Corrections for Confounding and Censoring
b) i. Frequency of promiscuous receptive anal intercourse with intrarectal tissue damage and bleeding..
ii.Frequency of injection with HIV-contaminated needles.
a) Absolute risks of disease are cumulative incidences or incidence density
rates that describe the frequency of occurrence of a phenomenon in a defined
population. To calculate absolute risk, we must divide the number of new cases
by either the number of persons in the population from which they came, or
the number of person-years or other measure of exposure accumulated by that
population. In a case-control study, we do not count that population or its
experience. Rather, we select a number of controls from that population and
compare their past exposures to those of the cases. When we simply classify
exposure as present or absent, the resulting 2x2 table resembles the table
of results from a cumulative incidence cohort study, so it is tempting to
perform the same calculations for absolute risk that are naturally performed
in that study. However, the results depend on the essentially arbitrary number
of controls chosen. For instance, if we were to attempt to estimate absolute
risk (cumulative incidence) from the previous table, we would obtain 1/6 for
nullparous ever-users, 1/9 for nullparous never-users, 4/9 for parous ever-users
and 1/3 for parous never-users. However, if twice the number of controls were
used and the extra controls were exposed in the same proportion as those shown,
the corresponding results would change to 1/11 for nullparous ever-users,
1/17 for nullparous never users, 2/7 for parous never users and 1/5 for parous
never users. The point is not that one set of numbers is wrong and the other
correct. Rather, both are arbitrary and depend on the choice of how many controls
to use in designing the study, that is, the investigator's choice, and not
on the reality of numbers of nullparous ever users, nullparous never users,
parous ever-users and parous never-uses in the population.
b)Using the numbers above, the "relative risk" associated with OC use would be mistakenly calculated as 1.50 in nullparous women and as 1.3 in parous women using the data above, but if the number of controls were doubled these would change respectively to 1.55 and 1.43. They would continue to increase indefinitely were additional similar controls added. On the other hand, under any of these circumstances the odds ratios would be 1.6 for either parous or nullparous women.