With active ingredients obtained for each commercial product and their respective toxicity ratings, the amounts of each pesticide type used were calculated in kilograms (kg) per hectare (ha). In view of the asymmetry encountered in distributions, the quantities of Pesticides Ib and II were added

and the sum was A66 classified based on the first (0) and sixth deciles (1.1) as follows: 0=0 kg/ha; 1≥0 but ≤1.1 kg/ha; and 2≥1.1 kg/ha. Covariates included were age and education (number of years of formal education) because previous studies have demonstrated their independent contribution to neurocognitive performance.31 Inferential analyses Significance was set at 10%. Loss to follow-up was analysed using either t tests or χ2 tests. Multivariable regression analyses were performed using generalised estimating equations,32 33 thus allowing effective estimates

of parameters with correlated data. Longitudinal associations involved use of the variable time in the equation,32 dichotomised as 0=T1 and 1=T2. Confounding was assessed, the criterion being a change in the value of the coefficients >10% with removal or addition of covariates. To test the study hypothesis, three product terms (dummy variable) were created,34 based both on the existing literature18 22 and on preliminary Spearman correlation analyses: Term 1=Use of IPM practices good/very good × organisational participation; Term 2=Use of IPM practices good/very good × use of Pesticides Ib and II>1.1 kg/ha; and Term 3=Use of Pesticides Ib and II>1.1 kg/ha × organisational participation. We began multivariable modelling with a saturated model (A) that included all study variables and product terms. In the later models (B, C and D), one product term was excluded at each stage. This technique was chosen to value the joint importance of the terms and later their individual significance (p<0.1) within the model.35 The quasi-information criterion ‘QICu’ was one criterion for model selection,32 aiming

for the lowest value along with parsimony and consistency with the prior literature.36 37 To confirm and interpret effect modification, multivariable analyses were stratified by organisational participation. Results Descriptive analysis Organisational participation was similar across times (table 1); however, a Batimastat variation was found in the type of organisation in which farmers participated (data not shown). Increases were observed in the proportion of individuals who participated in commercial potato production organisations (T1 31% to T2 35%) and in other types of organisations (13% to 23%), with decreases in conflict-resolution organisations (56% to 42%). In the last two categories, this change was statistically significant (p<0.001). Table 1 Descriptive characteristics of the study population The mean neurocognitive performance score was 4.4 at T1 (SD 1.4).