Characteristics and Outcomes of Patients Referred but Declined for ICU Admission in a Resource-Limited Setting in South Africa

Abstract

OBJECTIVES:

ICU beds are a limited resource and assessing net benefit from ICU admission remains incompletely understood. We sought to examine characteristics and outcomes of patients referred but declined for ICU admission, stratified by clinician-perceived likelihood of ICU benefit.

DESIGN:

Descriptive, retrospective cohort study.

SETTING:

Two public hospitals in the KwaZulu-Natal Department of Health, South Africa, from January 2022 to June 2022.

PATIENTS:

Adult ICU-referred patients classified into four triage cohorts by clinician-assessed acuity and ICU triage status: ICU admissions; high-acuity capacity-declined ICU referrals; high-acuity prognosis-declined ICU referrals; and low-acuity prognosis-declined ICU referrals.

INTERVENTIONS:

None.

MEASUREMENTS AND MAIN RESULTS:

We report descriptive statistics of ICU-referred patient characteristics stratified by triage cohorts and mortal status. Among 866 ICU referrals, 469 patients (54.2%) were admitted, and 397 patients (45.8%) were declined for ICU admission. ICU admissions had an observed all-cause 30-day hospital mortality of 27.7%. Among declined ICU referrals with nonmissing triage priority scores (n = 327), 250 patients (76.5%) were declined ICU admission for prognosis reasons, and 77 patients (23.5%) were declined for capacity reasons. Seventy-seven (9.7% of ICU referrals) high-acuity capacity-declined ICU referrals (“appropriate patient; no bed capacity; not admitted”) had a 26.0% mortality (74.0% survival), 111 (13.9%) high-acuity prognosis-declined ICU referrals (“too sick; not admitted”) had a 53.2% mortality (46.8% survival), and 139 (17.5%) low-acuity prognosis-declined ICU referrals (“too well; not admitted”) had a 7.2% mortality. Declined ICU referrals were less likely surgical/trauma patients and more likely medical patients coming from wards and emergency departments.

CONCLUSIONS:

In one sampling in the South African public healthcare system, many patients are referred but declined for ICU admission due to perceptions of being both “too well” or “too sick” to benefit and due to bed capacity, with a proportion of such patients experiencing potential clinical deviations from triage-expected trajectories.

KEY POINTS.

Question: What are the characteristics and outcomes of patients referred but declined for ICU admission in a resource-limited setting?

Findings: In this descriptive, retrospective cohort study, in one sampling in the South African public healthcare system, we found that many patients are referred but declined for ICU admission due to perceptions of being both “too well” or “too sick” to benefit and due to bed capacity. A proportion of such patients experience potential clinical deviations from triage-expected trajectories.

Meaning: How best to optimize utilization of scarce ICU resources—including who benefits from ICU admission—remains under active investigation.

ICU beds are a limited healthcare resource, especially, but not exclusively, in low- and middle-income countries (LMICs) (1, 2). To manage competing demands for, and supply of, available ICU beds, hospitals and clinical teams deploy intentional and/or unintentional triage—prioritization and rationing—strategies to admit patients most in need of and/or most likely to benefit from critical care. Some hospitals in the South African public healthcare system, for example, use ICU triage systems to evaluate patients for the likelihood of benefiting from ICU care, or lack thereof, and admit or decline ICU referrals accordingly, often with declination rates approaching 50% (3, 4). In well-resourced settings, allocation of ICU beds occurs but may be less overt or visible, with ICU admission rates for borderline patients sensitive to contemporary hospital capacity strain (5–7).

Multiple studies have sought to measure the outcomes of patients referred but declined, or delayed, for ICU admission as a result of acute ICU capacity strain and/or chronic ICU resource limitations (4, 7–19). However, there is an important missing nuance when declined ICU referrals are treated as a unified pool of patients for the purpose of outcomes assessment. ICU-referred patients may be “clinically appropriate” for ICU admission but declined admission due to acute or chronic lack of available ICU beds at the time of referral. Alternatively, ICU-referred patients may be declined ICU admission, with or without concomitant ICU bed shortage, because they are deemed “too well to benefit” (i.e., of low acuity) or “too sick to benefit” (i.e., of unrecoverably high acuity/comorbidity) (4, 20–22). Analytically pooling these patients across bimodal extremes of acuity and expected outcome runs the risk of masking important signals and opportunities for optimally matching patients with the most appropriate level of care for them. Further, in well-resourced settings, visualizing the counterfactual of patients overtly declined for ICU admission is difficult, given the reality of more subtle dynamic triage decisions (5–7).

We performed a descriptive, retrospective cohort study to examine characteristics and outcomes of patients referred but declined for ICU admission stratified by subjective, clinician-perceived likelihood of ICU benefit and acuity.

