Multivariate regression models were applied to assess the association between the health facility specific 5-year notification rate and health facility characteristics. Because of the (bimodal) distribution characteristics of the 5-year TB notification rates (see Fig. 1), different multivariate regression models were applied. First, a backward stepwise logistic regression model was built in which the dependent variable was either 1 (any TB cases reported between 2009 and 2013) or 0 (no TB cases at all between 2009 and 2013) to account for the 42 % of health facilities reporting no cases between 2009 and 2013. All variables for which the p-value was less than 0.2 were kept in the model. To assess the associations between notification rates and facility characteristics in those facilities from which cases were reported, the TB notification rate was first log-transformed and this transformed notification rate was analyzed using simple linear regression models. Here, backward stepwise linear regression was applied leaving all variables for which the p-value was less than 0.2 in the model. Time trends were calculated in- and excluding health facilities that did not provide data for all of the years 2009-2013. A p-value smaller than 0.05 was regarded as statistically significant.