Now, we've gone through the major ways you can measure mortality. We want to talk about other ways they're often used to model or estimate mortality. The two examples I'm going to give, IGME, which gives you an overview of how the UN tries to use data to estimate mortality, and the Lives Saved Tool or LiST, which is actually something that would be more likely to be used within an evaluation. In fact, we'll be teaching you how to use that model later in this course. IGME, this is a Inter-agency Group on Mortality Estimation. I want to explain the idea of their strategy for estimating child mortality. Now they do neonatal infant and under five. We're really going to focus on under five. The idea is that they try to compile and assess quality of all nationally representative datasets. The things we've already mentioned, vital registration, censuses, household surveys, sample registrations. They basically created a database to combine all this information for all low and middle-income countries. Secondly, they then take this data and recalculate data inputs and make adjustments. For example, with HIV, before we had the heavy treatment for HIV, in fact, HIV lead to biases in household surveys because women who had HIV in the past without treatment had died, their children were also at a higher risk of mortality. Therefore, you had this bias that children who died of AIDS were not reporting because her mother had also died. But there are methods to correct for that. What IGME does is they take all this information, they judge it on quality, they make corrections as appropriate, and then they fit a statistical model to get the smooth trend. We've seen that earlier in some of our slides of how that does. Then they extrapolate over time to say, "Here's the local trend," and say, "This is the current estimate of mortality in a country, even if there's no data for the current year." These are done yearly, it's updated, the information. They make some mythological changes they go along. They've also overtime expanded to do it by different age groups and by gender, and also some sub-national estimates, not only national. I'll give you a quick example of, here's two datasets that you can see. You have a country with lots of data. You see back in the 1970s, the data is less consistent, but over the time, the data becomes more and more consistent. The light blue line is IGME estimates. The gray is there uncertainty bounds around it. As you see, the information is we go down in time, mortality is going down, but also there's more and more consistency among the data points. But even there, there's a few outliers, but the uncertainty bounds around those estimates become less and less. Here's another low income country. Here, you can see we have complete divergence in terms of the information. We have many few data points and they're very inconsistent, and the UN estimates of mortality are very wide. If you look at the year of 2015, we're saying, well, the uncertainty bounds around the UN estimate or around any measured point is quite dramatic. The idea is one of the things that IGME is a good place for you to go and look and find out what data are available, kind of an expert opinion about how it is, the quality of the information, and some idea of what they think the uncertainty is around these measured points, whether it's vital registration or household surveys that are put into their overall estimate. Here's a dataset from Mozambique. In the previous slide, we showed you the best country, Bangladesh, with the most data, and then Angola, country that has very little data. Mozambique is more representative most low and middle-income countries. More data than Angola, but not like Bangladesh. You can see there's a consistent trend that starts out. There's wide variation in the past, but over time, you get more and more consistency between the different data sources. All of the data points are fairly consistent with the blue line, the UN's estimated trend in mortality. But the thing for you to remember here is, notice there's still lots of variation between a measured DHS mortality estimate in the year 2010 and '11. You can see from this graph, in the past, there's fewer surveys, fewer data point, there's wider spread in terms of the estimates. Over time, they become more consistent. The UN estimates, the uncertainty bound, the gray bounds get smaller and smaller over time. But a key point for you to remember or you recognize if you're thinking about measuring mortality is even here with DHS surveys, the UN assumes the survey is not the correct estimate. That you can see the survey points don't match their blue line. There's still uncertainty around those estimated values. Critical point for you if you are looking for baseline mortality in planning your survey, IGME at a national level would probably be a better source of that estimate than simply taking the most recent DHS or mixed data. Another approach to estimating under five mortality is using a model that unlike IGME, which tries to combine measured estimate, some mortality to get the best overall estimate, other models like the Lives Saved Tool or LiST, use changes. They try to model under five mortality using other variables. Specifically, what it tries to do is it estimates mortality changes or mortality by coverage of known effective interventions. The LiST tool started in about 2003, has now been copied by several other groups, EQUIST in UNICEF, One Health in WHO, that all use this information about building a baseline where you describe a country, and then as coverage changes occur, reestimate mortality based on cause of death structure and efficacy of interventions. The Lives Saved Tool is fairly complex and it takes into population dynamics, risk factors about nutrition and HIV, and interactions among interventions. It can be used both at the national and sub-national level. In fact, one of its major uses has been in evaluations where you're not able to estimate or measure mortality at baseline and inline because it requires too big of a sample survey to do, it's too costly, and the Lives Saved Tools can be used to fairly reliably estimate changes in under five mortality due to coverage changes of interventions. To summarize what we've talked about, measuring or estimating under five mortality for your valuation, let's go through. Generally, household surveys are the best option for countries where vital registration or vital statistics systems are weak, and that's going to be the case in nearly all low and middle-income countries. Full birth histories. We can either use full birth histories or summary birth histories in our surveys. Generally full birth histories, while they're time consuming and are more difficult to do, represent the current best practice. Also, you've got to remember that child deaths are rare, so recall period and sample sizes and certainty bounds are going to be large. As mortality goes down, you need bigger and bigger surveys to reliably measure mortality. If you're doing an evaluation of a program that says it's going to have a 10 percent reduction in mortality, well, if you're working in Bangladesh and under five mortality is 35, you're only going to be lowering it to 32. You're going to have to have quite an accurate and large survey to be able to measure that. One of the good advantages is that a single household survey, if large enough to reliably measure mortality, can do it both at baseline and inline of your survey. That's one real advantage that you don't have to pay the two surveys, you pay for one. The other approach that really has been used more in evaluation is modeling. You can have complex models that take multiple sources, like IGME, and that's a good way to have multiple information to try to estimate mortality. Also, the Lives Saved Tool is a slightly different approach that links coverage of interventions to estimate mortality changes. Both of these approaches of modeling is really what most of evaluations do in terms of producing estimates of impact of a program.
