Epidemiology is a game of catch-up. You are studying something that has already happened.

Rapid advances in technology have inadvertently given us the opportunity to develop tools to prevent infectious outbreaks before they occur. These tools could become powerful assets to defend public health and increase the resilience of health systems. However, we must be cautious when deciding how, when and where to deploy this new capability.

In my lifetime, I don’t think epidemiology has ever been more relevant than it is at this moment. It is a fascinating subject with a long history. Epidemiologists are innovators by nature. They are always looking for new, sometimes ingenious ways to track the emergence and spread of public health threats.

This was never truer than for the founder of modern epidemiology, John Snow. He is famous for thinking outside the box to understand how clusters of cholera infection developed around a local water pump in central London.1 This insight convinced the local authorities to remove the pump’s handle, thereby forcing locals to use alternative water sources, and the outbreak quickly abated. While this intervention seems obvious to us now, in 19th century London this was a giant leap forward for public health and the understanding of the spread of infectious diseases.

The holy grail of epidemiology, however, is to develop a method of detecting a threat to public health before people even begin to turn up to healthcare facilities with symptoms, and pre-emptively intervene to extinguish the threat before it takes hold. With rapid technological advances, this may soon become a reality, but the precise method may surprise you!

The emergence of internet epidemiology

During the course of the last year, some interesting research has emerged which suggested that trends in internet searches related to symptoms of illness retrospectively correlated to local clusters of COVID-19 infection.2 While intriguing, it could be seen as coincidental given the ever-increasing use of the internet and health awareness in the wider public.

However, what makes the findings most interesting is that the symptoms being researched by the public were not those you would most commonly associate with COVID-19 infection. Often, the query was related to gastrointestinal issues, and the peak of these searches appeared to occur weeks before any patient became unwell with the more common symptoms of COVID-19.3

If confirmed, these findings may present us with a new way of detecting possible new clusters of infection well in advance of any patient requiring treatment at a hospital, giving healthcare providers a crucial head start to prepare. As well, this method may allow us to find previously undetected clusters of infection, especially among more resilient populations, such as students, who are less likely to have become seriously unwell or sought hospital treatment.

There are multiple examples from across the last decade of further evidence that this method is of value. By applying a similar technique to search data available from late 2009, it is possible to retrospectively map clusters of H1N1 swine flu. Additional studies have suggested that these methods would also have alerted health authorities to recent emerging threats such as Zika in Columbia, Ebola in DRC and plague in Madagascar.3

This raises the possibility that if we had had the foresight to track internet search trends in the past, we would have had the opportunity to intervene in these outbreaks, prepare healthcare facilities more effectively to deal with what was coming, significantly reduce the impact of infection on patients, and perhaps even have prevented further infections in the community.

Seizing the opportunity

With this insight the question to consider now is should we work in partnership with large technology companies who have access to internet search data, to develop AI algorithms that can monitor emerging threats in real-time? Is it possible to scale this up to the international level, and bring in global health authorities to develop early warning systems for future pandemics and annual influenza waves?

While the scientific justification for such collaboration is compelling, implementing such a system would be far from straightforward.

Balancing public health with privacy

Despite the potential power of AI-based pandemic tracking, there is a serious legal and moral conflict that we must resolve. Tracing infection clusters by gathering and analysing search data without the knowledge or explicit consent of the user is a contentious issue. The right to data privacy is embedded in law in much of the developed world (I’m sure we have all had to adjust to the world of GDPR!).

We must therefore ask ourselves, “At what point do we cross the line from public interest into breach of individual rights to privacy?” There is probably a strong argument to allow us to use search data to locate clusters at a regional, city or perhaps even a neighbourhood level, but should we also be looking to identify households and individuals themselves? Is the potential to contain any new outbreak with such granularity worth the loss of liberty and privacy that would be necessary? And even if we were to identify individuals at risk, what then? Can we really impose measures in the name of public health based purely on what someone types into a search engine?

While an AI-based internet epidemiology approach would certainly have its supporters, an equal number would, quite legitimately, oppose further increased surveillance of their lives. As with most public initiatives, it is important to engage the public in this debate, and any changes must be taken slowly and carefully to ensure the balance between public interest and individual interest is struck.

The 21st century “John Snow” moment

John Snow revolutionized public health by thinking outside the box to understand why so many people in one London neighbourhood were becoming infected with cholera.1 Since then, the field of epidemiology has made significant progress in tracking and understanding the emergence and spread of infectious diseases. However, until now, we have yet to find that holy grail of detecting a threat before it becomes an issue.

Will the rise of “internet epidemiology” prove to be our giant leap forward, providing us with the tools to properly prepare health systems and significantly reduce the burden on healthcare providers and patients?

Where do the interests of public health and an individual’s right to privacy online align?

*According to the WHO, the most common symptoms of COVID-19 disease are fever, dry cough and fatigue. Other, less common symptoms include loss of taste or smell, nasal congestion, conjunctivitis, sore throat, headache, muscle or joint pain, skin rash, nausea or vomiting, diarrhea, and chills or dizziness.4

1. UCLA Department of Epidemiology. 2005. Father of modern epidemiology. Available at: https://www.ph.ucla.edu/epi/snow/fatherofepidemiology.html. Accessed April 2021.
2. Ahmad I, Flanagan R and Staller K. Increased Internet Search Interest for GI Symptoms May Predict COVID-19 Cases in US. Clin Gastroenterol Hepatol 2020;18(12): 2833–2834.
3. Ibid.
4. Aiken E, et al. Real-time estimation of disease activity in emerging outbreaks using internet search information. PLoS Comput Biol 2020;16:e1008117.
5. World Health Organization. 2020. Coronavirus disease (COVID-19). Available at: https://www.who.int/emergencies/diseases/novel-coronavirus-2019/question-and-answers-hub/q-a-detail/coronavirus-disease-covid-19. Accessed April 2021.