Admission, Triage, and Discharge Criteria for Oncological Patients in the Intensive Care Unit

Background and Justification The admission of a critically ill patient with cancer to the intensive care unit (ICU) is a complex decision at the intersection of evolving oncology therapeutics, expanding critical care capability, and medical ethics. While historical perceptions of poor outcomes persist, contemporary data demonstrate improved survival, particularly among patients with reversible acute illness and treatment-responsive underlying malignancy. This progress has, however, generated a different set of problems in regards to appropriate patient selection, equitable allocation of a scarce resource, and avoidance of non-beneficial or disproportionately burdensome care.

Substantial variability exists in ICU admission and discharge practice for patients with cancer across institutions and regions. This variability can lead to both under-utilisation of the ICU for patients with a high likelihood of benefit and over-utilisation for patients in the terminal phase of refractory disease.

A standardised framework is therefore required - one that moves beyond the outdated dichotomy of "cancer versus no cancer" and instead integrates the key prognostic determinants namely, reversibility of the acute illness, performance status and frailty, tumour biology and treatment trajectory, and the patient's own values and goals.

How admission and discharge decisions differ from those in non-oncological patients Several features distinguish this population and prevent direct adaptation of general ICU triage guidance, such as the Society of Critical Care Medicine framework for ICU admission, discharge, and triage, which does not address oncology-specific determinants in detail.

In non-cancer patients, ICU admission decisions rely principally on acute physiological derangement and its reversibility. In patients with cancer, a dual prognostic assessment is required: acute illness severity and organ failure are assessed with the usual severity assessment tools, while underlying cancer biology (histology, stage, molecular markers), treatment responsiveness, and performance status (ECOG or Karnofsky) must be assessed in parallel. Acute illness severity predicts short-term ICU mortality, whereas longer-term survival is determined predominantly by cancer-specific factors such as disease status and response to therapy.

In general ICU practice, reversibility refers to recovery of organ function. In oncology critical care, reversibility may additionally depend on delivering cancer-directed therapy while organ support is provided. Acute respiratory failure due to pulmonary lymphomatous infiltration may require mechanical ventilation while chemotherapy is administered; the reversible component is the tumour's response to treatment as well as the organ dysfunction itself.

Where ICU admission for a non-cancer patient with uncertain prognosis is often open-ended, a pre-defined time-limited trial with an agreed reassessment point is frequently the appropriate structure for a patient with cancer and an intermediate prognosis. Haematopoietic stem cell transplant recipients, recipients of CAR-T cell therapy, and patients with haematological malignancy and neutropenic sepsis may warrant priority for ICU admission rather than deferral.

Considerations without a general-ICU equivalent include whether cancer-directed therapy is curative or palliative in intent, whether the patient is receiving induction or late-line therapy, and whether this is a first or a repeated admission for progressive disease. Tumour lysis syndrome, superior vena cava syndrome with airway compromise, hyperleukocytosis with leukostasis, and malignant spinal cord compression with neurological deficit demand specific triage pathways.

For patients with cancer, discharge planning must additionally integrate the oncology treatment schedule and supportive care needs. For patients with limited prognosis despite ICU care, transition to palliative care is a defined destination requiring early palliative team involvement, patient and family acceptance of a change in goals, and structured handover. This pathway is used more frequently in oncology than in other ICU populations.

Why a consensus exercise, and why the Delphi method The evidence relevant to admission, triage, and discharge decisions in this population consists predominantly of observational cohorts, single-centre series, and indirect data. Randomised evidence bearing directly on who should be admitted, when a time-limited trial should be concluded, or when discharge is safe does not exist and is unlikely to be generated. Where evidence is limited or absent but decisions must nonetheless be made daily, formal consensus methods are the appropriate means of collating and structuring expert judgement, and are more reliable than the unstructured opinion of individuals or of an unstructured meeting. The method is described as "modified" because candidate statements are generated by the Steering Committee from a structured literature review rather than by open idea-generation among panellists in Round one; panellists may nonetheless propose additional statements during Round one.

Aim, intended audience, and scope of application The aim is to generate an international expert consensus statement on the criteria and procedures for ICU admission, triage, and discharge of adult patients with cancer. The intended audience comprises intensivists, medical and surgical oncologists, haematologists, palliative medicine and emergency physicians, critical care nurses and hospital administrators, and health policymakers. The intended geographical scope is global: for application in both high-resource and resource-limited health systems.

