OBJECTIVES: Appropriate design and efficient analytical strategy are generally considered as some of the prerequisites for a valid and reliable health outcomes research on non-randomized observational data. For rare outcomes, the case-control design is often presented as the most efficient, whereby, proper selection of controls is crucial. Using simulated data for the nested case-control design, we assess the relative efficiencies of two  sampling strategies for the controls- the version where controls can never be cases (design 1) and the recommended approach in which controls are sampled from the risk sets such that some controls can be future cases (design 2).   METHODS: In each simulation, we assumed an underlying hazard that follows a Weibull distribution with inputted values for the scale and shape parameters to generate 100 sets of cohorts of 4000 patients in treatment groups (i.e. treated and untreated). The process also involved an assumed hazard ratio for treatment and 3 factors that required adjustments. Designs 1 and 2 were then applied successively on each of the resulting datasets and then analysed to obtained for each design, the estimated odds ratio (OR- an approximate of the inputted hazard ratio) and its 1st and 3rd quartiles.RESULTS: We considered over 50 scenarios for hazard ratio that varied between 0.3 and 4.0. The absolute differences between the inputted hazard ratio and the estimated odd ratio ranged from 0.01-8.00 and from 0.01-0.50 for designs 1 and 2 respectively. The inputted hazard ratio was within the inter-quartile range of the OR in less than 5% of the runs with design 1 but more than 80% with design 2. CONCLUSIONS: Our study suggests that in nested case-control designed studies, if controls are not sampled from the appropriate risk sets, we can expect much larger bias in our estimates than with correct sampling