Once upon a time, spreading illnesses weren’t investigated until they hit crisis levels (see Outbreak); they’d wreak havoc in local areas but slide under the CDC radar until they were extreme and widespread enough to garner national attention. Today, researchers use dozens of different surveillance systems, ranging from clinic and hospital data to pharmaceutical purchases and social media posts. Tracking has improved. But to what result? In a recent speech given at the Munich Security Conference, philanthropist Bill Gates made it clear that we’re still unprepared for an epidemic. According to Gates, “epidemiologists say a fast-moving airborne pathogen could kill more than 30 million people in less than a year”. While we have insight into where disease has spread, understanding where illness is heading with enough time and context to manage that spread is still a hazy, fever-induced dream.
The Burden of Infectious Disease
Infectious diseases continue to be one of the greatest public health concerns across the globe. The economic burden of seasonal influenza amounts to $87.1 billion each year in the U.S. alone – 610,660 life-years lost, 3.1 million hospitalized days and $10.4 billion of direct medical costs – despite widespread public attention and billions invested in preventative measures. This does not account for the unpredictable threat of pandemic flu or the rapid emergence or re-emergence of diseases such as Zika and Ebola.
According to Gates, “epidemiologists say a fast-moving airborne pathogen could kill more than 30 million people in less than a year.”
The U.S. Centers of Disease Control spends more than $700 million each year trying to prevent and respond to influenza outbreaks, making it one of the best-tracked infectious diseases today. The current gold standard in outbreak tracking is provider-initiated reporting – syndromic surveillance from participating physician offices and confirmed virology from a laboratory reporting network. Initiatives to mine Twitter, Google and other social / search platforms have shown some promise in developing a more timely illness signal. Unfortunately, these models have suffered from low “signal-to-noise” ratio due to natural language processing and media effects—a challenge that led to the eventual decommissioning of Google Flu Trends in 2015. Public health officials have been willing to compromise on the immediacy of results, preferring strong signals even if they lag the spread of disease by weeks, but that has caused its own set of challenges.
Earlier identification and earlier response to outbreaks can significantly reduce the human and economic impact of these diseases. Public health policy depends on real time situational awareness – the right information at the right time, and most importantly, given to the right person – someone who can take action on that information. Sarah Park, State Epidemiologist for Hawaii, knows that the “data explosion” has happened. “There is a place for syndromic surveillance, but there needs to be a lot of expertise to sort out the noise, so that we can start acting and figuring out appropriate interventions.”
Think about interventions as being as predictable as seasonal flu shot campaigns. As parts of the country start to see the telltale signs of fall, the banners begin rising over local pharmacies – FLU SHOTS ARE HERE. In fact, some signs have been up since August, because nobody knows yet if flu season will be early or late, extreme or light. In essence, we’re hedging bets early in the season, making up for what we don’t know. But there’s one truth we do know: “The timing and the severity [of flu] are important. If we knew it was going to be an extremely severe year we would put out stronger messages,” states Christine Hahn, State Epidemiologist for Idaho.
But there are additional actions that can be taken. “The most practical utility [predictive surveillance] would be working with the vaccine manufacturers; if we knew the peak would be in January we might do a big effort in pushing the flu forward,” explains Hahn. Officials rely on situational awareness to drive policy decisions such as school closings, triage protocols, and reallocation of healthcare resources. The challenge is that appropriate policy interventions (and optimum health outcomes) assume that key epidemiological characteristics of the disease are already understood – e.g. severity of illness, contagiousness (R0, attack rate), and spread in the community. When a new outbreak emerges, this information must be estimated in real-time, without the benefits of thorough epidemiological investigation.
“The timing and the severity [of flu] are important. If we knew it was going to be an extremely severe year we would put out stronger messages.”
Case in Point: The 2009 H1N1 Pandemic
Eight years ago, a new pandemic took us by surprise, even taking the lives of those who are not usually at highest risk for flu mortality. The H1N1 influenza outbreak emphasized the weaknesses of existing illness surveillance capabilities and our preparedness. Because it started as a large outbreak of a clinically mild disease, often in a population that was generally healthy, it flew under the radar. One survey showed that less than 30% of adults with flu-like illness visited a physician during their illnesses, indicating that the CDC’s syndromic systems were grossly under representing influenza levels due to what’s called a ‘treatment seeking bias’.
Surveillance systems such as the CDC’s ILINet lagged in reporting time and didn’t stabilize to show the ‘actual’ illness until 4-6 weeks after the fact – not timely enough to inform intervention decisions. Per Iowa state epidemiologist Patricia Quinlisk, “When we see numbers rise before the chronic baselines, we don’t have the algorithms to know if they are really going up, or are we just seeing a seasonal variation.” ILINet is designed to address reporting delays by quickly detecting abnormal illness levels while waiting on confirmed virology from clinical labs. However, this information clearly did not surface in time to take quick action.
“We don’t have the algorithms to know if they are really going up, or are we just seeing a seasonal variation.”
What H1N1 2009 showed us above and beyond the reporting delays is that current surveillance systems are focused on detecting outbreaks, not managing them. The data from these systems has not been transformed into actionable information; instead it relies on knowledgeable humans (including the many state epidemiologists referred to in this article) to interpret the data and translate that into policies.
Where are We Headed Next?
To quickly and accurately interpret data in a way that can actually improve situational awareness, these epidemiologists need to know more, so they can do more. And they need it along three dimensions:
1. A new “ground truth”
Epidemiologists need more than just the data from those who enter the healthcare system – they need real-time, accurate data from consumers. They also need higher quality data or enough context on the epidemiological characteristics of diseases to make them actionable. Secondary attack rate in the household or schools or ages affected by disease are examples that could take them a large leap forward.
2. Automated data collection and insights
With richer data sets will come the need for more streamlined collection. We hear stories of sentinel providers using whiteboards to record the number of patients seen with influenza-like illness. There has to be a better way! With better collection comes the opportunity for subtler and telling insights – insights that can lead to more specific action plans and broad communication.
3. Better communications strategy
Speaking of communication – there is a huge opportunity here, but it comes with a new mindset regarding the circulation of information. State and federal agencies with insights can take a page out of the consumer marketing handbooks instead of utilizing government approved methods. Modern channels such as social media, text and other viral tactics can not only dispense information more quickly but can also ensure that the critical communication gets to the right people at the right time and place.
The Key to a Healthier Future
Imagine an outbreak in the future where data is gathered from the moment people feel ill. It gathers temperature and symptom data not just for a family, but individually for each member of a household, noting ages as well. The data is also tied to both a region and to an elementary school, as schools are often the primary spreaders of disease. More importantly, the same data collection tool works to collect information, as well as to disperse it, so that individuals, families and schools know if there’s a serious public health threat growing and can be told how to take action.
This is the future of illness surveillance, a future where diseases aren’t only tracked- they are managed and stopped in real time. This is what Kinsa has been working towards for the past 5 years, using smart thermometers to capture geolocated fever and symptom data the moment someone feels ill. The same smart thermometers are used to communicate with the ill, helping people to recover faster through earlier detection and diagnosis. This full communication cycle could help stop pandemics before they start.