Critical Care Trials Reporting a Mortality Effect
Image: Shutterstock / killa1
Introduction
Mortality is the ultimate outcome in critical care. It is also one of the most persuasive—and most easily overinterpreted—results a trial can report.
This collection brings together major adult acute and critical care randomised trials that reported a statistically significant reduction or increase in mortality. Trials are organised by physiological system and classified according to where the mortality finding arose: as a primary or secondary outcome, or in a subgroup or restricted analysis.
A mortality difference deserves attention, but its credibility depends on the trial’s design, analysis, consistency with related evidence and subsequent replication. Tables 9 and 10 therefore compare earlier mortality benefits and harms with the results of later large multicentre randomised trials. Only direct or reasonably close comparisons are included, with important differences in population, intervention, dose, timing and comparator stated explicitly.
Among the major, widely recognised trials, one of the largest relative mortality reductions was seen in PROSEVA: 28-day mortality was 16.0% with prone positioning versus 32.8% with supine positioning, representing a crude relative reduction of approximately 51% and an absolute reduction of 16.8 percentage points. The largest absolute reduction was reported in DESTINY II, where decompressive hemicraniectomy reduced six-month mortality from 73.0% to 32.7%—a difference of 40.3 percentage points—although most survivors remained dependent.
How to Read these Tables
For this collection, “statistically significant” includes a prespecified adjusted analysis or a prespecified Bayesian decision threshold where this formed the trial’s planned analysis. Orange labels identify important qualifications, including adjusted-only findings, prior-dependent Bayesian results, early stopping, subgroup analyses and absence of multiplicity adjustment. Inclusion does not imply that a finding is robust, reproducible or practice-changing.
Mortality Benefit
Mortality Harm
How Trials Were Selected
This is a curated educational collection rather than a systematic review or exhaustive catalogue. Candidate trials were identified through targeted literature searches, the Critical Care Reviews trials archive and ongoing literature surveillance, and cross-checking against major reviews, guidelines and previous compilations.
Eligible evidence comprised peer-reviewed reports of randomised trials in adults relevant to acute and critical care, including subsequently published analyses or follow-up of randomised cohorts. To qualify, a trial had to report a statistically significant between-group mortality difference as a primary or co-primary outcome or a secondary outcome, or in a subgroup or restricted analysis. Mortality had to be reported separately and not solely as one component of a composite outcome. Prespecified adjusted analyses and planned Bayesian decision thresholds were accepted, with important analytic qualifications identified in the tables.
Editorial judgement was used to focus the collection on trials considered major, influential or historically important to critical care. Relevant considerations included trial size, multicentre design, phase of development, methodological quality, influence on practice or guidelines, and importance in generating or resolving a major clinical controversy. No fixed sample-size threshold was applied, allowing smaller historically important trials to be included, but very small exploratory, highly specialised or methodologically weak studies were not routinely retained.
Trials were assigned according to the highest relevant level in the outcome hierarchy: primary or co-primary outcome before secondary outcome, and secondary outcome before subgroup analysis. Duplication was generally avoided, except where opposite-direction subgroup effects were supported or where later non-confirmation is specifically examined in Tables 9 and 10. The collection is therefore deliberately broad but not exhaustive. Inclusion does not establish that a mortality finding is robust, reproducible or practice-changing, while omission does not imply that a study is invalid.
Table 1. Statistically significant mortality benefit as a primary or co-primary outcome
Inclusion rule: adult randomised trials relevant to acute or critical care in which mortality or survival was a prespecified primary, co-primary, or principal mortality endpoint and the planned analysis crossed its prespecified statistical threshold.
| Trial | Population | Intervention | Comparator | Primary mortality / survival endpoint | Intervention result | Comparator result | Effect and qualification |
|---|---|---|---|---|---|---|---|
| Neuro | |||||||
| MARVEL2024 | Acute anterior-circulation large-vessel ischaemic stroke undergoing endovascular thrombectomy; n=1,680 | Intravenous methylprednisolone 2 mg/kg/day for 3 days plus thrombectomy (n=839) | Placebo plus thrombectomy (n=841) | All-cause death within 90 days; prespecified primary safety outcome | 23.2% | 28.5% | Adjusted RR 0.84 (95% CI 0.71–0.98); P=0.03. The primary efficacy outcome, 90-day modified Rankin Scale distribution, was neutral.primary safety outcomeneutral efficacy outcome |
| ENRICH2024 | Acute spontaneous supratentorial intracerebral haemorrhage suitable for surgery within 24 hours; n=300 | Early minimally invasive parafascicular haematoma evacuation plus medical management (n=150) | Guideline-based medical management alone (n=150) | Death within 30 days; primary safety endpoint | 14/150 (9.3%) | 27/150 (18.0%) | Bayesian risk difference −8.7 percentage points (95% CrI −16.4 to −1.0); posterior probability 0.987. Adaptive enrolment was restricted to lobar haemorrhage after the anterior-basal-ganglia futility rule was triggered.primary safety endpointadaptive trial |
| LASTE2024 | Anterior-circulation large-vessel ischaemic stroke with a large infarct of unrestricted size; n=333 randomised | Endovascular thrombectomy plus medical care (n=166 assigned) | Medical care alone (n=167 assigned) | All-cause death at 90 days; primary safety outcome | 36.1% | 55.5% | Adjusted RR 0.65 (95% CI 0.50–0.84); P<0.001. Nine randomised patients were excluded for consent or legal reasons, and recruitment stopped after external trials favoured thrombectomy.primary safety outcomeearly stop |
| Circulatory | |||||||
| DanGer-Shock2024 | STEMI-related cardiogenic shock; 355 patients in the primary analysis | Impella CP microaxial-flow pump plus standard care (n=179) | Standard care alone (n=176) | All-cause death at 180 days | 45.8% | 58.5% | HR 0.74 (95% CI 0.55–0.99); P=0.04. Severe bleeding and renal-replacement therapy were increased. |
| Resuscitation | |||||||
| Sack IAC-CPR1992 | Adult in-hospital cardiac arrest; 103 patients and 135 resuscitation attempts | Standard CPR plus interposed abdominal counterpulsation (48 patients) | Standard CPR (55 patients) | Survival to hospital discharge; one of three prespecified endpoints | 25% | 7% | P=0.02. Neurologically intact survival was 17% vs 6% and was not statistically significant.small single-centre trial |
| ALIVE2002 | Shock-resistant or recurrent ventricular fibrillation during out-of-hospital cardiac arrest; n=347 | Intravenous amiodarone during resuscitation (n=180) | Intravenous lidocaine (n=167) | Survival to hospital admission | 41/180 (22.8%) | 20/167 (12.0%) | OR 2.17 (95% CI 1.21–3.83); P=0.009. Survival to hospital discharge was not significantly different: 5.0% vs 3.0%; P=0.34.short-term survival endpointno discharge benefit |
| PAD Trial2004 | Definite out-of-hospital cardiac arrests occurring within 993 participating community units; 235 arrests | Structured volunteer response with CPR and automated external defibrillation (128 arrests) | Structured volunteer response with CPR alone (107 arrests) | Survival to hospital discharge | 30/128 (23.4%) | 15/107 (14.0%) | RR 2.00 (95% CI 1.07–3.77); P=0.03. The community unit, rather than the individual arrest, was the unit of randomisation and primary analysis.cluster randomised |
| VSE2013 | Adults with in-hospital cardiac arrest requiring vasopressors; n=268 | Vasopressin, epinephrine and methylprednisolone during CPR, followed by hydrocortisone for post-resuscitation shock (n=130) | Epinephrine plus matching placebos (n=138) | Survival to hospital discharge with Cerebral Performance Category 1 or 2; co-primary endpoint | 18/130 (13.9%) | 7/138 (5.1%) | OR 3.28 (95% CI 1.17–9.20); P=0.02. Return of spontaneous circulation, the other co-primary endpoint, was also higher. This was a bundled intervention conducted at three centres.co-primary endpointbundled strategy |
| PARAMEDIC22018 | Adult out-of-hospital cardiac arrest; n=8,014 | Parenteral adrenaline during resuscitation (n=4,012) | Saline placebo (n=3,995) | Survival at 30 days | 130/4,012 (3.2%) | 94/3,995 (2.4%) | Unadjusted OR for survival 1.39 (95% CI 1.06–1.82); P=0.02. Favourable neurological survival was not significantly increased. |
| ARREST2020 | Refractory VF/VT out-of-hospital cardiac arrest; 29 patients with primary-outcome data | Early ECMO-facilitated resuscitation pathway (n=14) | Standard advanced cardiac life support (n=15) | Survival to hospital discharge | 6/14 (43%) | 1/15 (7%) | Risk difference 36.2% (95% Bayesian CrI 3.7–59.2); posterior probability 0.9861. Bundled, single-centre phase 2 strategy stopped early for efficacy.Bayesianearly stop |
| DOSE VF2022 | Out-of-hospital cardiac arrest with ventricular fibrillation persisting after three standard shocks; n=405 | Double sequential external defibrillation, DSED (n=125), or vector-change defibrillation (n=144) | Continued standard defibrillation (n=136) | Survival to hospital discharge | DSED: 38/125 (30.4%); vector change: 31/143 (21.7%) | 18/135 (13.3%) | DSED adjusted RR 2.21 (95% CI 1.33–3.67); vector-change adjusted RR 1.71 (95% CI 1.01–2.88). Overall three-group P=0.009. Cluster-crossover recruitment stopped during the COVID-19 pandemic.cluster crossoverpremature stop |
| Airway | |||||||
| Rumbak early tracheotomy2004 | Medical ICU patients predicted to require mechanical ventilation for more than 14 days; n=120 | Percutaneous tracheotomy within 48 hours (n=60) | Continued translaryngeal intubation with tracheotomy planned for days 14–16 (n=60) | Hospital mortality; one of the principal clinical outcomes | 19/60 (31.7%) | 37/60 (61.7%) | P<0.005. Small single-centre trial with unusually high comparator mortality; the larger TracMan trial did not confirm a mortality benefit.small single-centre trialnot confirmed |
| Respiratory | |||||||
| Amato protective ventilation1998 | Adults with ARDS; n=53 | Protective ventilation using lower tidal volumes, higher PEEP and pressure limitation (n=29) | Conventional ventilation (n=24) | Death by day 28 | 11/29 (38%) | 17/24 (71%) | P<0.001. Stopped at the fifth interim analysis; hospital mortality was not significantly different.early stop |
| Esteban PCV vs VCV2000 | Adults with ARDS in 12 ICUs; n=79 | Pressure-controlled ventilation (n=37) | Volume-controlled ventilation (n=42) | In-hospital mortality | 19/37 (51%) | 33/42 (78%) | P=0.02. Small trial; mode effects were confounded by differences in delivered ventilation and were not confirmed by later evidence. |
| ARDSNet ARMA2000 | Mechanically ventilated adults with acute lung injury or ARDS; n=861 | Tidal volume 6 mL/kg predicted body weight (n=432) | Tidal volume 12 mL/kg predicted body weight (n=429) | Death before discharge home while breathing unassisted (up to day 180) | 31.0% | 39.8% | Absolute difference −8.8 percentage points (95% CI −15.3 to −2.4); P=0.007. |
| Fagon VAP strategy2000 | Mechanically ventilated adults with suspected ventilator-associated pneumonia; n=413 | Invasive quantitative-sampling management strategy (n=204) | Clinical, non-invasive management strategy (n=209) | Death by day 14; a principal prespecified clinical outcome | 33/204 (16.2%) | 54/209 (25.8%) | Absolute difference −9.6 percentage points; P=0.022. Day-28 mortality was not statistically significant. |
| Villar open-lung strategy2006 | Persistent ARDS in eight ICUs; 95 patients in the per-protocol analysis | Low tidal volume with PEEP set above the lower inflection point (n=50) | Conventional tidal volume and lower PEEP (n=45) | ICU and hospital mortality; principal outcomes | ICU: 16/50 (32.0%) | ICU: 24/45 (53.3%) | ICU mortality P=0.040; hospital mortality 34.0% vs 55.5%, P=0.041. Eight randomised patients were excluded for protocol/randomisation irregularities.per-protocol emphasis |
| ACURASYS2010 | Early severe ARDS; n=340 | 48-hour cisatracurium infusion (n=178) | Placebo (n=162) | Adjusted 90-day mortality | 31.6% | 40.7% | Prespecified adjusted HR 0.68 (95% CI 0.48–0.98); P=0.04. Unadjusted P=0.08; the larger ROSE trial was neutral.adjusted-positivenot confirmed |
| PROSEVA2013 | Early severe ARDS with PaO₂/FiO₂ <150 mm Hg; n=466 | Prolonged prone-positioning sessions (n=237) | Supine positioning (n=229) | Death by day 28 | 38/237 (16.0%) | 75/229 (32.8%) | HR 0.39 (95% CI 0.25–0.63); P<0.001. Day-90 mortality was also lower. |
| OXYGEN-ICU2016 | Adults expected to remain in one ICU for at least 72 hours; 434 analysed | Conservative oxygen therapy (n=216) | Conventional oxygen therapy (n=218) | ICU mortality | 25/216 (11.6%) | 44/218 (20.2%) | Absolute difference −8.6 percentage points (95% CI −15.0 to −1.7); P=0.01. Stopped after an unplanned interim analysis; larger oxygen-target trials were neutral.early stopnot confirmed |
| RECOVERY dexamethasone2021 | Adults hospitalised with COVID-19; n=6,425 | Dexamethasone 6 mg daily for up to 10 days (n=2,104) | Usual care (n=4,321) | All-cause death by day 28 | 482/2,104 (22.9%) | 1,110/4,321 (25.7%) | Age-adjusted rate ratio 0.83 (95% CI 0.75–0.93); P<0.001. Benefit was confined to patients receiving oxygen or invasive ventilation. |
| RECOVERY tocilizumab2021 | Hypoxic adults hospitalised with COVID-19 and systemic inflammation; n=4,116 | Tocilizumab plus usual care (n=2,022) | Usual care alone (n=2,094) | All-cause death by day 28 | 621/2,022 (31%) | 729/2,094 (35%) | Rate ratio 0.85 (95% CI 0.76–0.94); P=0.0028. Most participants also received systemic corticosteroids. |
| RECOVERY baricitinib2022 | Adults hospitalised with COVID-19; n=8,156 | Baricitinib plus usual care (n=4,148) | Usual care alone (n=4,008) | All-cause death by day 28 | 514/4,148 (12%) | 546/4,008 (14%) | Age-adjusted rate ratio 0.87 (95% CI 0.77–0.99); P=0.028. |
| CAPE COD2023 | Adults admitted to ICU with severe community-acquired pneumonia; n=795 analysed | Early hydrocortisone with response-guided taper (n=400) | Placebo (n=395) | Death by day 28 | 25/400 (6.2%) | 47/395 (11.9%) | Absolute difference −5.6 percentage points (95% CI −9.6 to −1.7); P=0.006. Stopped after the second planned interim analysis.early stop |
| Gastrointestinal | |||||||
