Index Supplemental Material

INDEX SUPPLEMENTAL MATERIAL:

Figure S1Flow diagram of retrieved trials.

Figure S2Bias risk assessment.

Figure S3All steroid trials evaluating mortality at longest follow up; forrest plot with subgroups according to risk of bias (a. fixed-effect model); trial sequential analyses (b. 10% relative risk reduction, and c. 20% relative risk reduction).

Figure S4All steroid trials evaluating mortality at longest follow up, stratified on short versus long course steroid treatment; forrest plots (a. fixed effect; b. random effect); trial sequential analyses (c-d. short course and e-f. long course treatment).

Figure S5All steroid trials evaluating mortality at longest follow up, stratified on disease severity; forrest plots (a. fixed effect; b: random effect); trial sequential analyses (c-d SIRS and sepsis and e-f severe sepsis and septic shock).

Figure S6Serious adverse events of all trials evaluating steroids for sepsis; forrest plot with subgroups according to risk of bias (a. fixed-effects model); trial sequential analysis(b.).

Figure S7All steroid trials evaluating mortality at 30 days follow up; forrest plots (a. fixed effect; b. random effect); trial sequential analyses (c. power 90% anticipated RRR 10%; d. power 80% anticipated RRR 20%).

Figure S8All steroid trials evaluating mortality at 90 days follow up:forrest plots (a. fixed effect; b. random effect); trial sequential analysis (c. power 90% anticipated RRR 10%).

Figure S9High dose steroid trials evaluating mortality at longest available follow up: forest plots (a. fixed effect; b. random effect); trial sequential analysis (c. power 90% anticipated RRR 10%).

Figure S10High dose steroid trials evaluating mortality at longest follow up; stratified on disease severity; forrest plots (a. fixed-effect; b: random-effects); trial sequential analyses (c-d. SIRS and sepsis and e. severe sepsis and septic shock).

Figure S11High dose steroid trials evaluating serious adverse events; forrest plot (a. random-effects model); trial sequential analysis (b. power 90% anticipated RRR 90%).

Figure S12Low dose steroid trials evaluating mortality at longest follow-up; forrest plot (a. fixed-effects model); trial sequential analyses of all low dose steroid trials (b. 10% relative risk reduction, c. 20% relative risk reduction); trial sequential analyses of lower risk of bias trials (d. power 90% anticipated RRR 90%; e. power 80% anticipated RRR 80%).

Figure S13Low dose steroid trials evaluating mortality at longest follow up, stratified on short versus long course steroid treatment; forrest plot random-effects.

Figure S14Low dose steroid trials evaluating mortality at longest follow up, stratified on disease severity; forrest plots (a. fixed-effect; b: random-effects); trial sequential analyses (c-d SIRS and sepsis and e-f severe sepsis and septic shock).

Figure S15Low dose steroid trials evaluating serious adverse events; forrest plot (a. random-effects model; trial sequential analysis (b. power 90%, anticipated RRR 10%).

Figure S16Low dose steroid trials evaluating mortality at 30 days follow-up; forrest plot (a. random-effects model); trial sequential analyses (b. power 90% anticipated RRR 10%; c. power 80% anticipated RRR 80%).

Figure S17 Error matrix plot (a. benefit of high dose steroids; b. harm of high dose steroids).

Figure S18 Error matrix plot (a. benefit of low dose steroids; b. harm of low dose steroids).

Table S1Definition of SIRS, sepsis, severe sepsis, and septic shock

Table S2Search strategy

Table S3Outcome measures according to GRADE

Table S4Excluded trials

Table S5Baseline characteristics included trials – extended version

Table S6Serious adverse events

Table S7GRADE assessment of outcomes of both high and low dose steroid trials

Table S8GRADE assessment of outcomes of high dose steroid trials

Table S9GRADE assessment of outcomes of low dose steroid trials

Figure S1: Flow of papers through review. Abstract not found:*[e1,e2]

Figure S2:Risk of bias summary of all included randomised clinical trials. The bias risk criteria are on the x-axis and the names of the first authors and year of the trial are on the y-axis. Green + = adequate. Red - = inadequate. Yellow ? = unclear. Other bias refers to either academic or funding bias.

