"confidence intervals for cohort studies in outbreak settings are useless"

Submitted by frank_castle on Wed, 04/29/2009 - 11:15

Dear all,
I have come to a statistical issue that is bugging me. It is heresy, I know. But truth must be told and I can just pray not to be cast into epi-hell.

Orthodox Epi tells us that studies are a tool to estimate population parameters from a sample (i.e. our study). Point estimates such as risk ratios, odds, means, etc. are accompanied by stability measures such as confidence intervals. Confidence intervals give us (depending on their p-values/alphas) the likelihood of the true location of the population RR/OR/mean etc. Ideally, in a study on, say coronary heart disease we would like to sample the total population but that is not possible. Hence the sample (=study) and the confidence intervals.

Enter outbreak investigations. Wedding reception with "barf-sûrprise".
Lets assume a cohort study with 97% participation.
This is, erm, our total population. Isn't it? I mean, we want to know about risk ratios for getting ill during this wedding. Not for all weddings in our total population (Germany: 84 million people).
Hence no need for confidence intervals in outbreak cohort studies with >90% participation.

QED?

‹ Outbreak investigations must always be case-control studies? p<0.001, CI 0.85 - 7.31 or "how can a confidence interval of a risk-ratio include 1 if the p-value is <0.001?" ›