Type of study. A mixed ecological study; ecological because it had the municipality as the unit of analysis, and mixed as it had, simultaneously, an exploratory character as well as a group comparison design .
Ethics: No personally identifiable information was recorded all information obtained from health surveillance systems were kept confidential. Consent was not required because this was a study that used secondary sources of information already published. This study was approved by the Institutional Review Board of University of Antioquia (2015-4690).
Case definition. We included all consecutive patients with self-poisoning by paraquat, admitted in all hospital centers of the Department of Antioquia and reported by e-Formulary or in a physical paper to the Regional Epidemiological Surveillance System (SIVIGILA by its acronyms in Spanish) from January 1st, 2010 to the December 31st, 2016. This report is mandatory for all health care providers and made by personnel trained from each hospital. Antioquia is one of the 32 departments of Colombia, located in the central northwestern part of Colombia with a population of 6,613,118 . This department can be subdivided into six regions: Magdalena Medio, Bajo Cauca, Urabá, Nordeste, Occidente, Norte, Oriente, Suroeste and Valle de Aburra.
Population. The number of citizens in each region was taken from the official census established by the National Administrative Department for Statistics in Colombia [15, 21].
Variables: The incidence rate was taken as the study's dependent variable and was calculated for those regions corresponding to report to the SIVIGILA. The independent variables examined by region were the following: (i) amount of area planted with permanent crops, (ii) quality of life-index calculated by National Administrative Department for Statistics , which characterizes the living conditions of Colombians by related to housing (i.e, materials used in walls, floors and presence of public services), people (i.e., education, health, child care, labor force, expenses and income, etc.), and households (i.e., possession of assets and perception of the boss or spouse on living conditions in the home), (iii) net enrollment rate, that is the relation between the students enrolled that are in the appropriate age range for each educational level and the population in age of such school, (iv) the poverty level, indicates the percentage of people whose income is below the poverty line meaning that cannot satisfy their most basic needs, (v) drinking water coverage, (vi) private health insurance, (vii) heavy metal poisoning rate, (viii) Gini index distribution (20), that indicates the concentration of land ownership; and (ix) rate per 100 000 inhabitants of cases reported to SIVIGILA with psychological violence, physical violence, deprivation and negligence or suicide attempt .
These variables were dichotomized using the following cut-offs: (i) planted area of 10%, (b) the quality of life variable of 60% (c) the poverty level of 30%, (d) drinking water coverage of 40%, (e) enrolment rate of 70%, (f) the Gini index of land of 0.7, (g) deprivation and negligence of 8.70, (h) the heavy metal poisoning rate of 0.3 and (i) private health insurance of 40%. These cut-off points were used as suggested by the department of national economic planning.
Database and analysis: Microsoft Excel 2016 was used to construct the database. Incidence rate ratios (IRR) were calculated using Poisson's regression. We obtain at the incidence rate ratio by exponentiating the Poisson regression coefficient. The Poisson regression coefficients are the difference between the logs of expected counts to incidence rate ratio. Modeling and variable selection in the multivariate analysis were made following Greenland's recommendations , We use The method of forward selection as follows : 1. Begin with no terms in the model. 2. Find the term that, when added to the model, achieves the largest value of R-squared. Enter this term into the model. 3. Continue adding terms until a preset limit on the maximum number of terms in the model is reached. All statistical analysis was done using Stata 11.0®.