Table 4.
Median posterior hazard ratio by agent, modified intention-to-treat.
| Outcome | Arm | Median posterior hazard ratio (95% CrI) |
|---|---|---|
| Recovery (HR > 1 is better) | Apremilast | 0.78 (0.52, 1.14) |
| Cenicriviroc | 0.88 (0.63, 1.23) | |
| Icatibant | 0.85 (0.60, 1.17) | |
| IC14 | 0.63 (0.40, 0.96) | |
| Celecoxib/famotidine | 0.50 (0.28, 0.90) | |
| Dornase alfa | 0.76 (0.48, 1.19) | |
| Death (HR < 1 is better) | Apremilast | 1.05 (0.66, 1.71) |
| Cenicriviroc | 1.24 (0.84, 1.83) | |
| Icatibant | 1.06 (0.71, 1.59) | |
| IC14 | 0.86 (0.48, 1.51) | |
| Celecoxib/famotidine | 1.67 (0.79, 3.58) | |
| Dornase alfa | 1.09 (0.61, 1.92) |
The posterior distributions of the cause-specific hazard ratios for recovery were computed using Bayesian proportional-hazard Weibull regressions. The cause-specific hazards were modeled as a function of study arm, adjusting for baseline COVID severity. Similarly, Bayesian proportional-hazard Weibull regressions were used to model the hazard functions for all-cause mortality. In these analyses, follow-up times were censored at 60 days if patients were still at risk for either outcome. Weakly informative priors were used for all models' parameters; see SAP for details. The table shows the medians of the posterior hazard ratio distributions for investigational agents compared with concurrently randomised controls, and 95% quantile-based credible intervals.