The research team, led by Hugo Belna of the Department of Intensive Care Medicine and Hyperbaric Medicine at CHU d’Angers, together with hematologist Corentin Orvain and colleagues, set out to answer a deceptively simple question: if a transplant recipient’s condition is worsening or improving in the first seventy-two hours of critical illness, does the direction of that change predict whether the patient will die in the ICU? The question matters because intensivists increasingly rely on the concept of a time-limited trial of intensive care, in which support is offered for a defined period and then reassessed. For that strategy to work, clinicians need reliable markers of trajectory, not just snapshots of severity on admission.

The study included 161 allogeneic hematopoietic cell transplantation recipients admitted to the ICU at Angers. To capture the dynamics of organ dysfunction, the investigators used two complementary tools. The first was the Sequential Organ Failure Assessment score, or SOFA, a widely validated composite index that grades dysfunction across six organ systems, including respiration, coagulation, liver function, cardiovascular performance, central nervous system status, and renal function, with each system scored from zero for normal function to four for severe failure. The second was the requirement for organ support, a practical and clinically grounded measure that includes interventions such as invasive mechanical ventilation for respiratory failure and vasoactive drugs or continuous renal replacement therapy for circulatory and kidney failure.

Because a meaningful trajectory can only be measured in patients who remain in the unit long enough for a second assessment, the analysis focused on the subset of 110 patients whose ICU length of stay was at least three days. In this group, the researchers compared the SOFA score recorded on the first day after admission with the score on day three, and calculated a delta SOFA, defined as the difference between the two measurements. They also tracked whether patients required new forms of organ support between days one and three, a change that often signals a decisive turn for the worse even when the numeric score shifts only modestly.

The findings confirmed what many clinicians suspected but had never formally quantified in this specific population. There were significant associations between the SOFA score on day three, the delta SOFA between days one and three, and the emergence of new organ support requirements between those same days, and ICU mortality. In other words, patients whose organ dysfunction remained severe or deepened during the first seventy-two hours, and those who needed to be placed on ventilators, vasopressors, or dialysis machines for the first time during that window, were significantly more likely to die before leaving the unit. The burden of organ failure, and the therapies needed to substitute for failing organs, emerged as the main driver of outcome in these transplant recipients.

Perhaps the most clinically consequential result, however, concerned which measurements carried the greatest predictive weight. The day three SOFA score and any change in organ support requirement between days one and three predicted ICU mortality better than the SOFA score recorded on day one or the delta SOFA itself. This asymmetry is intuitively appealing: a single late measurement that integrates the accumulated damage of the first days of critical illness appears to summarize prognosis more effectively than the arithmetic difference between two early snapshots. A patient admitted with a moderate score who deteriorates rapidly may end up with the same day three score as a patient admitted in worse condition who stabilized, yet their trajectories and their underlying physiology differ in ways the simple delta calculation does not fully capture.

Yet the study also delivers a sobering caveat. While the associations were statistically significant, the overall prediction accuracy of these measures was limited. Neither the day three SOFA score nor the trajectory of organ support could reliably separate survivors from non-survivors at the level of the individual patient. For clinicians hoping that a simple two-day score might provide a defensible threshold for continuing or withdrawing intensive support, the message is one of caution. The trajectory of organ failure is informative at the population level, but it is not precise enough to serve as a stand-alone decision rule for an individual bedridden patient whose family is asking whether more days in the unit will help.

This limitation has real implications for how the time-limited ICU trial is conducted in allogeneic transplant recipients. The approach, in which support is provided for a fixed period and then reassessed against objective markers, has gained traction precisely because early mortality predictions in this population have historically been so unreliable and so pessimistic. Older studies painted a uniformly bleak picture, leading some centers to deny ICU admission or limit support outright. More recent cohorts, including the one from Angers, suggest that outcomes are more heterogeneous than the older literature implied, and that a meaningful fraction of recipients who develop organ failure can still survive if support is maintained and reversible causes are treated.