| Villanueva transfusion strategy2013 | Adults with severe acute upper gastrointestinal bleeding; n=889 | Restrictive transfusion threshold, haemoglobin <7 g/dL (n=444) | Liberal threshold, haemoglobin <9 g/dL (n=445) | All-cause death within 45 days | 23/444 (5%) | 41/445 (9%) | HR 0.55 (95% CI 0.33–0.92); P=0.02. Single-centre trial; exsanguinating haemorrhage and selected cardiovascular disease were excluded. |
| Liver | |||||||
| Sort albumin in SBP1999 | Cirrhosis with spontaneous bacterial peritonitis; n=126 | Cefotaxime plus albumin 1.5 g/kg on day 1 and 1 g/kg on day 3 (n=63) | Cefotaxime alone (n=63) | Renal impairment and mortality; co-primary main endpoints | In hospital: 10% | In hospital: 29% | In-hospital P=0.01; 3-month mortality 22% vs 41%, P=0.03. |
| High-volume plasma exchange2016 | Adults with acute liver failure managed at three liver-transplant centres; n=182 | Standard medical treatment plus high-volume plasma exchange for 3 days (n=92) | Standard medical treatment alone (n=90) | Liver-transplantation-free survival during the hospital admission | Hospital survival: 58.7% | Hospital survival: 47.8% | HR for death, stratified for liver transplantation, 0.56 (95% CI 0.36–0.86); P=0.0083. Plasma exchange before subsequent transplantation did not improve post-transplant survival.open-label trialtransplant-stratified analysis |
| Renal | |||||||
| Ronco CVVH dose2000 | Critically ill adults with acute renal failure receiving CVVH; n=425 | Ultrafiltration 35 mL/kg/h (n=139) or 45 mL/kg/h (n=140) | Ultrafiltration 20 mL/kg/h (n=146) | Survival 15 days after stopping haemofiltration | 57% and 58% | 41% | 35 vs 20 mL/kg/h: P=0.0007; 45 vs 20: P=0.0013; 35 vs 45: P=0.87. Later RENAL and ATN intensity trials were neutral.not confirmed |
| Schiffl daily haemodialysis2002 | Critically ill adults with acute renal failure; n=160 | Daily intermittent haemodialysis | Alternate-day haemodialysis | Mortality during the index hospital course | 28% | 46% | P=0.01. Single-centre trial; subsequent larger renal-support intensity trials were neutral.not confirmed |
| ELAIN2016 | Critically ill adults with KDIGO stage 2 AKI plus enrichment criteria; n=231 | RRT within 8 hours of stage 2 AKI (n=112) | RRT at stage 3 AKI or conventional indication (n=119) | All-cause death by day 90 | 44/112 (39.3%) | 65/119 (54.7%) | HR 0.66 (95% CI 0.45–0.97); P=0.03. Single-centre; AKIKI and STARRT-AKI did not replicate a survival benefit.not confirmed |
| Sepsis | |||||||
| Lachman anti-LPS1984 | Obstetric or gynaecological septic shock; n=33 | Human plasma rich in anti-lipopolysaccharide IgG (n=14) | Conventional treatment (n=19) | Hospital mortality; principal efficacy outcome | 1/14 (7.1%) | 9/19 (47.4%) | Reported as statistically significant. Very small, old single-centre trial; later anti-endotoxin trials did not confirm benefit.small single-centre trialnot confirmed |
| Rivers EGDT2001 | Severe sepsis or septic shock presenting to one emergency department; n=263 | Six-hour early goal-directed therapy (n=130) | Standard therapy (n=133) | In-hospital mortality | 30.5% | 46.5% | Absolute difference −16.0 percentage points; P=0.009. ProCESS, ARISE and ProMISe subsequently found no benefit from protocolised EGDT.not confirmed |
| PROWESS2001 | Adults with severe sepsis; n=1,690 | Drotrecogin alfa activated for 96 hours (n=850) | Placebo (n=840) | All-cause death by day 28 | 210/850 (24.7%) | 259/840 (30.8%) | Relative risk reduction 19.4% (95% CI 6.6–30.5); P=0.005. PROWESS-SHOCK was neutral and the drug was withdrawn.not confirmedwithdrawn |
| Annane corticosteroids2002 | Catecholamine-dependent septic shock; n=300 randomised, with 229 corticotropin nonresponders in the prespecified primary population | Hydrocortisone 50 mg every 6 hours plus fludrocortisone 50 micrograms daily for 7 days (n=114 nonresponders) | Matching placebos (n=115 nonresponders) | Twenty-eight-day survival distribution among corticotropin nonresponders | 60/114 deaths (53%) | 73/115 deaths (63%) | Adjusted HR 0.67 (95% CI 0.47–0.95); P=0.02. The qualifying population was prespecified as the trial’s primary analysis; no benefit was found among corticotropin responders.prespecified restricted analysis |
| de Jonge SDD2003 | Adults admitted to one mixed ICU; n=934 | Selective digestive decontamination (n=466) | Standard care (n=468) | ICU and hospital mortality; primary endpoints | ICU: 69/466 (14.8%) | ICU: 107/468 (22.9%) | ICU RR 0.65 (95% CI 0.49–0.85), P=0.002; hospital mortality 24.2% vs 31.2%, P=0.02. Setting had low baseline antimicrobial resistance. |
| MONARCS afelimomab2004 | Severe sepsis with elevated IL-6; n=998 | Afelimomab anti-TNF F(ab′)₂ fragment | Placebo | All-cause death by day 28 | 43.6% | 47.6% | Unadjusted P=0.21; prespecified logistic model estimated adjusted absolute risk reduction 5.8%, P=0.041.adjusted-positive only |
| de Smet SOD/SDD2009 | Adults in 13 ICUs; n=5,939 | Selective oropharyngeal decontamination (n=1,904) or selective digestive decontamination (n=2,045) | Standard care (n=1,990) | Death by day 28 | SOD 26.6%; SDD 26.9% | 27.5% | Adjusted OR vs standard care: SOD 0.86 (95% CI 0.74–0.99); SDD 0.83 (0.72–0.97). Crude differences were not significant.adjusted-positive only |
| ETASS2013 | Adults with severe sepsis in multiple Chinese ICUs; n=361 | Thymosin α1 for seven days (n=181) | Saline control (n=180) | All-cause death by day 28 | 26.0% | 35.0% | RR 0.74 (95% CI 0.54–1.02); non-stratified P=0.062; log-rank P=0.049. The larger TESTS trial was neutral. |
| SAPS procalcitonin2016 | Critically ill adults starting antibiotics for presumed infection; n=1,546 | Procalcitonin-guided antibiotic discontinuation (n=761) | Standard care (n=785) | 28-day and 1-year mortality; primary safety outcomes | 149/761 (20%) | 196/785 (25%) | Absolute difference −5.4 percentage points (95% CI −9.5 to −1.2); P=0.0122. Mortality was designed as a non-inferiority safety endpoint; superiority was an unexpected finding.primary safety endpoint |
| APROCCHSS2018 | Adults with septic shock; 1,241 included in the corticosteroid comparison | Hydrocortisone plus fludrocortisone (n=614) | Placebo (n=627) | All-cause death by day 90 | 264/614 (43.0%) | 308/627 (49.1%) | RR 0.88 (95% CI 0.78–0.99); P=0.03. |
| MIST2021 | Severe presumed Gram-negative sepsis; n=202 | Intradermal heat-killed Mycobacterium w (n=101) | Saline placebo (n=101) | All-cause death by day 28 | 9/101 (8.9%) | 20/101 (19.8%) | Absolute difference −10.9 percentage points (95% CI −21 to −1); P=0.04. Small phase 2b trial. |
| EXIT-SEP2023 | Adults with Sepsis-3 sepsis in 45 Chinese ICUs; 1,760 in the primary analysis | Xuebijing injection plus standard care (n=878) | Saline placebo plus standard care (n=882) | All-cause death by day 28 | 165/878 (18.8%) | 230/882 (26.1%) | Absolute difference −7.3 percentage points (95% CI −11.2 to −3.4); P<0.001. Generalisability outside the studied formulation and setting is uncertain. |
| BLING III2024 | Critically ill adults with sepsis receiving piperacillin–tazobactam or meropenem; 6,981 in the primary analysis | Continuous β-lactam infusion (n=3,474) | Intermittent β-lactam infusion (n=3,507) | All-cause death by day 90 | 864/3,474 (24.9%) | 939/3,507 (26.8%) | Unadjusted primary OR 0.91 (95% CI 0.81–1.01), P=0.08; prespecified adjusted OR 0.89 (0.79–0.99), P=0.04.adjusted-positive only |
| SCREEN2025 | Adults admitted to 43 analysable hospital wards in five hospitals; n=60,055 | Visible qSOFA-based electronic sepsis alert and response pathway (n=29,442) | Masked alert / usual care (n=30,613) | In-hospital death within 90 days | 937/29,442 (3.2%) | 961/30,613 (3.1%) | Primary cluster/time-adjusted RR 0.85 (95% CI 0.77–0.93); P<0.001. Crude risk difference was 0.0% (95% CI −0.2 to 0.3).adjusted-positive onlystepped-wedge |
| TIGRIS2026 | Endotoxic septic shock with endotoxin activity 0.60–0.89; n=157 | Two polymyxin-B haemoadsorption sessions plus standard care (n=106) | Standard care alone (n=51) | All-cause death by day 28; Bayesian primary analysis | 41/106 (39%) | 23/51 (45%) | Prior-informed adjusted OR 0.67 (95% CrI 0.39–1.08); posterior probability of benefit 95.3%, crossing the 95% threshold. TIGRIS-only posterior probability was 79.7%; therefore the positive conclusion was prior-dependent.Bayesianprior-dependent |
| Trauma | |||||||
| CRASH-22010 | Adult trauma with, or at risk of, significant bleeding; n=20,127 | Tranexamic acid within eight hours of injury (n=10,060) | Placebo (n=10,067) | All-cause death in hospital within four weeks | 1,463/10,060 (14.5%) | 1,613/10,067 (16.0%) | RR 0.91 (95% CI 0.85–0.97); P=0.0035. Subsequent analyses support administration within three hours and possible harm when given later. |
| PAMPer2018 | Injured adults at risk of haemorrhagic shock during air-medical transport; n=501 | Two units of prehospital thawed plasma plus standard resuscitation (n=230) | Standard-care resuscitation (n=271) | All-cause death by day 30 | 23.2% | 33.0% | Absolute difference −9.8 percentage points (95% CI −18.6 to −1.0), P=0.03; HR 0.64 (95% CI 0.45–0.91), P=0.01. Cluster-randomised; benefit may depend on transport context. |
| Endocrine | |||||||
| Leuven I2001 | Predominantly postoperative adults in one surgical ICU; n=1,548 | Intensive insulin therapy, glucose 80–110 mg/dL (n=765) | Conventional insulin strategy (n=783) | All-cause ICU mortality | 35/765 (4.6%) | 63/783 (8.0%) | Sequentially adjusted P<0.04. Single-centre and stopped after interim monitoring; NICE-SUGAR later found increased mortality with intensive glucose control.early stopnot confirmed |
| Metabolic | |||||||
| de Silva activated charcoal2003 | Adults with yellow-oleander poisoning; n=401 | Multiple-dose activated charcoal for 72 hours (n=201) | Placebo after the initial charcoal dose (n=200) | Death during the index admission | 5/201 (2.5%) | 16/200 (8.0%) | Absolute difference −5.5 percentage points (95% CI −10.3 to −0.6); P=0.025. A later, larger mixed self-poisoning trial was neutral.not confirmed |
| Miscellaneous | |||||||
| Boyd oxygen-delivery optimisation1993 | High-risk adults undergoing major surgery; n=107 | Protocolised perioperative increase in oxygen delivery, including dopexamine (n=53) | Usual perioperative haemodynamic care (n=54) | All-cause death by day 28 | 3/53 (5.7%) | 12/54 (22.2%) | Absolute difference −16.5 percentage points; P=0.015. Small single-centre trial; later multicentre perioperative optimisation trials were neutral.not confirmed |
Interpretation: placement in this table is a statistical classification, not an endorsement of current practice. “Adjusted-positive only” means the unadjusted comparison did not cross P<0.05 but the trial’s prespecified adjusted primary-outcome model did. “Bayesian” denotes a prespecified posterior-probability threshold.
Table 2. Statistically significant mortality benefit as a secondary outcome in trials with a favourable primary outcome
Inclusion rule: adult randomised trials relevant to acute or critical care in which the prespecified primary efficacy outcome was statistically positive and a separate secondary, component, or supportive analysis found lower mortality or higher survival. Mortality-primary trials, trials with a neutral primary outcome, and subgroup-only mortality findings are excluded.
| Trial | Population | Intervention | Comparator | Positive primary endpoint | Mortality — intervention | Mortality — comparator | Mortality effect and qualification |
|---|---|---|---|---|---|---|---|
| Neuro | |||||||
| DESTINY II2014 | Malignant middle-cerebral-artery infarction in patients aged ≥61 years; n=112 | Decompressive hemicraniectomy plus intensive care (n=49) | Conservative intensive care (n=63) | Survival without severe disability (mRS 0–4) at 6 months: 38.5% vs 17.7%; OR 2.91 (95% CI 1.06–7.49); P=0.039 | 16/49 (32.7%) | 46/63 (73.0%) | Six-month absolute difference −40.3 percentage points; P<0.001. No survivor achieved mRS 0–2; most remained dependent.early stopdisability trade-off |
| ESCAPE2015 | Anterior-circulation large-vessel ischaemic stroke with small core and favourable collaterals; n=316 | Rapid endovascular thrombectomy plus standard care (n=165 randomised) | Standard care, including alteplase when eligible (n=151 randomised) | Favourable 90-day mRS shift: common OR 2.6 (95% CI 1.7–3.8); P<0.001 | 17/164 (10.4%) | 28/147 (19.0%) | Ninety-day RR 0.5 (95% CI 0.3–1.0); P=0.04. The trial stopped after an interim analysis prompted by external thrombectomy evidence.early stopborderline |
| RESCUEicp2016 | Severe traumatic brain injury with refractory intracranial pressure >25 mm Hg; 389 with 6-month outcome data | Last-tier decompressive craniectomy (n=201 analysed) | Ongoing medical care, usually including barbiturates (n=188 analysed) | Six-month GOS-E distribution differed: unordered χ²=30.69; P<0.001 after rejection of proportional odds | 26.9% | 48.9% | Absolute difference −22.0 percentage points. Death was a prespecified GOS-E category rather than a separately multiplicity-controlled mortality hypothesis; vegetative state and severe disability increased among survivors.component analysisdisability trade-off |
| ATTENTION2022 | Acute basilar-artery occlusion treated within 12 hours; n=340 | Endovascular thrombectomy plus medical care (n=226) | Best medical care (n=114) | mRS 0–3 at 90 days: 46% vs 23%; adjusted RR 2.06 (95% CI 1.46–2.91); P<0.001 | 37% | 55% | Ninety-day adjusted RR 0.66 (95% CI 0.52–0.82). Open-label, 2:1 trial conducted in China. |
| TENSION2023 | Anterior-circulation large-vessel stroke with established large infarct (ASPECTS 3–5); n=253 | Endovascular thrombectomy plus medical care (n=125) | Medical care alone (n=128) | Favourable 90-day mRS shift: adjusted common OR 2.58 (95% CI 1.60–4.15); P=0.0001 | 49/122 (40%) | 64/126 (51%) | Adjusted HR 0.67 (95% CI 0.46–0.98); P=0.038. Recruitment stopped at a prespecified interim review.early stopborderline |
| Circulatory | |||||||
| L’Her CPAP2004 | Patients aged ≥75 years with severe cardiogenic pulmonary oedema; n=89 | Standard treatment plus face-mask CPAP (n=43) | Standard medical treatment (n=46) | At 1 hour, respiratory rate and PaO₂/FiO₂ improved; severe complications 4 vs 17, P=0.002 | 3/43 (7%) | 11/46 (24%) | Mortality through 48 hours only: absolute difference −17 percentage points; P=0.017. No survival benefit persisted over the full hospital stay, and the report did not specify a modern single primary hierarchy.early timepoint onlyno sustained benefit |
| CULPRIT-SHOCK2017 | Acute myocardial infarction, multivessel disease and cardiogenic shock; n=706 | Culprit-lesion-only PCI, with staged PCI permitted (n=344) | Immediate multivessel PCI (n=341 in the primary analysis) | Death or severe renal failure requiring RRT by day 30: 45.9% vs 55.4%; RR 0.83 (95% CI 0.71–0.96); P=0.01 | 43.3% | 51.6% | Thirty-day RR 0.84 (95% CI 0.72–0.98); P=0.03. Mortality was a component analysis of the positive composite; mortality no longer differed at 1 year.composite componentno sustained benefit |
| Resuscitation | |||||||
| HACA2002 | Comatose survivors of witnessed VF/VT out-of-hospital cardiac arrest; n=275 | Mild hypothermia at 32–34°C for 24 hours (n=137) | Standard normothermic post-arrest care (n=138) | Favourable neurological outcome at 6 months: 75/136 (55%) vs 54/137 (39%); RR 1.40 (95% CI 1.08–1.81) | 56/137 (41%) | 76/138 (55%) | Six-month RR 0.74 (95% CI 0.58–0.95). Later temperature-control trials used different populations and comparators and did not reproduce a benefit from deeper cooling. |
