Associated factor of mortality rate amongst patients with AIDS and HIV-TB co-infections using zero inflated negative binomial method
Many data sets are characterized as count data with a preponderance of zeros. Data in the form of counts and proportions arise in many fields such as studies in medicine, public health, toxicology, epidemiology, sociology, psychology, engineering, agriculture and soon. When the dependent varia...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English |
Published: |
2014
|
Subjects: | |
Online Access: | http://eprints.uthm.edu.my/1269/1/24p%20MOHD%20ASRUL%20AFFENDI%20ABDULLAH.pdf http://eprints.uthm.edu.my/1269/ |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Many data sets are characterized as count data with a preponderance of zeros. Data in the form
of counts and proportions arise in many fields such as studies in medicine, public health,
toxicology, epidemiology, sociology, psychology, engineering, agriculture and soon. When the
dependent variable is a nonnegative count variable, a Poisson regression model is commonly
used to explain the relationship between the outcome variable and a set of explanatory variables.
However, if extra-zero Poisson counts are observed, it has been suggested that a zero-inflated
Poisson regression model is more appropriate than the classical Poisson regression model. One
frequently encountered problem in these data is that simple models such as the Poisson and the
Binomial models failed to explain the variation that exists. Often, data exhibit extra-dispersion
(over or under dispersion). Another complication in data in the form of counts and proportions is
that they are sometimes too sparse, that is smaller values have greater tendency to occur. In the
Poisson case counts that occur are generally small and in the binomial case the binomial
denominators are often small. Therefore, valid procedures are needed to detect departures from
the simple models. Hence, when a lot of extra zero exists, zero inflated Negative Binomial has
been suggested when overdispersion is present. It is more appropriate than the classical Negative
Binomial regression model. Hence, this thesis follows the general objective, that is to compare
Zero-Inflated Negative Binomial and Negative Binomial in identifying associated factors. The
specific objective is to fit a Zero-Inflated Negative Binomial death rate regression model for
mortality rate among AIDS/HIV co-infection patients and to compare Zero-Inflated Negative
Binomial death rate regression with Negative Binomial death rate, which is the best model when
a data existing zeroes values. It follows by to determine overdispersion in the model. Lastly, to
investigate the potential confounding factors affecting mortality rate among disease mapping co�infection patients among HIV-TB and AIDS. In this thesis, mortality rate is a subject of interest
as dependent variable according to age categories by years. The data are analyzed from AIDS
patients and HIV-TB mortality cases for comparing between Negative Binomial mortality and
Zero Inflated Negative Binomial Mortality (ZINBM) which is better. Beyond this substantive
concern, the choice should be based on the model providing the closest fit between the observed
and predicted values. Unfortunately, the literature presents anomalous findings in terms of
model superiority. In addition, the Akaike’s Information Criterion (AIC) and Bayesian
Information Criterion (BIC) values were used to compare the fit between models. The results
suggested that the literature are not entirely anomalous. However, the accuracy of the findings
depended on the proportion of zeros and the distribution for the non zeros. ZINBDR tend to be
the superior model, than the negative binomial model. The findings suggested there should be
consideration of the proportion of zeroes and the distribution for the nonzero when selecting a
model to accommodate zero-inflated data. |
---|