First a bit of background. Influenza needs no introduction. We all have had it and most of us get over the infection. But the disease is serious. It is serious for the elderly, people with chronic medical conditions and the immunocompromised. It caused an estimated 250 000 – 500 000 deaths worldwide every year. It is serious for the economy - lost productivity alone in the United States was $71-167 billion dollars. And the virus is rapidly evolving despite the latest vaccines and anti-virals, And it has great potential for pandemics as the recent “swine” flu scare has shown. For prevention and quick response to outbreaks, a national surveillance system is important. Current surveillance of the virus is based on a representative network of primary health doctors all around the country. They collect clinical symptom data as well as virology data to be sent to the labs. The problem with this approach is the associated 1-2 week reporting lag. In addition, an efficient health infrastructure is required.
To prevent reinvention of the wheel and to learn from previous research we started with a literature review. We searched the “usual suspects”: MEDLINE, EMBASE, SCOPUS, ISI Web of Knowledge and INSPEC. Because of the contemporary nature of the project, we used Google Blog Search and Google. For each relevant article we hand searched its references. We used a 2 pronged standarized search strategy because we predicted there might not be any previous research on social networks and influenza: 1) Social Networks AND influenza 2) Web Surveillance AND influenza
Here is a sample of our MEDLINE search strategy.
From our literature search we found 3 different approaches. Syndromic Surveillance: which uses indirect or proxy measures of influenza The most famous example is Google Flu Trends which uses sophisticated statistical methods to _model health seeking behaviour from _influenza specific search terms to estimate the _prevalence of influenza. Other examples are the Global Public Health Intelligence Network which uses an automated algorithm to monitor worldwide news reports. HealthMap is similar but integrates more sources of influenza information Primary Doctor web-based reporting – REAL FLU. This study required a sentinel network of GPs to report ILI electronically based on clinical symptom data Web based self reporting Pioneered by the Great Influenza Survey in the Netherlands for the 2003-2004 flu season. There were 13 300 active participants who filled an initial demographic questionnaire and were asked to fill out a weekly ILI symptom questionnaire via email. They found average correlation with their data and the data from the sentinel GP network.
Gripenet using the _same system superseded GIS and were implemented in the 2006-2007 flu season. With many participants in the Netherlands, Belgium and Portugal.
We are inspired by the citizen science movement which proposed that the collaborative work of non-professional citizens could help solve scientific problems. Example of this include: Our study is a pilot study funded by a summer studentship from the Health Research Council NZ to examine this possibility. We aim to improve on previous research in 3 key ways: Simplification – simple to a lay person and easy/non-obstructive to use 2) Organic Growth – using power of social networks to encourage participation 3) IP data – track influenza all across the world
The current issue is: Complicated ILI symptom quesitonnaire – 13Q With an equally complicated case definition by “Experts” (consensus vs evidence-based) Experts used was probably because there is no global or even regional consensus on influenza case definitions The solution we found was by an Australian researcher Thursky (2003) who proposed a symptom triad of Fever(subjective) cough and fatigue based on retrospective analysis. It was found to be effective and is actually found to be more sensitive and speciifc than data from the CDC. It had a s ensitivity (43.5-75.1%) , specificity (46.6-80.3%) and PPV ( 23.3-59.7%) . Equally important: the questions do not require a thermometer (with its associated problems) and is dead simple
The previous research had extensive media backing with TV and radio interviews, information posted to universities and schools and even competitions. Participation growth is thus dependant on the public relations onslaught. However on a student research budget this is _clearly not possible! But we choose Facebook because facebook applications are _inherently made to be spread… like….
The annoying Fortune Cookie. In my group of friends at least, I get these annoying and rather generic advice constantly. So we thought If a fortune cookie can spread… so can our project. So we programmed our application to take advantage of “viral” facebook mechanisms like: Newsfeeds, invitiations and profile badges which I will demonstrate later.
A third way in which we tried to improve is via the tracking of IP data. IP data can tell us where in the world down to the nearest city where the participant accessed facebook from. This gives us the potential for real time worldwide surveillance
Programming the application was relatively straight forward because Facebook provides API (Application Program Interface) which allows access to Facebook features using just a line of code. Official resources like the Developer Centre provides ample documentation and examples. We used PHP and MYSQL because they are open source. I am a medical student not a programmer but I was able to self-teach myself to (inelegantly) program the application based on freely availiable PHP or MYSQL tutorials on the net. Google = best friend.
The application need a front and explanation. So we built www.influenzatracking.com for referrals by the media. Media referrals are important because for the application to spread “organically” on facebook we first need a critical mass of users just like a nuclear reaction.
So this promotional page comes up when you type in www.influenzatracking.com.
Here is a quick overview about the project and its aims.
The long long PIF was required as part of the ethics approval process and explain in detail about the project.
Again, as part of our ethics approval we were required to have a consent form as shown.
Simple demographic questionnaire
There is the flu symptom questionnaire. It is simple and takes just 15s to fill out the questionnaire.
Each week, each patient would be given a weekly email notification asking them to fill out the applicaiton
So in summary. Participants are initially recruited by mass media to the website where participants can give consent and fill out a simple demographic questionnaire. After which they fill out their first flu symptom questionnaire. They would be reminded weekly via email to fill out subsequent questionnaires. “Viral” mechanisms built into the application is aimed at recruiting participants from within facebook.
Our beta-testing started on the 28/06/09 with 73 participants. At the end of our beta test, 52 participants have filled out more than 1 questionnaire. For a conversion rate of 71%. The demographics show that the typical participant is young, highly educated with a high vaccination rate with no gender bias. There were 26 application “fans”.
The median number of days between each questionnaire was 8.0. In the future, once we have more data, we planned to calculate the ratio of participants with ILI/number of total active participants in each city or country. The data would be displayed on a world map. This presentation marks the end of beta-testing and the start of making this project public with the associated PR drive.
So, please do join the project by going to dub dub dub influenzatracking.com. Also it would be great if you can blog or just pass on the word. We are also interested in forming research partnerships. So please do email or talk to our team if you are interested.