Modélisation de la sensibilité, spécificité et valeurs prédictives de la recherche d’une hématurie par bandelettes réactives dans le diagnostic de l’infection par Schistosoma haematobium

Since its recognition as an effective tool to identify Schistosoma haematobium infection, haematuria detection has been used in several large scale control programs. However, large variations in performance between age groups, gender, prevalence and areas were observed. Correlation between the degree of haematuria and the intensity of infection and between intensity of infection with age and/or gender could partly explained these variations. Other factors, such as the circadian cycles and day-to-day variations of haematuria and egg excretion, a concomitant pre sence of blood in urine for another reason than schistosomiasis and the intrinsic characteristics of the reagents strips also contribute to the observed variations. This study assessed the influence of age, gender and intensity of infection taken together on the performance of the diagnostic test by fitting logistic regression models. During a socio-economic study of schistosomiasis, carried out between 1989 and 1991 in a rice-growing area loca ted in Mali, West Africa, haematuria was detected by reagent strips among 2,873 six-year-old and more subjects randomly chosen from 14 villages. The "gold standard" of infection was provided by the parasite egg counts. Laboratory quality controls were implemented. Observed sensitivities, specificities, positive and negative predictive values were calculated by the standard two-by-two tables. The influence of age, gender, intensity of infection, measured in a continuous scale, on these parameters were studied by fitting two logistic models. The first one modelled sen sitivity and specificity. The second one modelled predictive values. Goodness-of-fit tests were performed. Large variations in the observed values of sensitivities, specificities and predictive values by age and gender were observed. The best fitted model for sensitivity and specificity included infection status, age, gender, intensity and an age by gender interaction. For the predictive values, the best model included haematuria, age, and an haematuria by age interaction. Given the large variations in the performances across the covariates strata defined by cross-tabulation of the age/gender and intensity categories, this modelling approach provided a more realistic appraisal of the perfor mances than using overall values obtained from a general twoby-two table. It allowed to assess the effect of the covariates of interest on the performances of the test and provided an adjustment of the performances on these covariates. The best fitted model was able to provide estimates of sensitivity for each age/gender stratum as a func tion of the intensity of infection, which could not be done by using the two-by-two table approach. A decrease in sensitivity adjusted for intensity of infection above 10 years of age among males is discussed. This modelling approach also uses all the information available and not only the data from the stratum of interest.