METHODS

Study Setting and Care Delivery Structure

South Africa ICU Capacity Strain (SA-ICU-Strain) Study Group is a longitudinal collaboration between the University of Pennsylvania Perelman School of Medicine and the KwaZulu-Natal Department of Health (Pietermaritzburg) and the University of KwaZulu-Natal School of Clinical Medicine (Durban) in South Africa. This study was conducted at Harry Gwala Regional Hospital and Grey’s Hospital, two KwaZulu-Natal Department of Health hospitals in the South African public healthcare system. The study hospitals serve as primary hospitals for the local populations and as mid-level regional referral hospitals for a network of district hospitals and community health centers (3, 23). ICU capabilities include nine ICU beds among 530 hospital beds (1.7%) and 11 ICU beds among 900 hospital beds (1.2%), respectively. The ICUs can provide mechanical ventilation, vasopressor infusions, and continuous hemodynamic and vital sign monitoring. While not prohibited, ward staffing and clinical infrastructure prevent routine use of mechanical ventilation and vasopressors, and vital signs can only be monitored on shift or spot checks, not continuously.

Clinicians in the emergency department, operating theater, wards, and other hospitals in the region may refer patients for consideration for ICU care, at which point they undergo a triage evaluation by the ICU team, which makes a decision on ICU admission vs. declination. As part of their usual practice, the study hospitals’ critical care teams use the 1999 Society of Critical Care Medicine (SCCM) ICU Triage Priority system (21, 24), a subjective, clinician assessment of perceived ICU need and potential ICU benefit, with the goal to only admit patients with priority I (needs ICU-specific therapies), priority II (needs intensive monitoring), and priority III (less likely to benefit from ICU), as bed capacity allows, and to avoid admitting patients with priority IV-A (too well to benefit from ICU) and priority IV-B (too sick [acutely and/or comorbidly] to benefit from ICU) regardless of bed capacity (3). Final SCCM ICU Triage Priority is ultimately determined subjectively by the on-call consultant (attending physician) at the time of the ICU referral. Of note, SCCM released updated ICU Admission, Discharge, and Triage Guidelines in 2016 (22). While relying on an updated literature base, the 2016 guidelines include the same broad principles of triage priority as in the 1999 guidelines. For example, 2016 ICU-priority 1 (“Critically ill patients who require life support for organ failure, intensive monitoring, and therapies only provided in the ICU environment”) aligns with 1999 priority I and priority II, 2016 ICU-priority 2 (“Patients, as described [in ICU-priority 1], with significantly lower probability of recovery and who would like to receive intensive care therapies”) aligns with 1999 priority III, and 2016 ICU-priority 5 (“Terminal or moribund patients with no possibility of recovery; such patients are in general not appropriate for ICU admission”) aligns with 1999 priority IV-B. The present retrospective study did not prospectively deploy the 1999 ICU Triage Priority system and does not overtly study that system specifically, but instead leverages the real use of the 1999 ICU Triage Priority system in the resource-limited study hospitals to measure characteristics and outcomes of patients declined for ICU admission for whom there is a contemporary clinician assessment of expected ICU need and benefit.

Study Population and Data Source

We studied all adult (age ≥ 18 yr) patients referred for ICU admission at the study hospitals from January 1, 2022, to June 30, 2022. The study data source was the Integrated Critical Care Electronic System (ICES) (25), which has served as the source for multiple prior studies (3, 23, 26–29). This ICU database, which is integrated into the critical care team clinical workflow, includes data on all referrals and admissions for ICU care (3, 25). Data are entered into ICES at the time of ICU referral, and ICU admissions have additional admission data captured and in-ICU outcomes recorded. Acute diagnosis and chronic comorbidity categories are determined in real-time by the clinical team, as described previously (3). Declined referrals and ICU survivors missing recorded patient identifiers (n = 95) were excluded, as were patients referred from other hospitals (or with an unknown referral origin; n = 119), as these patients often return to their referring hospital of origin and their hospital mortality status could therefore not be reliably assessed from records at the study hospitals alone.

Acuity Cohorts

ICU-referred patients were categorized into four triage cohorts by clinician-assessed acuity and ICU triage status. ICU admissions (“appropriate patient; admitted”) were considered de facto high acuity because, as above, the ICU-admitted patients by policy and practice are only among SCCM ICU Triage Priority I–III (3, 23); patients admitted to the ICU but with a missing SCCM ICU Triage Priority score (n = 63, 13.4%) were therefore still included. High-acuity capacity-declined ICU referrals were patients referred for ICU admission with SCCM ICU Triage Priority score of I, II, or III but declined due to bed availability (“appropriate patient; no bed capacity; not admitted”). High-acuity prognosis-declined ICU referrals were patients referred for ICU admission with SCCM ICU Triage Priority score of IV-B (too sick to benefit from ICU), but declined (“too sick; not admitted”). Low-acuity prognosis-declined ICU referrals were patients referred for ICU admission with SCCM ICU Triage Priority score of IV-A (too well to benefit from ICU) but declined (“too well; not admitted”). Patients were excluded from analyses if they were declined for ICU admission and had a missing SCCM ICU Triage Priority score (n = 70, 8.1%), as they could not be classified into a triage cohort.

