What determines the choice of the 'length of the recall period' in measuring mortality when undertaking SMART assessments .
Rogers
This is not a question specific to SMART surveys but to any retrospective mortality estimation technique. There is no hard and fast rule to deciding the length of recall period. Usually with SMART surveys (the RT stands for "relief and transition" which translates to emergency and immediate post-emergency) you will want to estimate mortality over a short and recent period. Short periods are better than long periods (i.e. to minimise recall bias).
The sample size (denominator) for the mortality estimator is person-time-at-risk. There are two components. These are persons (i.e. the number of people alive at the start of the recall period) and time (i.e. the number of days in the recall period). These are multiplied together to give person-days-at-risk. If you (retrospectively) follow-up 600 people for 90 days then you have 54,000 person-days-at-risk. Whether this is a big enough sample will depend on the precision required for the resulting estimate.
A simple and useful sample size formula is:
n(pdar) = u / (precision / 1.96)^2
where "u" is the expected rate and precision is the desired half width of the 95% confidence interval.
An example ... we expect a rate of 2 deaths / 10,000 / day (i.e. 0.0002) and want to estimate this with a precision of plus of minus 1 death / 10,000 / day. This gives:
n(pdar) = u / (precision / 1.96)^2
n(pdar) = 0.0002 / (0.0001 / 1.96)^2
n(pdar) = 76,832
If you use a cluster sampled design then you can expect a rather high design effect (e.g. because mortality due to infection will be clustered). The expected design effect (DEFF) is usually set at 2.0. The sample sizes calculated above should be multiplied by the expected design effect. Continuing the example we have:
n(pdar) = DEFF * [u / (precision / 1.96)^2]
n(pdar) = 2.0 * [0.0002 / (0.0001 / 1.96)^2]
n(pdar) = 153,664
This is a large number. There are two ways to get at this sample size : We can increase the number of people followed up or we can increase the length of the recall period. It is not usually sensible to increase the length of the follow-up period above about three months. Continuing the example:
n(persons followed for 90 days) = 153664 / 90 = 1708
This can be translated into households to be sampled. If (e.g.) the average household size is 4.5 persons then:
n(HH) = 1708 / 4.5 = 380
This is readily achievable so you may want to reduce the follow-up period and sample more households.
BEWARE : The sample for a SMART nutrition survey and a mortality survey will be different. If you use the same sample you risk underestimating mortality by excluding households in which all children have died. This is a survivor bias.
Another consideration is memorability of the start of the recall period. If you can use (e.g.) "since the start of the new year" or "since Christmas" then you are more likely to have deaths reported within the recall period and reduce recall bias.
You can find a review of some of these ideas in this [url=http://fex.ennonline.net/17/amethod]Field Exchange article[/url].
I hope this helps.
Mark Myatt
Technical Expert
Answered:
11 years agoThe recall period in Mortality surveys depend on the objective of conducting the survey and the starting recall point that the respondents can remember. Its also important to factor in any earlier mortality trends in the population of interest when deciding on the recall period.
Anonymous
Answered:
11 years ago