All statistical analyses were weighted. The weights used considered the sampling weight and were adjusted for non-response. Non-response was corrected by reweighting using the equal-quantile score method [26 ], based on socio-demographic and geographic data and the professional activity sector available in the sampling frame for all individuals. A calibration by raking ratio method was then applied, using the distributions by sex, age group, and professional activity sector in the target population, using the SAS macro Calmar [27 (link)].
We described population demographic characteristics with absolute and relative frequencies for categorical variables and with mean and standard deviation for continuous variables. We explored factors associated with hantavirus seropositivity by estimating seroprevalence ratios and 95% confidence intervals (CI) using weighted Poisson regression models with robust standard errors. The clustering of observations by geographic area was accounted for using a fixed area effect in all models. The outcome of interest was hantavirus seropositivity, and we used one model per factor of interest (the main exposure variable). Causal diagrams were made between hantavirus seropositivity and each factor of interest using DAGitty ([28 ,29 (link)]; diagrams not shown). Based on those causal diagrams, it was deemed reasonable to use unconditional models for the geographic area, age, and main profession (i.e., no important confounders were identified for those factors). Also, we identified age as a confounder of the association between seniority and hantavirus seropositivity; and main profession as a confounder of the association between average weekly exposure time in forest and hantavirus seropositivity. Nevertheless, the strong collinearity between age and seniority (Pearson correlation coefficient of 0.76), precluded a multivariable analysis testing both factors. Analyses were performed using Stata 14.
We described population demographic characteristics with absolute and relative frequencies for categorical variables and with mean and standard deviation for continuous variables. We explored factors associated with hantavirus seropositivity by estimating seroprevalence ratios and 95% confidence intervals (CI) using weighted Poisson regression models with robust standard errors. The clustering of observations by geographic area was accounted for using a fixed area effect in all models. The outcome of interest was hantavirus seropositivity, and we used one model per factor of interest (the main exposure variable). Causal diagrams were made between hantavirus seropositivity and each factor of interest using DAGitty ([28 ,29 (link)]; diagrams not shown). Based on those causal diagrams, it was deemed reasonable to use unconditional models for the geographic area, age, and main profession (i.e., no important confounders were identified for those factors). Also, we identified age as a confounder of the association between seniority and hantavirus seropositivity; and main profession as a confounder of the association between average weekly exposure time in forest and hantavirus seropositivity. Nevertheless, the strong collinearity between age and seniority (Pearson correlation coefficient of 0.76), precluded a multivariable analysis testing both factors. Analyses were performed using Stata 14.
Free full text: Click here
