Table 4.
The association between two alleles and HPV18HSIL and their interactions with HIV status using different methods for the example data set with 512 HIV-positive and 285 HIV-negative women with a total of 3682 visits and 2400 visits, respectively with complete information on HPV18HSIL.
| Logistic regression GEE | Linear regression GEE | Proposed two-step method |
|||||
|---|---|---|---|---|---|---|---|
| Allele | Stratum |
ÔR (or OR1/OR0) & 95% CIa |
p | β̂1 (or β̂3) and 95% CI | p | β̂1 (or β̂3) | p |
| DRB1501 | HIV− | – | – | 0.0067 (−0.0061, 0.0194) | 0.3042 | 0.0067 | 0.0465 |
| HIV+ | – | – | −0.0036 (−0.0075, 0.0003) | 0.0681 | −0.0036 | 0.8820 | |
| Interaction | – | – | −0.0103 (−0.0241, 0.0035) | 0.1445 | −0.0103 | 0.0710 | |
| BW4 | HIV− | – | – | −0.0015 (−.0044, 0.0014) | 0.3143 | −0.0015 | 0.0245 |
| HIV+ | 1.771 (0.3277, 9.574) | 0.5067 | 0.0025 (−0.0043, 0.0093) | 0.4655 | 0.0025 | 0.6800 | |
| Interaction | – | – | 0.0040 (−0.0034, 0.0114) | 0.2846 | 0.0040 | 0.4460 | |
“–” indicates model was not converged.
a
We provided confidence interval (CI) estimates for the parameter of interest when the GEE models were converged. However, one should be cautious in interpreting these CIs because the asymptotical consistency and normality may not be valid under rare events even if the models are converged, as we demonstrated in the simulations.