Thanks for the invitation to discuss our work. I have no conflicts of interest to declare. Mortality in critical care trials, given the nature of critical illness, mortality is the principal endpoint in randomized trials. It's common, it's patient-centered, it's objective, so less susceptible to ascertainment bias, and represents a net benefit of life-sustaining therapy. The most commonly used mortality is 28-day mortality, and it's sort of the basis of regulatory or FDA approval. And it seems to have arisen from high-profile randomized trials conducted in the late 1980s, early 1990s, where improvements in 14-day mortality were no longer evident for 28 days, so they extended it to that. The actual time points are arbitrary, but typically they're chosen in weekly or monthly increments. So commonly used time points can loosely be grouped into sort of short-term, 7, 14, or 28-day or 30-day mortality, medium-term, 60- or 90-day, longer-term, 180-day, one year, or even longer. Location-based measures are also common, particularly ICU discharge mortality and hospital discharge mortality. The time-based mortalities may have higher losses to follow-up, because if patients are discharged prior to their time point, either due to a lack of wish or need to return for follow-up care, or inadequate contact information, or death in another jurisdiction. Location-based mortalities have a higher follow-up, but are a little bit dependent on care processes. So for instance, if you have a change of care to palliative while in the ICU, but is discharged from the ICU prior to death, you're classified as an ICU survivor, or patients transferred to long-term weaning centers, which are more common in the United States, classified as hospital survivors despite still being on mechanical ventilation. So both have some advantages and disadvantages. The natures of deaths may also vary by time point. So early deaths may be more directly related to illness or conditions studied in the trial. Later, sort of medium-term deaths may reflect more the withdrawal of life supports due to an inability to wean from life supports, or there are complications for prolonged ICU stay or immobility. And patients typically value sort of sustained long-term mortality improvements more than transient short-term improvements. Regulators seek mortality measures where they can be confident that differences have clearly stabilized, while trialists and funders seek mortality measures that minimize losses to follow-up and maximize differences between intervention and control groups to maximize trial validity and minimize costs. So all of them have sort of advantages and disadvantages. And survival curves can have variable mortality patterns. Some survival curves seem to sort of stabilize. Others seem to sort of continue to diverge. Others sort of diverge and then reconverge. Or others can still cross sort of multiple times. So the objectives of our study was to sort of assess the proportion of deaths occurring at the different time points and try to get a better handle of these. Assess when mortality differences appear to stabilize and determine whether these relationships differ by disease condition, either sepsis versus ARDS or other conditions. So what we did is we conducted a systematic review and we included all randomized control trials in critical care patients that reported more than one mortality outcome among the following mortalities. ICU discharge, hospital discharge, or any of these time points, 7, 14, 28, 30, 60, 90, 180 day or one year mortality. And they could be reported either explicitly or extracted from survival curves. And we only took trials where there was at least a minimum of 100 enrolled patients. We collected these over almost a 30 year period from 1990 to 2018 and included the critical care trials in five general medical journals as well as seven critical care specialty journals. By including trials reporting mortality at more than one time point, each trial sort of acts as its own control to calculate paired differences within each trial. So that allows us to compare numbers of deaths at different time points. And it also changes the mortality risk difference between the time points, allowing calculation with incremental risk difference. And that sort of represents the survival curve divergence between time points to identify when further separation of survival curves is no longer detectable. And then these incremental risk differences among the trials were then pooled using random effects meta-analysis to determine whether the pooled incremental risk difference, to determine pooled incremental risk differences. So we, in terms of our results, we reviewed over two and a half thousand citations and identified 343 RCTs over those 28 year period that enrolled over 228,000 patients. The median enrolled patients was 324. About a third of them were sepsis trials. 