The technical details of the study design deserve attention when weighing its conclusions. As a retrospective, single-center analysis, it reflects the practices, case mix, and support capabilities of one French academic hospital, and its findings may not generalize perfectly to centers with different transplantation protocols, donor types, or intensity of care. The restriction of the trajectory analysis to patients staying at least three days also introduces a form of survivor bias, since patients who died within the first forty-eight hours were necessarily excluded from the day one to day three comparisons. The authors acknowledge these constraints, and the study is presented as a brief report intended to generate hypotheses rather than to settle definitively how prognostication should be performed.

What the study does establish is a clearer framework for thinking about critical illness after allogeneic hematopoietic cell transplantation. Organ failure, measured either by the SOFA composite or by the concrete need for mechanical ventilation, vasopressors, or renal replacement therapy, is the dominant determinant of ICU survival, outweighing the mere fact of prior transplantation. Serial assessment over the first three days adds some information, with the day three score and new support requirements outperforming admission values, but the incremental prognostic value of tracking the trajectory is modest. For now, the study supports a pragmatic approach: admit these patients, provide a genuine trial of intensive support, reassess organ function and support needs at seventy-two hours, and combine those objective measurements with clinical judgment, discussions with the hematology team, and the values of patients and families rather than relying on any single score to dictate the limits of care.

Subject of Research: Evolution of organ failures and their prognostic value in allogeneic hematopoietic cell transplantation recipients admitted to the intensive care unit

Article Title: Evolution of organ failures in allogeneic hematopoietic cell transplantation recipients admitted to the intensive care unit

Article References: Belna, H., Taillantou-Candau, M., Kouatchet, A., Moreau, C., Riou, J., Klemencie, M., Francois, S., Thépot, S., Mokeddem, C., Giltat, A., Hunault, M., Asfar, P., Tanguy-Schmidt, A., & Orvain, C. (2026). Evolution of organ failures in allogeneic hematopoietic cell transplantation recipients admitted to the intensive care unit. Annals of Hematology. https://doi.org/10.1007/s00277-026-07306-4

Image Credits: AI Generated

DOI: 10.1007/s00277-026-07306-4

Keywords: allogeneic hematopoietic cell transplantation, intensive care unit, SOFA score, organ failure, organ support, ICU mortality, prognosis, critical care, hematology, mechanical ventilation, renal replacement therapy, time-limited trial

The study, led by Bin Liu and Chungen Xing of the Second Affiliated Hospital of Soochow University together with colleagues at Chengdu Medical College-affiliated hospitals, assembled a cohort of 5,847 adult patients with gastrointestinal bleeding drawn from four publicly available critical care databases spanning 2008 to 2022: the MIMIC-IV database, the eICU Collaborative Research Database, the Northwestern Medicine ICU database, and the Salzburg Intensive Care database. All four repositories are hosted on the PhysioNet platform and contain de-identified electronic health record data released under institutional review board approval. To be included, patients needed at least twelve blood urea nitrogen observations during the first day of their ICU stay, a demanding requirement that ensured the researchers could map laboratory values onto an hourly grid and reconstruct genuine temporal patterns rather than isolated snapshots.

The cohort was seriously ill. ICU mortality ranged from 18.0 percent to 19.7 percent across the four databases, meaning roughly one in five patients died during their intensive care stay. Men made up between 56.5 and 58.7 percent of each cohort, with women comprising the remaining 41.3 to 43.5 percent. Gastrointestinal bleeding is one of the most common emergencies that brings patients into intensive care, and mortality prediction in this population has long relied on static scoring systems such as the Glasgow-Blatchford Score, the AIMS65 score, and the Rockall score, all of which incorporate admission laboratory values but cannot respond to how a patient evolves after arrival.