Figure S3 All steroid trials evaluating mortality at longest follow up. a. Forrest plot with subgroups according to risk of bias (fixed-effect model). b+c. Trial Sequential Analyses. b. A diversity-adjusted information size (DIS) of 21,755 patients was calculated based on an anticipated relative risk reduction (RRR) of 10% (Diversity 73%, event proportion of 41% in the control arm, α=0.05 (two-sided), β=0.10 (power 90%)).The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries when all trials were analysed. c. A DIS of 3,977 patients was calculated based on an anticipated RRR of 20%(Diversity 73%, event proportion of 41% in the control arm, α=0.05 (two-sided), β=0.20 (power 80%)).The blue cumulative z-curve was constructed using a random-effects model and crossed the boundary for futility.

Figure S4 All steroid trials evaluating mortality at longest follow up stratified on treatment duration (short < 4 days and long ≥ 4 days) versus placebo. a.+b. Forrest plots (a: fixed-effect model; b: random-effects model). c+dTrial sequential analyses of all trials evaluating short course steroids for sepsis: c.A diversity-adjusted information size (DIS) could not be calculated based on an anticipated relative risk reduction of 10% (Diversity 71%, event proportion in the control arm 33.7%,α=0.05 (two sided), β 0.10 (power 90%)) . The blue cumulative z-curve was constructed using a random-effects model. d. A DIS of 5118 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity 71%, event proportion in the control arm 33.7%, α=0.05 (two-sided), β=0.20 (power 80%)).The blue cumulative Z score did not cross any of the boundaries. e+f. Trial sequential analyses of all trials evaluating long course steroids for sepsis: e. A DIS of 19178 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity 72%, event proportion in the control arm 44.0%, α=0.05 (two sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries after all trials were analysed. f. A DIS of 3,722 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity 72%, event proportion in the control arm 44.0%, α=0.05 (two-sided), β=0.20 (power 80%)). Nearly futility for an intervention effect of RRR 20% was found.

Figure S5 a+bForrest plots (a. fixed effect, b. random effect) evaluating all trials stratified based on disease severity: Severe Inflammatory Response Syndrome (SIRS) and sepsis versus severe sepsis and septic shock.c+d: Trial sequential analysis based on all trials evaluating SIRS and sepsis. c. A diversity-adjusted information size (DIS) of 30,895 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity 74%, event proportion in the control arm 34.2%, α=0.05 (two sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model. d. A DIS of 5,605 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity 74%, event proportion in the control arm 34.2%, α=0.05 (two-sided), β=0.20 (power 80%)). The blue cumulative Z score did not cross any of the boundaries. e+fTrial sequential analyses based on all trials evaluating severe sepsis and septic shock e. A DIS of 14,808 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity 75%, event proportion of 53% in the control arm, α=0.05 (two-sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries after all trials were analysed. f. A DIS of 2,762 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity 75%, event proportion of 53% in the control arm, α=0.05 (two-sided), β=0.20 (power 80%)). The blue cumulative z-curve did not cross any of the boundaries after all trials were analysed.

Figure S6:Serious adverse events of all trials evaluating steroids for sepsis. a. Forrest plot with subgroups according to risk of bias (fixed-effect model). b. Trial Sequential Analysis of serious adverse events of all trials evaluating steroids for sepsis. A diversity-adjusted information size (DIS) of 52,784 patients was calculated (Diversity 88%, event proportion of 40.4 % in the control arm, anticipated relative risk reduction 10%, α=0.05 (two-sided), β=0.10 (power 90 %)).The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries when all trials were analysed.

Figure S7a+bForrest plots (a. fixed effect, b. random effect) evaluating mortality in all steroid trials with 30 days follow up stratified on risk of bias. c.+d. TSA analysis evaluating mortality in all steroid trials with 30 days follow up. c. A Diversity adjusted information size (DIS) of 5,875 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity 0%, event proportion of 41.3% in the control arm, α=0.05 (two-sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries after all trials were analysed. d.A DIS of 1075 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity 0%, event proportion in the control arm 41.3%, α=0.05 (two-sided), β=0.20 (power 80%)). The cumulative z-curve (blue line), constructed using a random-effects model, crossed the boundary for futility (using an anticipated intervention effect of 20% relative riskreduction).

