Page 304 - Contributed Paper Session (CPS) - Volume 2
P. 304
CPS1871 Natividad J.M. et al.
2. Methodology
The data were obtained from the HIV Regional Epidemiology Surveillance
Unit (HIV RESU) of Department of Health (DOH). The data include all HIV cases
in each municipality/city of NCR for the year 2014. The sexually active
population data (15 to 85 years old) in each NCR municipality/city was from
the year 2010 since it is the most recent Census of the Population and Housing
(CPH) from the Philippine Statistics Authority (PSA) available. Moreover, DOH
has stated that most HIV cases are men. In July 2015, 94% of the 682 cases
registered in HIV/AIDS Registry of the Philippines (HARP) were male. Reported
from January to May of 2016 are 3,132 new cases of HIV involving male to
male sex and sex with both males and females. This led to the only covariate
used in this study which was the sexually active male population of NCR from
the year 2010.
Standardized Incidence Ratio (SIR) is the most common statistic used to
estimate the relative risk in disease mapping (Samat & Ma'arof, 2013). In
disease mapping, suppose that the areas to be mapped is divided into m sub-
areas (i = 1,2,…, m). The common risk r is defined to be
=
where y is the total count of HIV cases and N is the total population exposed
to risk in NCR. The estimator of relative risk for region i with respect to the
common risk relative risk r is
= ; =
where is the count of cases in region i, is the population exposed to risk
in region i and is the expected count which is computed with respect to the
common risk r. SIR is greatly affected when the expected count is small and
very small spatial units are involved. There are more variations of SIR in small
cities compared to large cities (Tango, 2010). Hence, SIR does not always
provide an appropriate measure for disease mapping. This happens when the
difference in population exposed to risk among areas are large, and hence,
causing a misleading estimation of the relative risk.
Utilizing Bayesian models provide more stable relative risk estimate due to
prior information, shrinkage, and spatial smoothing. There are two Bayesian
models applied in this study, namely, Poisson-Gamma model with covariate
and Bayesian CAR model to cope with the drawback of the SIRs.
For the Poisson-gamma model, is the observed count of HIV cases in
th
the i area, and has a Poisson distribution. The parameter of interest is is
the relative risk that quantifies whether the area has a higher risk or lower
occurrence of cases than that expected from the reference rates, the intercept
term is denotes the baseline log relative risk of disease across the region
0
being studied, and that serve as the random effects used for smoothing.
With the addition of the covariate which is the percentage of sexually active
1
293 | I S I W S C 2 0 1 9