| ResQTrial2011 | Adults with non-traumatic out-of-hospital cardiac arrest across 46 emergency medical service agencies; 2,470 provisionally randomised and 1,653 in the final assigned-treatment analysis | Active compression–decompression CPR plus an impedance-threshold device (n=840) | Standard manual CPR (n=813) | Survival to hospital discharge with favourable neurological function: 75/840 (9%) vs 47/813 (6%); OR 1.58 (95% CI 1.07–2.36); P=0.019 | One-year survival: 74/840 (9%) | One-year survival: 48/813 (6%) | One-year survival P=0.03. Pulmonary oedema was more frequent with the intervention: 11% vs 7%; P=0.015.secondary survival endpointdevice combination |
| Respiratory | |||||||
| Brochard NIV1995 | Severe acute exacerbation of COPD; n=85 | Face-mask non-invasive ventilation plus standard treatment (n=43) | Standard treatment alone (n=42) | Endotracheal intubation: 11/43 (26%) vs 31/42 (74%); P<0.001 | 4/43 (9%) | 12/42 (29%) | In-hospital absolute difference −19.3 percentage points; P=0.02. The highly selected trial excluded patients needing immediate intubation. |
| Nava NIV weaning1998 | Intubated patients with COPD who failed a T-piece weaning trial; n=50 | Early extubation to non-invasive pressure support (n=25) | Continued invasive pressure-support weaning (n=25) | Higher probability of weaning during ventilation, P=0.002; mean ventilation 10.2 vs 16.6 days, P=0.021 | 2/25 (8%) | 7/25 (28%) | Sixty-day survival 92% vs 72%; P=0.009 by time-to-event analysis. Small, selected COPD trial. |
| Plant ward NIV2000 | Mild-to-moderate acidotic COPD exacerbation on general respiratory wards; n=236 | Early ward-based non-invasive ventilation plus standard care (n=118) | Standard care alone (n=118) | Meeting prespecified intubation criteria: 18/118 (15%) vs 32/118 (27%); P=0.02 | 12/118 (10%) | 24/118 (20%) | In-hospital absolute difference −10.2 percentage points; P=0.05. Benefit was not evident in the most severely acidotic stratum.borderline |
| Hilbert NIV2001 | Immunosuppressed patients with pulmonary infiltrates, fever and early hypoxaemic respiratory failure; n=52 | Intermittent face-mask non-invasive ventilation (n=26) | Supplemental oxygen without ventilatory support (n=26) | Endotracheal intubation: 12/26 (46%) vs 20/26 (77%); P=0.03 | 13/26 (50%) | 21/26 (81%) | Hospital absolute difference −30.8 percentage points; P=0.02. A later, larger immunocompromised-patient trial did not reproduce the mortality effect.small single-centre trialnot confirmed |
| Auriant NIV2001 | Acute hypoxaemic respiratory failure after lung resection; n=48 | Nasal-mask non-invasive ventilation plus standard therapy (n=24) | Standard therapy alone (n=24) | Endotracheal ventilation: 5/24 (20.8%) vs 12/24 (50.0%); P=0.035 | 3/24 (12.5%) | 9/24 (37.5%) | In-hospital absolute difference −25.0 percentage points; P=0.045.small single-centre trialborderline |
| Ferrer hypoxaemic NIV2003 | Severe acute hypoxaemic respiratory failure without COPD or cardiogenic pulmonary oedema; n=105 | Non-invasive ventilation plus high-concentration oxygen (n=51) | High-concentration oxygen alone (n=54) | Intubation: 13/51 (25%) vs 28/54 (52%); P=0.010 | 9/51 (18%) | 21/54 (39%) | ICU mortality P=0.028; cumulative 90-day survival P=0.025. Adjusted OR for 90-day death 0.39; P=0.017. |
| Confalonieri hydrocortisone2005 | Severe community-acquired pneumonia; n=46 | Hydrocortisone bolus then 10 mg/hour for 7 days (n=23) | Placebo plus standard care (n=23) | By day 8: improved PaO₂/FiO₂, P=0.002; lower MODS score, P=0.003; less delayed shock, P=0.001 | 0/23 (0%) | 7/23 (30%) | Hospital mortality P=0.009. Preliminary group-sequential trial with multiple primary variables and no reported multiplicity adjustment.small trialmultiple endpoints |
| Nava prophylactic NIV2005 | High-risk patients after a successful spontaneous-breathing trial and planned extubation; n=97 | Immediate prophylactic non-invasive ventilation (n=48) | Standard medical therapy (n=49) | Reintubation: 4/48 (8%) vs 12/49 (24%); adjusted risk difference −16%; P=0.027 | 3/48 (6%) | 9/49 (18%) | Crude ICU-mortality difference P=0.064; a prespecified mediation model estimated a total mortality-risk reduction of 10 percentage points, P=0.01. Hospital mortality was not significant.model-derived only |
| Ferrer early post-extubation NIV2006 | Patients at increased risk of post-extubation respiratory failure; n=162 | Non-invasive ventilation for 24 hours after extubation (n=79) | Conventional oxygen management (n=83) | Post-extubation respiratory failure: 13/79 (16%) vs 27/83 (33%); P=0.029 | 2/79 (3%) | 12/83 (14%) | ICU mortality P=0.015. Overall 90-day survival was not significantly different; the 90-day benefit appeared only in the hypercapnic subgroup.ICU endpoint onlyno sustained overall benefit |
| Meduri early ARDS2007 | Early severe ARDS within 72 hours; n=91 | Prolonged low-dose methylprednisolone infusion (n=63) | Placebo (n=28) | ≥1-point improvement in lung-injury score by day 7: 69.8% vs 35.7%; P=0.002 | 13/63 (20.6%) | 12/28 (42.9%) | ICU mortality absolute difference −22.3 percentage points; P=0.03. Small 2:1 trial with baseline imbalance in catecholamine-dependent shock; larger corticosteroid trials have produced mixed results.small trial |
| ABC2008 | Mechanically ventilated medical ICU patients; n=336 | Paired daily spontaneous-awakening and spontaneous-breathing trials (n=167) | Daily spontaneous-breathing trials with usual sedation care (n=168) | Ventilator-free days to day 28: 14.7 vs 11.6; difference 3.1 days (95% CI 0.7–5.6); P=0.02 | 74/167 (44%) | 97/168 (58%) | One-year HR 0.68 (95% CI 0.50–0.92); P=0.01. Twenty-eight-day mortality was not significantly different. |
| Ferrer hypercapnic post-extubation NIV2009 | Chronic respiratory disease with hypercapnia during a spontaneous-breathing trial; n=106 | Non-invasive ventilation for 24 hours after extubation (n=54) | Conventional oxygen therapy (n=52) | Respiratory failure within 72 hours: 8/54 (15%) vs 25/52 (48%); OR 5.32 for control vs NIV (95% CI 2.11–13.46); P<0.0001 | 6/54 (11%) | 16/52 (31%) | Ninety-day survival favoured NIV; log-rank P=0.0146. Highly selected hypercapnic chronic-respiratory-disease population. |
| Patel helmet NIV2016 | Adults with ARDS requiring face-mask non-invasive ventilation; n=83 | Helmet-delivered non-invasive ventilation (n=44) | Face-mask non-invasive ventilation (n=39) | Intubation: 8/44 (18.2%) vs 24/39 (61.5%); difference −43.3% (95% CI −62.4 to −24.3); P<0.001 | 15/44 (34.1%) | 22/39 (56.4%) | Ninety-day difference −22.3 percentage points (95% CI −43.3 to −1.4); P=0.02. Single-centre trial stopped early after 83 patients.early stopsingle centre |
| DEXA-ARDS2020 | Established moderate-to-severe non-COVID ARDS; n=277 | Dexamethasone 20 mg daily then 10 mg daily (n=139) | Usual care without protocolised corticosteroid (n=138) | Ventilator-free days by day 28: 12.3 vs 7.5; difference 4.8 days (95% CI 2.57–7.03); P<0.0001 | 29/139 (20.9%) | 50/138 (36.2%) | Day-60 difference −15.3 percentage points (95% CI −25.9 to −4.9); P=0.0047. Recruitment stopped at 277 of 314 planned because enrolment was slow.under-enrolled |
| ACTT-12020 | Adults hospitalised with COVID-19 and lower-respiratory-tract involvement; n=1,062 | Remdesivir for up to 10 days (n=541) | Placebo (n=521) | Median recovery 10 vs 15 days; recovery rate ratio 1.29 (95% CI 1.12–1.49); P<0.001 | Day 15: 6.7% | Day 15: 11.9% | Day-15 HR 0.55 (95% CI 0.36–0.83). The definitive day-29 comparison was not significant: 11.4% vs 15.2%; HR 0.73 (95% CI 0.52–1.03).early timepoint onlyno day-29 benefit |
| REMAP-CAP IL-6 antagonists2021 | Critically ill adults with COVID-19 receiving organ support; initial efficacy cohort n=803 | Tocilizumab (n=350) or sarilumab (n=45) | No immune modulator / standard care (n=397) | Median organ-support-free days: 10, 11 and 0; adjusted OR vs control 1.64 (95% CrI 1.25–2.14) and 1.76 (1.17–2.91); superiority >99.9% and 99.5% | 28.0%; 22.2% | 35.8% | Adjusted OR for hospital survival: 1.64 (95% CrI 1.14–2.35) for tocilizumab and 2.01 (1.18–4.71) for sarilumab; posterior superiority 99.6% and 99.5%. Adaptive Bayesian platform with concurrent-domain adjustment.Bayesianadaptive platform |
| Liver | |||||||
| Monescillo early TIPS2004 | Acute variceal bleeding with hepatic venous pressure gradient ≥20 mm Hg; n=52 | Early bare-stent TIPS (n=26) | Somatostatin, sclerotherapy and secondary prophylaxis (n=26) | Treatment failure: 3/26 (12%) vs 13/26 (50%); P=0.003 | 1 year: 31% | 1 year: 65% | One-year absolute difference −34 percentage points; P=0.01. In-hospital mortality was 11% vs 31%, P=0.02. Single-centre trial used invasive haemodynamic selection and an older control regimen.small single-centre trial |
| García-Pagán early TIPS2010 | High-risk acute variceal bleeding: Child–Pugh C 10–13 or B with active bleeding; n=63 | Covered TIPS within 72 hours (n=32) | Vasoactive therapy plus endoscopic band ligation (n=31) | Failure to control bleeding or rebleeding: 1/32 (3%) vs 14/31 (45%); P=0.001 | 4/32 (12.5%) | 12/31 (38.7%) | Deaths P=0.01; one-year actuarial survival 86% vs 61%, P<0.001. Small trial stopped early after interim benefit.early stopsmall trial |
| Sepsis | |||||||
| Morelli esmolol2013 | Septic shock with persistent heart rate ≥95/min after haemodynamic optimisation and requiring high-dose norepinephrine; n=154 | Esmolol infusion titrated to a heart rate of 80–94/min (n=77) | Standard treatment alone (n=77) | Heart-rate target achieved in all intervention patients; median 96-hour heart-rate AUC −28/min vs −6/min; P<0.001 | 38/77 (49.4%) | 62/77 (80.5%) | Twenty-eight-day adjusted HR 0.39 (95% CI 0.26–0.59); P<0.001. Single-centre, open-label phase 2 trial with exceptionally high comparator mortality; mortality was not the powered primary endpoint.phase 2single-centre |
Interpretation: these are secondary mortality findings and should not be given the same confirmatory weight as a prespecified primary mortality endpoint. Unless stated otherwise, P values and confidence intervals are nominal and were not protected by a trial-wide multiplicity procedure. “Component analysis” denotes mortality extracted from a broader positive composite or ordinal endpoint; “model-derived only” denotes a significant adjusted or mediated estimate despite a non-significant crude mortality comparison.
Table 3. Statistically significant mortality benefit as a secondary outcome in trials with a neutral or unfavourable primary outcome
Inclusion rule: adult randomised trials relevant to acute or critical care in which the prespecified primary efficacy outcome was not statistically positive and a separate secondary overall, time-specific, or cause-specific mortality endpoint favoured the intervention. Mortality-primary trials, trials with a positive primary outcome, and subgroup-only mortality findings are excluded.
| Trial | Population | Intervention | Comparator | Neutral or unfavourable primary endpoint | Mortality — intervention | Mortality — comparator | Mortality effect and qualification |
|---|---|---|---|---|---|---|---|
| Neuro | |||||||
| DECIMAL2007 | Malignant middle-cerebral-artery infarction in patients aged 18–55 years; n=38 | Early decompressive craniectomy plus standard medical care (n=20) | Standard medical care alone (n=18) | mRS 0–3 at 6 months: 5/20 (25.0%) vs 1/18 (5.6%); P=0.18 | 5/20 (25.0%) | 14/18 (77.8%) | Six-month absolute difference −52.8 percentage points; P<0.0001. A very small sequential trial; functional independence was not statistically improved.small sequential trial |
| DESTINY2007 | Malignant middle-cerebral-artery infarction in patients aged 18–60 years; n=32 | Early decompressive hemicraniectomy plus intensive care (n=17) | Conservative intensive care (n=15) | mRS 0–3 at 6 months: 8/17 (47%) vs 4/15 (27%); P=0.23 | 30 days: 2/17 (12%) | 30 days: 8/15 (53%) | Thirty-day mortality favoured surgery; P=0.02. Mortality formed the first step of the sequential monitoring plan, and the trial ended after the pooled European analysis.sequential designsmall trial |
| HAMLET2009 | Space-occupying middle-cerebral-artery infarction randomised within 96 hours; n=64 | Decompressive surgery plus best medical treatment (n=32) | Best medical treatment alone (n=32) | mRS 0–3 at 1 year: 8/32 (25%) vs 8/32 (25%); absolute difference 0% (95% CI −21 to 21) | 7/32 (22%) | 19/32 (59%) | One-year absolute reduction 38 percentage points (95% CI 15–60); P=0.002. Survival increased without improving the prespecified disability threshold; benefit was concentrated among patients treated within 48 hours.survival–disability trade-off |
| CLEAR III2017 | Spontaneous intraventricular haemorrhage managed with an external ventricular drain; n=500 | Intraventricular alteplase (n=249) | Intraventricular saline (n=251) | mRS 0–3 at 180 days: 48% vs 45%; RR 1.06 (95% CI 0.88–1.28); P=0.554 | 46/249 (18%) | 73/251 (29%) | Day-180 adjusted HR 0.60 (95% CI 0.41–0.86); P=0.006. More alteplase recipients survived with mRS 5: 17% vs 9%; RR 1.99 (95% CI 1.22–3.26).survival–disability trade-off |
| TICH-22018 | Adults with acute spontaneous intracerebral haemorrhage treated within 8 hours; n=2,325 | Tranexamic acid 1 g bolus then 1 g over 8 hours (n=1,161) | Placebo (n=1,164) | Ninety-day mRS shift: adjusted OR 0.88 (95% CI 0.76–1.03); P=0.11 | Day 7: 101/1,161 (9%) | Day 7: 123/1,164 (11%) | Day-7 adjusted OR 0.73 (95% CI 0.53–0.99); P=0.0406. No sustained day-90 benefit: 250/1,161 (22%) vs 249/1,164 (21%); HR 0.92; P=0.37.early timepoint onlyno day-90 benefit |
| MISTIE III2019 | Spontaneous supratentorial intracerebral haemorrhage ≥30 mL; n=506 | Image-guided catheter evacuation plus alteplase (n=255) | Standard medical care (n=251) | mRS 0–3 at 365 days: 45% vs 41%; adjusted absolute difference 4% (95% CI −4 to 12); P=0.33 | Day 7: 2/255 (1%) | Day 7: 10/251 (4%) | Day-7 P=0.02. Day-30 mortality was not statistically different (P=0.07); a prespecified adjusted 365-day Cox model favoured MISTIE (HR 0.67, 95% CI 0.45–0.98; P=0.037), whereas the unadjusted log-rank P value was 0.084.early timepointmodel-dependent long-term signal |
| Circulatory | |||||||
| SHOCK1999 | Acute myocardial infarction complicated by cardiogenic shock; n=302 | Emergency revascularisation (n=152) | Initial medical stabilisation (n=150) | Thirty-day all-cause mortality: 46.7% vs 56.0%; absolute difference −9.3% (95% CI −20.5 to 1.9); P=0.11 | 6 months: 50.3% | 6 months: 63.1% | Six-month absolute difference −12.8 percentage points; P=0.027. This was a later survival endpoint after the primary 30-day comparison did not reach statistical significance.delayed benefit |
| RUSSLAN2002 | Left-ventricular failure complicating acute myocardial infarction; n=504 | Six-hour levosimendan infusion in four dose groups | Placebo infusion | Clinically significant hypotension or myocardial ischaemia was similar across groups; P=0.319 | Day 14: 11.7% | Day 14: 19.6% | Fourteen-day HR 0.56 (95% CI 0.33–0.95); P=0.031. Results pooled four levosimendan doses; the retrospective 180-day result was borderline: 22.6% vs 31.4%; P=0.053.pooled dose groupssecondary endpoint |
