Page 65 - Contributed Paper Session (CPS) - Volume 8

CPS2184 M Lutfor Rahman
network logistic regressions to explore the factors influencing TB infection
possibilities.
The crude estimate shows that age, diabetes mellitus, contact type,
sleeping together, and eating together are important covariates. The multiple
logistic regression reveals that index characteristics- biologic product and
symptomatic period (days) and contact characteristics age, contact type, and
exposure duration are important factors. However, simple and multiple logistic
regressions do not consider the dependency of responses. The hierarchical
logistic regression considers the dependency of responses while modelling TB
network data. The hierarchical logistic regression finds that the variables age,
diabetes mellitus, and contact type are significant for TB infection in contacts.
The crude, multiple, and hierarchical models do not consider the structure of
TB network. At this stage, the network logistic regression appeared to be a
good tool for analyzing TB network data. The network logistic regression
shows that only age and contact type are much valuable to interpret TB
network data.
The current study was supplemented by two models viz. hierarchical and
network logistic models. The benefit of these models was visible as they
provide more precise list of predictors for TB infection. Hierarchical model
indicates age, diabetes mellitus, and contact type being important and further
network logistic analysis limits the predictors to age and contact type to be
central for TB infection.
To sum up the discussion, we emphasize on the factors age and contact
type (household or casual) to be the most important factors for TB infection
when people interact TB patients in the real life. In older age, people are
vulnerable to any disease including TB as their immune system become
weaker, thus older people are more likely to have TB infection than the
younger people. Contact type- particularly household contacts appeared to
be more exposed to TB infection than others as found in all methods. The
other risk factors e.g. eating together, sleeping together are broadly
represented by household contacts. None of the exposure characteristics i.e.
exposure site (big or small) and ventilation facilities (yes or no) are found to
be important for TB infection among contacts.
The current work can be extended to dynamic network system where for
each of the time point in a follow up study, the number of contacts (edges)
and number of vertices (nodes), number of infected individuals can also be
predicted. If dynamic network logistic regression is in use, it would be possible
to forecast mean degree (number of nodes on average) and network size for
a particular future time point. The similar approach, particularly network
logistic regression, can be replicated in the investigation of other infectious
diseases.

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