>Dear alllstat members,
>
>My special thanks to those who replied my query re: the 95% CI for PAR%.
>Here is the summary of the mails which i received as reply :
>
>(1) For well-constructed confidence intervals for PAR see StatsDirect from
>http://www.statsdirect.com/update.htm which gives you a free 10-day
>trial and full access to all functions - see Analysis_Miscellaneous_Risk
>(Prospective) menu item and its help.
>
>EXPOSED UNEXPOSED
>OUTCOME YES a b
>NO c d
>
>Outcome rate exposed (Pe) = a/(a+c)
>Outcome rate not exposed (Pu) = b/(b+d)
>Relative risk (RR) = Pe/Pu
>Risk difference (RD) = Pe-Pu
>Estimate of population exposure (Px) = (a+c)/(a+b+c+d)
>Population attributable risk % = 100*(Px*(RR-1))/(1+(Px*(RR-1)))
>
>Walter's approximate variance formula is used to construct the
>confidence interval for population attributable risk (Walter, 1978;
>Leung and Kupper, 1981).
>Walter SD. Calculation of attributable risks from epidemiologic data.
>International Journal of Epidemiology 1978;7:175-82.
>
>Leung HM, Kupper LL. Comparison of confidence intervals for attributable
>risk. Biometrics 1981;37:293-302.
>
>****
>(2) If you have an estiamte of the log(RR) with its s.e. then simulate
>say 10000 samples from a normal dist with this mean and sd. If you further
>have an observed exposure prevalence, then the log-odds of this has
>variance 1/(np(1-p)), so you simulate 10000 samples from a normal with mean
>log-prevalence odds and this variance.
>For each pair compute the PAR from the realised values of log(RR) and
>log(odds),
>and take the 250th and 9750th of the observations.
>
>Alternativly you can use 27 pages of notes to derive the approx s.d. of the
>PAR using the delta method.
>
>*********
>(3) If the estimation procedure is simple and you/your computer can afford
>to run in 1000+
>times, establish the CI by simulations from the model fit to your data. No
>need for any
>theoretical twists and turns, nor for any software other than what you use
>for estimation.
>
>
>
>Thank you all again.
>
>
>Best regards,
>
>
>Leah
|