Table 3.
For selected endpoints in the SEARCH Study, point estimates, 95% confidence intervals, and efficiency comparisons when estimating the intervention effect in Stage 2 with the unadjusted estimator and with TMLE using Adaptive Prespecification. All approaches adjusted for individual-level missingness in Stage 1.
| Breaking the matches | Keeping the matches | ||||
|---|---|---|---|---|---|
| Stage 2 | Effect (95% CI) | Efficiency | Effect (95% CI) | Efficiency | |
| HIV incidence | Unadjusted | 0.98 (0.66, 1.45) | 1 | 0.98 (0.78, 1.24) | 3.1 |
| TMLE | 0.96 (0.73, 1.26) | 2.1 | 0.96 (0.8, 1.17) | 4.6 | |
| TB incidence | Unadjusted | 0.79 (0.64, 0.98) | 1 | 0.79 (0.69, 0.92) | 2.2 |
| TMLE | 0.8 (0.67, 0.95) | 1.4 | 0.8 (0.69, 0.91) | 2.6 | |
| Hypertension control | Unadjusted | 1.19 (1.1, 1.3) | 1 | 1.19 (1.11, 1.28) | 1.7 |
| TMLE | 1.18 (1.1, 1.26) | 1.6 | 1.19 (1.11, 1.27) | 1.8 | |
| Viral suppression | Unadjusted | 1.15 (1.11, 1.2) | 1 | 1.15 (1.11, 1.2) | 1 |
| TMLE | 1.16 (1.13, 1.2) | 1.1 | 1.15 (1.11, 1.2) | 1 | |
Efficiency: Variance estimate for the unadjusted effect estimator breaking the matches used for randomization, divided by the variance estimate of another approach (e.g., TMLE with Adaptive Prespecification, keeping the matches used for randomization).