Figure S8 a+bForrest plots of all trials with 90 days mortality data (a. fixed effect, b. random effects). Only two low dose steroid trials provided 90 days mortality data. c. TSA analysis of all trials evaluating mortality with 90 days follow up. Insufficient data was available to perform a trial sequential analyses.

Figure S9: All-cause mortality at longestfollow-up in high dose steroids (>500 mg hydrocortisone or equivalent) versus placebo. a+bForest plotsof with subgroups according to the risk of bias in the trials (a. fixed-effect model, b. random-effects model); c.Trial sequential analysis of lower risk of bias trials. A diversity adjusted information size of 31,163 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity 78%, event proportion of 37.3% in the control arm,α=0.05 (two-sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the borders.

Figure S10 a+b. Forrest plots (a. fixed effect, b. random effect) evaluating all high dosesteroid trials (>500 mg hydrocortisone or equivalent) stratified based on disease severity: Severe Inflammatory Response Syndrome (SIRS) and sepsis versus severe sepsis and septic shock.c+d. Trial sequential analyses based on all high dose steroid trials evaluating SIRS and sepsis. c.A diversity-adjusted information size (DIS) of 28,497 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity=75%, event proportion of 36.2 % in the control arm, α=0.05 (two-sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries after all trials were analysed. d. A DIS of 5,182 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity 20%, event proportion of 36.2% in the control arm, α=0.05 (two-sided), β=0.20 (power 80%)). The cumulative z-curve (blue line) was constructed using a random-effects model and did not cross any of the boundaries after all trials were analysed. e. Trial sequential analysis based on all high dose steroid trials evaluating severe sepsis and septic shock could not be performed because of lack of data, also to less data was available when an analysis anticipating a RRR of 20% was attempted.

Figure S11. a. Forest plot on serious adverse events in the trials with high dose steroids (>500 mg hydrocortisone or equivalent) with subgroup according to bias risk. b.Trial sequential analysis of serious adverse events. A diversity adjusted information size (DIS) of 9,665 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity=44%, event proportion of 43.4% in the control arm, α=0.05 (two-sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries.

Figure S12: Low dose steroid trials (≤500 mg hydrocortisone or equivalent) mortality at longest available follow-up a. forest plot according to risk of bias subgroups (fixed-effect model).b+c.Trial sequential analyses (TSA) of the effect of low dose steroid use on mortality at longest follow-up.b. A diversity-adjusted information size (DIS) of 16,044 patients was calculated based on an anticipated relative risk reduction (RRR) of 10% (Diversity70%, event proportion of 46% in the control arm, α=0.05 (two-sided), β=0.10 (power 90%)).The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the borders after all trials were analysed. c. A DIS of 2,951 patients was calculated based on an anticipated RRR of 20%(Diversity70%, event proportion of 46% in the control arm, α=0.05 (two-sided), β=0.20 (power 80%)).The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries when all trials were analysed.

d+e. Trial sequential analyses of low dose steroid trials with lower risk of bias. d. A diversity-adjusted information size (DIS) of 9,226 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity=59%, event proportion of 52.4% in the control arm). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries. e. A DIS of 1,718 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity 59%, event proportion in the control arm 52.4%, α=0.05 (two-sided), β=0.20 (power 80%)).The cumulative z-curve (blue line), constructed using a random-effects model, crossed the boundary for futility (using an anticipated intervention effect of 20% relative riskreduction).

Figure S13: Forrest plot of mortality at longest follow-up of all trials evaluating low dose steroid (≤500 mg hydrocortisone or equivalent) use with subgroups according to short (<4 days) or long (≥4 days) course low dose steroid use (random-effects model).

