Table 2.

Bias in the average estimated RR using three control sampling methods

True RRProportion exposedAverage RR
Full cohort designCase-control designs
Method 1: incidence density sampling without replacementMethod 2: incidence density sampling with replacementMethod 3: pure control sampling
1.0010%0.991.001.000.99
30%1.001.001.001.00
50%1.001.001.001.00
1.5010%1.501.511.511.63
30%1.501.501.491.62
50%1.501.501.491.62
2.0010%2.002.022.012.34
30%2.002.001.982.33
50%2.002.001.982.34
  • NOTE: For a variant prevalence of 10%, the average simulated number of cases when the true RR was 1.0, 1.5, or 2.0 was 339.0 ± 16.5, 353.3 ± 16.4, and 365.5 ± 15.9, respectively. Case-control ratios for the simulated full cohorts when the true RRs were 1.0, 1.5, and 2.0 were 1:4.9, 1:4.7, and 1:4.5. For a variant prevalence of 30%, the average simulated number of cases when the true RRs were 1.0, 1.5, and 2.0 was 339.5 ± 16.4, 381.6 ± 16.4, and 417.8 ± 17.2, respectively. Case-control ratios for the simulated full cohorts when the true RRs were 1.0, 1.5, and 2.0 were 1:4.9, 1:4.2, and 1:3.8. For a variant prevalence of 50%, the average simulated number of cases when the true RRs were 1.0, 1.5, and 2.0 was 339.8 ± 16.3, 408.7 ± 17.1, and 469.2 ± 18.2, respectively. Case-control ratios for the simulated full cohorts when the true RRs were 1.0, 1.5, and 2.0 were 1:4.9, 1:3.9, and 1:3.3.