The analytical approach was deliberately layered. Generalized additive models, a flexible statistical framework capable of capturing nonlinear relationships without forcing them into rigid parametric shapes, were used to characterize how the concentration and duration of blood urea nitrogen abnormalities jointly related to outcomes. Adjusted logistic regression then evaluated fifteen distinct blood urea nitrogen ranges, allowing the team to ask not just whether high values were dangerous but exactly which thresholds mattered and for how long they needed to persist. Restricted cubic splines examined time-weighted blood urea nitrogen exposure, integrating both magnitude and persistence into a single measure. Finally, the researchers ran meta-analyses across the four databases, performed sensitivity analyses restricted to patients with denser measurement sets, and formally compared static models built on admission values alone against dynamic models that incorporated the full hourly trajectory.

The headline finding is strikingly simple. Patients whose blood urea nitrogen stayed below 30 milligrams per deciliter for more than twelve hours had substantially lower odds of ICU death, with odds ratios ranging from 0.40 to 0.62 across the databases. In other words, keeping this value in the low range through the critical first day was associated with roughly a forty to sixty percent reduction in the odds of mortality compared with patients whose values did not remain in that zone. At the opposite extreme, blood urea nitrogen of 50 milligrams per deciliter or higher sustained for more than eighteen hours markedly increased mortality odds, with odds ratios between 1.55 and 3.39. The dose-response logic is intuitive: the longer a patient remained in the dangerous range, the greater the accumulated risk.

The time-weighted analysis refined this picture further. Prolonged time spent within the 5 to 30 milligrams per deciliter band was associated with lower mortality odds, while greater cumulative exposure to the 40 to 70 milligrams per deciliter range increased risk. Crucially, the associations were nonlinear, meaning the relationship between urea nitrogen exposure and death was not a straight line but a curve with thresholds and inflection points. Equally important for the credibility of the findings, the meta-analysis detected no between-database heterogeneity, and the results remained broadly consistent when the analysis was repeated in subsets of patients with more densely sampled laboratory measurements, addressing the concern that sparse sampling might have distorted the hourly trajectories.

Blood urea nitrogen is a deceptively rich biomarker. The liver produces ammonia during protein metabolism, the urea cycle converts it into urea, and the kidneys excrete it, so blood levels rise with kidney dysfunction, with dehydration and reduced renal perfusion, and with the digestion of blood itself inside the gastrointestinal tract, since digested hemoglobin delivers a large protein load to the gut. In a bleeding patient, an elevated value therefore weaves together several ominous threads at once: hypovolemia, impaired organ perfusion, possible acute kidney injury, and ongoing or severe blood loss. Prior research has already linked elevated blood urea nitrogen to mortality in general ICU populations, in sepsis, in acute heart failure with cardiorenal syndrome, and in acute pancreatitis, and ratios combining urea nitrogen with albumin or creatinine have been proposed as prognostic tools across a range of conditions.

What this study adds is the temporal dimension. A single admission value tells you where a patient started; a trajectory tells you where they are heading. When the researchers compared static models built on admission blood urea nitrogen alone against dynamic models incorporating the hourly trajectory data, the improvement was measurable and statistically significant. The area under the receiver operating characteristic curve, a standard measure of discrimination, rose from 0.678 to 0.701 in the static models to between 0.718 and 0.748 in the dynamic models, with all differences reaching significance on DeLong testing at P values of 0.05 or below. The dynamic models also showed improved reclassification of patients into risk categories and lower Brier scores, indicating better-calibrated overall predictions rather than merely better ranking.

The clinical implications are immediate and practical. Blood urea nitrogen is already drawn repeatedly in virtually every ICU patient, so no new test, no new equipment, and no additional cost would be required to implement trajectory-based monitoring. What would change is interpretation: instead of treating each laboratory result as an isolated data point, early warning protocols could flag patients whose values climb into or persist within the higher ranges during the first day, prompting earlier endoscopic evaluation, more aggressive resuscitation, closer renal monitoring, or escalation of care. Conversely, patients whose values settle and remain below 30 milligrams per deciliter for more than half a day might be identified as lower risk, potentially guiding earlier de-escalation and better allocation of scarce intensive care resources.