| CRISTAL2013 | ICU patients presenting with hypovolaemic shock; n=2,857 | Colloid solutions for fluid resuscitation during the ICU stay (n=1,414) | Crystalloid solutions (n=1,443) | Twenty-eight-day mortality: 359/1,414 (25.4%) vs 390/1,443 (27.0%); RR 0.96 (95% CI 0.88–1.04); P=0.26 | Day 90: 434/1,414 (30.7%) | Day 90: 493/1,443 (34.2%) | Ninety-day RR 0.92 (95% CI 0.86–0.99); P=0.03. Exploratory secondary endpoint in an open-label pragmatic comparison involving several different colloid and crystalloid solutions.exploratory endpointheterogeneous fluids |
| Respiratory | |||||||
| FLORALI2015 | Non-hypercapnic acute hypoxaemic respiratory failure; n=310 | High-flow nasal oxygen (n=106) | Standard oxygen (n=94) or non-invasive ventilation (n=110) | Intubation by day 28: 40/106 (38%) vs 44/94 (47%) vs 55/110 (50%); overall P=0.18 | 90 days: 13/106 (12.3%) | 22/94 (23.4%); 31/110 (28.2%) | Overall 90-day survival P=0.02. HR for death was 2.01 (95% CI 1.01–3.99; P=0.046) with standard oxygen and 2.50 (1.31–4.78; P=0.006) with NIV, each versus high-flow oxygen. Open-label, three-arm trial; later studies have not consistently reproduced the survival signal.three-arm trialnot confirmed |
| COV-BARRIER2021 | Adults hospitalised with COVID-19 receiving standard care; n=1,525 | Baricitinib plus standard care (n=764) | Placebo plus standard care (n=761) | Progression to high-flow oxygen, NIV, invasive ventilation, or death by day 28: 27.8% vs 30.5%; OR 0.85 (95% CI 0.67–1.08); P=0.18 | Day 28: 62/764 (8.1%) | Day 28: 100/761 (13.1%) | Day-28 HR 0.57 (95% CI 0.41–0.78); nominal P=0.0018. Mortality was a key secondary endpoint, but formal success of the primary endpoint was not achieved.nominal secondary analysis |
| Nutrition | |||||||
| Refeeding Syndrome Trial2015 | Critically ill adults who developed refeeding syndrome within 72 hours of starting nutritional support; n=339 | Protocolised caloric restriction during correction of refeeding syndrome (n=169) | Continued standard caloric intake (n=170) | Mean days alive after ICU discharge through day 60: 44.8 vs 39.9; difference 4.9 days (95% CI −2.3 to 13.6); P=0.19 | Day-60 survival: 149/164 (91%) | Day-60 survival: 128/163 (78%) | Day-60 survival P=0.002; overall survival time also favoured caloric restriction, log-rank P=0.002. Survival was an individual component of the neutral composite primary outcome.composite component |
| Renal | |||||||
| STOP-AKI2018 | Critically ill adults with sepsis-associated acute kidney injury; adaptive dose-finding trial n=301 | Recombinant human alkaline phosphatase 1.6 mg/kg for 3 days (n=111) | Placebo (n=116) | Mean daily endogenous creatinine-clearance AUC over days 1–7: treatment difference 9.5 mL/min (95% CI −23.9 to 25.5); P=0.47 | Day 28: 16/111 (14.4%) | Day 28: 31/116 (26.7%) | Day-28 absolute difference −12.3 percentage points (95% CI −22.7 to −1.9); P=0.02; day-90 P=0.03. Exploratory phase-2, dose-selected comparison; the subsequent phase-3 REVIVAL trial reported identical 28-day mortality of 27.9% in both groups.phase 2 dose selectionnot confirmed |
| Haematological | |||||||
| TRICC1999 | Euvolaemic critically ill adults with haemoglobin <9 g/dL; n=838 | Restrictive transfusion threshold of 7 g/dL (n=418) | Liberal transfusion threshold of 10 g/dL (n=420) | Thirty-day mortality: 78/418 (18.7%) vs 98/420 (23.3%); absolute difference −4.7% (95% CI −10.2 to 0.8); P=0.11 | Hospital: 93/418 (22.2%) | Hospital: 118/420 (28.1%) | Hospital mortality absolute difference −5.8 percentage points; reported P=0.05. The corresponding rounded 95% CI crossed the null (−11.7 to 0.3 percentage points), making this a borderline nominal finding.borderline P=0.05CI crosses null |
| Sepsis | |||||||
| CITRIS-ALI2019 | Sepsis with acute respiratory distress syndrome; n=167 | Intravenous vitamin C 50 mg/kg every 6 hours for 96 hours (n=84) | Placebo (n=83 randomised; n=82 in mortality analysis) | Mean modified SOFA change to 96 hours: between-group difference −0.10 (95% CI −1.23 to 1.03); P=0.86; the co-primary biomarker comparisons were also neutral | Day 28: 25/84 (29.8%) | Day 28: 38/82 (46.3%) | Absolute difference −16.6 percentage points (95% CI −31.1 to −2.0); χ² P=0.03. Mortality was one of 46 secondary outcomes with no multiplicity adjustment; later vitamin-C trials did not confirm a survival benefit.46 secondary outcomesno multiplicity adjustmentnot confirmed |
| Trauma | |||||||
| PROPPR2015 | Severely injured patients predicted to require massive transfusion; n=680 | Plasma:platelets:red cells in a 1:1:1 ratio (n=338) | Plasma:platelets:red cells in a 1:1:2 ratio (n=342) | All-cause mortality at 24 hours: 12.7% vs 17.0%; P=0.12; at 30 days: 22.4% vs 26.1%; P=0.26 | Exsanguination: 31/338 (9.2%) | Exsanguination: 50/342 (14.6%) | Death from exsanguination within 24 hours differed by −5.4 percentage points (95% CI −10.4 to −0.5); P=0.03. This was cause-specific mortality; both all-cause mortality co-primary endpoints were neutral.cause-specific mortalityneutral all-cause mortality |
| WOMAN2017 | Women with clinically diagnosed postpartum haemorrhage; n=20,060 randomised | Tranexamic acid plus usual care (n=10,036 in bleeding-death analysis) | Placebo plus usual care (n=9,985 in bleeding-death analysis) | All-cause death or hysterectomy: 534/10,051 (5.3%) vs 546/10,009 (5.5%); RR 0.97 (95% CI 0.87–1.09); P=0.65 | Bleeding death: 155/10,036 (1.5%) | Bleeding death: 191/9,985 (1.9%) | Death due to bleeding RR 0.81 (95% CI 0.65–1.00); P=0.045. All-cause mortality was not statistically different (2.3% vs 2.6%; P=0.16); the stronger effect with treatment within 3 hours is a subgroup finding and is not used here.cause-specific mortalityborderline nominal finding |
| PATCH-Trauma2023 | Adults with major trauma and suspected trauma-induced coagulopathy treated within advanced trauma systems; n=1,310 | Prehospital tranexamic acid 1 g followed by a 1-g infusion over 8 hours (n=661 assigned) | Matching placebo (n=646 assigned) | Favourable functional survival at 6 months: 307/572 (53.7%) vs 299/559 (53.5%); RR 1.00 (95% CI 0.90–1.12); P=0.95 | Day 28: 113/653 (17.3%) | Day 28: 139/637 (21.8%) | Twenty-eight-day RR 0.79 (95% CI 0.63–0.99). Six-month mortality was not significantly different: 19.0% vs 22.9%; RR 0.83 (95% CI 0.67–1.03).secondary endpointno sustained benefit |
Interpretation: these findings require more caution than a prespecified primary mortality endpoint because the primary efficacy outcome was not statistically positive. Unless stated otherwise, P values and confidence intervals are nominal and were not protected by a trial-wide multiplicity procedure. Early-timepoint, delayed, cause-specific, model-dependent, and failure of subsequent confirmation are identified explicitly; no subgroup-only mortality result is used to qualify a trial for this table.
Table 4. Statistically significant mortality benefit in a subgroup analysis
Inclusion rule: adult randomised trials relevant to acute or critical care in which the designated overall analysis did not establish a mortality benefit, but mortality was significantly lower within a patient subgroup or a treatment-by-subgroup interaction supported a differential mortality effect. Trials already classified by a statistically significant overall primary or secondary mortality result are not repeated here.
| Trial | Population | Intervention | Comparator | Overall or primary result | Mortality — intervention subgroup | Mortality — comparator subgroup | Subgroup mortality effect and qualification |
|---|---|---|---|---|---|---|---|
| Circulatory | |||||||
| ATHOS-32020 subgroup analysis | Catecholamine-resistant vasodilatory shock; parent trial modified intention-to-treat n=321 | Intravenous angiotensin II plus standard vasopressors (n=163) | Placebo plus standard vasopressors (n=158) | MAP response at 3 hours: 69.9% vs 23.4%; P<0.001. Overall 28-day mortality: 46% vs 54%; HR 0.78 (95% CI 0.57–1.07); P=0.12 | High renin: 28/55 (50.9%) | High renin: 51/73 (69.9%) | Above-median baseline renin: HR 0.56 (95% CI 0.35–0.88); P=0.012; treatment-by-renin interaction P=0.048. Retrospective biomarker analysis with a sample-derived threshold; not prospectively validated.post hoc biomarker subgroupinteraction P=0.048 |
| 65 Trial2020 | ICU patients aged ≥65 years with vasodilatory hypotension; n=2,598 randomised, n=2,463 in the primary analysis | Permissive hypotension: MAP target 60–65 mm Hg | Usual-care vasopressor management | Ninety-day mortality: 500/1,221 (41.0%) vs 544/1,242 (43.8%); absolute difference −2.85% (95% CI −6.75 to 1.05); P=0.154 | Chronic hypertension: 214/560 (38.2%) | Chronic hypertension: 253/571 (44.3%) | Adjusted OR 0.67 (95% CI 0.51–0.88); interaction P=0.047. Prespecified subgroup, but not adjusted for multiplicity. An adjusted overall sensitivity model also crossed the nominal threshold (OR 0.82, 95% CI 0.68–0.98), unlike the designated unadjusted primary analysis.prespecified subgroupno multiplicity adjustment |
| Resuscitation | |||||||
| ALPS2016 | Shock-refractory ventricular fibrillation or pulseless ventricular tachycardia after out-of-hospital cardiac arrest; primary analysis n=3,026 | Parenteral amiodarone or lidocaine during resuscitation | Saline placebo | Survival to discharge: 24.4% with amiodarone, 23.7% with lidocaine, and 21.0% with placebo; amiodarone P=0.08 and lidocaine P=0.16 versus placebo | Bystander-witnessed: 72.3% amiodarone; 72.2% lidocaine | Bystander-witnessed: 77.3% | Survival was 27.7% with amiodarone and 27.8% with lidocaine versus 22.7% with placebo; absolute benefit 5.0 percentage points (P=0.04) and 5.2 points (P=0.03), respectively. Prespecified subgroup; heterogeneity by witnessed status was marginal (P=0.05).prespecified subgroupmarginal interaction |
| Respiratory | |||||||
| Prone–Supine I2001 | Mechanically ventilated adults with acute lung injury or ARDS; n=304 | Prone positioning for ≥6 hours daily for 10 days (n=152) | Supine positioning (n=152) | Ten-day mortality: 32/152 (21.1%) vs 38/152 (25.0%); RR 0.84 (95% CI 0.56–1.27). ICU-discharge and 6-month mortality were also neutral | PaO₂/FiO₂ ≤88: 23% | PaO₂/FiO₂ ≤88: 47% | Ten-day RR 0.49 (95% CI 0.25–0.95) in the most severely hypoxaemic subgroup. Post-hoc analysis across several severity cuts, limited to an early mortality timepoint; later prolonged-proning trials used materially different protocols.post hoc subgroup10-day endpoint |
| FACTT2017 latent-class analysis | ARDS Network FACTT participants classified retrospectively into hypoinflammatory and hyperinflammatory subphenotypes; parent trial n=1,000 | Conservative fluid-management strategy | Liberal fluid-management strategy | Parent-trial 60-day mortality: 25.5% vs 28.4%; P=0.30; ventilator-free days favoured conservative management | Hyperinflammatory: 40% | Hyperinflammatory: 50% | The treatment-by-subphenotype interaction was significant for 90-day mortality (P=0.0039). Direction reversed in the hypoinflammatory group: 26% with conservative versus 18% with liberal management. Statistical support rests on a retrospective latent-class interaction, not a prospectively enrolled phenotype-specific trial.post hoc latent classqualitative interaction |
| HARP-22018 latent-class analysis | ARDS patients retrospectively classified into hypoinflammatory and hyperinflammatory subphenotypes; parent trial n=540 | Simvastatin 80 mg daily | Placebo | Parent-trial ventilator-free days: 12.6 vs 11.5; P=0.21. Overall 28-day mortality: 22.0% vs 26.8%; RR 0.8 (95% CI 0.6–1.1) | Hyperinflammatory: 32% | Hyperinflammatory: 45% | Twenty-eight-day survival differed within the hyperinflammatory subgroup; P=0.008. The formal treatment-by-subphenotype interaction for mortality was not significant (P=0.14). A later high-IL-18 analysis produced a similar subgroup-specific signal but again lacked a significant interaction.post hoc latent classinteraction not significant |
| Nutrition | |||||||
| VITdAL-ICU2014 | Critically ill adults with vitamin-D deficiency; final analysis n=475 | Enteral vitamin D₃ 540,000 IU loading dose, then 90,000 IU monthly (n=237) | Placebo (n=238) | Median hospital stay: 20.1 vs 19.3 days; P=0.98. Overall hospital mortality: 67/237 (28.3%) vs 84/238 (35.3%); HR 0.81; P=0.18 | 25(OH)D ≤12 ng/mL: 28/98 (28.6%) | 25(OH)D ≤12 ng/mL: 47/102 (46.1%) | Hospital-mortality HR 0.56 (95% CI 0.35–0.90); P=0.01; interaction P=0.04. Predefined subgroup in a single-centre trial; the later multicentre VIOLET trial did not confirm benefit in severe deficiency.predefined subgroupnot confirmed |
| Renal | |||||||
| SMART2019 sepsis analysis | Critically ill adults with sepsis identified within the cluster-randomised SMART trial; n=1,641 in the secondary cohort | Balanced crystalloids (n=824) | 0.9% saline (n=817) | Parent-trial MAKE30: 14.3% vs 15.4%; adjusted OR 0.90 (95% CI 0.82–0.99); P=0.04. Overall 30-day in-hospital mortality: 10.3% vs 11.1%; P=0.06 | Sepsis: 217/824 (26.3%) | Sepsis: 255/817 (31.2%) | Adjusted OR 0.74 (95% CI 0.59–0.93); P=0.01. Prespecified subgroup but published as a secondary analysis of a single-centre, open-label, cluster-crossover trial; the subgroup interaction for the parent composite outcome was not conventionally significant.prespecified subgroupsecondary cohort analysis |
| Sepsis | |||||||
| HA-1A sepsis trial 1991 | Adults with sepsis and suspected Gram-negative infection; n=543. The culture-confirmed Gram-negative bacteraemia subgroup comprised 200 patients. | A single 100 mg intravenous dose of HA-1A plus usual care | Albumin placebo plus usual care | Overall 28-day mortality: 39% versus 43%; P=0.24 | Gram-negative bacteraemia: 32/105 (30%) | Gram-negative bacteraemia: 45/92 (49%) | Twenty-eight-day mortality was lower in the culture-confirmed subgroup; P=0.014. This was a culture-defined subgroup and no formal treatment-by-bacteraemia interaction was reported. The subsequent CHESS trial did not confirm the benefit. culture-defined subgroup not confirmed |
| KyberSept2001 | Adults with severe sepsis or septic shock; n=2,314 | High-dose antithrombin III, 30,000 IU over 4 days (n=1,157) | Albumin placebo (n=1,157) | Twenty-eight-day mortality: 38.9% vs 38.7%; P=0.94 | No concomitant heparin: 44.9% | No concomitant heparin: 52.5% | Ninety-day absolute difference −7.6 percentage points in the no-heparin subgroup (n=680); nominal P=0.03. Heparin exposure was not randomised; high-dose antithrombin increased bleeding when combined with heparin.concomitant-treatment subgroupnominal analysis |
| OPTIMIST2003 | Severe sepsis; primary efficacy cohort with INR ≥1.2 n=1,754, plus a lower-INR safety cohort n=201 | Tifacogin infusion for 96 hours | Placebo | High-INR 28-day mortality: 301/880 (34.2%) vs 296/874 (33.9%); P=0.88 | INR <1.2: 10/83 (12.0%) | INR <1.2: 27/118 (22.9%) | Unadjusted P=0.051, but the prespecified adjusted model gave P=0.03 in the lower-INR cohort. A separate post-hoc no-heparin analysis in high-INR patients was borderline (34.6% vs 42.7%; P=0.05). Neither finding had a clearly significant interaction, and bleeding was more frequent with tifacogin.adjusted-only signalno clear interaction |
