The primary objective was to estima

The primary objective was to estimate anal cancer incidence
overall and over time among HIV-infected MSM, HIV-infected
other men (ie, heterosexual, injection drug use [IDU], or other
HIV risk factor), and HIV-infected women. We compared

incidence rates to the reference groups of HIV-uninfected men
and women. For each cohort member, the start of follow-up
(ie, baseline) was defined as the cohort entry date, or 1 January
1996, whichever came last. Individuals were followed until
there was an anal cancer diagnosis, loss to follow-up, death,
or up to 31 December 2007, whichever occurred first.
We initially computed anal cancer incidence rates over the
entire study period. Adjusted rate ratios (RRs) were calculated
for HIV-infected MSM and other men compared with HIVuninfected
individuals (reference group) using Poisson
models with terms for calendar era (1996–1999, 2000–2003,
and 2004–2007), age at entry into a calendar era by 5-year age
groups, race/ethnicity (white, black, other), and cohort. Multiple
imputation [16] was used to impute missing race/ethnicity
on the basis of an individual’s age, HIV status, calendar period
of entry into cohort, anal cancer status, and cohort. We also
computed anal cancer incidence rates for HIV-infected and
uninfected women, although adjusted RRs were not computed
because no cases were observed among HIV-uninfected
women.
We then computed cohort-specific anal cancer incidence
rates for HIV-infected individuals. Using linear regression, the
cohort-specific incidence rate was modeled as a function of
each cohort’s prevalence of MSM. Each of the 13 observations
representing an individual cohort was weighted according to
the amount of person-time contributed by that cohort. We
report the R2 as a measure of the amount of variability in
cohort-specific rates explained by the cohort-specific prevalence
of MSM.
In subsequent analyses, we evaluated whether anal cancer
incidence rates have changed over time (1996–1999, 2000–2003,
and 2004–2007) for HIV-infected MSM, other men, and
women. Denominators for these rates were obtained by dividing
each subject’s person-time into the corresponding calendar eras;
to obtain numerators, anal cancer diagnoses made during
a given era were assigned to that era. P values comparing anal
cancer incidence rates by era (2000–2003 as reference) were
obtained from a Poisson regression model, and a global
P value was based on the likelihood ratio statistic. Multivariable
Poisson models were then used to evaluate changes in
RRs over time for HIV-infected MSM and other men compared
with HIV-uninfected men, adjusting for race/ethnicity,
age at entry into calendar era in 5-year groups, and cohort.
Because only 3 of the 13 cohorts contributed data on HIVuninfected
individuals, we also report standardized incidence
ratios (SIRs) for HIV-infected MSM, other men, and women,
adjusting for age and race/ethnicity, using national US Surveillance,
Epidemiology and End Results (SEER) program rates
for 1996–2005 [17].