Collin

SMC coverage was calculated among the study population by area type: urban, peri-urban, and rural. A multiple logistic model was fitted for the two binary outcomes and associations between these two outcomes, and the set of potential confounding variables (previously described) were investigated. Only postintervention data were used. For the first outcome (receiving the visit of a CHW), the level of analysis was the household, whereas for the second (being administered a course of SP+3AQ), it was the child. In the latter case, a random intercept for households was included in the model to account for the hierarchical structure of the data—children are nested in households. MultiCollinearity was ruled out by using the Collin package (STATA Corp.) and verifying that variance inflation factors did not exceed 4. The statistical significance of the Wald χ² test was checked to assess goodness-of-fit. To ensure consistency between the analyses and comparability of the results, the same set of covariates was included in the different models, whatever the outcome or the type of analysis (impact evaluation, sensitivity analysis, and coverage assessment). Therefore, models are not as parsimonious as they could be, but their interpretation is more straightforward and coherent.

The descriptive analysis of qualitative and quantitative variables, was followed by a bivariate analysis to identify significant variables associated with penile nodules. The variables were included in a multivariate logistic regression model to identify independent diagnoses associated with penile nodules. The modelling strategy was exploratory and the diagnostic variables were retained in the model except obsessive compulsive disorder, anorexia and bulimia which were too rare to be included in a multivariate model (15, 5 and 3 persons, respectively). Colinearity was ruled out using the collin package (STATA, College Station, Texas) and ensuring that variance inflation factors were <4. The Hosmer-Lemeshow goodness-of-fit test was used to test the model.
A similar procedure was used for bivariate and multivariate analyses of explanatory variables associated with repeated incarcerations.
Principal component analysis was performed and a loading plot of the different diagnoses was performed in order to look for potential clusters of variables.
Stata13 (College Station, Texas, USA) was used for the analysis.