To answer your questions in turn ... (1) Yes. The technique come from biology / ecology to answer such questions as "How many beetles in this forest?" or "How many fish in the sea?". There is no expectation of exhaustivity. There are a number of assumptions ... Closed population - not usually a problem for SAM but can be if central location screening attracts "outsiders". Reliable identification and matching - Good training and proper collection of identifying data should ensure this. Equal catchability - may be a problem if central location screening attracts "outsiders". independence - Not usually a problem if different methods, staff, and informants are used. What you should see here is that central location screening risks violating the closed population and equal catchabilty assumptions and may also, by its public nature, violate independence. The short answer is you can use central location screening but be careful. Note : You can identify and make rough corections for violations. See: http://www.brixtonhealth.com/CRCaseFinding.pdf for more information. (2) Sample size is not a straightforward issue. This has been discussed previously on this forum. The standard estimator for the number of cases (N) in the population sampled by a capture-recapture study is: N = [((M - 1) * (C - 1)) / (R - 1)] - 1 Where: M = number found by one method
C = number found by other method
R = number found by both methods The estimator is unbiased if: (M + C) > N and: R > 7 Provided you satisfy these constraints then you should be OK. You have to make guesses for N and the two sensitivities to do this. See: http://www.brixtonhealth.com/CRCaseFinding.pdf for more information. (3) You should attempt to cover as much of the area as possible. This will allow you to make a thorough map and avoid the bias that will come from excluding boundary areas. (4) The usual approach is to decide on the spatial resolution of the survey. I do not usually go above a quadrat side length of 13 km (that gives 169 square kilometers). You can use more innovative sampling schemes and more complicated analysis to go bigger but retain resolution but I will not go into this here. You then draw the grid taking care not to introduce a bias by (e.g.) putting all sampling locations in valleys or by avoiding program area boundaries. You than sample as many communities as possible in a single day from each quadrat. I have seen quadrat sizes vary from 3-by-3 (Rwanda) and 13-by-13 (Niger). The number of communities per quadrat ranges from a mean of 3 to a mean of 5. (5) Active case-finding usually only works well for SAM cases. I did some work on this during development of CSAS using WHM case-definitions and found something like 100% sensitivity for WHM < 70%, 70% sensitivity for WHM 70% - 74%, and 40% sensitivity for WHM 75% - 79% (I can find the exact figures if needed). This may improve a little for MUAC case-definitions but probably not enough to be useful. My advaice for moderate cases is to go house-to-house. MAM cases are more common than SAM cases so you don't need to sample as many communities. I am sure you will be able to think of alternatives if you need to do both SAM and MAM ... for example ... do active cases finding in 3 villages and door-to-door in a further 2 villages (first and third village) from each quadrat I hope this is useful. let me know if I missed anything.
|