Primary Outcome and Outcome Ascertainment

The primary outcome was 30-day all-cause in-hospital mortality, with time zero at the time of ICU referral. The ICES database does not follow patients who are declined ICU admission and does not capture post-ICU outcomes for patients discharged alive from the ICU, and no local hospital electronic medical record system exists. To assess hospital outcomes for patients referred but declined for ICU admission and ICU survivors after transfer to the ward, four investigators (J.I., S.A.A., N.N., Z.F.) performed a manual review of paper mortuary records, which are recorded for all in-hospital deaths.

We considered patients to have a “potential clinical deviation from a triage-expected trajectory” if they were high-acuity capacity-declined hospital survivors, high-acuity prognosis-declined hospital survivors, or low-acuity prognosis-declined decedents. This assumes not that “all patients” declined for ICU admission due to capacity constraints or too-high acuity are expected to die in the hospital, or that “all patients” declined for ICU admission due to being too well are expected to survive to hospital discharge, but that the “typical such patients” are “more likely” to have that respective binary outcome if unable to access ICU resources.

Descriptive Statistics

We report descriptive statistics on patient characteristics—demographics, ICU referral and admission details, acute diagnoses, and chronic comorbidities—and observed ICU and hospital outcomes, stratified by triage cohort, and within triage cohorts, by mortality status (all-cause 30-d hospital survivors vs. decedents). We then performed a descriptive report of the patients with potential clinical deviations from triage-expected trajectories among ICU declinations.

The study protocol was approved by the Biomedical Research Ethics Committee of the University of KwaZulu-Natal (“Class Approval for a Critical Care Database,” October 21, 2024, Protocol No. BCA211/14, Durban, South Africa) and by the Institutional Review Board of the University of Pennsylvania (“Association of ICU Capacity Strain and Mortality in a Resource-Limited Setting,” Protocol No. 824688, January 31, 2019, Philadelphia, PA). The procedures followed were in accordance with the ethical standards of the Institutional Review Boards and with the Declaration of Helsinki of 1975, and with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for reporting observational studies (30).

RESULTS

Patient Characteristics Across Triage Cohorts

Among 866 ICU referrals, 469 (54.2%) were admitted, and 397 (45.8%) were declined for ICU admission. The primary analytic cohort (n = 796) included 469 ICU admissions (58.9%) and 327 declined ICU referrals (41.1%) with nonmissing SCCM Triage Priority scores (Fig. 1). Tables 1 and 2 report patient characteristics by triage cohort. ICU admissions had an observed all-cause 30-day hospital mortality of 27.7%. Consistent with known local hospital practice and prior studies (3, 23, 26, 28, 29, 31), zero patients were admitted who were assessed as ICU Triage Priority IV-A/too well to benefit or IV-B/too sick to benefit. Among declined ICU referrals with nonmissing triage priority scores (n = 327), 77 patients (23.5%) were declined ICU admission for capacity reasons, and 250 patients (76.5%) were declined ICU admission for prognosis reasons. Seventy-seven (9.7% of all analyzed ICU referrals) high-acuity capacity-declined ICU referrals had a 26.0% mortality (74.0% survival), 111 (13.9%) high-acuity prognosis-declined ICU referrals had a 53.2% mortality (46.8% survival), and 139 (17.5%) low-acuity prognosis-declined ICU referrals had a 7.2% mortality (92.8% survival).

Figure 1.

Cohort enrollment and ICU referral triage categorization. ICU-referred patients were categorized into four triage cohorts by clinician-assessed acuity and ICU triage status: ICU admissions; high-acuity capacity-declined ICU referrals (“appropriate patient; no bed capacity; not admitted”); high-acuity prognosis-declined ICU referrals (“too sick; not admitted”); and low-acuity prognosis-declined ICU referrals (“too well; not admitted”). SCCM = Society of Critical Care Medicine.

TABLE 1.

