Abstract

Background: Helminths infections (i.e., S. haematobium, S. mansoni and Hook worm) affect more than a quarter of world’s population, with consequences for nutritional and educational development of infected individuals. They are a common cause for morbidity, especially among children in undeveloped countries. Control strategies require better understanding of helminths epidemiology. As such, appropriate statistical methods to model infection prevalence and intensity are crucial. This study aimed at joint modeling of helminths infection prevalence and intensity using robust methods in order to understand the epidemiology. Methods: Zero altered models were fitted and applied to two datasets (one from Malawi and another from Zambia). Malawi data were collected in a cluster ran domized study, in Chikhwawa district in 2004, with 18 villages randomised to intervention and control arms. Zambia data were collected from school children in a cross-sectional study in Lusaka province in 2004. A range of Zero Inflated (ZI)models (ZI-Poisson [ZIP] and ZI-Negative Binomial [ZINB]) and Hurdle mod els (Poisson Logit Hurdle [PLH] and Negative Binomial Logit Hurdle [NBLH]) were developed for infection analysis, adjusted for age, sex, education level, treatment arm, occupation, and polyparasitism, among others. Model estimation was based on maximum likelihood estimation (MLE) and model selection was based on Akaike Information Criteria (AIC). Exponential and Spherical variogram models were used to estimate residual spatial effects. Results: Chikhwawa study had a total of 1, 642 participants. Overall, 55.4 % were female and mean age of study population was 32.4 with SD = 22.8. Prevalence was as follows: S. haematobium = 19.4 %, S. mansoni = 5.0 % and Hookworm = 22.9 %. Schistosomiasis and Hookworm infections were highly aggregated in a vi relatively small and heavily infected population proportion. A large proportion of individuals were non-egg excretors (S. haematobium = 85.8 %, S. mansoni = 95.7 % and Hookworm = 80 %) for outcomes of interest hence data had a large number of zeros. Data showed overdispersion evidence with p-value <0.0001. NBLH model offered the best fit to data with lowest AIC = 3, 482. Schistosomiasis infection was associated with age (RR = 0.96); with the highest intensity in school-age children. Fishing (RR = 0.73) and working in gardens (RR = 1.21) along the Shire River were also clear risk factors. Hookworm showed a high intensity in older people than in younger children and also high intensity in males than in females (RR =1.19). Intervention reduced both infection intensity and prevalence in intervention arm as compared to control arm. Residual spatial effects showed some degree of spatial dependence across the study area. The Zambia study had a total of 2040 participants. Overall, 50.4 % were female and mean age of study population was 9.98 with SD = 2.14. Urinary Schistoso miasis prevalence was 9.6 %. NBLH model offered the best fit with lowest AIC = 3, 230. Schistosomiasis prevalence was associated with age (AOR = 0.69), sex(AOR = 1.17), altitude (0.37), NDVI (AOR = 1.04) and temperature (AOR =0.99). Infection intensity was associated with age (RR = 0.55), sex (RR = 1.28), altitude (RR = 0.11), temperature (RR = 0.84), and NDVI (RR = 1.07). Conclusion: Helminths were highly localized, with small section of people harboring parasites; showing heterogeneous infection risk for both Malawi and Zambia settings. Joint modeling approach allowed identification of risk factors for infection presence and severity hence provide a platform to design combative control efforts. NBLH offered best-fit to data with capability to handle overdispersion, excess zeros and capture true zeros in the data. Its implementation and interpretation, ease of components, and its direct link with observed data make it a valuable alternative for analysing zero inflated count data.

More details

School	: School of Natural and Applied Sciences
Issued Date	: 2012

Download full document