The authors are careful about the limits of their design. This was a retrospective cohort study, so the associations describe risk stratification rather than proving that any particular intervention that lowers blood urea nitrogen would improve survival. The requirement for at least twelve measurements in the first twenty-four hours selects a monitored subset of patients, and although sensitivity analyses in denser-measurement subsets were reassuring, the findings will need prospective validation before they reshape guidelines. The datasets came from de-identified records under PhysioNet data use agreements, the study was reported following the TRIPOD statement for prediction research, and the authors declared no conflicts of interest. Even with those caveats, the message is compelling: one of the cheapest, oldest tests in medicine, read as a moving curve rather than a single number, may help clinicians see which bleeding patients are quietly sliding toward death while there is still time to intervene.

Subject of Research: Hourly blood urea nitrogen trajectories as predictors of ICU mortality in critically ill patients with gastrointestinal bleeding

Article Title: Hourly Blood Urea Nitrogen Trajectories and ICU Mortality Risk in Critically Ill Patients with Gastrointestinal Bleeding: A Multicenter Retrospective Cohort Study

Article References: Liu, B., Zheng, X., Xu, Z., Yan, H., Du, Y., Yong, Z., & Xing, C. (2026). Hourly Blood Urea Nitrogen Trajectories and ICU Mortality Risk in Critically Ill Patients with Gastrointestinal Bleeding: A Multicenter Retrospective Cohort Study. Advances in Therapy. https://doi.org/10.1007/s12325-026-03796-w

Image Credits: AI Generated

DOI: 10.1007/s12325-026-03796-w

Keywords: blood urea nitrogen, gastrointestinal bleeding, ICU mortality, risk stratification, dynamic trajectories, critical care, MIMIC-IV, eICU, retrospective cohort study, predictive modeling, acute kidney injury, biomarkers

The study, led by Lavienraj Premraj and Christopher Camarda of Monash University and Monash Health in Australia, together with an international team of neurocritical care investigators, focused on patients with non-traumatic acute brain injury: intracerebral hemorrhage, aneurysmal subarachnoid hemorrhage, and acute ischemic stroke. These conditions strike more than a million people worldwide each year, and many of the most severely affected require invasive mechanical ventilation. In the ICU, the management of oxygenation is a daily balancing act. Too little oxygen risks starving vulnerable brain tissue — the penumbral zones surrounding a hemorrhage or an infarct — of the fuel it needs to survive. Too much oxygen, a condition known as hyperoxia, has been repeatedly linked in observational studies to oxidative stress, vasoconstriction, and worse outcomes.

Despite the stakes, the specific targets have remained stubbornly uncertain. The European Society of Intensive Care Medicine consensus panel currently recommends maintaining the partial pressure of arterial oxygen, or PaO2, between 80 and 120 millimeters of mercury in ventilated patients with brain injury. But that recommendation rests on limited prospective evidence. Major randomized trials, including ICU-ROX, UK-ROX, and most recently the LOGICAL trial in cardiac arrest patients, have generally failed to show that conservative oxygen strategies improve survival or neurological recovery, leaving clinicians without a clear mandate. None of those trials monitored oxygen directly inside brain tissue, and none was designed to pinpoint the PaO2 level associated with the lowest mortality in brain-injured patients specifically.

To break this impasse, the research team turned to Bayesian inference, a statistical approach that formally combines prior evidence with new data rather than treating each dataset in isolation. The final model was fitted to the Australian and New Zealand Intensive Care Society Adult Patient Dataset, a binational registry capturing 99 percent of Australian ICU admissions, which yielded 11,264 eligible ventilated patients: 4,846 with intracerebral hemorrhage, 4,319 with aneurysmal subarachnoid hemorrhage, and 2,099 with acute ischemic stroke. Crucially, the priors — the starting assumptions of the Bayesian machinery — were derived from two independent cohorts: the MIMIC-IV critical care database from the United States, contributing 1,789 patients, and the European ENIO observational study, contributing 544. The model was fitted sequentially, with posterior estimates from each cohort updating the priors for the next, allowing effect estimates to accumulate transparently across populations without assuming that any single dataset tells the whole story.