Patient Characteristics at ICU Referral by ICU Triage Cohort

Patient Characteristics at ICU Referral	ICU Admissions	Declined ICU Referralsa (n = 327, 41.1%)
		ICU Capacity-Declined/High Acuity	ICU Prognosis-Declined/High Acuity	ICU Prognosis-Declined/Low Acuity
	“Appropriate Patient; Admitted”	“Appropriate Patient; No Bed Capacity; Not Admitted”	“Too Sick; Not Admitted”	“Too Well; Not Admitted”
Patients, n	469 (58.9)	77 (9.7)	111 (13.9)	139 (17.5)
Age, yr, mean (sd)	41.6 (16.2)	43.1 (17.0)	48.9 (17.5)	39.7 (15.7)
Male gender, n (%)	268 (57.1)	38 (49.4)	53 (47.8)	71 (51.1)
Black race, n (%)	400 (91.3)	61 (83.6)	92 (92.0)	123 (89.5)
Society of Critical Care Medicine ICU Triage Priorityb, n (%)
I (needs ICU-specific therapy)	222 (54.7)	33 (42.9)	n/a	n/a
II (needs intensive monitoring)	129 (31.8)	20 (26.0)	n/a	n/a
III (less likely to benefit from ICU)	55 (13.6)	24 (31.2)	n/a	n/a
IV-A (too well to benefit from ICU)	0 (0.0)	n/a	n/a	139 (100.0)
IV-B (too sick to benefit from ICU)	0 (0.0)	n/a	111 (100.0)	n/a
Primary ICU referral diagnosisc, n (%)
Trauma	159 (38.4)	13 (17.3)	21 (19.1)	26 (19.3)
Infection	65 (15.7)	12 (16.0)	28 (25.5)	14 (10.4)
Other (nontrauma/noninfection)b	190 (45.9)	50 (66.7)	61 (55.5)	95 (70.4)
Referral source, n (%)
Emergency department	181 (38.6)	33 (42.9)	54 (48.7)	41 (29.5)
Operating theater	161 (34.3)	9 (11.7)	7 (6.3)	31 (22.3)
Ward	127 (27.1)	35 (45.5)	50 (45.1)	67 (48.2)
Referring service, n (%)
Surgical	372 (80.2)	56 (72.7)	60 (54.6)	113 (81.3)
Medical	92 (19.3)	21 (27.3)	50 (45.5)	26 (18.7)
Pre-referral hospital length of stay, median calendar days (IQR)	1 (1–2)	1 (1–2)	1 (1–3)	1 (1–2)
ICU length of stay, median calendar days (IQR)	4 (2–6)	n/a	n/a	n/a
ICU mortality, all-cause at 30 d from ICU referral, n (%)	94 (20.2)	n/a	n/a	n/a
Hospital mortality, all-cause at 30 d from ICU referral, n (%)	130 (27.7)	20 (26.0)	59 (53.2)	10 (7.2)
Hospital survival, at 30 d from ICU referral, n (%)	339 (72.3)	57 (74.0)	52 (46.8)	129 (92.8)

IQR = interquartile range, n/a = not applicable by definition.

^aWith nonmissing Society of Critical Care Medicine Triage Priority scores.

^bAssigned by the consultant (attending physician) at the time of ICU referral.

^cThe primary indication for ICU admission as determined by the admitting ICU team as part of real-time routine care.

Variables are reported as complete cases unless otherwise noted.

TABLE 2.

Comorbidities and Acute Diagnoses at ICU Referral by ICU Triage Cohort

Patient Characteristics at ICU Referral	ICU Admissions	Declined ICU Referralsa (n = 327, 41.1%)
	High Acuity	ICU Capacity-Declined/High Acuity	ICU Prognosis-Declined/High Acuity	ICU Prognosis-Declined/Low Acuity
	“Appropriate Patient; Admitted”	“Appropriate Patient; No Bed Capacity; Not Admitted”	“Too Sick; Not Admitted”	“Too Well; Not Admitted”
Patients, n	469 (58.9)	77 (9.7)	111 (13.9)	139 (17.5)
Chronic comorbiditiesb, n (%)
HIV	85 (18.1)	17 (22.1)	26 (23.4)	20 (14.4)
Chronic respiratory disease	17 (3.6)	3 (3.9)	5 (4.5)	3 (2.2)
Chronic cardiovascular disease	27 (5.8)	8 (10.4)	27 (24.3)	11 (7.9)
Diabetes	55 (11.7)	10 (13.0)	22 (19.8)	12 (8.6)
Chronic kidney disease	8 (1.7)	6 (7.8)	9 (8.1)	6 (4.3)
Chronic neurologic disease	9 (1.9)	2 (2.6)	4 (3.6)	3 (2.2)
Acute diagnoses and end-organ dysfunction at ICU admissionc, n (%)
Concern for infectiond	336 (71.6)	25 (32.5)	39 (35.1)	21 (15.1)
Sepsise	153 (32.6)	10 (13.0)	28 (25.2)	12 (8.6)
Trauma	162 (34.5)	13 (16.9)	21 (18.9)	26 (18.7)
Acute respiratory process	179 (38.2)	30 (39.0)	50 (45.1)	30 (21.6)
Acute cardiovascular process	87 (18.6)	25 (32.5)	32 (28.8)	21 (15.1)
Acute kidney injury	31 (6.6)	8 (10.4)	13 (11.7)	11 (7.9)
Acute metabolic derangements	197 (42.0)	32 (41.6)	52 (46.9)	52 (37.4)
Acute neurologic dysfunction (e.g., encephalopathy)	88 (18.8)	7 (9.1)	32 (28.8)	14 (10.1)
Acute bleeding	34 (7.3)	8 (10.4)	1 (0.9)	19 (13.7)

^aWith nonmissing Society of Critical Care Medicine Triage Priority scores.

