Page 59 - Contributed Paper Session (CPS) - Volume 8

CPS2184 M Lutfor Rahman
which is one of the ten leading infectious diseases causes deaths of 1.3 million
people around the globe in 2016. It is imperative to identify the leading factors
responsible for spreading the TB infection as well as to quantify the likelihood
of normal individuals being infected in a natural environment, particularly
taking into account the TB networks.
Binary logistic regression assumes independency between observations.
When there is nonindependence in the data, namely due to the presence of a
hierarchical structure, logistic mixed models should be used to model binary
outcomes (Mendes 2013). Is relatively frequent, but not theoretically
adequate, the use of binary logistic regression even if in the presence of a
hierarchical structure. Mendes (2013) used logistic model for interpreting TB
data and to relate TB infection with the risk factors, but their model does not
consider the dependency structure of the data.
Additionally, a more challenging structure can be presented, considering
not only a non-independency between observations but also the index-
contact network relationship. In this case, a more complex statistical approach
but with lower methodological limitations could be based on network analysis,
popularly known as social network analysis (SNA), rooted in the mathematical
graph theory. The SNA has established itself as a powerful tool for studying
structure and complex dynamics of systems (Lusseau et.al. 2008). It has been
employed across the disciplines prominently in studies of transportation
systems (Sen et.al. 2013), infectious disease spread Rothenberg et al. 1995;
PastorSatorras et.al. 2001; Bell et.al. 1999; Cook et al. 2007; computer viruses
through the internet (Newman, 2002), animal behaviour (Wey, 2019);
leadership evaluation (Hoppe et.al. 2010). The methodological and
performance of networks tools evaluation have been reported by Freeman
(1978); Brandes (2001); and Butts (2008). Further, social network analysis has
extensively been reviewed by Wasserman and Faust (1994); Martínez-López et
al. (2009). This study aims to model transmission network of an infectious
disease (event under analyses - infection), particularly considering the
casestudy of Tuberculosis (TB) network in Portugal. Three levels of risk factors
were considered: index patients, contacts and characteristics of exposure. In
methodological terms, three different models were used and the results were
compared.

2. Methodology
The main interest in this study was to model the binary response (infected
or not infected). Three levels of risk factors were considered: index-patients,
contacts and characteristics of exposure: 1) index-patients: biologic products
(Bronchoalveolar lavage, Sputum), chest radiography (Without cavitation,
Cavitation), symptomatic period (in days); 2) contacts: age (years), sex
(Male/Female), diabetes (Yes/No), Chronic renal failure (Yes/No); 3) exposure:

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