The analysis adjusted for a parsimonious set of clinically relevant confounders, including age, sex, Glasgow Coma Scale score on admission, and chronic cardiovascular or respiratory comorbidity. In the final model, estimates were further adjusted using the Australia and New Zealand Risk of Death score, with the oxygen-related components of the score deliberately removed to avoid statistical circularity, and random intercepts for hospital site and admission year accounted for institutional and temporal clustering. Thresholds were then identified using a region of practical equivalence: the widest continuous interval of PaO2 values within 4 percent of the minimum predicted mortality, a criterion that had to hold in 99 percent of the posterior samples. This stringent definition was designed to ensure that the resulting ranges were robust rather than artifacts of statistical noise.

The results revealed a striking heterogeneity across brain injury subtypes. For subarachnoid hemorrhage, the relationship between early PaO2 and ICU mortality retained a pronounced U-shape even after adjustment, with mortality rising steeply at both low and high oxygen tensions. The optimal range for these patients was 100 to 190 millimeters of mercury. For acute ischemic stroke, the curve was flatter but the optimal band was narrower still, at 100 to 160 millimeters of mercury. Intracerebral hemorrhage proved the most forgiving, with a broad safe zone stretching from 80 to 230 millimeters of mercury. In every case, the derived ranges were substantially wider than the 80 to 120 millimeter window endorsed by the ESICM, yet patients managed within these broader ranges showed essentially the same probability of ICU mortality as those falling within the narrower guideline-defined normoxia.

The danger zones, by contrast, were consistent and sobering. Extreme hyperoxia, defined as a PaO2 above 299 millimeters of mercury, was associated with increased ICU mortality across all three brain injury subtypes, with posterior probabilities ranging from 84 percent for intracerebral hemorrhage to 99 percent for subarachnoid hemorrhage. Hypoxia below the lower limit of the safe range also carried elevated risk, particularly in subarachnoid hemorrhage patients, where the odds of ICU death rose by roughly 28 percent. Even mild hyperoxia — oxygen levels above the upper boundary of the derived safe range but below the extreme threshold — was very likely harmful in patients with subarachnoid hemorrhage and ischemic stroke. Exploratory analyses using day-three blood gas values from the MIMIC-IV and ENIO cohorts reinforced these patterns, and showed that patients who developed or persisted in hyperoxia between day one and day three fared worse than those who avoided it entirely.

The mechanistic story behind these numbers is biologically plausible. In subarachnoid hemorrhage, elevated arterial oxygen may amplify an already substantial oxidative burden, promoting cellular and endothelial injury that increases the risk of vasospasm and delayed cerebral ischemia, both of which are strongly linked to death. In ischemic stroke patients treated with mechanical thrombectomy, high oxidative stress is already present, and excess oxygen may worsen reperfusion injury rather than rescue the penumbra. Notably, the study’s finding that higher PaO2 does not reliably translate into better brain tissue oxygenation echoes mechanistic work showing that arterial oxygen content is a poor proxy for oxygen delivery to ischemic tissue, particularly when cerebral autoregulation is disrupted.

The authors are careful to spell out the limitations. This was an observational study, so causation cannot be proven, and residual confounding may persist despite robust adjustment. The analysis relied on a single representative PaO2 value from the first 24 hours rather than continuous oxygen exposure, and none of the datasets included direct measurements of brain tissue oxygen tension. Functional neurological outcomes — arguably the endpoint that matters most to patients and families — could not be assessed. The choice of the 4 percent mortality band for defining the safe zone is, as the authors acknowledge, ultimately arbitrary, though they imposed a high standard of certainty by requiring consistency across 99 percent of posterior samples. Randomized trials, including the ongoing Mega-ROX Brains program, will be needed to confirm whether these liberalized ranges are truly safe.

