Okay, I'd like to move on now to talking about the survey part, but also talking about the possibility that there are downsides to surveillance. For example, one of my former students was working on the Guinea worm eradication program in Chad and they had a couple of outbreaks of Guinea worm, in areas, where Guinea worm hadn't been seen for years. Guinea worm is a parasitic disease that you acquire by essentially drinking a little larvae of the parasite, that parasite then grows inside your intestine and normally pierces your intestine and ends up in a blood vessel in one of the extremities classically in your lower legs. When that worm sort of produces a lot of eggs, it creates this bulbous sore on one's leg or on one's arm and that sore tends to burn. It just burns and makes people want to put that sore in water and when it does the bulb bursts open and out go the larvae and eggs and the transmission cycle begins again. Well, here they found cases, which take a year to transmit. In areas where they hadn't seen any cases before, and by going and looking they realized that dogs also carry guinea worms. Until just a couple of years ago we never knew that. Now, controlling dog transmission is a central part of this final effort to eradicate Guinea worm. This is a really nice example of how when we have surveillance, that's really sensitive and really active we can start to understand the minutia of transmission and learn things that can't be learned when it's masked by many cases and only the big picture images. So surveillance, it has great advantages, but it turns out surveillance has some disadvantages or negatives as well. For example, if there are some sort of changing behaviors in society such as people more likely or less likely to come to get medical care we could see a change in incidents that really isn't related to a change in disease it's just a change of incidence detected in our surveillance data. There also is a strong tendency that surveillance data will not represent all of the population well. So here we are, we're back in that Tanzanian camp. The same camp that we looked at with the low birth-weight events reducing over time. Again looking at three and a half years of surveillance data we see that the incidence of dysentery was very high for two years and then seems to almost disappear. Makes you wonder why that is, it turns out that in January of 2000, the old Medical Coordinator, a lovely Italian doctor, decided he was going to leave, a new young doctor from Togo arrived, his name was was Dr. Ami great guy and he said "Wait, wait, wait. You are telling me, we've had two episodes per 1000 per month, for two consecutive years. You are telling me that this camp has been the center of the worst Shigella dysenteriae outbreak anywhere in Africa for two years continuously? I don't believe it. I want every time a patient comes in, I want you instead of treating them as an outpatient, I want you to make them an inpatient, give them ORS, but I don't want you to give them anything else until I've seen their poop, and seen if there's blood in there poop." In doing so he eradicated dysentery. And the reason was because everyone in the camp knew if you walked up to the outpatient window and you said I have diarrhea you would get oral re-hydration solution or ORS. If you walked up to the outpatient window and said I've got diarrhea with blood in it, you got ORS but also a really strong antibiotic. In this case, it was nalidixic acid and everyone wanted that strong antibiotic. Here he just didn't change the definition of dysentery, he changed the application of medical services, for a case of dysentery. In doing so, our surveillance data would suggest there was a dramatic drop when in fact the incidence hasn't changed at all, surveillance, It's susceptible to changing behaviors. This graphic is coming from the US government and it's showing that number of new cases of HIV, between 1982 and 2009 and it turns out that in 2002 we changed our definition, we expanded our definition of what it took to be considered an HIV positive case. When we change that definition the number of recorded cases spiked. So here we have an apparent spike in HIV but actually, the true incidence didn't dramatically spike. We just changed our definition, and in doing so, we have this artifact in our surveillance data. There are certain events that almost certainly are never or rarely going to be captured by surveillance data. Here we are, we're looking at the cover of The New York Times from back in 2013, and there is, a group of rebel soldiers who have taken prisoner some of the government soldiers, and are standing over them and about to execute them, and indeed they did execute them in front of a New York Times reporter, who then after taking this picture wrote a story about it. This sort of event will virtually never be captured by ongoing surveillance data because A, the people involved have incentives to hide it. B. It tend to happen in areas where there's not a lot of people watching or witnessing, and the bodies might get buried and never be found. So, we would expect certain events are very very hard for surveillance data to identify. Here is a little image and it's from the website of the group Women Under Siege which is talking about violence in Burma, but this organization was formed to try to make a record of rapes happening in Syria by armed combatants in the goal of telling the outside world how rape was being used as a weapon in Syria. Most of the reports that came in and they come in from other countries are coming in by social media. So I had a graduate student a few years ago, and she went out to the region on behalf of Women Under Siege. She was fluent in Arabic and she attempted to confirm 70 something social media reports that mostly had come from medical workers in hospitals, of rapes that had happened, that got reported by the social media website, and she could only confirm two. So, this reminds us that, not only is rape very hard to detect by surveillance but that we are evolving to have kinds of