^bAs determined by the admitting ICU team as part of real-time routine care.

^cAcute active diagnoses and processes present at the time of ICU admission (but not necessarily the primary indication for ICU admission) as determined by the admitting ICU team as part of real-time routine care.

^dConcern for infection includes infection as the primary indication for ICU admission or admission sepsis flag (both as determined by the admitting ICU team as part of real-time routine care) or receipt of antimicrobials at the time of ICU admission.

^eSepsis includes concern for infection, as above, and admission sepsis flag (as determined by the admitting ICU team as part of real-time routine care) or meeting criteria by quick Sequential Organ Failure Assessment score or Systemic Inflammatory Response Syndrome score when available.

High-acuity capacity-declined ICU referrals, compared with patients admitted to the ICU, were: less likely male (49.4 vs. 57.1%); similar age (mean 43.1 vs. 41.6 yr); less likely referred for trauma (17.3 vs. 38.4%); less likely referred from the operating theater (11.7 vs. 34.3%) and more likely referred from the emergency department (42.9 vs. 38.6%) or ward (45.5 vs. 27.1%); and more likely medical referrals (27.3 vs. 19.3%).

High-acuity prognosis-declined ICU referrals, compared with patients admitted to the ICU, were: older (mean 48.9 vs. 41.6 yr); less likely male (47.8 vs. 57.1%); less likely referred for trauma (19.1 vs. 38.4%) and more likely referred for infection (25.5 vs. 15.7%); less likely referred from the operating theater (6.3 vs. 34.3%) and more likely referred from the emergency department (48.7 vs. 38.6%) or ward (45.1 vs. 27.1%); and more likely medical referrals (45.5 vs. 19.3%) and less likely surgical referrals (54.6 vs. 80.2%).

Low-acuity prognosis-declined ICU referrals, compared with patients admitted to the ICU, were: less likely male (51.1 vs. 57.1%); similar age (mean 39.7 vs. 41.6 yr); less likely referred for trauma (19.3 vs. 38.4%); and less likely referred from the operating theater (22.3 vs. 34.3%) or the emergency department (29.5 vs. 38.6%) and more likely referred from the ward (48.2 vs. 27.1%).

Among the all-referred population, male patients were admitted more frequently (54.0% of ICU referrals but 57.1% of ICU admissions and 49.5% of ICU declinations) and had higher priority ICU Triage Priority scores than female patients (40.4 vs. 28.3% priority I), but ICU Triage Priority scores were similar between male (28.2% priority I) and female (26.3% priority I) patients when trauma referrals (83.6% male predominance) are excluded.

Potential Clinical Deviations From Triage-Expected Trajectories Among Declined ICU Referrals

Tables 3 and 4 report patient characteristics by triage cohort stratified by hospital mortality status. Among 327 declined ICU referrals, patients with potential clinical deviations from triage-expected trajectories included: 57 (74.0%) high-acuity capacity-declined ICU referral survivors, 52 (46.8%) high-acuity prognosis-declined ICU referral survivors, and 10 (7.2%) low-acuity prognosis-declined ICU referral decedents.

TABLE 3.