Even with those caveats, the study represents a genuine advance in a field where practice has long outpaced evidence. By triangulating three international datasets and nearly 14,000 patients, it provides the first empirical evaluation of existing PaO2 recommendations in ventilated brain-injured patients, and it does so with a statistical framework that honors uncertainty rather than hiding it. For intensivists, the practical message is twofold: the anxiety-inducing narrowness of the current guideline window may be unnecessary, particularly for patients with intracerebral hemorrhage, but the upper limits are not negotiable — pushing arterial oxygen above roughly 300 millimeters of mercury is very likely to harm, whatever the underlying brain pathology. As neurocritical care moves toward personalized, physiology-guided ventilation, this work offers a data-driven map of where the true safe harbor lies.

Subject of Research: Optimal arterial oxygen tension thresholds in mechanically ventilated patients with non-traumatic acute brain injury

Article Title: Early Arterial Oxygen Tension Thresholds for Mechanically Ventilated Patients with Non-traumatic Acute Brain Injury: A Bayesian Analysis of Three Large Neurocritical Care Datasets

Article References: Early Arterial Oxygen Tension Thresholds for Mechanically Ventilated Patients with Non-traumatic Acute Brain Injury: A Bayesian Analysis of Three Large Neurocritical Care Datasets. (n.d.). https://doi.org/10.1007/s12028-026-02659-0

Image Credits: AI Generated

DOI: 10.1007/s12028-026-02659-0

Keywords: arterial oxygen tension, PaO2 thresholds, acute brain injury, mechanical ventilation, neurocritical care, subarachnoid hemorrhage, intracerebral hemorrhage, ischemic stroke, Bayesian analysis, hyperoxia, ICU mortality, ESICM guidelines

The historical record explains why skepticism took root. Studies conducted in the 1980s and 1990s reported in-hospital mortality approaching 90 percent among patients with acute myeloid leukemia who required invasive mechanical ventilation. Against such figures, many clinicians questioned whether intensive care offered anything more than a prolonged and painful death. Yet the evidence landscape has steadily transformed. Refinements in patient selection, combined with sweeping advances in supportive care, have progressively improved outcomes, and more recent studies report in-ICU mortality rates of approximately 45 percent for critically ill patients with acute leukemia. That trajectory, from near-certain fatality to a coin-flip chance of survival, represents one of the quiet revolutions in hematology and intensive care.

A major new analysis has now brought unprecedented statistical weight to this transformation. An individual participant data meta-analysis by Chean and colleagues, published in Intensive Care Medicine, examines the outcomes of critically ill adults with acute leukemia using data from 2,003 patients treated in 55 intensive care units across 19 countries, with an inclusion period spanning more than two decades, from January 2000 to July 2023. The scale and duration of the study fill a long-standing knowledge gap created by earlier single-center investigations and limited sample-size analyses that focused mainly on patients younger than 60 years, a group that no longer reflects the real-world demographics of the disease.

What distinguishes this meta-analysis methodologically is its use of the individual participant data approach. Unlike conventional meta-analyses that rely on aggregated results extracted from published papers, an IPD meta-analysis goes back to the raw data of each contributing study and reanalyzes them. This allows researchers to improve both the quantity and quality of the evidence, to include information that was never reported in the original publications, and to standardize analyses and outcome definitions across studies. The method demands that investigators contact the corresponding author of every included trial or cohort, an arduous process that explains why such analyses remain relatively rare but carry exceptional authority when completed.

The findings confirm what many specialists hoped but could not previously prove with such rigor: ICU mortality has decreased over time among patients requiring mechanical ventilation. The authors of the accompanying editorial, Guillaume Berton and Norbert Vey of Aix-Marseille University and Institut Paoli-Calmettes in Marseille, attribute this trend to improvements on two fronts simultaneously. On the hematology side, new chemotherapy regimens have emerged, while supportive care has advanced dramatically through antimicrobial prophylaxis and improved management of febrile neutropenia, the dangerous fever that strikes patients whose infection-fighting white cells have been wiped out by both disease and treatment.