surveillance information. Newspaper reports of outbreaks, social media tabulations, Facebook entries on certain subjects that are perhaps either difficult to confirm or maybe even manipulatable. In terms of being manipulatable, here is a pretty dramatic example of that. We are looking at a graphic that is comparing major killing events in Iraq, that were recorded by the newspaper monitoring group, Iraq Body Count, with the Wikileaks release of the Iraq war logs back in 2010. What we're looking at are on our x-axis, categories of killing events broken down by how many people were killed. In the first column 20 or more people were killed, in the next vertical column, it was 10 to 19 people killed, going all the way over to the right where only one person was killed in some violent event. Then on our y-axis, we have the likelihood that an event recorded in Wikileaks was reported by this newspaper monitoring group, and you will see that as shown by the darkest color portion of that bar. If 20 or more people were killed most of the time both of these surveillance systems, the US government collecting rumors that was then later released in the form of Wikileaks War Logs, and Iraq Body Counts newspaper monitoring. Most of the time both surveillance systems had an event if 20 or more people were killed. But if we look over on the right-hand side, if only one person was killed, which by the way we believe to be the main killing events in Iraq over the window from 2003 to 2010. Most of the time, probably over 80 percent of the time, when it was in one dataset, it wasn't recorded in the other datasets. So here, this surveillance is very insensitive to a single person being shot as they're out on the street, is very very sensitive to a car bombs going off and killing 20 people, and that gives the impression when you read the newspaper that most violence in Iraq over this period of war was from car bombs and explosive events and big battles, when in fact the Lions Air violence was one or two people being shot in some killing event. Thus when people take that Iraq Body Count data and analyze it, they make all kinds of conclusions like this New England Journal of Medicine article, that car bombs and explosives are doing most of the killing, when that's not true at all, it's just that the surveillance system they are using is so biased that it makes it look that way. So surveys on the other hand have certain advantages, they tend to be more complete in terms of the information you want about the related social factors. If you want to know if an Ebola case had gone to a funeral, if you want to know if that Ebola case is coming from a household where another Ebola case had happened, well, that's easy to capture in a household survey, so tends to be more complete. It also tends to be more representative, you're typically with a survey having a random sample of the population. So you control the information that's collected with it and as a result of it being statistically representative, you can do statistical testing and see, "Oh, is it really more likely?" As was the case in Sierra Leone, that Muslim families are more likely to have a case of Ebola than a Christian family. I can do that statistical testing with a survey, but I couldn't do it with surveillance that's incomplete and got bias in it. The disadvantages are; something's got to be pretty serious in order for me to get reasonable information with a survey. My rule of thumb is, if 2 percent of households aren't experiencing something, then you probably shouldn't even bother asking about it with a survey. It takes a lot of skill to do a survey, and if people don't believe the results, if they have a different viewpoint, they might just reject the findings, this happens all the time. So there have been a couple of occasions when dozens of surveys, either nutrition surveys or mortality surveys, were done in a crisis and each time when summaries are tabulated afterwards, there's enormous evidence that the quality of the surveys is very very poor, for example, a woman named Leslie Boss et al looked at a series of surveys done in Somalia by various NGOs and found that the quality was so poor, that it was essentially impossible to combine the results and draw any conclusions. Similarly, Paul Spiegel and Peter Salama did an analysis of a 126 nutritional surveys in Ethiopia, back around two thousand when there was a period of acute food shortage. They found that only half of those surveys even tried to take a sample and guess what, on those occasions when people didn't take a formal sample but just went out and stopped and measured people wherever they felt inspired, like clockwork, they measured much higher levels of malnutrition than in those surveys that involved a sample, and only six of a 126 had enough places, enough clusters, enough children measured, to have any statistically meaningful results. Oh my gosh, what a waste of resources, what a waste of time to be doing all those surveys that are so low quality, they probably don't mean very much and they probably don't guide services very meaningfully or well. To summarize our thoughts about surveillance versus surveys, surveillance has huge advantages in that it tends to be timely and believed and can be medically confirmed in most cases, it's almost always incomplete, it's very open to biases, for certain things like executions of soldiers or rape events that people don't want to talk about, it's just not going to capture very much. So it's really best for arguing priorities, like we saw in Northern Kurdistan when diarrhea was accounting for 75 percent of all deaths. Even if that data was incomplete, it still telling us diarrhea is the main problem. Surveys tend to be more complete, but harder for people to believe. Therefore they tend to be more valuable, either for advocating that something is a crisis in general or for non-controversial things. If you do a survey about the fraction of people that have a metal roof, you will never get any blowback or pressure and have people tell you they don't believe the results.