Patient Characteristics by ICU Triage Cohort and Mortal Status

Patient Characteristics at ICU Referral	ICU Admissions		Declined ICU Referralsa
	High Acuity		ICU Capacity-Declined/High Acuity		ICU Prognosis-Declined/High Acuity		ICU Prognosis-Declined/Low Acuity
	“Appropriate Patient; Admitted”		“Appropriate Patient; No Bed Capacity; Not Admitted”		“Too Sick; Not Admitted”		“Too Well; Not Admitted”
	Survivors	Decedents	Survivorsd	Decedents	Survivorsd	Decedents	Survivors	Decedentsd
Patients, n	339 (72.3)	130 (27.7)	57 (74.0)	20 (26.0)	52 (46.8)	59 (53.2)	129 (92.8)	10 (7.2)
Age, yr, mean (sd)	39.1 (15.2)	48.0 (17.2)	40.5 (15.3)	50.6 (20.0)	49.2 (18.0)	48.6 (17.3)	39.1 (15.4)	48.3 (17.3)
Male gender, n (%)	191 (56.3)	77 (59.2)	28 (49.1)	10 (50.0)	25 (48.1)	28 (47.5)	65 (50.4)	6 (60.0)
Black race, n (%)	285 (91.1)	115 (92.0)	46 (85.2)	15 (79.0)	42 (93.3)	50 (90.9)	113 (90.4)	10 (100.0)
Society of Critical Care Medicine ICU Triage Priorityb, n (%)
I (needs ICU-specific therapy)	170 (59.7)	52 (43.0)	23 (40.4)	10 (50.0)	n/a	n/a	n/a	n/a
II (needs intensive monitoring)	91 (31.9)	38 (31.4)	17 (29.8)	3 (15.0)	n/a	n/a	n/a	n/a
III (less likely to benefit from ICU)	24 (8.4)	31 (25.6)	17 (29.8)	7 (35.0)	n/a	n/a	n/a	n/a
IV-A (too well to benefit from ICU)	0 (0.0)	0 (0.0)	n/a	n/a	n/a	n/a	129 (100.0)	10 (100.0)
IV-B (too sick to benefit from ICU)	0 (0.0)	0 (0.0)	n/a	n/a	52 (100.0)	59 (100.0)	n/a	n/a
Primary ICU referral diagnosisc, n (%)
Trauma	127 (44.1)	32 (25.4)	11 (19.6)	2 (10.5)	9 (17.3)	12 (20.7)	24 (19.2)	2 (20.0)
Infection	30 (10.4)	35 (27.8)	8 (14.3)	4 (21.1)	13 (25.0)	15 (25.9)	13 (10.4)	1 (10.0)
Other (nontrauma/noninfection)	131 (45.5)	59 (46.8)	37 (66.1)	13 (68.4)	30 (57.7)	31 (53.5)	88 (70.4)	7 (70.0)
Referral source, n (%)
Emergency department	136 (40.1)	45 (34.6)	27 (47.4)	6 (30.0)	20 (38.5)	34 (57.6)	36 (27.9)	1 (10.0)
Operating theater	121 (35.7)	40 (30.8)	8 (14.0)	1 (5.0)	5 (9.6)	2 (3.4)	30 (23.3)	5 (50.0)
Ward	82 (24.2)	45 (34.6)	22 (38.6)	13 (65.0)	27 (51.9)	23 (39.0)	63 (48.8)	4 (40.0)
Referring service, n (%)
Surgical	274 (82.0)	98 (75.4)	44 (77.2)	12 (60.0)	30 (58.8)	30 (50.9)	105 (81.4)	8 (80.0)
Medical	60 (18.0)	32 (24.6)	13 (22.8)	8 (40.0)	21 (41.2)	29 (49.2)	24 (18.6)	2 (20.0)
Pre-referral hospital length of stay, median calendar days (IQR)	1 (1–2)	1 (1–2)	1 (1–1)	2 (0.5–10)	1 (1–4)	1 (1–3)	1 (1–2)	1 (1–2)
ICU length of stay, median calendar days (IQR)	4 (2–6)	5 (2–8)	n/a	n/a	n/a	n/a	n/a	n/a
Length of stay from ICU referral to hospital death, median calendar days (IQR)	n/a	5 (2–10)	n/a	3 (0–8.5)	n/a	1 (0–3)	n/a	6.5 (5–13)

IQR = interquartile range, n/a = not applicable by definition.

^aWith nonmissing Society of Critical Care Medicine Triage Priority scores.

^bAssigned by the consultant (attending physician) at the time of ICU referral.

^cThe primary indication for ICU admission as determined by the admitting ICU team as part of real-time routine care.

^dPatients with potential clinical deviations from triage-expected trajectories.

Variables are reported as complete cases unless otherwise noted.

TABLE 4.