Perhaps the most striking demographic shift documented in the analysis is the changing age profile of patients reaching the ICU. The proportion of patients older than 65 years admitted to intensive care nearly doubled over the study period, rising from 18 percent before 2010 to 33 percent after 2010. This is not a coincidence of referral patterns but a direct consequence of therapeutic innovation. For decades, elderly patients with acute leukemia were often offered palliative care because intensive cytotoxic chemotherapy, the gold standard for younger patients, was considered too toxic for bodies with diminished functional reserves. That paradigm has been upended by non-intensive approaches, most notably the combination of hypomethylating agents with the BCL-2 inhibitor venetoclax, which has demonstrated a median overall survival of 14.7 months and has become the standard of care for AML patients ineligible for intensive chemotherapy.

The therapeutic revolution extends further. Although the Chean meta-analysis excluded post-transplant ICU admissions, advances in allogeneic stem-cell transplantation now allow this potentially curative procedure to be offered to increasingly older patients, with evidence supporting matched sibling, unrelated, and haploidentical donor options in those over 60. Together, these developments signal that treatment for older AML patients has moved decisively from palliative approaches to strategies with curative intent, and the resulting lengthening of survival will only reinforce the role of intensive care in this population. But with more therapeutic possibilities come new dilemmas, above all patient selection, which must rest on a careful evaluation of frailty and is tightly interwoven with the triage decisions made before ICU admission is even offered.

The meta-analysis also delivered a finding that unsettles prior assumptions: older age, defined as above 65 years, emerged as an adverse predictor of in-ICU mortality. This contrasts with earlier studies that had reported in-ICU mortality rates for elderly patients similar to those of younger cohorts. The editorial authors suggest the discrepancy may reflect variable levels of expertise among the participating centers, illustrated by differences in outcomes according to the national economic level of the countries involved. Such disparities may influence not only the quality of intensive care management but also which patients are selected for admission in the first place, raising uncomfortable questions about equity of access to critical care for older patients with leukemia.

The stakes of that triage question are amplified by what happens after patients leave the unit. A recent report highlighted in the editorial found that a significant proportion of AML patients require modifications to their antileukemic treatment following ICU discharge, altering their long-term prognosis, with older age increasing this risk. In other words, the ICU episode does not exist in isolation; it can derail the optimal cancer therapy on which survival ultimately depends. Conversely, one important limitation of the Chean analysis is that treatment response was not considered, even though response is a major determinant of survival and of the success of organ failure management. Future studies, the editorial argues, should incorporate modern response criteria, including the residual disease assessments now routinely used in hematology practice.

Finally, the editorial closes with a human dimension that data alone cannot capture: patients’ own wishes. Goals-of-care discussions are key to determining the most appropriate management strategy for critically ill patients with cancer, yet they remain relatively infrequent in practice, a gap that demands further research. Meanwhile, the treatment landscape continues to accelerate. More potent antileukemic agents are inducing deeper and longer responses, but each advance carries novel toxicities: venetoclax can trigger tumor lysis syndrome, targeted molecular therapies against FLT3, IDH, and menin can provoke differentiation syndrome causing respiratory, hemodynamic, and renal failure, and the spread of immunotherapy brings risks exemplified by cytokine release syndrome and immune cell-associated neurotoxicity seen with T-cell engagers and CAR T cells in acute lymphoblastic leukemia. As these therapies expand, the editorial concludes, more than ever a continuous close collaboration between intensivists and hematologists is paramount to ensure proper management of these increasingly complex, increasingly survivable patients.

Subject of Research: Intensive care unit access and outcomes for critically ill adults with acute leukemia

Article Title: Beyond classical boundaries: rethinking ICU access in acute leukemia

Article References: Berton, G., & Vey, N. (2026). Beyond classical boundaries: rethinking ICU access in acute leukemia. Intensive Care Medicine. https://doi.org/10.1007/s00134-026-08623-y

Image Credits: AI Generated

DOI: 10.1007/s00134-026-08623-y

Keywords: acute leukemia, intensive care unit, acute myeloid leukemia, ICU mortality, mechanical ventilation, venetoclax, hypomethylating agents, allogeneic stem-cell transplantation, tumor lysis syndrome, individual participant data meta-analysis, elderly patients, goals of care