Social Media Interventions for Precision Public Health: Promises and Risks

"Questions remain about when it is appropriate to couple tools for digital phenotyping with targeted communication interventions to influence health behaviours."

In 2013, a series of methods were published demonstrating the possibility of using Facebook 'likes' to predict aspects of personality and demographics. To understand the role that similar approaches might play in preventive medicine, this article examines studies in which social media data are used to predict or model health-related behaviours and outcomes. It then explores how these methods might be operationalised in the design of precision behavioural interventions, and how the effects of these interventions might be amplified or lead to unintended consequences when delivered in a networked public.

Traditional behaviour change approaches might see a government or public health organisation address problems of vaccine coverage by conducting a survey on vaccine hesitancy to guide the design of a communication intervention, for example. Social media presents an opportunity to identify and deliver personalised digital interventions in an integrated way. Ethical questions arise with the possibility of deploying such online personalised interventions at scale, without consent, and where targets of the intervention are unaware they are being manipulated.

Individual-level studies that predict attitudes, behaviours, and health outcomes of people link social media user data to validated survey instruments or health records, often using much smaller cohorts. This approach could work across major social media platforms and make it possible to detect phenomena like vaccine hesitancy and refusal.

At a population level, studies examining associations between what can be observed on social media and health outcomes have so far been limited to high-prevalence conditions and behaviours like vaccine coverage; thus, it is not yet clear whether social media data can be used to reliably model rarer outcomes. Population-level studies can be operationalised to complement traditional public health surveillance with faster and less costly information, but they tend to produce shallow information and are blunt instruments for designing communication interventions.

As the researchers explain, the challenges associated with delivering and evaluating population-level digital behaviour change interventions come from the networked nature of online social spaces. These challenges relate to:

• Evaluation: We are only starting to grapple with the experimental designs needed to test such interventions in trials and in natural settings.
• Implementation: Social networks may supress or amplify the effects of behaviour change interventions in unpredictable ways. For example, trials testing the effects of messaging interventions aimed at influencing vaccination attitudes often fail to show an effect on behaviour, but this may be because they are tested on individuals in artificial environments rather than in the social spaces where information credibility and beliefs are socially constructed.

Risks and unintended consequences of use of automated behaviour change tools include:

• Short term: Increased use of these methods may represent an erosion of privacy and with it, a perceived threat to individual autonomy.
• Medium term: The perceived erosion of privacy that comes with public knowledge of expanded surveillance can drive behaviours underground and hence make social media a less reliable signal of behaviour.
• Long term: New research efforts in the area could be adapted for use in commercial or political applications, including by organisations unconstrained by the ethical standards required within academic environments.

"Given the capacity to scale precision behaviour change interventions to societal levels, clear governance structures are now needed to allow for their safe and ethical use. Within academia, ethics reviews will need to consider not only transparency and alignment with participant values but also the broader impact that reporting may have on society."

That said, there are opportunities. Methods for iteratively refining predictive models to better target Facebook users may permit direct communication with people who have been traditionally hard to reach, and to reach them well before they visit a clinic or hospital.

The researchers conclude: "While the research area is still in its infancy, examples from outside published research leave little doubt that we can take advantage of social media to deliver fully automated, targeted, and cost-effective behaviour change interventions at scale....[However, u]ntil researchers have the capacity to evaluate them in well-designed studies demonstrating that the benefits outweigh the risks, we recommend caution in their deployment in preventive medicine and public health."