Comorbidities and Acute Diagnoses by ICU Triage Cohort and Mortal Status

Patient Characteristics at ICU Referral	ICU Admissions		Declined ICU Referralsa
	High Acuity		ICU Capacity-Declined/High Acuity		ICU Prognosis-Declined/High Acuity		ICU Prognosis-Declined/Low Acuity
	“Appropriate Patient; Admitted”		“Appropriate Patient; No Bed Capacity; Not Admitted”		“Too Sick; Not Admitted”		“Too Well; Not Admitted”
	Survivors	Decedents	Survivors	Decedents	Survivors	Decedents	Survivors	Decedents
Patients, n	339 (72.3)	130 (27.7)	57 (74.0)	20 (26.0)	52 (46.8)	59 (53.2)	129 (92.8)	10 (7.2)
Chronic comorbiditiesb, n (%)
HIV	58 (17.1)	27 (20.8)	11 (19.3)	6 (30.0)	14 (26.9)	12 (20.3)	18 (14.0)	2 (20.0)
Chronic respiratory disease	10 (3.0)	7 (5.4)	2 (3.5)	1 (5.0)	2 (3.9)	3 (5.1)	2 (1.6)	1 (10.0)
Chronic cardiovascular disease	16 (4.7)	11 (8.5)	6 (10.5)	2 (10.0)	10 (19.2)	17 (28.8)	11 (8.5)	0 (0.0)
Diabetes	35 (10.3)	20 (15.4)	7 (12.3)	3 (15.0)	11 (21.2)	11 (18.6)	11 (8.5)	1 (10.0)
Chronic kidney disease	2 (0.6)	6 (4.6)	4 (7.0)	2 (10.0)	6 (11.5)	3 (5.1)	6 (4.7)	0 (0.0)
Chronic neurologic disease	8 (2.4)	1 (0.77)	1 (1.8)	1 (5.0)	1 (1.9)	3 (5.1)	3 (2.3)	0 (0.0)
Acute diagnoses and end-organ dysfunction at ICU admissionc, n (%)
Concern for infectiond	229 (67.6)	107 (32.3)	15 (26.3)	10 (50.0)	18 (34.6)	21 (35.6)	20 (15.5)	1 (10.0)
Sepsise	90 (26.6)	63 (48.5)	6 (10.5)	4 (20.0)	12 (23.1)	16 (27.1)	12 (9.3)	0 (0.0)
Trauma	130 (38.6)	32 (24.6)	11 (19.3)	2 (10.0)	9 (17.3)	12 (20.3)	24 (18.6)	2 (20.0)
Acute respiratory process	124 (36.6)	55 (42.3)	16 (28.1)	14 (70.0)	20 (38.5)	30 (50.9)	25 (19.4)	5 (50.0)
Acute cardiovascular process	60 (17.7)	27 (20.8)	14 (24.6)	11 (55.0)	15 (28.9)	17 (28.8)	20 (15.5)	1 (10.0)
Acute kidney injury	11 (3.2)	20 (15.4)	6 (10.5)	2 (10.0)	7 (13.5)	6 (10.2)	11 (8.5)	0 (0.0)
Acute metabolic derangements	125 (36.9)	72 (55.4)	23 (40.4)	9 (45.0)	29 (55.8)	23 (39.0)	46 (35.7)	6 (60.0)
Acute neurologic dysfunction (e.g., encephalopathy)	72 (21.2)	16 (12.3)	4 (7.0)	3 (15.0)	12 (23.1)	20 (33.9)	12 (9.3)	2 (20.0)
Acute bleeding	23 (6.8)	11 (8.5)	7 (12.3)	1 (5.0)	1 (1.9)	0 (0.0)	19 (14.7)	0 (0.0)

^aWith nonmissing Society of Critical Care Medicine Triage Priority scores.

^bAs determined by the admitting ICU team as part of real-time routine care.

^cAcute active diagnoses and processes present at the time of ICU admission (but not necessarily the primary indication for ICU admission) as determined by the admitting ICU team as part of real-time routine care.

^dConcern for infection includes infection as the primary indication for ICU admission or admission sepsis flag (both as determined by the admitting ICU team as part of real-time routine care) or receipt of antimicrobials at the time of ICU admission.

^eSepsis includes concern for infection, as above, and an admission sepsis flag (as determined by the admitting ICU team as part of real-time routine care) or meeting criteria by the quick Sequential Organ Failure Assessment score or Systemic Inflammatory Response Syndrome score when available.

Among high-acuity capacity-declined ICU referrals, survivors compared with decedents were: younger (mean 40.5 vs. 50.6 yr); more likely referred from the emergency department (47.4 vs. 30.0%) and operating theater (14.0 vs. 5.0%) and less likely referred from the ward (38.6 vs. 65%); and more likely surgical referrals (77.2 vs. 60.0) and less likely medical referrals (22.8 vs. 40.0%).

Among high-acuity prognosis-declined ICU referrals, survivors compared with decedents were: of similar age (mean 49.2 vs. 48.6 yr) and primary ICU referral diagnosis; less likely referred from the emergency department (38.5 vs. 57.6%) and more likely referred from the ward (51.9 vs. 39.0%); and more likely surgical referrals (58.8 vs. 50.9) and less likely medical referrals (41.2 vs. 49.2%).

Among low-acuity prognosis-declined ICU referrals, decedents compared with survivors were: older (mean 48.3 vs. 39.1 yr); more likely referred from the operating theater (50.0 vs. 23.3%) and less likely referred from the emergency department (10.0 vs. 27.9%) or ward (40.0 vs. 48.8%); and of similar primary ICU referral diagnosis and surgical vs. medical referral.

DISCUSSION

In this descriptive, retrospective cohort study, we examined characteristics and outcomes of patients referred but declined for ICU admission stratified by clinician-perceived likelihood of ICU benefit and acuity. The primary findings of this study include that: there is a substantial group of patients (45.8%) referred but declined for ICU admission; ICU declinations occur both due to bed capacity and, in the majority (76.5%), clinical prognosis determinations, and among the latter, for both perceptions of being “too well” and “too sick” to benefit; characteristics of ICU-admitted and ICU-declined triage cohorts differ; and, while triage proved accurate in many ways (i.e., “too sick” patients had substantially higher mortality than “too well” patients), nontrivial proportions of patients have potential clinical deviations from triage-expected trajectories (e.g., 46.8% of “too sick” declinations who survive). Among studies seeking to measure the outcomes of patients referred but declined, or delayed, for ICU admission (4, 7, 10–19), this is the first to our knowledge that reports outcomes of declined ICU referrals by reasons for ICU declination (i.e., capacity constraints, too sick to benefit, or too well to benefit).