Relationship to existing documents This is a new consensus statement and not an update or adaptation of any existing document. Where a consensus statement of this panel diverges from that framework or from other published guidance, the divergence and its rationale will be identified explicitly in the resulting manuscript.

Objectives Primary objective To develop an international consensus statement providing explicit criteria and a procedural framework for the admission, triage, and discharge of adult critically ill patients with cancer in the ICU.

Secondary objectives

• To define the conduct of time-limited trials of ICU therapy - indications, duration, pre-specified reassessment criteria, and documentation - as the principal structured response to prognostic uncertainty.
• To provide criteria for safe and appropriate ICU discharge, including transition to ward-based care, palliative care units, or hospice.
• To identify and publish the questions for which the panel finds insufficient evidence or fails to reach consensus, as a prioritised research agenda for oncology critical care.

Methods Scope and population The statement applies to adults (≥18 years) with active haematological malignancy or solid tumour who present with, or develop, a critical illness requiring consideration of ICU admission, or who are admitted to the ICU and being considered for discharge. Both emergency and planned admissions are included.

Registration, protocol availability, and amendments This protocol will be posted at www.oncoccs.com before the first invitation to vote is issued. The protocol is version-controlled. Any amendment made after registration will be recorded.

Steering Committee: composition, selection, and role The Steering Committee (SC) comprises 12 to 14 members: intensivists with oncology critical care practice, medical oncologists, palliative medicine physicians, and an independent Delphi methodology expert. Members were identified by the Principal Investigator and the OCCS on the basis of clinical and scholarly standing in oncology and critical care. All SC members will be named in the manuscript, with their discipline, country, and role.

The SC is responsible for scope definition, the structured literature review, drafting and revision of candidate statements, adjudication of the consensus process, and preparation of the manuscript.

SC members will not vote in any Delphi round and will not be members of the expert panel. Where the SC exercises judgement that affects the content of the final statement - revising the wording of a statement between rounds, merging or splitting items, or retaining an item that failed to reach consensus. The SC will not reinstate any statement rejected by the panel without unanimous SC agreement and explicit reporting of that fact.

The independent Delphi methodology expert administers the surveys, holds the participant key, performs the anonymised analysis, and prepares the between-round feedback. The methodology expert does not draft or revise statements and does not vote.

Expert panel: eligibility criteria and panel size Eligibility requires all of the following: (i) a substantial clinical practice in the care of critically ill patients with cancer, in intensive care, oncology, haematology, palliative medicine, or emergency medicine; (ii) at least five years of post-training practice; and (iii) demonstrated scholarly contribution to the field - peer-reviewed publication, invited lecturing, or a leadership role in a relevant professional society or guideline body. Panellists will be selected by the Steering Committee against these criteria.

The target panel size is 30-40 voting members to preserve an adequate number of respondents after attrition across three rounds and disciplinary and regional representation. A minimum of 24 completed responses will be required for any round to be analysed.

Composition will be reported against the following requirements:

• Representation from all six World Health Organization regions.
• Representation of both high-income and low- or middle-income health systems, and of both academic and non-academic practice settings.
• Multidisciplinary composition spanning intensive care, medical oncology, haematology, palliative medicine, and emergency medicine.

Recruitment of the panel Candidate panellists will be identified from the author lists of publications retrieved in the structured literature review and from the membership and faculty lists of relevant professional societies. Invitations will be issued by e-mail from a single project address under the centralised oversight of the Steering Committee. Up to two reminders will be sent to non-responders.

Invitees will not be asked to nominate additional panellists, except that an invitee who declines may propose a colleague of equivalent standing as a replacement; any such substitution requires SC approval against the eligibility criteria and will be recorded. The number of invitations issued, the number accepted, the number declining, and the number not responding will be reported. Participation is voluntary; consent is implied by completion of a survey questionnaire, and the information pack sent to each panellist will state this explicitly.

Preparatory research and evidence summaries Candidate statements will be generated from a structured review of the literature. The search will cover PubMed/MEDLINE, Cochrane Central Register of Controlled Trials from inception to the date of the search, using a combination of MeSH terms and free-text keywords; the full search strings, the dates on which each search was run, and the yield at each stage will be reported in the supplementary material.

