A Poisson model with a random effect for individual
was developed and fit using a generalized linear latent
and mixed model (GLLAMM) in Stata version 11 [21]
to assess associations between independent variables and
the incidence of TB. Variables included as fixed effects
were: age (centered at the mean), duration of employment
(duration of employment in the facility minus the
total person-time in years), occupation (categorical variable
with administrative staff as the reference group),
HIV status (categorical with HIV-negative as reference
group; HIV-positive, and HIV status not recorded), history
of work in a TB ward, general ward, pediatric ward,
outpatient department, primary health care center, stores/
workshop, and number of wards HCWs had worked in
(1 ward vs. >1 ward). Adaptive Gaussian quadrature
with 10 points was used to estimate the parameters of
the models. Model selection was performed comparing
the Akaike information criterion (AIC) and the model
with the lowest AIC was selected. Variables were retained
in the final model if the p-value was