| SAFE2011 severe-sepsis analysis | Severe-sepsis subgroup within the SAFE fluid-resuscitation trial; n=1,218 | 4% albumin (n=603) | 0.9% saline (n=615) | Parent-trial 28-day mortality: 20.9% vs 21.1%; RR 0.99 (95% CI 0.91–1.09); P=0.87 | Severe sepsis: 185/603 (30.7%) | Severe sepsis: 217/615 (35.3%) | Unadjusted RR 0.87 (95% CI 0.74–1.02); P=0.09; heterogeneity P=0.06. Significance arose only in a multivariable complete-case model (919/1,218): adjusted OR 0.71 (95% CI 0.52–0.97); P=0.03.adjusted-only signalcomplete-case model |
| VASST2008 | Adults with septic shock receiving norepinephrine; n=778 infused | Low-dose vasopressin added to open-label vasopressors (n=396) | Norepinephrine added to open-label vasopressors (n=382) | Twenty-eight-day mortality: 35.4% vs 39.3%; RR 0.90 (95% CI 0.75–1.08); P=0.26 | Less severe shock, day 90: 69/193 (35.8%) | Less severe shock, day 90: 83/180 (46.1%) | RR 0.78 (95% CI 0.61–0.99); P=0.04 in the prospectively stratified 5–14 μg/min norepinephrine stratum. The test for heterogeneity between severity strata was not significant (P=0.10).prespecified stratuminteraction not significant |
| ALBIOS2014 | Adults with severe sepsis; n=1,818 | 20% albumin plus crystalloid, targeting serum albumin ≥30 g/L | Crystalloid alone | Twenty-eight-day mortality: 31.8% vs 32.0%; P=0.94. Ninety-day mortality: 41.1% vs 43.6%; RR 0.94; P=0.29 | Septic shock, day 90: 243/558 (43.6%) | Septic shock, day 90: 281/563 (49.9%) | RR 0.87 (95% CI 0.77–0.99); P=0.03; heterogeneity P=0.03. Exploratory subgroup; fuller covariate adjustment attenuated the estimate to RR 0.88 (95% CI 0.77–1.01). A subsequent septic-shock albumin trial did not show a mortality benefit.exploratory subgroupadjustment-sensitive |
| Anakinra phase-3 reanalysis2016 | Adults from a previously neutral phase-3 sepsis trial with archived data available for reanalysis; n=763 | Intravenous anakinra | Placebo | The parent trial did not show an overall 28-day survival benefit; among patients without concurrent hepatobiliary dysfunction and DIC, survival was 71.4% vs 70.8%; P=0.88 | HBD plus DIC: 9/26 (34.6%) | HBD plus DIC: 11/17 (64.7%) | HR for death 0.28 (95% CI 0.11–0.71); P=0.0071. Retrospective organ-failure phenotype comprising only 43 patients; requires prospective confirmation.post hoc subgroupn=43 |
| EUPHRATES2018 post-hoc analysis | Endotoxaemic septic shock; post-hoc per-protocol subgroup with MODS >9, EAA 0.60–0.89, and two completed treatments; n=194 | Two polymyxin-B haemoperfusion treatments plus standard care (n=88) | Sham haemoperfusion plus standard care (n=106) | Parent-trial 28-day mortality: 84/223 (37.7%) vs 78/226 (34.5%); P=0.49; the prespecified MODS >9 population was also neutral | 23/88 (26.1%) | 39/106 (36.8%) | Adjusted OR 0.52 (95% CI 0.27–0.99); P=0.047; adjusted survival HR 0.56 (95% CI 0.33–0.95); P=0.03. The unadjusted mortality comparison was not significant (P=0.11); post-hoc biomarker restriction and per-protocol selection make the signal model-dependent.post hoc per protocoladjusted-only signal |
| Trauma | |||||||
| EPO Critical Care2007 | Medical, surgical and trauma ICU patients enrolled 48–96 hours after admission; parent trial n=1,460 | Epoetin alfa 40,000 units weekly for up to 3 weeks (n=733 overall) | Placebo (n=727 overall) | Red-cell transfusion incidence was not reduced: RR 0.95 (95% CI 0.85–1.06). Overall adjusted day-29 mortality HR 0.79 (95% CI 0.56–1.10) | Trauma, day 29: 3.5% | Trauma, day 29: 6.6% | Prespecified trauma analysis: adjusted HR 0.37 (95% CI 0.19–0.72); day-140 adjusted HR 0.40 (95% CI 0.23–0.69). The treatment-by-admission-group interaction was not significant (P=0.16), and thrombotic events increased overall: HR 1.41 (95% CI 1.06–1.86).prespecified subgroupinteraction P=0.16thrombotic harm |
| CRASH-32019 | Adults with traumatic brain injury treated within 3 hours; 9,127 in the primary analysis, including 5,615 with mild-to-moderate injury | Tranexamic acid 1 g loading dose followed by 1 g over 8 hours | Placebo | Head-injury death within 28 days: 855/4,613 (18.5%) vs 892/4,514 (19.8%); RR 0.94 (95% CI 0.86–1.02) | Mild-to-moderate TBI: 166/2,846 (5.8%) | Mild-to-moderate TBI: 207/2,769 (7.5%) | Head-injury-death RR 0.78 (95% CI 0.64–0.95); heterogeneity by injury severity P=0.030. No benefit was evident in severe TBI: RR 0.99 (95% CI 0.91–1.07).prespecified subgroupinteraction P=0.030cause-specific mortality |
| Metabolic | |||||||
| BICAR-ICU2018 | Critically ill adults with severe metabolic acidaemia; n=389 | 4.2% sodium bicarbonate targeting arterial pH >7.30 (n=195) | No protocolised bicarbonate (n=194) | Death by day 28 or ≥1 organ failure at day 7: 128/195 (66%) vs 138/194 (71%); P=0.24. Overall 28-day mortality: 45% vs 54%; P=0.07 | AKIN stage 2–3: 42/92 (46%) | AKIN stage 2–3: 57/90 (63%) | Twenty-eight-day mortality P=0.0166 in the prespecified acute-kidney-injury stratum; the composite primary outcome also differed within the stratum (P=0.0462). Open-label subgroup result; BICARICU-2 subsequently found no 90-day mortality benefit in this targeted phenotype (62.1% vs 61.7%; P=0.91).prespecified subgroupnot confirmed |
Interpretation: subgroup findings are inherently less secure than an overall prespecified comparison. A subgroup-specific P value does not prove that treatment effects differ between subgroups; the formal interaction test is therefore reported whenever available. Prespecified strata, post-hoc analyses, multiplicity, model dependence, very small samples, and failure of subsequent confirmation are identified explicitly. These signals are best treated as hypothesis-generating unless confirmed in a trial prospectively restricted to the proposed responder population.
Table 5. Statistically significant mortality harm as a primary or co-primary outcome
Inclusion rule: adult randomised trials relevant to acute or critical care in which mortality or survival was a prespecified primary, co-primary, or principal mortality endpoint and the planned analysis crossed its prespecified statistical threshold in favour of the comparator. Trials in which mortality was only a secondary endpoint, a component of a broader primary endpoint, or a subgroup finding are classified in later tables.
| Trial | Population | Intervention | Comparator | Primary mortality / survival endpoint | Intervention result | Comparator result | Effect and qualification |
|---|---|---|---|---|---|---|---|
| Neuro | |||||||
| MRC CRASH2004 | Adults with clinically significant head injury and Glasgow Coma Scale score ≤14; n=10,008 randomised | Methylprednisolone 2 g over one hour, then 0.4 g/hour for 48 hours (n=5,007) | Matched placebo (n=5,001) | All-cause death within two weeks | 1,052/4,985 (21.1%) | 893/4,979 (17.9%) | RR 1.18 (95% CI 1.09–1.27); P=0.0001. Recruitment was stopped after the excess-mortality signal; six-month mortality was also higher with corticosteroids.early stop |
| Circulatory | |||||||
| Hayes oxygen-delivery trial1994 | Heterogeneous critically ill adults at increased risk of death who did not reach supranormal haemodynamic goals with fluid alone; n=100 randomised | Dobutamine titrated to supranormal cardiac-index, oxygen-delivery and oxygen-consumption targets (n=50) | Usual haemodynamic care; dobutamine only for cardiac index <2.8 L/min/m² (n=50) | In-hospital mortality; principal clinical endpoint | 27/50 (54%) | 17/50 (34%) | Absolute increase 20 percentage points (95% CI 0.9–39.1); P=0.04. The trial was stopped at interim review; ICU mortality was also 50% versus 30%.early stopsmall trial |
| Resuscitation | |||||||
| SSSP-22017 | Zambian adults with sepsis and hypotension; 209 included in the analysis, 89.5% living with HIV | Six-hour protocol of intravenous-fluid boluses, vasopressors and transfusion when indicated (n=106) | Usual clinician-directed haemodynamic care (n=103) | In-hospital mortality | 51/106 (48.1%) | 34/103 (33.0%) | RR 1.46 (95% CI 1.04–2.05); P=0.03. Day-28 mortality was also higher. Almost all participants were managed on general wards without mechanical ventilation, limiting transferability to highly resourced ICUs.resource-limited setting |
| Respiratory | |||||||
| BALTI-22012 | Intubated adults within 72 hours of ARDS onset; n=326 randomised, n=324 with primary-outcome data | Intravenous salbutamol 15 μg/kg ideal body weight/hour for up to seven days (n=161 analysed) | Saline placebo (n=163 analysed) | All-cause death by day 28 | 55/161 (34%) | 38/163 (23%) | RR 1.47 (95% CI 1.03–2.08); P=0.03. Recruitment stopped after the second interim analysis with fewer than one quarter of the planned sample; 12-month mortality was not significantly different.early stop |
| OSCILLATE2013 | Adults with early moderate-to-severe ARDS; n=548 | Early high-frequency oscillatory ventilation using an open-lung, high-mean-airway-pressure strategy (n=275) | Protocolised lung-protective conventional ventilation (n=273) | In-hospital mortality | 129/275 (47%) | 96/273 (35%) | RR 1.33 (95% CI 1.09–1.64); P=0.005. Stopped after 548 of the planned 1,200 participants because mortality was higher with oscillation.early stop |
| ART2017 | Adults with moderate-to-severe ARDS of less than 72 hours’ duration; n=1,010 | Lung-recruitment manoeuvre followed by compliance-titrated PEEP (n=501) | Conventional low-PEEP strategy (n=509) | All-cause death by day 28 | 277/501 (55.3%) | 251/509 (49.3%) | HR 1.20 (95% CI 1.01–1.42); P=0.041. Six-month mortality, barotrauma and early haemodynamic instability were also increased; the primary result was statistically fragile.borderlinebundled strategy |
| RECOVERY higher-dose corticosteroids2023 | Adults hospitalised with hypoxic COVID-19 receiving no oxygen or simple oxygen only; n=1,272 | Dexamethasone 20 mg daily for five days, then 10 mg daily for five days (or equivalent) (n=659) | Usual care, usually including low-dose corticosteroids (n=613) | All-cause death by day 28 | 123/659 (19%) | 75/613 (12%) | Rate ratio 1.59 (95% CI 1.20–2.10); P=0.0012. The comparison was stopped early for harm. This result applies to patients without ventilatory support; the later ventilatory-support cohort was neutral.early stopoxygen-status specific |
| Sepsis | |||||||
| TNFR:Fc fusion protein1996 | Adults with septic shock; n=141 | Single infusion of TNFR:Fc 0.15 mg/kg (n=30), 0.45 mg/kg (n=29), or 1.5 mg/kg (n=49) | Placebo (n=33) | All-cause death by day 28; dose-response analysis | 30%, 48%, and 53% | 10/33 (30%) | Mortality rose across placebo and increasing doses; dose-response P=0.02. The pooled comparison of all active doses against placebo was not significant (P=0.13), so classification depends on the prespecified trend analysis.dose-response onlysmall phase 2 trial |
| 546C882004 | Adults with septic shock requiring vasopressors; n=797 | Intravenous non-selective nitric-oxide-synthase inhibitor 546C88 for up to 72 hours (n=439) | Placebo infusion (n=358) | All-cause death by day 28 | 259/439 (59%) | 174/358 (49%) | P<0.001. The data-monitoring board stopped the study early; cardiovascular adverse events accounted for much of the between-group imbalance.early stop |
| Trauma | |||||||
| DCLHb traumatic shock trial1999 | Adults with severe traumatic haemorrhagic shock; 112 enrolled and 98 infused | Diaspirin cross-linked haemoglobin as an adjunct to standard resuscitation (n=52 infused) | Saline plus standard resuscitation (n=46 infused) | All-cause death by day 28 | 24/52 (46%) | 8/46 (17%) | OR 4.1 (95% CI 1.6–10.0); P=0.003. The trial was stopped early; the published efficacy comparison was based on the 98 participants who received an infusion.early stopinfused analysis |
| Endocrine | |||||||
| Takala growth-hormone trials1999 | Adults with prolonged critical illness after cardiac surgery, abdominal surgery, trauma or acute respiratory failure; two concurrent trials, total n=532 | High-dose recombinant human growth hormone until ICU discharge or for up to 21 days | Matched placebo | In-hospital mortality in each trial | Finnish: 39%; multinational: 44% | Finnish: 20%; multinational: 18% | Finnish RR 1.9 (95% CI 1.3–2.9) and multinational RR 2.4 (1.6–3.5); P<0.001 in each trial. Two independent randomised comparisons were reported together.two concurrent trials |
| NICE-SUGAR2009 | Medical and surgical ICU adults expected to require at least three days of critical care; n=6,104 | Intensive glucose control targeting 81–108 mg/dL (n=3,054 allocated) | Conventional glucose control targeting ≤180 mg/dL (n=3,050 allocated) | All-cause death by day 90 | 829/3,010 (27.5%) | 751/3,012 (24.9%) | OR 1.14 (95% CI 1.02–1.28); P=0.02. Severe hypoglycaemia was also substantially more frequent with intensive control. |
Interpretation: placement in this table is a statistical classification, not proof that every element of a complex strategy caused the excess deaths. Early stopping can exaggerate effect estimates, and several trials tested bundled interventions or highly specific populations. TNFR:Fc qualifies through its prespecified dose-response analysis despite a non-significant pooled active-versus-placebo comparison. Mortality results that were secondary, component-only, or subgroup analyses are reserved for Tables 6–8.
Table 6. Statistically significant mortality harm as a secondary outcome in trials with a favourable primary outcome
Inclusion rule: adult randomised trials relevant to acute or critical care in which the prespecified primary efficacy outcome was statistically favourable and a separate secondary, component, or supportive analysis found higher mortality or lower survival. Mortality-primary trials, trials with a neutral or unfavourable primary outcome, and subgroup-only mortality findings are excluded.
| Trial | Population | Intervention | Comparator | Positive primary endpoint | Mortality — intervention | Mortality — comparator | Mortality effect and qualification |
|---|---|---|---|---|---|---|---|
| Circulatory | |||||||
| POISE2008 | Adults aged ≥45 years with, or at risk of, atherosclerotic disease undergoing non-cardiac surgery; n=8,351 | Extended-release metoprolol succinate started 2–4 hours before surgery and continued for 30 days (n=4,174) | Matching placebo (n=4,177) | Cardiovascular death, non-fatal myocardial infarction, or non-fatal cardiac arrest by day 30: 244/4,174 (5.8%) vs 290/4,177 (6.9%); HR 0.84 (95% CI 0.70–0.99); P=0.0399 | 129/4,174 (3.1%) | 97/4,177 (2.3%) | Thirty-day all-cause mortality HR 1.33 (95% CI 1.03–1.74); P=0.0317. The primary benefit was driven by fewer myocardial infarctions, while stroke also increased (1.0% vs 0.5%; HR 2.17, 95% CI 1.26–3.74; P=0.0053).secondary endpointbenefit–harm trade-off |
Interpretation: this is a secondary mortality finding and should not be given the same confirmatory weight as a prespecified primary mortality endpoint. The mortality comparison was nominal and sits within a wider benefit–harm profile: fewer perioperative myocardial infarctions were offset by more deaths and strokes.
