Work Related Update (Might be boring, but at least it shows I'm alive)

Yesterday and today were spent cleaning the database, and just as I thought it was done, it turns out there is more missing information that they want to go back and look up in the paper records- something out of my control. We’re shooting for a complete database and meeting for writing up potential study findings tomorrow.

Yesterday had a lot of rain- Ugandan weather is either nice, or it’s a downpour like you wouldn’t believe (last time I was here it was hailing bullets). Fortunately, I’ve been lucky enough to be inside each time it rained (thunderstorms Monday night, with clearing and sudden showers Tuesday during the day). I ate out yesterday, at an Indian restaurant across the street (I’ve been there before). In general, the largest population after the native Ugandans are Indians, and so Indian food is a pretty good bet. I had a vegetable dish, just because I am once again feeling the lack of vegetables in my diet.

I think when I get home, a lot of eating is in order. I actually am curious what my weight is now too.

In terms of the paper, the intervention seems to have worked pretty well (I mentioned that there was a study going on that ended at the same time the spraying happened so using the number of people showing up with malaria was a bad idea. However, if we use % of positive blood smears as our outcome instead, the percentage of people with positive malaria smears showing up probably won’t be affected, so we have a good outcome measure). I took a look at the % positive malaria lab results today, and you can see the same effect, a drop from about 40% positive to something almost sub 5%.

The major issue to deal with in this study is to account for all the biases, and things that could make the study results flawed. Usually the way this problem is approached is by asking, “what is the ideal way I would have done this experiment” and then look at how much your reality deviated from the ideal- each deviation is a potential source of bias/error which in turn can make your study results flawed and untrue. For example:

Ideal study: Find everyone in the Kihihi district (where this anti-mosquito spraying campaign, hereafter referred to as “IRS,” took place). Measure each person daily and capture every single case of malaria by taking a daily blood sample and looking at it under the microscope, to look for malaria parasites. This way you know exactly how many cases of malaria there are at any point in time. Do your spraying campaign, count malaria cases afterwards.

Now go back in time, to a world where you didn’t do your IRS campaign, so that all the weather conditions, distributions of people, conditions of roads, everything is the same, and instead of doing your spray campaign, don’t do anything, and see how many malaria cases happened afterwards.

Any difference in malaria cases between the first world and the second world (both identical except one world got the IRS, the other didn’t) would be the causal effect of spraying.

Single biggest limitation of the current study:

1) Can’t go back in time (I should work on this one, and make the big bucks).

Why this makes problems: I don’t have good comparison groups- I’d like to compare a sprayed population to a non-sprayed population to prove that IRS works or doesn’t work.

If I could go back in time, I would have 2 groups of identical people to compare, which would be great. The next best thing is to compare 2 groups of nearly identical people in one world, and give IRS to one but not the other, and look at malaria cases. I don’t have that either. The only comparison I get to make is a group of people without spray (before) and the same group of people after IRS (after).

The whole before-after comparison issue is fraught with problems, mainly the problem of all sorts of things changing over time (making things less and less identical between my before and after population). I can go about looking at my before and after population in a number of different ways:

1) Age: Make sure the age distribution of the before population is similar to the after population

2) Gender: Check before vs. after populations on this too

3) Parish: Make sure that people showing up to the clinic are representing their parish the same before and after spraying

4) Do all the above, and in combination (ie: make sure that the percent of children under the age of 5 that are male who showed up before spraying is the same as the percent of children under the age of 5 that are male who showed up after spraying)

5)…

There are a lot of checks I should run, comparing the before and after populations. If I can show that they are mostly identical, and also indicate that all these other variables (rainfall, humidity, road conditions, etc.) either don’t matter or were the same as well, then the paper should be in good shape.

In general, all the limitations that I have to think about now are the same limitations that I’ll have to work on in general if I plan to use surveillance data in the future. It’s nice to know that the work is applicable and has future implications.