Appendix Table 3.
Unmeasured Variable and Covariateb | High-Risk Population | Low-Risk Population | ||||||
A = 0 | A = 1 | A = 0 | A = 1 | |||||
Y = 1 | Y = 0 | Y = 1 | Y = 0 | Y = 1 | Y = 0 | Y = 1 | Y = 0 | |
U = 0 | ||||||||
L = 1 | 260 | 1,473 | 37 | 211 | 402 | 2,276 | 57 | 326 |
L = 0 | 318 | 1,799 | 210 | 1,192 | 491 | 2,781 | 325 | 1,842 |
U = 1 | ||||||||
L = 1 | 780 | 638 | 112 | 91 | 260 | 213 | 37 | 31 |
L = 0 | 953 | 779 | 631 | 516 | 317 | 260 | 210 | 172 |
a The causal graph for this scenario is depicted in Appendix Figure 3. In the example data, the true causal effects of treatment A, and measured covariate L on the outcome Y are null.
bA is an indicator for treatment (1 if treated, 0 otherwise), L is an indicator for a measured covariate (such as CD4 count), Y is an indicator for mortality, and U is an indicator for an unmeasured covariate, such as underlying immune function.