Suppose you are looking at the prevalence of a certain disease and you can detect this disease in fecal samples. How does having a bunch of pooled fecal samples (i.e., taking a fecal sample from 5 individuals and pooling them together to get "one" sample) affect the variance compared to having a bunch of individual samples where each sample represents one person.

I am doing a series of meta-analyses on the prevalence of a certain disease. The meta-analyses where the disease is detected in pooled fecal samples tends to have a lower heterogeneity compared to the meta-analyses where the disease is detected in individual fecal samples (individual-level sampling). What could explain this?

  • $\begingroup$ Because "pooling" is used in so many different but similar-sounding ways in statistics, it would be better to refer to this as "compositing." The answer depends on whether the aliquots (the components of each composite) can be considered statistically independent and it also depends on how much the measurement of any sample contributes to its variance. $\endgroup$ – whuber Jan 23 at 21:55

Not sure in what sense you mean "pooling". In epidemiology, there's a concept of composite sampling. That is, for a rare disease and sensitive assay, you can collate a fraction of biospecimen and provide the diagnostic for 10 to 100 unique samples and, pending a positive result, subsample from the cluster to identify which participant was prevalent for disease. The process goes back to Dorfman who in 1943 suggested the approach as a cost effective way to test US military recruits for syphilis. For prevalence estimates, it's not necessary to subsample.

Meta-analyses also usually discuss pooled analyses where a weighted regression is used to combine the results from a number of studies that are considered to be "sufficiently homogeneous". Personally, I don't think this is the point of a good meta-analysis, since a bunch of underpowered studies does not one adequately powered study make. There's lots of reasons (early termination, publication bias, exploratory analyses published as confirmatory analyses) "the literature" is not a good database, though it contains some very good records.

Ultimately, if the choice boils down to an epidemiologic study that uses composite sampling or individual sampling, the composite study is superior because you can come up with the exact same prevalence estimate (unbiasedness) for either a lot less money or a lot more participants. I wager the researchers who are smart enough to composite also chose to recruit a bigger N. I very much doubt the heterogeneity differs significantly for an individual sampled vs compositely sampled study of the same N.

  • $\begingroup$ Sorry, I meant I am doing a meta-analysis of studies which have used compositee samples for their prevalence estimate. I have another meta-analysis which used studies that have done individual-level sampling. $\endgroup$ – aspire94 Jan 23 at 22:06
  • $\begingroup$ @aspire94 glad to clear it up. My main point is that composite studies will tend to have a larger N, and is something you can confirm. Larger N means more reliable estimate of prevalence, and better homogeneity And they're probably run better, because composite sampling is an intelligent thing to do. $\endgroup$ – AdamO Jan 23 at 22:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.