TABLE 2.
Cox proportional hazards models for covariates associated with cytomegalovirus (CMV) disease and covariates associated with death in children at high risk for CMV disease and/or those with CMV disease and death
| Covariates associated with CMV disease | Covariates associated with death in children that were at high risk for CMV disease and/or who developed CMV disease | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| CMV disease (n = 30, %) | No CMV disease (n = 119, %) | Univariate (HR, 95% CI) | Model A (aHR, 95% CI) | Model B (aHR, 95% CI) | Died (n = 29, %) | Survived (n = 50, %) | Univariate (HR, 95% CI) | Model C (aHR, 95% CI) | Model D (aHR, 95% CI) | Model E (aHR, 95% CI) | |
| Cyclosporin A | 0 (0.0) | 2 (1.7) | No estimate | – | – | 1 (3.4) | 0 (0.0) | 7.80 (0.97, 62.5) | 13.2 (1.25, 140.0) | – | – |
| PTLD | 11 (36.7) | 5 (4.2) | 5.43 (2.54, 11.6) | 4.55 (1.90, 10.9) | 6.06 (2.61, 14.1) | 3 (10.3) | 12 (24.0) | 0.53 (0.16, 1.77) | – | – | – |
| Acyclovir prophylaxis | 11 (36.7) | 24 (20.2) | 3.08 (1.41, 6.70) | 3.37 (1.33, 8.56) | 4.86 (2.06, 11.5) | 2 (6.9) | 11 (22.0) | 0.58 (0.13, 2.45) | – | – | – |
| Biliary atresia | 22 (73.3) | 58 (48.7) | 2.54 (1.13, 5.73) | 1.87 (0.72, 4.88) | – | 19 (65.5) | 36 (72.0) | 1.18 (0.53, 2.65) | – | – | – |
| Under 18 months | 13 (43.3) | 31 (26.1) | 2.26 (1.08, 4.72) | 1.85 (0.75, 4.53) | 2.24 (0.96, 5.24) | 20 (69.0) | 24 (48.0) | 2.37 (1.04, 5.41) | 1.99 (0.62, 6.39) | – | – |
| Care in private | 17 (56.7) | 71 (59.7) | 0.43 (0.20, 0.91) | 0.36 (0.16, 0.80) | 0.37 (0.17, 0.83) | 18 (62.1) | 32 (64.0) | 0.43 (0.20, 0.94) | 0.33 (0.12, 0.93) | 0.29 (0.11, 0.81) | 0.28 (0.10, 0.80) |
| Azathioprine | 3 (10.0) | 13 (10.9) | 1.82 (0.55, 6.09) | – | 3.51 (0.98, 12.6) | 2 (6.9) | 4 (8.0) | 3.67 (0.80, 16.8) | 3.21 (0.64, 16.0) | – | 4.87 (0.90, 26.4) |
| High risk for CMV disease | 17 (56.7) | 48 (40.3) | 1.73 (0.84, 3.57) | – | – | 28 (96.6) | 38 (76.0) | 3.98 (0.54, 29.3) | – | 4.69 (0.59, 37.3) | – |
| Black | 18 (60.0) | 83 (69.7) | 0.94 (0.45, 1.98) | – | – | 20 (69.0) | 28 (56.0) | 1.90 (0.86, 4.20) | – | – | 2.86 (0.58, 14.1) |
| CMV prophylaxis | 8 (26.7) | 54 (45.4) | 0.46 (0.20, 1.09) | 0.45 (0.18, 1.15) | – | 13 (44.8) | 28 (56.0) | 0.66 (0.31, 1.41) | – | – | 1.15 (0.40, 3.28) |
| Valganciclovir prophylaxis | 10 (33.3) | 54 (45.4) | 0.57 (0.26, 1.25) | – | – | 11 (37.9) | 30 (60.0) | 0.49 (0.23, 1.07) | 0.55 (0.20, 1.49) | – | – |
| EBV infection | 26 (86.7) | 95/114 (83.3) | 1.11 (0.38, 3.18) | – | – | 18/23 (78.3) | 45 (90.0) | 0.73 (0.27, 1.97) | – | 0.36 (0.11, 1.16) | 0.35 (0.10, 1.29) |
| Recipient CMV positive | 13 (43.3) | 66 (55.5) | 0.71 (0.34, 1.46) | – | – | 1 (3.4) | 12 (24.0) | 0.25 (0.03, 1.85) | – | – | 0.18 (0.02, 1.49) |
Note: Statistically significant associations are shown in bold font. Models A and B reflect the two best‐performing models in the Cox proportional hazards analysis of covariates that were associated with CMV disease: Model A included all covariates which in the univariate analysis had a p‐value of <.1, and which did not violate the proportional hazards assumption in the multivariate model; Model B was a stepwise proportional Cox regression model inclusive of covariates which did not violate the proportional hazards assumption. In the model comparison, Model A had an Akaike information criterion (AIC) of 225.6 and a corrected AIC weight (AICcWt) of 40.2%, whereas Model B had an AIC of 224.8 and an AICcWt of 59.8%, reflecting that Model B encapsulated the available data most efficiently. Models C, D, and E were the three best‐performing models in the Cox proportional hazards analysis of covariates associated with death in children at high risk of CMV disease or with CMV disease: Model C included all covariates that had a p‐value of <.1 in the univariate analysis and did not violate the proportional hazards assumption in the multivariate model; Model D was a stepwise proportional Cox regression model inclusive of covariates that did not violate the proportional hazards assumption; Model E was derived through least absolute shrinkage and selection operator (LASSO) regression and included only covariates that did not violate the proportional hazards assumption. In the model comparison, Model C had an AIC of 346.8 and an AICcWt of 37.2%, Model D had an AIC of 346.5 and an AICcWt of 43.8%, and Model E had an AIC of 348.5 and an AICcWt of 16.2%.
Abbreviations: aHR, adjusted hazard ratio; CI, confidence interval; EBV, Epstein‒Barr virus; HR, hazard ratio; PTLD, post‐transplant lymphoproliferative disease.