Figure 5. Timing of deaths during the first 6 d.

For the purposes of survival analysis, the clock has been started at randomisation and stops either at death or discharge (assumed to be 40 d if discharged alive sooner than 40 d).