to NCD preventive behaviours during the prelockdown phase were related to changing developments in response to the pandemic, such as the identification of Nigeria's first COVID-19 case (Figure 7) and the suspension of major football leagues (Figure 8). Over time, tweets on NCD preventive behaviours with the most engagement in the form of retweets and likes were those that called for caution in protecting health, those concerned with the high percentage of pandemicrelated deaths in people with underlying NCDs, and those that projected alarm over increasing COVID-19 cases in Lagos (Figures 9 and 10).
As the lockdown period eased (Figure 4), negative sentiments decreased and conversations moved from alarm to health advice and challenging health misinformation. Finally, toward the post-lockdown phase ( Figures 5 and 6), conversation spikes included interests in weight loss and mental health. Posts about depression also became more prominent.

Identifying atypical collaborators for strengthened public health responses
Using the data extracted, we traced the evolution of the conversation and identified atypical yet influential actors involved in disseminating messaging around NCD preventive behaviours in the context of COVID-19 to the public. Across all four lockdown periods examined, we consistently noted that the conversations were driven more widely by individuals than by accounts which had been marked as organisational accounts by Twitter. Messaging came from famous Nigerian personalities including social media influencers, comedians, musicians, footballers, and prominent businesspeople; with more robust engagement than messaging from health organisations. Notably, we found that social media influencer accounts like @Aproko_doctor with a similar number of followers as the Nigerian Centre for Disease Control (NCDC) (1.1 million at the time) had significantly more retweets and likes from their social media content compared to that of the NCDC.
We were also able to evaluate the engagement of traditional actors. While having significantly less engagement than tweets from influential individuals, organisational actors including the global and national accounts of the World Health Organization, the Lagos State Ministry of Health, UNICEF Nigeria, Nigerian Centre for Disease Control, and the Nigerian Federal Ministry of Health played a role in sharing health messaging. Businesses such as Konga Nigeria and the Nigerian Exchange Group; government agencies  including the Nigerian Police Force and media agencies such as the Daily Trust Newspapers also shared health messaging during this time period.

The importance of contextually-informed public health measures
Traditional survey methods have a significant time lag between the collection and interpretation of data with the work of interpreting large data sets resting primarily with researchers that may lack insights into the context behind the data. By contrast, the use of social media analysis enabled not only the capturing of public perceptions in near real-time, but also contextual information, such as replies, clarifying tweets, pictures, retweets and other forms of engagement, to ground-truth the meanings inferred from these perceptions.
Data analysed identified a conflict between the stayat-home measures to interrupt disease transmission on the one hand and the demand for public space opportunities for recreational physical activity on the other (Figures 10-15). This is reflected in tweets making observations about people meant to be indoors who were out on the streets playing sports instead ( Figure 16).
Additionally, the government's suspension of the use of various public spaces and admonitions to stay calm and obey lockdown orders stood in stark contrast  to the public's concerns about the feasibility of lockdown for poorer Nigerians and the perceived elitist nature of banning interstate travel by road while permitting flights (See Figures 9 and 10). This ability to understand how high level responses may impact the public and sub-populations e.g. low-income groups, when responding to rapidly evolving situations could help to improve crisis responses.

Implications for practice
Our findings suggest that in rapidly evolving crises, social media analysis could be considered as part of a complementary set of tools needed to inform the design of robust responses. Themes that arise from social media conversations can inform the tailoring of crisis responses to various sub-populations and their needs. For example, Figure 6. The evolution of sentiments on NCD preventive behaviours in the context of COVID-19 from blogs, news and Twitter from January to August 2020. an early identification of a demand for public spaces in the context of the closure of these spaces during lockdowns could inform alternative arrangements to support physical and mental health. Changes in public concerns over time, such as in the pre-lockdown, lockdown and post-lockdown periods as identified above, could be used to evolve public health messaging for appropriateness to changing needs.       Social media can also be valuable in understanding key stakeholders driving discourse in the context of crisis. Our approach made it easy to identify influential yet atypical disseminators of health messaging, particularly social media influencers. Some of these individuals would otherwise have been difficult to identify. These actors could prove valuable in the design of robust and timely multi-sectoral responses to health issues. Our findings also suggest it is worth exploring how to support such digital actors with evidence-informed communication that links to the relevant health authority sources to avoid sharing mis-information. This is due to their prominent role in communicating with the public and the higher levels of engagement they achieve in comparison to traditional health authorities.
Identifying influential actors also provides insights on how to frame messages on health-related topics to ensure a stronger reach and engagement. For example, we found that the public responded better to the use of humour, satire and everyday lived experiences deployed by individual influencers to share health information compared to more technical advisories deployed by the institutional accounts. Such information could be used in near real time tailoring of public health messaging on behaviour change in times of crisis.
Our analysis also focused specifically on NCD preventive behaviours in the context of the COVID-19 lockdown, particularly given the inadequate consideration of lockdowns on other facets of health promotion. However in the context of COVID-19, misinformation, disinformation and/or denial of the pandemic could have had an impact on people's response to lockdown measures. In a survey of respondents from English speaking African countries, Osuagwu et al. (2021) noted prevalent misinformation around how to prevent and to treat COVID-19. Exposure to misinformation was found to vary based on education, employment status, age, sex and knowledge of the clinical symptoms of COVID-19. During a crisis, social media can also support the identification of digital sources of misinformation and/or disinformation in order to improve access to evidence-based information.
Social media analytics capture the perception of people with active social media accounts, a factor that is limited by social media bans and access to the internet in Nigeria. In 2021, only 15.8% of the Nigerian population accessed social media, with a significant proportion joining in 2020 (Kemp 2021). We also found more social media data from Lagos, Ibadan, Abuja and Kaduna, all major Nigerian cities. Another potential limitation of using social media analytics is that it is better at flagging social media conversations in popular global languages such as English than discussions in indigenous languages. For these reasons, this approach should be considered complementary to existing approaches, and may be comparatively more valuable in cities.
Finally, social media analytics is dependent on publicly shared information, which itself is determined by various factors. For example, people may be more willing to share on informative topics (Lee et al. 2011) and on topics that spark strong sentiments such as emergencies (Simon et al. 2015) than on stigmatised or taboo topics. However, in anticipation of future rapidly evolving health disruptions, more research is warranted to explore the ways that this tool could be used in data-poor contexts as part of the public health toolbox to understand evolving public sentiments, to frame messages more effectively, and to engage traditional as well as atypical multisectoral actors.

Figure 16.
A trending theme during the lockdown period was the use of public space for physical activity, sometimes inviting punishment from government forces enforcing lockdown as seen above (also see Lawanson et al. 2020).