Introduction There are many sample designs for case-control studies. Three of them were simulated to investigate the properties of their risk estimators when the aim of the study is to analyse the space along with other covariates. They are: the case-base sampling, where all controls are selected at the beginning while the cases are sampled during the study as they occurs; the survivor sampling, in which both cases and controls are sampled at the end of the study; and the risk-set sampling, where both cases and controls are sampled during the study.

Methods A realistic at risk population was created by sampling individuals from the empirical spatial distributions derived from governmental census information of a Brazilian city. Two epidemic scenarios were built, a transmissible and a nontransmissible disease. We used the generalised additive models to estimate the risks in each different study, fitting semiparametric models with the geographical coordinates and other covariates as age, income, gender and study.

Results The results suggest that the estimated spatial risks are similar in the three sample designs, but the standard deviations vary in the space and, the widest variation occurs in the survivor sampling (for the nontransmissible disease) and in the case-base sampling (for the transmissible disease). The parametric estimates that are closest to the initially defined were attained by the risk-set sampling, at the nontransmissible disease scheme.

Conclusion We conclude that the best risk estimates are attained by sampling the controls at the same time of the cases, as the epidemic occurs.