Table 7. Statistically significant mortality harm as a secondary outcome in trials with a neutral or unfavourable primary outcome
Inclusion rule: adult randomised trials relevant to acute or critical care in which the prespecified primary efficacy outcome was not statistically favourable and a separate secondary overall, time-specific, cause-specific, component, or supportive mortality analysis found higher mortality or lower survival with the intervention. Trials in which the qualifying mortality finding was the primary analysis, trials with a statistically favourable primary outcome, and subgroup-only mortality findings are excluded.
| Trial | Population | Intervention | Comparator | Neutral or unfavourable primary endpoint | Mortality — intervention | Mortality — comparator | Mortality effect and qualification |
|---|---|---|---|---|---|---|---|
| Neuro | |||||||
| MAST-I1995 | Acute ischaemic stroke treated within 6 hours; 2×2 factorial trial, n=622 | Intravenous streptokinase 1.5 million units, alone or with aspirin | No streptokinase, with the embedded aspirin allocation retained | Six-month death or severe disability was not reduced: OR 0.9 (95% CI 0.7–1.3) | Day 10: 27% | Day 10: 12% | Ten-day case fatality OR 2.7 (95% CI 1.7–4.3); 2P<0.00001. This was the streptokinase main effect in a factorial trial and the excess was early; the six-month composite remained neutral.early mortalityfactorial main effect |
| ECASS I1995 | Acute hemispheric ischaemic stroke treated within 6 hours; n=620 | Intravenous alteplase 1.1 mg/kg (n=313) | Placebo (n=307) | Barthel Index and modified Rankin Scale at 90 days showed no significant difference in the intention-to-treat analysis | Day 90: 22.4% | Day 90: 15.8% | Ninety-day mortality was higher with alteplase; log-rank P=0.04. The trial used a higher alteplase dose than current practice, and important protocol violations complicated the efficacy analyses.secondary safety outcomeearly thrombolysis era |
| MAST-E1996 | Acute ischaemic stroke treated within 6 hours; n=310 | Intravenous streptokinase 1.5 million units (n=156) | Placebo (n=154) | Death or severe disability at 6 months: 124/156 (79.5%) vs 126/154 (81.8%); no significant difference | Day 10: 53/156 (34.0%) | Day 10: 28/154 (18.2%) | Ten-day mortality difference P=0.002, driven mainly by haemorrhagic transformation. Six-month mortality was not statistically different (46.8% vs 38.3%; P=0.13), and the trial stopped early.early mortalitystopped early |
| EPO Stroke2009 | Acute middle-cerebral-artery ischaemic stroke treated within 6 hours; n=522 | Erythropoietin 40,000 IU intravenously at baseline, 24 hours, and 48 hours (n=256) | Placebo (n=266) | Barthel Index at day 90 showed no treatment benefit; P=0.45 | 42/256 (16.4%) | 24/266 (9.0%) | OR 1.98 (95% CI 1.16–3.38); P=0.01. Alteplase was permitted and prespecified interaction analyses raised particular concern about combined exposure.secondary safety outcometreatment interaction concern |
| Mourvillier hypothermia2013 | Comatose adults with severe community-acquired bacterial meningitis; n=98 | Cooling to 32–34°C for 48 hours followed by passive rewarming (n=49) | Standard temperature management (n=49) | Unfavourable Glasgow Outcome Scale result at 3 months: 86% vs 74%; P=0.13 | 25/49 (51%) | 15/49 (31%) | RR 1.99 (95% CI 1.05–3.77); P=0.04. The trial stopped early at the data-monitoring board’s request because of excess mortality; the adjusted mortality HR was 1.76 (95% CI 0.89–3.45).stopped earlysecondary mortality outcome |
| Eurotherm32352015 | Adults with traumatic brain injury and intracranial pressure >20 mm Hg despite first-line treatment; n=387 | Titrated hypothermia at 32–35°C plus standard care (n=195) | Standard care alone (n=192) | Six-month GOS-E shifted towards worse outcomes with hypothermia: adjusted common OR 1.53 (95% CI 1.02–2.30); P=0.04 | 68/195 (34.9%) | 51/192 (26.6%) | Six-month death HR 1.45 (95% CI 1.01–2.10); P=0.047. Recruitment was suspended early, and mortality was a secondary time-to-event analysis accompanying an unfavourable primary functional result.unfavourable primary outcomestopped early |
| Respiratory | |||||||
| Post-extubation NIV2004 | Adults who developed respiratory failure within 48 hours after elective extubation; n=221 | Non-invasive ventilation plus standard medical therapy (n=114) | Standard medical therapy (n=107) | Reintubation: 48% vs 48%; RR 0.99 (95% CI 0.76–1.30) | ICU: 28/114 (25%) | ICU: 15/107 (14%) | RR 1.78 (95% CI 1.03–3.20); P=0.048. Reintubation was delayed with NIV (median 12 hours vs 2 hours 30 minutes; P=0.02), a plausible contributor to harm.borderline nominal findingdelayed reintubation |
| LOCO₂2020 | Mechanically ventilated adults with acute respiratory distress syndrome; 201 analysed | Conservative oxygen targets: PaO₂ 55–70 mm Hg and SpO₂ 88–92% (n=99) | Liberal oxygen targets: PaO₂ 90–105 mm Hg and SpO₂ ≥96% (n=102) | Day-28 mortality: 34/99 (34.3%) vs 27/102 (26.5%); difference 7.8 percentage points (95% CI −4.8 to 20.6) | Day 90: 44/99 (44.4%) | Day 90: 31/102 (30.4%) | Day-90 difference 14.0 percentage points (95% CI 0.7–27.2); adjusted HR 1.62 (95% CI 1.02–2.56). The trial stopped early for safety concerns and low probability of primary-endpoint benefit.later mortality endpointstopped early |
| CYCOV2021 | Severe COVID-19 pneumonia requiring venovenous ECMO; n=34 | Early CytoSorb cytokine adsorption during ECMO (n=17) | ECMO without cytokine adsorption (n=17) | Adjusted mean log IL-6 concentration at 72 hours showed no difference: mean difference 0.30 (95% CI −0.70 to 1.30); P=0.54 | Day 30: 14/17 (82%) | Day 30: 4/17 (24%) | Thirty-day survival was 3/17 (18%) vs 13/17 (76%); P=0.0016. This was a very small, single-centre, open-label trial and survival was secondary.small trialsecondary survival endpoint |
| VENT-AVOID2024 | Acute hypercapnic COPD exacerbation either failing NIV or difficult to wean from invasive ventilation; n=113 in two separately randomised strata | Extracorporeal carbon-dioxide removal plus standard care; NIV stratum n=26 | Standard care alone; NIV stratum n=22 | Ventilator-free days through day 5 did not differ across strata; P=0.36. In the NIV stratum the median was 5 days in both groups | Day 60, NIV stratum: 43% | Day 60, NIV stratum: 6% | Day-60 mortality P=0.01; in-hospital mortality was 22% vs 0% (P=0.02). This was a small, prespecified, separately randomised NIV stratum rather than a post-randomisation subgroup; the trial ended early because of slow enrolment.separately randomised stratumstopped early |
| SESAR2025 | Adults with early moderate-to-severe acute respiratory distress syndrome receiving invasive mechanical ventilation in 37 French ICUs; n=687 | Inhaled sedation with sevoflurane for up to 7 days (n=346) | Intravenous sedation with propofol for up to 7 days (n=341) | Ventilator-free days through day 28 were lower with sevoflurane: median 0.0 (IQR 0.0–11.9) vs 0.0 days (IQR 0.0–18.7); median difference −2.1 days (95% CI −3.6 to −0.7); standardised HR 0.76 (95% CI 0.50–0.97) | Day 90: 183/346 (52.9%) | Day 90: 151/341 (44.3%) | Ninety-day survival was 47.1% vs 55.7%; hazard ratio for death 1.31 (95% CI 1.05–1.62). Seven-day mortality was also higher with sevoflurane: 19.4% vs 13.5%; RR 1.44 (95% CI 1.02–2.03).key secondary endpointunfavourable primary outcome |
| Gastrointestinal | |||||||
| PROPATRIA2008 | Predicted severe acute pancreatitis; n=298 | Multispecies probiotic preparation for 28 days (n=153) | Placebo (n=145) | Infectious complications: 46/153 (30%) vs 41/145 (28%); RR 1.06 (95% CI 0.75–1.51); P=0.76 | 24/153 (16%) | 9/145 (6%) | RR 2.53 (95% CI 1.22–5.25); P=0.01. Bowel ischaemia occurred in nine probiotic recipients and none receiving placebo (P=0.004); eight cases were fatal.secondary safety outcomebowel ischaemia signal |
| Nutrition | |||||||
| REDOXS2013 | Mechanically ventilated critically ill adults with multi-organ failure; 2×2 factorial trial, n=1,223 | Early high-dose enteral and parenteral glutamine (n=611) | No glutamine (n=607) | Day-28 mortality: 198/611 (32.4%) vs 165/607 (27.2%); P=0.049, above the prespecified P<0.044 efficacy threshold | Hospital: 227/611 (37.2%) | Hospital: 188/607 (31.0%) | Hospital mortality P=0.02; six-month mortality was also higher, 259 (43.7%) vs 218 (37.2%); P=0.02. These were glutamine main-effect analyses in a factorial trial.factorial main effectsecondary mortality endpoints |
| Renal | |||||||
| AKIKI 22021 | Critically ill adults with KDIGO stage-3 acute kidney injury, prolonged oliguria or blood urea nitrogen >112 mg/dL, and no urgent indication for kidney replacement therapy; n=278 | More-delayed kidney replacement therapy until a mandatory indication or blood urea nitrogen 140 mg/dL (n=141) | Delayed strategy with initiation soon after randomisation (n=137) | Days alive and free of kidney replacement therapy through day 28: median 10 (IQR 0–24) vs 12 (0–25); P=0.93 | Day 60: 77/141 (55%) | Day 60: 60/137 (44%) | Prespecified multivariable HR 1.65 (95% CI 1.09–2.50); P=0.018. The unadjusted mortality comparison did not reach significance (P=0.071), so the qualifying result is model-dependent.adjusted model onlycrude P=0.071 |
| Haematological | |||||||
| BART2008 | High-risk cardiac surgery; n=2,331 | Aprotinin (n=781 randomised; n=780 in bleeding analysis) | Tranexamic acid (n=770) or aminocaproic acid (n=780) | Massive postoperative bleeding: 74/780 (9.5%) vs 93/770 (12.1%) and 94/780 (12.1%); aprotinin vs pooled lysine analogues RR 0.79 (95% CI 0.61–1.01) | Day 30: 6.0% | Day 30: 3.9% and 4.0% | Aprotinin vs pooled lysine analogues RR 1.53 (95% CI 1.06–2.22). Neither individual pairwise comparison crossed significance; the trial stopped early after the mortality signal.pooled comparator analysisstopped early |
| TITRe22015 | Adults with postoperative haemoglobin <90 g/L after non-emergency cardiac surgery; n=2,003 | Restrictive red-cell transfusion threshold of 75 g/L (n=1,000) | Liberal threshold of 90 g/L (n=1,003) | Serious infection or ischaemic event within 3 months: 35.1% vs 33.0%; OR 1.11 (95% CI 0.91–1.34); P=0.30 | Day 90: 42/1,000 (4.2%) | Day 90: 26/1,003 (2.6%) | HR 1.64 (95% CI 1.00–2.67); P=0.045. Mortality was a prespecified secondary outcome and the confidence interval sat at the null boundary.borderline nominal findingsecondary endpoint |
| Sepsis | |||||||
| 6S2012 | Adults with severe sepsis requiring fluid resuscitation in ICU; n=798 in the modified intention-to-treat analysis | 6% hydroxyethyl starch 130/0.42 up to 33 mL/kg/day (n=398) | Ringer’s acetate (n=400) | Death or dialysis dependence at day 90: 202/398 (51%) vs 173/400 (43%); RR 1.17 (95% CI 1.01–1.36); P=0.03 | Day 90: 201/398 (51%) | Day 90: 172/400 (43%) | Mortality component RR 1.17 (95% CI 1.01–1.36); P=0.03. Only one patient in each group was dialysis-dependent, so death accounted for nearly the entire significant primary composite.mortality componentprimary composite unfavourable |
| Schädler CytoSorb2017 | Mechanically ventilated adults with severe sepsis or septic shock and acute lung injury or ARDS; 97 analysed | CytoSorb haemoperfusion for 6 hours daily for up to 7 days (n=47) | No haemoperfusion (n=50) | Normalised plasma IL-6 over study days 1–7 showed no between-group difference among 75 evaluable patients; P=0.15 | Day 60: 21/47 (44.7%) | Day 60: 13/50 (26.0%) | Unadjusted P=0.039. Baseline kidney-replacement therapy was imbalanced, and after adjustment for morbidity and baseline imbalances haemoperfusion was not associated with mortality (P=0.19).unadjusted analysis onlyadjusted result neutral |
| OASIS2015 | Adults with severe sepsis enrolled within 24 hours of the first acute organ dysfunction; n=305 | Oral talactoferrin 1.5 g every 8 hours for up to 28 days (n=153) | Placebo (n=152) | Day-28 mortality: 38/153 (24.8%) vs 27/152 (17.8%); P=0.117 | Hospital: 43/153 (28.1%) | Hospital: 27/152 (17.8%) | Hospital mortality P=0.037; three-month mortality was also higher, 46/153 (30.1%) vs 31/152 (20.4%); P=0.036. The trial was stopped early after interim safety and futility review.later mortality endpointsstopped early |
| Trauma | |||||||
| VICTORY2026 | Adults with deep second- or third-degree burns covering ≥20% total body surface area and requiring grafting; n=238 | Intravenous vitamin C 50 mg/kg every 6 hours for 96 hours (n=120) | Placebo (n=118) | Day-28 death or persistent organ dysfunction: 49/120 (40.8%) vs 35/118 (29.7%); adjusted RR 1.28 (95% CI 0.99–1.65); P=0.06 | Day 28: 18/120 (15.0%) | Day 28: 9/118 (7.6%) | Adjusted RR 1.96 (95% CI 1.32–2.90); P=0.001. Hospital mortality was also higher (23.3% vs 16.1%; adjusted RR 1.44, 95% CI 1.03–2.00; P=0.03); the primary result crossed the prespecified futility/harm boundary and recruitment stopped early.mortality componentstopped early |
| Miscellaneous | |||||||
| PROXI2009; mortality follow-up 2012 | Adults undergoing acute or elective laparotomy; 1,386 in the original outcome analysis and 1,382 with long-term vital status | 80% inspired oxygen during surgery and for 2 postoperative hours (n=685) | 30% inspired oxygen (n=701) | Surgical-site infection within 14 days: 131/685 (19.1%) vs 141/701 (20.1%); OR 0.94 (95% CI 0.72–1.22); P=0.64 | Median 2.3 years: 159/685 (23.2%) | Median 2.3 years: 128/701 (18.3%) | HR 1.30 (95% CI 1.03–1.64); P=0.03. This was a delayed follow-up analysis; the apparent excess was concentrated in patients undergoing cancer surgery, whereas the non-cancer interaction estimate was neutral.delayed follow-upeffect heterogeneity |
Interpretation: these findings require more caution than a statistically significant prespecified primary mortality analysis because the qualifying mortality result was secondary, supportive, or a component of another endpoint. Unless stated otherwise, P values and confidence intervals are nominal and were not protected by a trial-wide multiplicity procedure. Early and delayed endpoints, pooled-comparator analyses, model-dependent or unadjusted findings, mortality components, and separately randomised strata are identified explicitly; no post-randomisation subgroup-only mortality result is used to qualify a trial for this table.
Table 8. Statistically significant mortality harm in a subgroup analysis
Inclusion rule: adult randomised trials relevant to acute or critical care in which the designated overall analysis did not establish mortality harm, but mortality was significantly higher within a patient subgroup or a treatment-by-subgroup interaction supported a differential mortality effect. Trials already classified by a statistically significant overall primary or secondary mortality result are not repeated here. A neutral trial may appear in both subgroup tables when a qualitative interaction supports opposite-direction effects in different subgroups.
