| dc.description.abstract | Introduction
AIDS surveillance has been the cornerstone of national efforts to monitor the spread of HIV infection in the world and to target HIV-prevention programs and health-care services. HIV case surveillance provides data to better characterize populations in which HIV infection has been newly diagnosed, including persons with evidence of recent HIV infection such as adolescents and young adults. The aim of this work after summarizing the properties of the two discriminating methods is to explore the convergence (give same results) and divergence (give different results) of the two analytical methods when they are used to classify participants as “HIV positive” or “HIV negative” using the symptoms of HIV.
Methods
The study was conducted in Chimanimani district, Manicaland Province of Zimbabwe in 2005.This was secondary data analysis of data from baseline study that utilised the Household survey of HIV-prevalence and behaviour in Chimanimani District, Zimbabwe which sort to quantify the magnitude of HIV and AIDS problem among children and adults; determine the knowledge ,attitudes, behaviour and practice of the general population; identify prevention and care programmes and human rights issues concerning HIV and AIDS among the general public and provide evidence-based information to policy makers on HIV and AIDS preventive mitigatory needs.
Results
The data was analysed using discriminant analysis (DA) and logistic regression (LR). The results of the DA showed that 87.4% of the cases were correctly classified as either HIV-
vii
positive or HIV-negative whilst LR managed to classify 89.1% of the same cases. LR identified 11 variables that include swollen lymph nodes, burning urine, clothes too large which were not picked by DA as significant. Comparing the results obtained from logistic regression and discriminant analysis indicate that the two techniques gave almost the same percentage of correct classification, different error rates and kappa coefficients but have a very high overall kappa coefficient of 0.98. The Logistic Regression model had positive coefficients whilst the Discriminant Analysis model had negative coefficients but the two can correctly identify the almost the same number of patients who are HIV-positive or HIV- negative. This means that the clinician can apply any model to classify a patient.
Overall HIV prevalence among the 15-49 year age group was 15.1%, 95% CI = [13.1-16.9].
Conclusions
The logistic model has proved to be an efficient tool for classifying patients as HIV-positive or negative. It has shown that it can correctly classify 89.1% of the patients who come presenting with HIV symptoms whilst the discriminant mode can correctly classify 87.4% of the patients. LR has a lower error rate (10.9) and a higher Predictive Value Positive (79.6) compared to DA’s 12.6 error rate and 30.5 Predictive Value Positive. The results of Logistic regression model were closer to those of discriminant analysis model and any model can be used. | |
