Dr. A Coomarasamy posted:
>Three therapeutic (surgical) approaches are compared to unblock
>(recanalise) the fallopian tube in women with confimed blocked tubes.
>Pregnancy rates are compared following the procedures, follow up varying
>from months to years. we wanted to compare the proportion of women who got
>pregnant with each treatment and whether the proportions were statistically
>(and clinically )significantly different. When a woman has 2 or 3 or more
>pregnancies following the procedure, this reflects that the procedure was a
>'long-term' cure and therefore we wanted to compare the pregnancies
achieved
>with each procedure rather than merely 'how many got pregnant/ how many
>didn't (ie not counting the number of pregnancies a woman had, but just
>whether she was 'cured' or 'not cured'). When a woman has 2 or more
>pregnancies, and each pregnancy is counted, the data is not independent
>anymore - is it sensible to do this? if so how do you compare the
>proportions? what would be the denominators?
Difficult to see how the number of children desired, thus number of attempts
to become pregnant again, could be estimated so as to make a meaningful
denominator. Perhaps a stratified analysis in which only those expressing
desire for multiple pregnancies enter the analysis to compare procedures? A
chi-square test for trend or some other nonparametric test comes to mind,
all else being equal. Weiler H, A Significance Test for Simultaneous Quantal
& Quantitative Responses, TECHNOMETRICS 1964;6(3):273-85 also might offer
the sort of approach needed?
Good luck,
David Birnbaum, PhD, MPH
Clinical Assistant Professor
Dept. of Health Care & Epidemiology
University of British Columbia, Canada
|