Development of candidate statements

The steering group will convert the evidence summaries and identified gaps into candidate statements grouped into pre-specified clinical domains, subject to the following rules:

• Each statement addresses a single, independently answerable proposition; bundled propositions combining more than one clinical question will be split before Round 1.
• Statements are worded so that agreement and disagreement are both meaningful positions, and so that a statement cannot be endorsed merely because it is unobjectionable.

The complete Round 1 instrument, the domain structure, will be published as supplementary material.

Structure and objectives of the consensus rounds

A minimum of three rounds will be conducted. Each round has a distinct, pre-specified objective:

Step Objective Items presented Round 1 Obtain first-pass agreement across the full statement set; elicit proposed rewording and additional statements All candidate statements Round 2 Confirm the stability of statements that reached consensus in Round 1; re-vote on revised statements; vote on statements newly proposed by panellists in Round 1 Statements reaching consensus in Round 1 (for confirmation), revised statements, and new statements Round 3 Resolve statements that remain without consensus or that showed unstable agreement across Rounds 1 and 2 Unresolved and unstable statements only Round 4 (conditional) Conducted only if instability persists on more than 10% of items after Round 3; otherwise not conducted Unstable statements only

Statements are retired from further voting once they have met the consensus definition in two consecutive rounds. A statement reaching consensus in a single round is not retired. Statements failing to reach consensus after the final round are recorded as "no consensus" and are not carried forward.

Voting instrument, response options, and administration Surveys will be administered electronically on a secure survey platform. Responses will be collected individually and asynchronously; at no point will responses be collected in a group setting. Each statement will be presented with its evidence summary, a seven-point Likert response (1 = strongly disagree; 4 = neither agree nor disagree; 7 = strongly agree), and a free-text field.

Free-text comment is optional for every item and required for any vote of 1-3 on a statement, so that disagreement is accompanied by its reasoning and can be acted upon between rounds. In Round 1 only, panellists may propose additional statements for inclusion in subsequent rounds. Panellists may abstain from any item on grounds of insufficient expertise; abstentions are excluded from the denominator and reported separately.

Statements will be presented grouped by clinical domain in a fixed order rather than randomised, because the domains follow the clinical sequence of admission, triage, and discharge and random presentation would impair comprehension.

Surveys will be conducted in English. A plain-language summary will accompany the information pack, each round will remain open for a minimum of three weeks with two reminders, and panellists in any time zone or with limited connectivity may request an extension or a document-based version of the ballot. Any adaptation granted will be recorded.

Definition of consensus and its rationale For Likert-scale items, agreement is defined as a score of 5-7 and disagreement as a score of 1-3; 4 denotes neither agreement nor disagreement and is included in the denominator. All thresholds are specified a priori.

Category Likert-scale items Strong consensus (opinion unlikely to change with time) Median 6-7 with ≥75% of votes in the 6-7 range, or median 1-2 with ≥75% of votes in the 1-2 range Consensus Median ≥5 with ≥75% of votes in the 5-7 range, or median ≤3 with ≥75% of votes in the 1-3 range No consensus All items not meeting the above criteria

After closing of each round, and along with the subsequent round, every panellist will receive: the anonymised distribution of votes for each statement, expressed as the number and percentage in each response category; the median and interquartile range; so that individual position can be compared with the group; an anonymised thematic synthesis of the free-text comments; and, for every statement whose wording has changed, a side-by-side display of the previous and revised text with the change highlighted.

Anonymity Anonymity is planned by design and applies among panellists and between panellists and the Steering Committee. Panellists will not know the identity of other panellists until the study is complete, at which point those consenting will be named in the collaborative group. The Steering Committee receives aggregate and free-text data with identifiers removed and at no point has access to identifiable individual responses.

Processing and analysis of responses Quantitative responses will be summarised as counts and percentages in each response category, with median and interquartile range for Likert items and percentage selection for multiple-choice items. Free-text responses will be analysed by two members of the Steering Committee. Steering group will decide whether an item is revised, split, merged, or retained unchanged for the next round.

Stability of responses will be assessed by comparing the distribution of responses for each statement between two consecutive rounds. To avoid sparse cells at a panel size of 30-40, responses will be collapsed into three ordered categories (1-3, 4, 5-7) before non-parametric chi-square (χ²) testing; where expected cell counts fall below five, Fisher's exact test will be substituted. A p value \<0.05 indicates instability. An item achieving agreement in one round and falling into disagreement in the next is classified as unstable and is revised and re-presented rather than counted as having reached consensus.