13% were ARDS and the majority were sort of other variable interventions. Mortality was the primary outcome in 50% of the trials and was most commonly measured at 28 or 30 days in 78% of the trials. So the risk of bias was low. So more than 90% of the trials had allocation concealment, intention to treat analysis, recumbent of the plant sample size and low losses to follow up. And about half of the interventions were blinded. So if you look at the number of trials and duration of follow up has increased quite a bit over time. So in each successive five year time point, the total number of trials increased. The last time point's only three years. And follow up also increased. So more trials followed patients for 60 to 90 days or more than 90 days. The follow up by time point was quite high. So it was the median follow up was 100% including the 25% lower quartile. Although it dropped off a little bit at the later time points. But it was fairly high for the trials included. The mortality in these trials increased quite rapidly. So compared to 108 80 day deaths, a third of the deaths occurred by seven days and about half of them by 14 days. And this shows the number of deaths relative to ICU mortality. It turns out that ICU mortality was similar to about 28 30 day mortality. When you compare the trials to each other. And similarly, hospital mortality was similar to 60 day mortality. Now if you looked at the incremental mortality, so the survival curves continued to diverge until at least day 60. So this shows the divergence of the survival curves between each time point. So the first time point between zero and seven days is about two and a half percent. The survival curves diverged and then went decreased around 1% between day seven and 14 and 14 and 30 days. And between 30 and 60 was less than the percent and then between 60 and 90, it was small and no longer statistically significant. So this is just a tabular form what was shown in the graph before. So between zero and seven days, the survival curves diverged by two and a half percent. And then it went down to 1%. And then, so the point here is that we could detect survival curve divergence until at least day 60. And even between day 60 and day 90, there was a further divergence, but it no longer reached statistical significance. Now, we anticipated initially that relatively few RCTs would have a P of less than 0.05. So decided a priori to include RCTs with P of less than 0.2 at any time point as our primary analysis. However, we did sensitivity analysis where we looked at RCTs with a P of less than 0.05. And that's sort of the rightward column. The divergence of the mortality curves is roughly the same pattern, but the numbers are bigger. So between zero and seven days, instead of being 2.4%, it's 3.8%. But again, you could detect that until day 60 and possibly until day 90. On the left column is all trials. Because that includes the number of trials where there was no differences, the effects are diluted. And so the differences are smaller and can only be detected up to day 30, but generally the same pattern. Furthermore, when we looked between the different disease groups, and this may be hard to see, so we looked at between sepsis trials, ARDS trials, and non-sepsis, non-ARDS trials, the differences were more or less the same overall and between each one of those subgroups. And there was no consistent differences. And also when we looked at over the decades, so trials done in the 1990s, trials done in the 2000s and the 2010s, it was also relatively similar results for each incremental time point with no consistent differences. In terms of the clinical impact, it was actually quite limited. So only 13 of the 43 or 30% of the interventions that reduced mortality at P0.05 were recommended by guidelines. So for sepsis, that includes corticosteroids, lactic-guided resuscitation, and norepinephrine versus dopamine for septic shock. For ARDS, a respiratory failure included a number of interventions to reduce tidal volume, higher PEEP, pronged ventilation, non-invasive ventilation, and ECMO, and typically interventions that sort of prevented the harms from mechanical ventilation. And there was a number of other interventions and guidelines from non-sepsis, non-ARDS trials. There was also a few interventions that were initially recommended, but then not supported by follow-up trials, including early goal-directed therapy, activated propion C, early paralysis, and intensive insulin therapy. But overall, the clinical impact of the interventions that reduced mortality was quite limited. So in summary, 28 or 30-day mortality approximates ICU mortality, and 60-day mortality approximates hospital mortality, with the caveat that very few US-based RCTs reported ICU or hospital mortality. So this may only be applicable in non-USA RCTs down in Europe, Canada, Australia, and other areas. An increasing survival curve divergence can be measured until at least day 60, and possibly at day 90. And the study does have a number of limitations, although the overall number of included RCTs was large. Some of the binary comparisons are small. As I mentioned, a few USA RCTs reported ICU or hospital mortality, and smaller numbers of RCTs reported mortality beyond 90 days. We also pooled mortality over diverse group of interventions and study context, which may have obscured subtle differences in course based on where the study was conducted and what was compared. So there may be differences, but there just wasn't enough statistical power to detect that. So in conclusions, our ICU and hospital mortality can be used as surrogates for 28-day, 30-day, and 60-day mortality, respectively, at least in non-USA-based RCTs. And RCTs in critical care should consider increasing minimum follow-up to 60 to 90 days to maximize survival curve separation and increase statistical power. And longer follow-up has already been occurring to some extent, as you can see, over the decades. So I just wanna acknowledge my co-authors, and includes John Marshall, who will be recognized as a Lifetime Achievement Award tomorrow. Thank you very much, and I guess we'll take questions at the end.

mortality

critical care trials

measurement

time points

systematic review