Table A1.
Parameter estimates for the cross-sectional regression of Table 1 EGFR = HIV infection, serum albumin and BUN in the Bronx WIHS using mixed models.
| Variable | Working Correlation Structure | |||||
|---|---|---|---|---|---|---|
| Independence | Equicorrelation | |||||
| Point Estimate | 95% CI 1 | Z-Value (p) 1 | Point Estimate | 95% CI 1 | Z-Value (p) 1 | |
| HIV Infection (βHIV,CS) |
−2.04 | (−5.04, −1.04) | −4.02 (<0.0001) | −3.99 | (−7.04 −0.93) | −2.55 (0.01) |
| Albumin Per g/dL (βALB,CS) |
−6.21 | (−7.30 −5.11) | −11.04 (<0.0001) | −9.89 | (−11.03, −8.73) | −16.90 (<0.0001) |
| BUN Per mg/dL (βBUN,CS) |
−1.87 | (−1.95, −1.79) | −44.68 (<0.0001) | −1.22 | (−1.30, −1.13) | −29.08 (<0.0001) |
| Akaike Information Criteria (AIC) | 99,374.5 | 94,934.5 | ||||
1 The confidence interval and p-values for independence working correlation structure in particular but also arguably for equicorrelation as well overestimate the precision of the parameter estimates. Unlike GEE, mixed models are not robust to misspecification of the correlation structure.