The stopping rule requires demonstration of stability across two consecutive rounds, irrespective of whether consensus or divided opinion is observed. Statements with persistent but stable division will be reported as: "No consensus statement can be made owing to divided expert opinion." Attrition bias will be assessed by comparing response distributions between rounds and, where feasible, between completers and non-completers of the subsequent round.

Analyses will be performed using Microsoft Office 365 Excel (Microsoft Corp., Redmond, WA, USA) and IBM SPSS Statistics for Windows, Version 26.0 (IBM Corp., Armonk, NY, USA). No inferential comparison between subgroups of panellists is planned; any such analysis, if undertaken, will be reported as exploratory and unplanned.

Incentives No payment, honorarium, reimbursement, or other incentive will be offered to Steering Committee members or panellists. The only acknowledgement of participation is authorship or collaborator status as set out under Manuscript Preparation and Authorship, and the criteria for this are stated in the invitation.

Data management During the study, anonymised data will be stored securely by the methodology expert. On completion, anonymised datasets will be retained by the first and corresponding authors and made available on reasonable request, subject to institutional approval. The participant key will be destroyed once the collaborative group byline has been finalised.

Ethical considerations The study involves healthcare professionals and collects no patient-identifiable clinical data. Institutional ethics committee approval will not be obtained. Participation is voluntary, panellists may withdraw at any round without giving a reason, and consent is implied by completion of a survey questionnaire.

Formulation of the Final Consensus Statements

Final consensus statements will be derived from items achieving consensus and stability. Wording will be calibrated to the degree of agreement obtained:

• Statements achieving strong consensus will use directive language (for example, "should" or "should not").
• Statements achieving consensus of moderate strength will use conditional language (for example, "may" or "may be considered").
• Items failing to achieve consensus will be published as such, with the response distribution and a summary of the arguments advanced on each side.

Every published statement will carry an explicit label identifying it as expert consensus and will state the percentage agreement and median on which it rests.

Planned Reporting The manuscript will be prepared in accordance with the ACCORD checklist, and a completed checklist will accompany the submission.

Results to be reported

• The date on which the consensus exercise opened, the opening and closing date of each round, the time taken for analysis between rounds, and any extension or delay, with its reason.
• For each round: the number of panellists invited, the number responding, the response rate, and the sociodemographic and professional composition of respondents - discipline, country, WHO region, practice setting and gender.
• For each statement and each round: the number and percentage of votes in each response category, the median, the interquartile range, and the number of abstentions.
• A thematic account of the free-text comments underlying each unresolved or revised statement.
• Every statement modified, split, merged, added, or removed during the process, with the round at which the change was made and the reason for it, presented as a statement-evolution table.
• A flow diagram showing the number of statements entering and leaving each round, and the disposition of every statement.
• Every deviation from this registered protocol, with its reason. Discussion elements to be reported
• Methodological strengths and limitations, addressing at a minimum: response rates and attrition; the representativeness of the panel by region, discipline, the potential for between-round feedback to induce conformity rather than genuine reconsideration; the influence of Steering Committee decisions on statement wording; and the absence of formal evidence appraisal.
• Whether the consensus statements are consistent with existing published literature and guidance, including the SCCM admission, discharge, and triage framework; where they diverge, the reasons this process may have reached a different position.

Other information to be reported

• Endorsing organisations and the role of each. The OCCS is the sponsoring society; endorsement will be sought from other national and international societies after the final statements are agreed, and any such organisation's role - endorsement only, or participation in the process - will be stated.
• Conflicts of interest of all Steering Committee members and panellists, financial and intellectual, declared before participation on a standard form, updated at the conclusion of the process, published in full as a supplementary document, and accompanied by a description of how conflicts were managed, including any recusal from voting on specific statements.
• Use of AI-assisted technology, in the form set out in the Declaration of the Use of Artificial Intelligence, declared in the cover letter, the Methods, and the Acknowledgements.
• Funding: none received for the conduct or publication of this study; no funder has had any role in design, conduct, analysis, or reporting. Should any funding be obtained subsequently, the source and the funder's role will be stated.