| Trial | Population | Intervention | Comparator | Overall or primary result | Mortality — intervention subgroup | Mortality — comparator subgroup | Subgroup mortality effect and qualification |
|---|---|---|---|---|---|---|---|
| Neuro | |||||||
| ASK1996 | Moderate-to-severe acute ischaemic stroke treated within 4 hours; n=340 | Intravenous streptokinase 1.5 million units | Placebo | Death or Barthel Index <60 at 3 months showed a nonsignificant unfavourable trend: RR 1.08 (95% CI 0.74–1.58) | Treatment >3 hours: higher mortality | Treatment >3 hours: reference | For treatment after 3 hours, mortality RR 1.98 (95% CI 1.18–3.35); within 3 hours, RR 1.11 (95% CI 0.38–3.21). The reported timing interaction (P=0.04) applied to the combined death-or-disability outcome rather than mortality alone; subgroup death counts were not tabulated.time subgroupmortality interaction not reported |
| RIGHT-22019 ICH analysis | Prespecified intracerebral-haemorrhage subgroup in an ambulance-based trial of presumed ultra-acute stroke; n=145 | Transdermal glyceryl trinitrate 5 mg daily for 4 days (n=74) | Sham dressing (n=71) | Parent-trial 90-day modified Rankin Scale: adjusted common OR 1.25 (95% CI 0.97–1.60); P=0.083. Within ICH: adjusted common OR 1.87 (95% CI 0.98–3.57) | In hospital: 34/74 (45.9%) | In hospital: 21/71 (29.6%) | Adjusted OR 2.26 (95% CI 1.03–4.95). Day-90 mortality was not significant: 35/73 (47.9%) vs 23/71 (32.4%); adjusted HR 1.50 (95% CI 0.86–2.62). Prespecified diagnostic subgroup, but small and not supported by the later mortality endpoint.prespecified subgroupin-hospital endpoint only |
| SAFE2007 TBI follow-up | Traumatic-brain-injury subgroup from the SAFE fluid-resuscitation trial; 460 randomised and 420 with 24-month vital status | 4% albumin; n=214 with vital status | 0.9% saline; n=206 with vital status | Parent-trial 28-day mortality: 20.9% vs 21.1%; RR 0.99 (95% CI 0.91–1.09); P=0.87 | TBI, 24 months: 71/214 (33.2%) | TBI, 24 months: 42/206 (20.4%) | RR 1.63 (95% CI 1.17–2.26); P=0.003. In severe TBI the corresponding result was 61/146 (41.8%) vs 32/144 (22.2%); RR 1.88 (95% CI 1.31–2.70); P<0.001. This was a post-hoc long-term follow-up of a baseline subgroup.post hoc follow-upbaseline TBI subgroup |
| BaSICS2021 | Critically ill adults requiring fluid challenges and at risk of acute kidney injury; n=10,520, including 483 with traumatic brain injury | Balanced crystalloid solution | 0.9% saline | Overall 90-day mortality: 1,381/5,230 (26.4%) vs 1,439/5,290 (27.2%); HR 0.97 (95% CI 0.90–1.05); P=0.47 | TBI: 77/246 (31.3%) | TBI: 50/237 (21.1%) | HR 1.48 (95% CI 1.03–2.12); interaction P=0.02. Prespecified subgroup, but TBI was recorded only as a yes/no baseline diagnosis and detailed injury characteristics were unavailable.prespecified subgroupinteraction P=0.02 |
| ICU-ROX2022 brain-pathology analysis | Post-hoc subgroup with acute brain pathology other than hypoxic–ischaemic encephalopathy; mortality data n=214 | Conservative oxygen therapy (n=110) | Usual oxygen therapy (n=104) | Parent-trial 180-day mortality: 35.7% vs 34.5%; OR 1.05 (95% CI 0.81–1.37); ventilator-free days were also neutral | Day 180: 38/110 (34.5%) | Day 180: 28/104 (26.9%) | Absolute difference 7.6 percentage points (95% CI −4.7 to 19.9); P=0.23. The qualifying signal is interaction-only: P=0.02 versus the suspected hypoxic–ischaemic-encephalopathy subgroup, but P=0.53 versus patients without neurological disease.post hoc subgroupwithin-subgroup P=0.23interaction-only signal |
| Circulatory | |||||||
| SOAP II2010 | Adults with shock requiring vasopressors; n=1,679, including a prespecified cardiogenic-shock subgroup of 280 | Dopamine; cardiogenic shock n=135 | Norepinephrine; cardiogenic shock n=145 | Overall 28-day mortality: 52.5% vs 48.5%; OR 1.17 (95% CI 0.97–1.42); P=0.10 | Cardiogenic shock: higher Kaplan–Meier mortality | Cardiogenic shock: reference curve | Twenty-eight-day Kaplan–Meier comparison P=0.03. The formal treatment-by-shock-type interaction was not significant (P=0.87), and subgroup-specific mortality counts were not reported in the main trial table.prespecified subgroupinteraction P=0.87 |
| Respiratory | |||||||
| FACTT2017 latent-class analysis | ARDS Network FACTT participants retrospectively classified as hypoinflammatory or hyperinflammatory; parent trial n=1,000 | Conservative fluid-management strategy | Liberal fluid-management strategy | Parent-trial 60-day mortality: 25.5% vs 28.4%; P=0.30; ventilator-free days favoured conservative management | Hyperinflammatory, day 90: 50% | Hyperinflammatory, day 90: 40% | Treatment-by-subphenotype interaction P=0.0039. Direction reversed in the hypoinflammatory group: 18% with conservative versus 26% with liberal management. Statistical support rests on a retrospective latent-class interaction; within-class effect estimates were not the prespecified trial analysis.post hoc latent classqualitative interaction |
| LIVE2019 | Moderate-to-severe ARDS classified at the bedside as focal or non-focal; n=400 in the primary analysis | Ventilation personalised to the investigator-assigned lung morphology | Standard low-PEEP lung-protective ventilation | Ninety-day mortality showed no difference: HR 1.01 (95% CI 0.61–1.66); P=0.98 | Misclassified morphology: 26/40 (65%) | Misclassified morphology: 18/57 (32%) | HR 2.8 (95% CI 1.5–5.1); P=0.012; interaction P<0.001. Misclassification was identified by post-hoc central review of baseline imaging, so the result does not test a prospectively deployable classifier and is vulnerable to subgroup selection.post hoc reclassificationinteraction P<0.001 |
| Gastrointestinal | |||||||
| SUP-ICU2019 high-severity analysis | Acutely admitted ICU adults at risk of gastrointestinal bleeding; post-hoc SAPS II >53 subgroup with complete score data, n=1,137 | Pantoprazole 40 mg intravenously daily (n=579) | Placebo (n=558) | Parent-trial 90-day mortality: 510/1,642 (31.1%) vs 499/1,640 (30.4%); RR 1.02 (95% CI 0.91–1.13); P=0.76 | SAPS II >53: 272/579 (47.0%) | SAPS II >53: 229/558 (41.0%) | RR 1.13 (95% CI 1.00–1.29); treatment-by-severity interaction P=0.049. The threshold was selected post hoc, SAPS II data were missing for many participants, and analyses accounting for missing scores attenuated the association.post hoc thresholdborderline interactionmissing-data sensitive |
| Nutrition | |||||||
| EFFORT Protein2023 | Mechanically ventilated, nutritionally high-risk critically ill adults; n=1,329 randomised, with an a priori acute-kidney-injury subgroup | Higher protein prescription, ≥2.2 g/kg/day | Usual protein prescription, ≤1.2 g/kg/day | Time to discharge alive by day 60: HR 0.91 (95% CI 0.77–1.07); P=0.27. Overall 60-day mortality: 222/642 (34.6%) vs 208/648 (32.1%); RR 1.08 (95% CI 0.92–1.26) | Baseline AKI: higher 60-day mortality | Baseline AKI: reference | RR 1.4 (95% CI 1.1–1.8); interaction P=0.02. A priori subgroup, but the trial stopped early for futility with 1,329 of 4,000 planned participants, subgroup event counts were not reported in the main text, and multiplicity was not controlled.a priori subgroupstopped earlyno multiplicity adjustment |
| Sepsis | |||||||
| ADDRESS2005 | Adults with severe sepsis at lower predicted risk of death; n=2,640, including a recent-surgery subgroup with single-organ dysfunction | Drotrecogin alfa activated, 24 μg/kg/hour for 96 hours | Placebo | Twenty-eight-day mortality: 243/1,333 (18.2%) vs 221/1,307 (16.9%); RR 1.08 (95% CI 0.91–1.27); trial stopped for futility | Recent surgery plus single-organ dysfunction: 67/323 (20.7%) | Recent surgery plus single-organ dysfunction: 44/313 (14.1%) | Nominal P=0.03. The subgroup comprised patients with surgery in the preceding 30 days and single-organ dysfunction; no clearly significant treatment interaction was reported, and serious bleeding was increased overall.post hoc clinical subgroupinteraction not reported |
| ANDROMEDA-SHOCK2020 post-hoc analysis | Early septic shock with normal capillary refill time 2 hours after randomisation; post-randomisation subgroup n=184 | Resuscitation targeting lactate reduction (n=82) | Resuscitation targeting capillary-refill-time normalisation (n=102) | Parent-trial 28-day mortality: 92/212 (43.4%) vs 74/212 (34.9%); HR for capillary-refill targeting 0.75 (95% CI 0.55–1.02); P=0.06 | Normal CRT at 2 hours: 33/82 (40%) | Normal CRT at 2 hours: 23/102 (23%) | P=0.009; adjusted OR 3.3 (95% CI 1.5–7.1); P=0.003. Normal capillary refill at 2 hours was measured after treatment began, creating a post-randomisation subgroup susceptible to selection and mediation bias.post hoc subgrouppost-randomisation classifier |
Interpretation: subgroup findings are inherently less secure than an overall prespecified comparison. A subgroup-specific P value does not prove that treatment effects differ between subgroups; the formal interaction test is therefore reported whenever available. Interaction-only rows do not imply a statistically significant comparison within the displayed subgroup. Prespecified strata, post-hoc analyses, multiplicity, model dependence, missing data, absent arm-level counts, and post-randomisation classifiers are identified explicitly. These signals are best treated as hypothesis-generating unless confirmed in a trial prospectively restricted to the proposed at-risk population.
Table 9. Statistically significant mortality benefits not confirmed by a subsequent large multicentre RCT
Inclusion rule: adult randomised trials relevant to acute or critical care that reported a statistically significant mortality reduction in a primary, secondary, subgroup or restricted analysis, followed by a larger multicentre randomised trial of the same or a closely related treatment strategy that did not confirm the mortality benefit. “Not confirmed” includes a neutral result or a point estimate favouring harm; it does not by itself prove that the original finding was false.
| Initial trial | Initial population | Intervention and comparator | Initial mortality benefit | Subsequent trial | Subsequent population | Subsequent mortality result | Interpretation and qualification |
|---|---|---|---|---|---|---|---|
| Neuro | |||||||
| rFVIIa phase 2b 2005 | Spontaneous intracerebral haemorrhage diagnosed within 3 hours; n=399 | rFVIIa 40, 80 or 160 micrograms/kg versus placebo, within 4 hours of onset | 90-day mortality: 18% versus 29%; P=0.02 Secondary |
FAST
2008 · large multicentre RCT
FASTEST 2026 · large multicentre RCT |
FAST: acute spontaneous intracerebral haemorrhage treated within 4 hours; n=841.
FASTEST: selected acute spontaneous intracerebral haemorrhage treated within 2 hours; n=626. |
FAST: no significant survival benefit at 90 days; placebo mortality was 19%
FASTEST: 180-day mortality 20/328 (6.1%) versus 22/298 (7.4%); RR 0.83 (95% CI 0.46–1.48) |
FAST and FASTEST both confirmed that rFVIIa reduced haematoma expansion but did not establish an improvement in clinical outcome or survival. FASTEST tested 80 micrograms/kg within 2 hours and was stopped early for futility. |
| Circulatory | |||||||
| LIDO2002 | Severe low-output heart failure requiring intravenous inotropic support; n=203 | Levosimendan infusion versus dobutamine infusion | 180-day mortality: 26% versus 38%; HR 0.57 (95% CI 0.34–0.95); P=0.029Secondary follow-up | SURVIVE2007 · large multicentre RCT | Acute decompensated heart failure requiring intravenous inotropic support; n=1,327 | 180-day mortality: 26% versus 28%; HR 0.91 (95% CI 0.74–1.13); P=0.40 | The exact active comparator was retained, but the larger trial enrolled a broader acute-heart-failure population and used 180-day mortality as its primary outcome. |
| Resuscitation | |||||||
| HACA2002 | Comatose survivors of witnessed out-of-hospital cardiac arrest with initial ventricular fibrillation or pulseless ventricular tachycardia; n=275 | Cooling to 32–34°C for 24 hours versus usual temperature management | 6-month mortality: 41% versus 55%; RR 0.74 (95% CI 0.58–0.95)Secondary | TTM22021 · large multicentre RCT | Comatose adults after out-of-hospital cardiac arrest of presumed cardiac or unknown cause; n=1,850 | 6-month mortality: 50% versus 48%; RR 1.04 (95% CI 0.94–1.14); P=0.37 | TTM2 compared 33°C with protocolised normothermia and early fever treatment, whereas the earlier control group did not receive active fever prevention. TTM2 also included a broader arrest population. |
| Airway | |||||||
| Rumbak early tracheotomy2004 | Medical ICU patients predicted to require mechanical ventilation for more than 14 days; n=120 | Percutaneous tracheotomy within 48 hours versus continued translaryngeal intubation with tracheotomy planned for days 14–16 | Hospital mortality: 19/60 (31.7%) versus 37/60 (61.7%); P<0.005Principal outcome | TracMan2013 · large multicentre RCT | Mechanically ventilated adults considered likely to require at least another 7 days of ventilation across 72 UK critical care units; n=909 | Thirty-day mortality: 30.8% with early versus 31.5% with deferred tracheostomy; absolute risk reduction 0.7 percentage points (95% CI −5.4 to 6.7) | TracMan did not confirm the large mortality reduction. It compared tracheostomy within 4 days with deferral until after day 10 if still required; only 44.9% of the deferred group ultimately underwent tracheostomy. The timing, selection criteria and comparator strategy therefore differed from the smaller single-centre Rumbak trial. |
| Respiratory | |||||||
| ACURASYS2010 | Early severe ARDS with PaO2/FiO2 <150 mm Hg; n=340 | 48-hour cisatracurium infusion versus placebo; deep sedation in both groups | Adjusted 90-day HR 0.68 (95% CI 0.48–0.98); P=0.04; crude mortality 31.6% versus 40.7%, P=0.08Primary · adjusted | ROSE2019 · large multicentre RCT | Early moderate-to-severe ARDS with PaO2/FiO2 <150 mm Hg; n=1,006 | 90-day in-hospital mortality: 42.5% versus 42.8%; difference −0.3 percentage points (95% CI −6.4 to 5.9); P=0.93 | ROSE tested the cisatracurium strategy against usual care that encouraged lighter sedation, rather than deep sedation plus placebo. It was stopped early for futility. |
| FLORALI 2015 | Adults with non-hypercapnic acute hypoxaemic respiratory failure and PaO₂/FiO₂ ≤300 mm Hg; n=310 | High-flow nasal oxygen versus standard oxygen or non-invasive ventilation | 90-day mortality: 12% versus 23% versus 28%; standard oxygen versus high-flow oxygen HR 2.01 (95% CI 1.01–3.99), P=0.046; non-invasive ventilation versus high-flow oxygen HR 2.50 (95% CI 1.31–4.78), P=0.006 Secondary |
HIGH
2018 · large multicentre RCT
SOHO-COVID 2022 · large multicentre RCT SOHO 2026 · large multicentre RCT |
HIGH: immunocompromised adults with acute hypoxaemic respiratory failure; n=776.
SOHO-COVID: adults with COVID-19 respiratory failure; n=711. SOHO: adults with acute hypoxaemic respiratory failure from any cause; n=1,110 in the analysis. |
HIGH: 28-day mortality 35.6% versus 36.1%; HR 0.98 (95% CI 0.77–1.24); P=0.94
SOHO-COVID: 28-day mortality 10% versus 11%; P=0.60 SOHO: 28-day mortality 14.6% versus 14.6%; difference −0.05 percentage points (95% CI −4.21 to 4.10); P=0.98 |
Three subsequent multicentre trials did not reproduce the FLORALI survival advantage. Their populations differed, but SOHO provided the closest large direct comparison in general acute hypoxaemic respiratory failure and found identical 28-day mortality. |
| OXYGEN-ICU 2016 | ICU adults expected to remain for at least 72 hours; 480 randomised and 434 included in the modified intention-to-treat analysis | Conservative oxygen targeting PaO₂ 70–100 mm Hg or SpO₂ 94–98% versus conventional oxygen allowing PaO₂ up to 150 mm Hg and SpO₂ 97–100% | ICU mortality: 25/216 (11.6%) versus 44/218 (20.2%); RR 0.57 (95% CI 0.37–0.90); P=0.01 Primary Stopped early |
ICU-ROX
2019 · large multicentre RCT
HOT-ICU 2021 · large multicentre RCT UK-ROX 2025 · large multicentre RCT |
ICU-ROX: mechanically ventilated ICU adults; n=1,000.
