Page 307 - Contributed Paper Session (CPS) - Volume 2

CPS1871 Natividad J.M. et al.
Table 1. Relative risks under SIR, Poisson-Gamma model, and CAR
Poisson- Bayesian
HIV Population Gamma CAR
NCR Cities/ Cases Exposed Relative Relative
Municipality (yi) to Risk (Ni) SIR Risk Risk
Caloocan 159 1,017,262 0.5721 0.5685 0.5818
Las Piñas 68 390,644 0.6371 0.6279 0.6414
Makati 206 398,383 1.8925 1.884 1.8530
Malabon 49 246,188 0.7285 0.7121 0.7088
Mandaluyong 149 237,586 2.2953 2.281 2.2040
Manila 348 1,173,946 1.0849 1.082 1.0840
Marikina 97 306,470 1.1584 1.147 1.1580
Muntinlupa 64 312,838 0.7488 0.7378 0.7511
Navotas 16 167,691 0.3492 0.6651 0.4232
Parañaque 85 422,804 0.7358 0.7282 0.7482
Pasay 109 282,651 1.4114 1.399 1.3870
Pasig 188 476,765 1.4432 1.436 1.4340
Pateros 12 45,149 0.9728 0.8948 1.0760
Quezon City 526 1,948,715 0.9879 0.9861 0.9905
San Juan 26 92,993 1.0233 0.9883 1.0930
Taguig 113 443,167 0.9332 0.9248 0.9490
Valenzuela 70 399,792 0.6408 0.6330 0.6488

Bayesian CAR is the better model since it has less DIC (118.5) compared to
the Poisson-Gamma model with DIC 149.4. This further supports that Bayesian
CAR model is the most appropriate model in estimating relative risk of HIV
infection in NCR for it also takes into account spatial autocorrelation.
Under Bayesian CAR model, there were eight cities observed having
relative risk greater than one which means these areas have higher occurrence
of HIV cases, whereas the remaining places experienced less than expected.
The relative risks varied from 0.5786 to 2.1970. Mandaluyong still had the
highest HIV relative risk estimate with 2.1970. Furthermore, its neighboring
areas like Makati (1.8530), Pasig (1.4340), Pasay (1.3870), Marikina (1.1580),
Manila (1.0840), San Juan (1.0930), and Pateros (1.0760) had high estimates of
HIV relative risk. The municipality of Pateros, which is a nearby area of
Mandaluyong, obtained a relative risk estimate greater 1 due to spatial
autocorrelation under Bayesian CAR as compared to the SIR and Poisson-
Gamma model. HIV relative risk of Mandaluyong is associated with those in
the neighboring cities due to the implications of the positive spatial
autocorrelation.

296 | I S I W S C 2 0 1 9