Figure S14 a+bForrest plots (a. fixed effect, b. random effect) evaluating all low dose trials (≤500 mg hydrocortisone or equivalent) stratified based on disease severity: Severe Inflammatory Response Syndrome (SIRS) and sepsis versus severe sepsis and septic shock.c+d. Trial sequential analyses based on all low dose trials evaluating SIRS and sepsis. c.A diversity-adjusted information size (DIS) could not be calculated based on an anticipated relative risk reduction of 10% (Diversity=63%, event proportion in the control arm 24.8%,α=0.05 (two sided), β=0.10 (power 90%))d. A DIS of 6,034 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity=63%, event proportion in the control arm 24.8%,α=0.05 (two sided), β=0.20 (power 80%)). The blue cumulative z-curve was constructed using a random effects model and did not cross any of the boundaries. e+f. Trial sequential analyses based on all low dose trials evaluating severe sepsis and septic shock e. A DIS of 8,070 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity 56%, event proportion of 54.6% in the control arm, α=0.05 (two-sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries after all trials were analysed. f. A DIS of 1,510 patients was calculated based on an anticipated relative risk reduction of20% (Diversity 56%, event proportion in the control arm 54.6%, α=0.05 (two-sided), β=0.20 (power 80%)).The cumulative z-curve (blue line), constructed using a random-effects model, crossed the boundary for futility (using an anticipated intervention effect of 20% relative riskreduction).

Figure S15: Serious adverse eventsof low dose steroid (≤500 mg hydrocortisone or equivalent) use a. forest plot withsubgroups according to the risk of bias in the trials.b.trial sequential analysis including all trials. A diversity adjusted information size could not be calculated due to insufficient data.

Figure S16:a. Forrest plot of mortality (random effects model) at 30 days follow-up of low dose steroids (≤500 mg hydrocortisone or equivalent) use according to risk of bias subgroups. b+c. Trial sequential analyses of mortality at 30 days including all trials evaluating low dose steroids versus placebo. b. A diversity-adjusted information size (DIS) of 8734 patients was calculated based on an anticipated relative risk reduction of 10% (Diversity 27%, event proportion of 39.2% in the control arm, α=0.05 (two sided), β=0.10 (power 90%)). The blue cumulative z-curve was constructed using a random-effects model and did not cross any of the boundaries after all trials were analysed. c. A DISof 1,594 patients was calculated based on an anticipated relative risk reduction of 20% (Diversity 27%, event proportion of 39.2% in the control arm, α=0.05 (two sided), β=0.20 (power 80%) The blue cumulative z-curve was constructed using a random effects model and crossed the inner border after all trials were analysed, indicating futility for an intervention effect with 20% relative risk reduction.

Figure S17The Manhattan error matrix plot for overview of benefit and harm of all trials evaluating high dose steroids (>500 mg hydrocortisone or equivalent) use. Trials and their outcomes with a. benefit and b. harm according to the three dimensions of systematic error (bias, measured by level of evidence), random error (measured by standard error (SE)), and design error (measured by grading the outcome measures from a patients’ perspective).Two trials had standard errors larger than 80 and were therefore not shown.

Figure S18The Manhattan error matrix plot for overview of benefit and harm of all trials evaluating low dose steroids (≤500 mg hydrocortisone ore equivalent) use. Trials and their outcomes with a. benefit and b. harm according to the three dimensions of systematic error (bias, measured by level of evidence), random error (measured by standard error (SE)), and design error (measured by grading the outcome measures from a patients’ perspective).Three trials had standard errors larger than 80 and were therefore not shown.

Condition / Definition
Systemic inflammatory response syndrome (SIRS) / The presence of two or more of the following: temperature > 38°C or < 36°C, heart rate > 90 beats/minute, respiratory rate > 20 breaths/minute or PaCO2 < 4.3 kPa , and white blood cell count > 12 cells/mm3 or < 4 cells/mm3 or > 10% immature (band) forms.
Sepsis / SIRS in response to an infectious process.
Severe sepsis / Sepsis associated with organ dysfunction,hypoperfusion, or hypotension. Hypoperfusion and perfusion abnormalities may include, but are not limited to lactic acidosis, oliguria, or an acute alteration in mental status.
Septic shock / Sepsis with hypotension, despite adequate fluid resuscitation, along with presence of perfusion abnormalities that may include, but are not limited to lactic acidosis, oliguria, or an acute alteration in mental status.

Table S1 Definitions for sepsis and organ failure as stated by the American College of Chest Physicians/Society of Critical Care Medicine in 1992 [e3].