CASE STUDY
It used to be that epidemiologists at the CDC only knew that you had the flu when you visited your doctor and the doc- tor then reported that information to other agencies. By the time this information made its way to the CDC or state health departments, it was often several days old, which meant that the medical community was always playing catch-up. And of course, your doctor was only aware of your illness if you showed up at his office, not when you decided to tough it out at home. Nowadays, however, before you ever go to the doc- tor, you are likely to search the Internet for âflu symptoms,â tweet your suddenly declining health status, or Instagram yourself lying in bed. Study the case below, which shows how the Internet and social media have impacted the field of epi- demiology, and then answer the questions that follow.
Google Used to Predict Influenza Outbreaksâ2008
In 2008, Google launched Google Flu Trends, an applica- tion that compiles aggregated data from key word searches
for clinical terms, such as thermometer, chest congestion, muscle aches, or flu symptoms. Google reports the data on a website, which then provides an early-warning system for the locations of new flu outbreaks. Because the data are collected from searches performed each day, trends in flu symptoms become apparent much more quickly than when they are based on data reported during office vis- its or in lab reports from physicians around the country. When the CDC compared actual cases over the course of a year with Googleâs findings, the data from the two sources matched.
Initially, Google was only compiling information about flu trends in the United States and Canada. But after the H1N1 virus appeared in Mexico in 2009, the CDC asked Google to go back and look at Internet searches conducted by people in Mexico during that time. Evaluation of the data indicated that Google detected an uptick several days before the CDC did (Figure 23.1).
1,200 1,000 800 600 400 200 0
Influenza activity
1113151719212325272931 2 4 6 8 1012141618202224262830 2
Suspected Confirmed
Mar
Apr
May
Date of illness onset
Intense
High Moderate
Low Minimal
Jul Aug Sep
Oct Nov
Dec Jan
Feb Mar
Apr May Jun
Figure 23.1 2009 Inflenza activity (a) Data from the Centers for Disease Control, based on physical examination by health care providers, shows a peak in influenza cases in late April. (b) Google âFlu Trendsâ shows an influenza peak in February.
No. of cases
Case Study Exercise 23 Morbidity and Mortality Weekly Report 199
Epidemic tracking based on Internet searches reflects what is called collective intelligence. It works because indi- viduals using their personal computers tend to search for terms related to their immediate needs and intentions, and they generally do this before presenting in a doctorâs office or emergency room. The methodology used in the Google search was published in the prestigious science journal Nature, and another independent study has been published about a similar search analysis conducted by Yahoo, showing that it was effective in predicting flu trends.
Some people worry that data collected from Internet searches may compromise individualsâ privacy. However,
Google maintains that Flu Trends cannot be used to iden- tify individual users because the data are anonymous and are aggregated before being presented. Another potential drawback is that this data collection method is less likely to be useful for tracking epidemics in societies having a low percentage of computer ownership, namely, developing countries. However, considering the high stakes involved in identifying an epidemic quickly, Internet search term analy- sis holds great promise for public health. And, unlike most health innovations, itâs free!
3. How might data collected in this way not be representative of a particular population?
4. Explain why the two graphs in Figure 23.1 do not show influenza activity peaking at the same time.
6. In 2013, Google Flu Trends vastly overestimated the number of cases of the flu in the United States. One explanation for this was that there was a greater than usual number of media reports about influenza early in the flu season. Based on your knowledge of how Flu Trends works, explain why Google missed the mark in 2013
| CASE STUDY |
| It used to be that epidemiologists at the CDC only knew that you had the flu when you visited your doctor and the doc- tor then reported that information to other agencies. By the time this information made its way to the CDC or state health departments, it was often several days old, which meant that the medical community was always playing catch-up. And of course, your doctor was only aware of your illness if you showed up at his office, not when you decided to tough it out at home. Nowadays, however, before you ever go to the doc- tor, you are likely to search the Internet for âflu symptoms,â tweet your suddenly declining health status, or Instagram yourself lying in bed. Study the case below, which shows how the Internet and social media have impacted the field of epi- demiology, and then answer the questions that follow. Google Used to Predict Influenza Outbreaksâ2008 In 2008, Google launched Google Flu Trends, an applica- tion that compiles aggregated data from key word searches for clinical terms, such as thermometer, chest congestion, muscle aches, or flu symptoms. Google reports the data on a website, which then provides an early-warning system for the locations of new flu outbreaks. Because the data are collected from searches performed each day, trends in flu symptoms become apparent much more quickly than when they are based on data reported during office vis- its or in lab reports from physicians around the country. When the CDC compared actual cases over the course of a year with Googleâs findings, the data from the two sources matched. Initially, Google was only compiling information about flu trends in the United States and Canada. But after the H1N1 virus appeared in Mexico in 2009, the CDC asked Google to go back and look at Internet searches conducted by people in Mexico during that time. Evaluation of the data indicated that Google detected an uptick several days before the CDC did (Figure 23.1). 1,200 1,000 800 600 400 200 0 Influenza activity 1113151719212325272931 2 4 6 8 1012141618202224262830 2 Suspected Confirmed Mar Apr May Date of illness onset Intense High Moderate Low Minimal Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Figure 23.1 2009 Inflenza activity (a) Data from the Centers for Disease Control, based on physical examination by health care providers, shows a peak in influenza cases in late April. (b) Google âFlu Trendsâ shows an influenza peak in February. |
No. of cases
Case Study Exercise 23 Morbidity and Mortality Weekly Report 199
Epidemic tracking based on Internet searches reflects what is called collective intelligence. It works because indi- viduals using their personal computers tend to search for terms related to their immediate needs and intentions, and they generally do this before presenting in a doctorâs office or emergency room. The methodology used in the Google search was published in the prestigious science journal Nature, and another independent study has been published about a similar search analysis conducted by Yahoo, showing that it was effective in predicting flu trends.
Some people worry that data collected from Internet searches may compromise individualsâ privacy. However,
Google maintains that Flu Trends cannot be used to iden- tify individual users because the data are anonymous and are aggregated before being presented. Another potential drawback is that this data collection method is less likely to be useful for tracking epidemics in societies having a low percentage of computer ownership, namely, developing countries. However, considering the high stakes involved in identifying an epidemic quickly, Internet search term analy- sis holds great promise for public health. And, unlike most health innovations, itâs free!
3. How might data collected in this way not be representative of a particular population?
4. Explain why the two graphs in Figure 23.1 do not show influenza activity peaking at the same time.
6. In 2013, Google Flu Trends vastly overestimated the number of cases of the flu in the United States. One explanation for this was that there was a greater than usual number of media reports about influenza early in the flu season. Based on your knowledge of how Flu Trends works, explain why Google missed the mark in 2013
