Page 64 - Contributed Paper Session (CPS) - Volume 8

CPS2184 M Lutfor Rahman
Table 5: Estimates of parameters in hierarchical and network logistic regression
analysis of TB networks in Portugal
Characteristi Categor Hierarchical Logistic Network Logistic
cs y Regression (Model Regression
Coefficie p- Adjusted Coefficie p- Adjuste
nt valu OR (95% nt valu d
e CI) e OR

Intercept -2.539 0.00 -8.801 0.00
0 0
Contacts’ characteristics
Age in years 0.033 0.00 1.034 0.0925 0.04 1.0969
0 (1.015,1.053 5 *
)*
Diabetes No 1 1
mellitus Yes 0.788 0.03 2.198 1.128 0.48 3.0895
5 (1.059,4.566 3
)* NS
Exposure characteristics
Contact type Casual 1 1
1.346 0.00 3.842 3.638 * 0.04 38.016
Househol 1 (1.766,8.359) 2 *
d **
Sleeping No 1
together Yes 1.438 NS 0.49 4.210
3

In multiple logistic regressions, Nagelkereke R is 0.167, in hierarchical
2
2
regression R is not available and in network logistic regression (NLR) pseudo-
R2 is 0.580 which indicates that the variation in having TB infection is better
explained by network logistic regression as the NLR takes into account
information on index-contact relation in addition to the exposure
characteristics. However, there could be lack of information fitting all the
models as the value of R is not strong enough in all models. One of the
2
reasons of poor performance of this network logistic regression
approximation is that there could be some interaction terms those were not
been considered in the model. This can be considered as a limitation of the
network logistic regression process.

4. Discussion and Conclusion
This study has taken into account the network analysis of Tuberculosis
patients particularly considering a TB network from Portugal. In this endeavor,
we have compared the estimates from crude, multiple, hierarchical, and

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