FIGURE 4.

Between‐group differences in outcomes according to various baseline characteristics. Differences in days alive without life support (DAWOLS) and mortality at Day 90 with 95% confidence intervals according to the variables assessed at baseline (including predicted risks of mortality at Day 90 using the internal prediction model) with 95% confidence intervals. Values are presented as the treatment effects of 12‐mg dexamethasone, that is, positive differences indicate higher values in the 12‐mg group and vice versa. For both outcomes, predicted values and 95% confidence intervals are truncated at the lowest/highest possible values (0/90 days and 0/100%, respectively). The p and S values from the likelihood ratio tests assessing evidence in favour of heterogeneous treatment effects are displayed below each plot. For the continuous variables, predicted differences are only displayed for the central 90% of values in the data due to the large uncertainty at the extreme values with limited data. Figure S5 displays predicted differences across all observed values in the datasets as specified in the statistical analysis plan. ¹⁵