Infection and morbidity due to schistosomiasis: Determinants and impact of treatment

Abstract

Schistosomiasis is one of the most widespread parasitic infections of man, and is second only to malaria in socioeconomic and public health importance. The present study examined the role of differences in exposure patterns to canal water among ages and genders in explaining the observed variation in infection rates among them in an endemic community. The study further assessed the burden of morbidity and abnormal laboratory findings due to schistosomiasis and the impact of treatment on this morbidity in the same community. Predictive models for infection were built using logistic regression and the emerging methodology of Neural Networks (NNs) and their performance was compared. Information collected on 9093 participants for two years included: an interview for demographic variables and risky behavior and urine examination. A subsample of the participants, 1882, had undergone clinical and ultrasound examinations twice twelve months apart. When other confounding variables were controlled, variation in the exposure to canal water had a small impact on the odds of prevalent or incident infection of males relative to females, and of the {dollar}\le{dollar}20 year age group relative to the older age group. Hematuria and proteinuria were significantly positively associated with infection. Many other plausible lesions were also positively associated with infection although the associations were not statistically significant. Nevertheless, the criteria set forth to select suitable ultrasonographic lesion(s) to be assessed in the monitoring of control programs for morbidity due to schistosomiasis were not met by any of the lesions. An important finding is that the predictive performance of the new NNs model compared favorably with that of the logistic regression model when the same variables were used. Moreover, when all the variables in the data set were used, the performance of the NNs based model improved substantially, while many of these variables could not be included in a logistic regression for technical reasons. The epidemiological implications of these findings in the design of future studies of schistosomiasis and planning of control strategies are discussed.