HOT-ICU: ICU adults with acute hypoxaemic respiratory failure; n=2,928. UK-ROX: mechanically ventilated ICU adults receiving supplemental oxygen; n=16,500. |
ICU-ROX: 180-day mortality 35.7% versus 34.5%; OR 1.05 (95% CI 0.81–1.37)
HOT-ICU: 90-day mortality 42.9% versus 42.4%; RR 1.02 (95% CI 0.94–1.11); P=0.64 UK-ROX: 90-day mortality 35.4% versus 34.9%; adjusted risk difference 0.7 percentage points (95% CI −0.7 to 2.0); P=0.28 |
Three substantially larger trials using different conservative oxygen strategies did not confirm the OXYGEN-ICU mortality benefit. UK-ROX was the largest and found no reduction in 90-day mortality. |
| Nutrition | |||||||
| VITdAL-ICU2014 | Critically ill adults with 25-hydroxyvitamin D ≤20 ng/mL; n=475; mortality signal in the prespecified subgroup ≤12 ng/mL | Enteral vitamin D3 540,000 IU, then 90,000 IU monthly, versus placebo | Severe-deficiency subgroup hospital mortality: 28.6% versus 46.1%; HR 0.56 (95% CI 0.35–0.90); interaction P=0.04Subgroup | VIOLET2019 · large multicentre RCT | Critically ill adults at high risk of death with confirmed vitamin D deficiency; 1,360 randomised and 1,078 in the primary analysis | 90-day mortality: 23.5% versus 20.6%; difference 2.9 percentage points (95% CI −2.1 to 7.9); P=0.26 | VIOLET administered the same loading dose earlier but did not continue monthly dosing. Benefit was not seen with greater baseline deficiency, and the trial stopped for futility. |
| Renal | |||||||
| Ronco dose trial2000 | Critically ill adults with acute renal failure receiving continuous venovenous haemofiltration; n=425 | Effluent 35 or 45 mL/kg/hour versus 20 mL/kg/hour | Survival 15 days after stopping haemofiltration: 57% and 58% versus 41%; P=0.0007 and P=0.0013Primary | RENAL2009 · large multicentre RCT | Critically ill adults with acute kidney injury requiring continuous renal-replacement therapy; n=1,508 | 90-day mortality: 44.7% versus 44.7%; OR 1.00 (95% CI 0.81–1.23); P=0.99 | RENAL tested 40 versus 25 mL/kg/hour using continuous haemodiafiltration. The modality and time point differed, but it directly addressed higher-intensity continuous therapy. |
| Schiffl daily haemodialysis 2002 | Critically ill adults with acute renal failure requiring intermittent haemodialysis; n=160 | Daily haemodialysis versus alternate-day haemodialysis | Mortality: 28% versus 46%; P=0.01 Primary | ATN 2008 · large multicentre RCT | Critically ill adults with acute kidney injury plus sepsis or failure of at least one non-renal organ; n=1,124 | 60-day mortality: 53.6% versus 51.5%; OR 1.09 (95% CI 0.86–1.40); P=0.47 | ATN found no survival advantage from more intensive renal support. It used intermittent haemodialysis, sustained low-efficiency dialysis or continuous haemodiafiltration according to haemodynamic stability, so it was a broader intensity trial rather than an exact replication of daily intermittent haemodialysis. |
| ELAIN2016 | Single-centre, predominantly surgical ICU cohort with KDIGO stage 2 AKI plus plasma NGAL >150 ng/mL; n=231 | RRT within 8 hours of KDIGO stage 2 versus at stage 3 or an absolute indication | 90-day mortality: 39.3% versus 54.7%; HR 0.66 (95% CI 0.45–0.97); P=0.03Primary | STARRT-AKI2020 · large multicentre RCT | Multinational critically ill cohort with severe AKI and no urgent indication for RRT; n=2,927 | 90-day mortality: 43.9% versus 43.7%; RR 1.00 (95% CI 0.93–1.09); P=0.92 | The later trial was much broader and allowed standard-strategy patients to avoid RRT unless conventional indications developed. ELAIN’s biomarker-enriched, mainly surgical population and delayed trigger were materially different. |
| STOP-AKI 2018 | International adaptive phase 2a/2b trial in adults with sepsis-associated acute kidney injury; n=301 overall; selected-dose comparison n=227 | Human recombinant alkaline phosphatase 1.6 mg/kg once daily for three days versus placebo | 28-day mortality: 16/111 (14.4%) versus 31/116 (26.7%); difference −12.3 percentage points (95% CI −22.7 to −1.9); P=0.02 Secondary | REVIVAL 2024 · large multicentre RCT | Adults with sepsis-associated acute kidney injury; n=649 in the efficacy analysis | 28-day mortality: 27.9% versus 27.9%; 90-day mortality: 33.9% versus 34.8% | REVIVAL tested ilofotase alfa against placebo in a larger international trial and found identical 28-day mortality. It was stopped for futility on the primary mortality endpoint. |
| Sepsis | |||||||
| HA-1A sepsis trial 1991 | Adults with sepsis and suspected Gram-negative infection; n=543; culture-confirmed Gram-negative bacteraemia subgroup n=200 | A single 100 mg intravenous dose of HA-1A versus albumin placebo, with usual care in both groups | Gram-negative bacteraemia subgroup: 28-day mortality 32/105 (30%) versus 45/92 (49%); P=0.014 Subgroup | CHESS 1994 · large multicentre RCT | Adults with septic shock and suspected Gram-negative infection; n=2,199 overall; 621 met all criteria and had confirmed Gram-negative bacteraemia | Confirmed Gram-negative bacteraemia: 14-day mortality 109/328 (33%) versus 95/293 (32%); P=0.864 | CHESS enrolled a much larger septic-shock population and did not confirm the mortality benefit in patients with confirmed Gram-negative bacteraemia. The mortality time point was 14 rather than 28 days. |
| Rivers EGDT2001 | Single-centre emergency-department patients with severe sepsis or septic shock; n=263 | Six-hour early goal-directed therapy using central venous oxygen saturation targets versus standard therapy | In-hospital mortality: 30.5% versus 46.5%; P=0.009Primary | ARISE2014 · large multicentre RCT | Emergency-department patients with early septic shock in 51 centres; n=1,600 | 90-day mortality: 18.6% versus 18.8%; RR 0.98 (95% CI 0.80–1.21); P=0.90 | Usual care had adopted earlier fluids, antibiotics and haemodynamic assessment. Two other large trials, ProCESS and ProMISe, were also neutral. |
| PROWESS2001 | Severe sepsis with at least one acute organ dysfunction; n=1,690 | Drotrecogin alfa activated 24 micrograms/kg/hour for 96 hours versus placebo | 28-day mortality: 24.7% versus 30.8%; relative risk reduction 19.4%; P=0.005Primary | PROWESS-SHOCK2012 · large multicentre RCT | Adults with septic shock requiring ongoing vasopressors after fluid resuscitation; n=1,697 | 28-day mortality: 26.4% versus 24.2%; RR 1.09 (95% CI 0.92–1.28); P=0.31 | The drug, dose and duration were replicated in a narrower septic-shock population. The subsequent large multicentre RCT failed to show benefit, leading to worldwide withdrawal of drotrecogin alfa. |
| de Jonge SDD 2003 | Adults admitted to a medical-surgical ICU; n=934 | Selective digestive decontamination using oral and enteral polymyxin E, tobramycin and amphotericin B plus four days of intravenous cefotaxime versus standard care | ICU mortality: 69/466 (14.8%) versus 107/468 (22.9%); P=0.002. Hospital mortality: 113/466 (24.2%) versus 146/468 (31.2%); P=0.02 Primary | SuDDICU 2022 · large multicentre RCT | Mechanically ventilated adults in a cluster-crossover trial across 19 ICUs; n=5,982 | Hospital mortality: 753/2,791 (27.0%) versus 928/3,191 (29.1%); OR 0.91 (95% CI 0.82–1.02); P=0.12 | SuDDICU tested a closely related SDD regimen in a much larger, multicentre cluster-crossover trial. Its point estimate favoured SDD, but the mortality reduction was not statistically significant. |
| EUPHAS 2009 | Severe sepsis or septic shock after emergency surgery for intra-abdominal infection; n=64 | Two sessions of polymyxin B haemoperfusion plus conventional therapy versus conventional therapy alone | 28-day mortality: 32% versus 53%; unadjusted HR 0.43 (95% CI 0.20–0.94) Secondary Stopped early |
ABDOMIX
2015 · large multicentre RCT
EUPHRATES 2018 · large multicentre RCT TIGRIS 2026 · large multicentre RCT |
ABDOMIX: septic shock after emergency surgery for peritonitis; n=243.
EUPHRATES: septic shock with endotoxin activity ≥0.60; n=450. TIGRIS: endotoxic septic shock with endotoxin activity 0.60–0.89 and multiorgan failure; n=157. |
ABDOMIX: 28-day mortality 27.7% versus 19.5%; P=0.14
EUPHRATES: 28-day mortality 37.7% versus 34.5%; RR 1.09 (95% CI 0.85–1.39); P=0.49 TIGRIS: 28-day mortality 41/106 (38.7%) versus 23/51 (45.1%); adjusted OR 0.67 (95% CrI 0.39–1.08); posterior probability of benefit 95.3%. At 90 days, the adjusted OR was 0.54 (95% CrI 0.32–0.87), with a 99.4% posterior probability of benefit |
ABDOMIX and EUPHRATES did not confirm the EUPHAS mortality benefit. TIGRIS subsequently crossed its prespecified Bayesian threshold, using an informative prior derived from a selected EUPHRATES subgroup. The later evidence is therefore mixed rather than uniformly negative, and the populations were not exact replications of postoperative abdominal sepsis. |
| SIC2007 | Severe SIRS, sepsis or septic shock with APACHE III score >70; n=249 randomised | Sodium selenite 1,000 microgram bolus then 1,000 micrograms/day for 14 days versus placebo | Per-protocol 28-day mortality: 42.4% versus 56.7%; OR 0.56 (95% CI 0.32–1.00); P=0.049; intention-to-treat P=0.109Per-protocol only | SISPCT2016 · large multicentre RCT | Severe sepsis or septic shock in 33 German ICUs; n=1,089; 2×2 factorial design | 28-day mortality: 28.3% versus 25.5%; P=0.30 | SISPCT used the same loading dose and daily sodium-selenite dose for up to 21 days. Its intention-to-treat result did not reproduce the earlier per-protocol signal. |
| ETASS 2013 | Adults with severe sepsis in multiple Chinese ICUs; n=361 | Thymosin α1 for seven days versus saline control | 28-day mortality: 26.0% versus 35.0%; RR 0.74 (95% CI 0.54–1.02); non-stratified P=0.062; log-rank P=0.049 Primary · log-rank | TESTS 2025 · large multicentre RCT | Adults with Sepsis-3 sepsis at 22 Chinese centres; 1,106 randomised and 1,089 included in the modified intention-to-treat analysis | 28-day mortality: 127/542 (23.4%) versus 132/547 (24.1%); HR 0.97 (95% CI 0.76–1.24); P=0.82 | TESTS used the same seven-day treatment duration in a much larger, double-blind trial and did not confirm a mortality benefit. ETASS crossed significance only in its log-rank analysis; its risk-ratio confidence interval included no effect. |
| CITRIS-ALI2019 | Sepsis with ARDS present for less than 24 hours; n=167 | Intravenous vitamin C 50 mg/kg every 6 hours for 96 hours versus placebo | 28-day mortality: 29.8% versus 46.3%; difference −16.6 percentage points; P=0.03Secondary · 1 of 46 | LOVIT2022 · large multicentre RCT | ICU adults with sepsis receiving vasopressors; n=872 | 28-day mortality: 35.4% versus 31.6%; RR 1.17 (95% CI 0.98–1.40); death or persistent organ dysfunction was higher with vitamin C, RR 1.21 (95% CI 1.04–1.40) | The dose and 96-hour duration were replicated, but LOVIT enrolled vasopressor-dependent sepsis without requiring ARDS. CITRIS-ALI’s mortality result was secondary, without adjustment for its many secondary outcomes; its primary outcomes were neutral. |
| Endocrine | |||||||
| Leuven I2001 | Mechanically ventilated adults in a single surgical ICU, predominantly after cardiac surgery; n=1,548 | Intensive insulin targeting 80–110 mg/dL versus insulin only above 215 mg/dL, targeting 180–200 mg/dL | ICU mortality: 4.6% versus 8.0%; sequentially adjusted P<0.04; hospital mortality 7.2% versus 10.9%, P=0.01Primary | NICE-SUGAR2009 · large multicentre RCT | Mixed medical and surgical ICU adults expected to require at least 3 days of treatment; n=6,104 | 90-day mortality: 27.5% versus 24.9%; OR 1.14 (95% CI 1.02–1.28); P=0.02 | The later trial reversed the mortality direction and increased severe hypoglycaemia. Its conventional target was ≤180 mg/dL, substantially tighter than Leuven I’s control strategy, and its population was much broader. |
| Metabolic | |||||||
| BICAR-ICU2018 | Critically ill adults with severe metabolic acidaemia; n=389; mortality signal in the prespecified AKIN stage 2–3 stratum, n=182 | 4.2% sodium bicarbonate targeting arterial pH ≥7.30 versus no bicarbonate | AKIN 2–3 subgroup 28-day mortality: 42/92 (46%) versus 57/90 (63%); P=0.0166Subgroup | BICARICU-22025 · large multicentre RCT | Severe metabolic acidaemia with moderate-to-severe acute kidney injury at 43 French ICUs; n=640 | 90-day mortality: 62.1% versus 61.7%; difference 0.4 percentage points (95% CI −7.2 to 8.0); P=0.91 | This was the closest direct confirmation of the original AKI subgroup and used the same pH target. The mortality benefit disappeared, although bicarbonate still reduced use of kidney replacement therapy. |
Interpretive note: “Not confirmed” means that the later trial did not establish the earlier mortality benefit. A neutral subsequent large multicentre RCT is not proof of equivalence, and differences in population, comparator, timing, dose or co-interventions may explain some discordance. Primary publications are linked from both the initial and subsequent trial names; every link opens in a new tab.
Table 10. Statistically significant mortality harms not confirmed by a subsequent large multicentre RCT
Inclusion rule: adult randomised trials relevant to acute or critical care that reported a statistically significant mortality increase in a primary, secondary, subgroup or restricted analysis, followed by a subsequent large multicentre randomised trial of the same or a closely related treatment strategy that did not confirm the mortality harm. “Not confirmed” includes a neutral result or a point estimate favouring benefit; it does not by itself establish safety or prove that the original finding was false.
| Initial trial | Initial population | Intervention and comparator | Initial mortality harm | Subsequent trial | Subsequent population | Subsequent mortality result | Interpretation and qualification |
|---|---|---|---|---|---|---|---|
| Respiratory | |||||||
| LOCO₂ 2020 | Mechanically ventilated adults with acute respiratory distress syndrome; 201 analysed | Conservative oxygen targets of PaO₂ 55–70 mm Hg and SpO₂ 88–92% versus liberal targets of PaO₂ 90–105 mm Hg and SpO₂ ≥96% | 90-day mortality: 44.4% versus 30.4%; adjusted HR 1.62 (95% CI 1.02–2.56) Later endpoint Stopped early | HOT-ICU 2021 · large multicentre RCT | ICU adults with acute hypoxaemic respiratory failure receiving at least 10 L/min oxygen or FiO₂ ≥0.50; n=2,928 | 90-day mortality: 42.9% versus 42.4%; RR 1.02 (95% CI 0.94–1.11); P=0.64 |
The much larger later trial did not reproduce the LOCO₂ mortality signal. HOT-ICU used PaO₂ targets of 60 versus 90 mm Hg and enrolled broader acute hypoxaemic respiratory failure, whereas LOCO₂ was restricted to mechanically ventilated ARDS and stopped early.
There are numerous other trials exploring different oxygenation strategies, with most showing little between-group difference. |
| Haematological | |||||||
| TITRe22015 | Adults with postoperative haemoglobin <90 g/L after non-emergency cardiac surgery; n=2,003 | Restrictive red-cell transfusion threshold of 75 g/L versus liberal threshold of 90 g/L | 90-day mortality: 4.2% versus 2.6%; HR 1.64 (95% CI 1.00–2.67); P=0.045SecondaryBorderline | TRICS III2017 · large multicentre RCT | Adults at moderate-to-high operative risk undergoing cardiac surgery; n=5,243 | Mortality by discharge or day 28: 3.0% versus 3.6%; OR 0.85 (95% CI 0.62–1.16) | TRICS III did not reproduce the excess mortality and established non-inferiority for its primary composite. Randomisation occurred before surgery and its liberal thresholds varied by location, so the strategies were closely related but not identical to postoperative-only TITRe2. |
| Sepsis | |||||||
| 6S2012 | Adults with severe sepsis requiring fluid resuscitation in ICU; 798 in the modified intention-to-treat analysis | 6% hydroxyethyl starch 130/0.42 in balanced crystalloid versus Ringer’s acetate | 90-day mortality: 51% versus 43%; RR 1.17 (95% CI 1.01–1.36); P=0.03Mortality component | CHEST2012 · large multicentre RCT | Heterogeneous ICU adults requiring fluid resuscitation; n=7,000 | 90-day mortality: 18.0% versus 17.0%; RR 1.06 (95% CI 0.96–1.18); P=0.26 | CHEST did not reproduce a statistically significant mortality increase, but HES increased renal-replacement therapy. It enrolled a broader, lower-risk ICU population and compared HES 130/0.4 in saline with saline; 6S studied severe sepsis and used balanced solutions. Worrying signals of harm exist in both trials |
| Endocrine | |||||||
| NICE-SUGAR2009 | Medical and surgical ICU adults expected to require at least 3 days of critical care; n=6,104 | Intensive glucose control targeting 81–108 mg/dL versus conventional control targeting ≤180 mg/dL | 90-day mortality: 27.5% versus 24.9%; OR 1.14 (95% CI 1.02–1.28); P=0.02Primary | TGC-Fast2023 · large multicentre RCT | Critically ill adults not receiving early parenteral nutrition; n=9,230 | 90-day mortality: 10.5% versus 10.1%; RR 1.04 (95% CI 0.92–1.17); P=0.51 | TGC-Fast did not reproduce the mortality increase, but it also showed no clinical benefit from tight control. It excluded early parenteral nutrition, used a computer-guided protocol and compared 80–110 mg/dL with a more liberal 180–215 mg/dL target in a markedly lower-mortality population. |
Interpretive note: “Not confirmed” is used in the replication sense: the later trial did not establish the earlier mortality harm. A neutral subsequent large multicentre RCT is not proof of equivalence or safety, and differences in population, comparator, timing, dose or co-interventions may explain some discordance.