Manuscript Preparation and Authorship The Steering Committee will synthesise the evidence summaries and the statements achieving consensus into a cohesive manuscript. Authorship will follow ICMJE criteria: a named writing committee comprising members who meet all four criteria, together with a collaborative group byline listing all panellists who completed the required voting rounds, indexed as collaborators. Individual contributions will be described using the CRediT taxonomy. The criteria for inclusion in the collaborative byline - completion of a specified minimum number of rounds - will be stated in the invitation, so that panellists know at the outset what participation entails. The manuscript will be circulated to the full panel for review and approval before submission.

Declaration of the Use of Artificial Intelligence Artificial intelligence (AI)-assisted technology was used in the preparation of this protocol and language edition. This section records that use prospectively so that the declaration required at submission is drawn from a contemporaneous record rather than reconstructed. The declaration follows the ICMJE Recommendations, which require disclosure at submission of any use of AI-assisted technologies, description of that use in the cover letter and in the appropriate section of the manuscript, and retention of full human accountability for the content; the position statements of COPE and WAME are consistent with this.

Tools used, and the purpose of each use Tool and developer Version and period of use Purpose and verification Claude (Anthropic PBC, San Francisco, CA, USA) Claude Opus 5; date of use 01.08.2026 Drafting and editing of the protocol text; document formatting. All output reviewed, corrected, and approved by the Principal Investigator and the Steering Committee; all references independently verified against PubMed

Reporting of AI use in the manuscript The manuscript will carry the declaration in line with ICMJE guidance: a statement at submission in the cover letter; a description in the Methods of any AI use bearing on the generation of study materials, including the drafting of candidate statements and evidence summaries; and a statement in the Acknowledgements of AI use for writing and language editing. The wording will name the tool, its developer, the model version, the period of use, and the purpose, and will state that no AI tool contributed to panel selection, voting, analysis, or interpretation, and that the authors take responsibility for the final content. If the target journal specifies a different declaration format, that format will be followed and the substance retained. Any additional AI use arising later in the project will be added to the table of tools and purposes above at the time it occurs.

Planned Timeline Period Activity Output Months 1-2 Steering Committee formation; scope and clinical questions finalised; protocol registered Registered protocol Months 3-5 Structured literature review; evidence summaries; drafting of candidate statements Evidence summaries; candidate statements Months 5-6 Panel constitution; conflict-of-interest declarations; recruitment of patient representatives; piloting of the Round 1 instrument Final panel roster; piloted Round 1 instrument Month 7 Delphi Round 1 Round 1 results; feedback report Month 8 Analysis, statement revision, Delphi Round 2 Round 2 results; feedback report Month 9 Analysis, statement revision, Delphi Round 3 Final consensus statements Month 10 Drafting of the full consensus statement document Complete draft Month 11 Panel review, final revisions, ACCORD checklist completion, submission Submitted manuscript

Dissemination and Implementation

• Publication: submission to a peer-reviewed critical care or oncology journal, with the completed ACCORD checklist as a supplementary file.
• Executive summary: a one-page infographic and clinical algorithm for bedside use.
• Society channels: launch at the OCCS Annual Congress; publication on the OCCS website; dissemination through society newsletters and webinars.
• Translation and local adaptation: institutions adopting the statement are encouraged to adapt it to local resource constraints and to document the adaptation.

Review and Update The Steering Committee will undertake a formal review and update every five years, or sooner in response to major shifts in cancer therapeutics or critical care evidence. Any update will state why it was needed and will cite this statement as the document being updated.

Anticipated Limitations

• The output reflects expert opinion informed by a structured but non-systematic review of the literature, and carries the limitations inherent to consensus methods, including sensitivity to the composition of the panel.
• Consensus is not evidence. Agreement among experts may be high on questions where the underlying data are weak or absent; the published statements will make the evidential basis of each item explicit for this reason.
• The consensus thresholds are conventional but arbitrary, and items falling marginally below threshold will be reported transparently rather than discarded silently.
• Between-round feedback may induce conformity to the majority rather than genuine reconsideration; the free-text record will be used to distinguish reasoned position change from drift toward the mode.
• Statement wording is controlled by the Steering Committee, which introduces a source of influence that anonymised voting does not neutralise. All wording changes will be published so that this influence is visible.
• Panel composition, although internationally diverse, cannot represent every health system; applicability in severely resource-limited settings will require local adaptation.
• Conducting the process in English restricts participation and may under-represent non-anglophone practice.
• Attrition across rounds may bias the final result; attrition will be quantified and analysed rather than assumed to be non-differential.