Table 1 Data from a hypothetical cohort study with multifactorial selection in exposure (E) originating from C (known source) and U (unknown source; Fig. 2C). The relative effect of E on the disease outcome D is constant across strata of C and U (RRED = 2.0). U influences the effect of C on D (RRCD = 1.5 when U = 0 but 4.0 when U = 1). Lack of adjustment for U therefore leads to bias in the estimated exposure effect also when C is adjusted for. Stratification for C may lead to false conclusions regarding the heterogeneity in the exposure effect
From: A novel framework for classification of selection processes in epidemiological research
U | C | E | D = 1, n | D = 0, n | Risk | RRED1 | RRCD2 |
|---|---|---|---|---|---|---|---|
0 | 0 | E = 1 | 20 | 80 | 20% | ||
E = 0 | 90 | 810 | 10% | 2.0 | |||
1 | E = 1 | 90 | 210 | 30% | 1.5 | ||
E = 0 | 105 | 595 | 15% | 2.0 | 1.5 | ||
1 | 0 | E = 1 | 40 | 160 | 20% | ||
E = 0 | 80 | 720 | 10% | 2.0 | |||
1 | E = 1 | 400 | 100 | 80% | 4.0 | ||
E = 0 | 200 | 300 | 40% | 2.0 | 4.0 | ||
Totals | E = 1 | 550 | 550 | 50% | |||
E = 0 | 475 | 2425 | 16% | 2.32 | |||
0 | E = 1 | 60 | 240 | 20% | |||
E = 0 | 170 | 1530 | 10% | 2.0 | |||
1 | E = 1 | 490 | 310 | 61% | |||
E = 0 | 305 | 895 | 25% | 2.41 |