Critical care is a high-intensity, high-cost resource. As with all high-intensity interventions, admission to an ICU carries both benefits and risks that vary based on patient-level factors. In an optimal scenario, only patients for whom benefits exceed risks would be admitted to an ICU (7–9, 21, 22), thereby matching patients with the right care in the right location, averting unnecessary ICU-associated risks, facilitating patient- and family-centered care, improving cost efficiency of critical care, and preserving ICU capacity for those most likely to benefit and for surge periods with acutely greater demand.

In LMICs and in other resource-constrained settings, acknowledging the realities of chronic shortfalls of critical care resources relative to total, longitudinal demand, there is recognition that: not all patients that “need” critical care will have access to an ICU bed; facilitating nonwasteful use of scarce ICU resources may require a restriction to higher acuity patients than might be admitted in settings of more abundant resources; and that some patients may have a prognosis promising or poor enough that ICU admission is unlikely to provide meaningful benefit.

In this real-world context, in one practice setting in South Africa, we document that over 45% of patients referred for critical care are declined for ICU admission, a majority of these ICU declinations are for prognosis (76.5%) rather than capacity (23.5%) reasons, and that nontrivial proportions of these ICU-declined patients have potential clinical deviations from triage-expected trajectories: ICU-acceptable and high-acuity/too-sick patients surviving outside of the ICU, and low-acuity/too-well patients dying outside of the ICU. Such potential clinical deviations from triage-expected trajectories are likely to be analytically missed if ICU referrals are studied as a single, homogeneous pool. For example, the Triage decision making for the Elderly in European Intensive Care Units (ELDICUS) study in largely well-resourced settings in Europe documented increasing ICU refusal rates among elderly referred patients, but not whether those declinations (or the observed mortality differences) were due to different proportions of patients assessed as too sick or differential age-group–sensitivity to capacity strain (7). In another study in a resource-limited setting in South Africa, over 90% of ICU referrals were declined for prognosis reasons (56.3% for “futility” and 35.1% for “no need”), but clinical outcomes were not reported by ICU declination reason (4). This should fuel continued investigation into how best to assess prognosis with and without ICU care—the ICU net benefit—and how best to align patients with level-of-care to optimize outcomes and resource utilization (Fig. 2). In total, understanding who benefits from ICU admission and how to allocate scarce ICU beds optimally remain incompletely understood and under active investigation.

Figure 2.

Possibilities and counterfactuals among triage cohorts and mortal status. Each triage cohort and observed mortal status can be explained by possibilities and counterfactuals spanning triage priority and ICU referral outcomes. This uncertainty and observed potential clinical deviations from triage-expected trajectories should fuel continued investigation into how best to assess prognosis with and without ICU care—the ICU net benefit—and how best to align patients with level-of-care to optimize outcomes and resource utilization.

The SCCM ICU Triage Priority score system (21) and its successors strive to provide structure to ICU triage decisions and demonstrate reliability and validity, especially when protocolized (24), but are ultimately, in real use, still substantially subjective assessments at great risk of conscious and unconscious bias from the deciding clinician. We found that declined ICU referrals, compared with ICU admissions, were less likely surgical and trauma patients coming from operating theaters and more likely medical patients coming from wards and emergency departments. High-acuity prognosis-declined ICU referrals, in particular, were notably older than all other cohorts. These findings suggest that the “young, trauma or surgical, male patient” may often be prioritized, in both SCCM ICU Triage Priority scoring and with respect to those actually admitted, over the “older, medical patient,” a similar phenomenon observed across diverse resource settings (7, 14).

In our study, surprisingly, 46.8% of those who declined ICU admission for being too sick survived. Potential explanations include that the “too sick to benefit” assessment was incorrect, that such patients were fortunate to survive despite not being admitted and similar patients should still be considered for ICU admission in future instances, or that they never needed ICU-level care at all. Of note, compared with those admitted to the ICU, high-acuity capacity-declined ICU referrals had overall similar observed hospital mortality (26.0 vs. 27.7%). Despite this observed similarity, we cannot draw the conclusion from this finding that there is no ICU net benefit for this study population, as this is not an apt comparison. First, this is, at least in part, in the context of high-acuity, capacity-declined ICU referrals being of overall lower perceived acuity compared with ICU admissions: less likely triage priority I and priority II, and more likely priority III. Second, there are undoubtably additional unrecorded differences in clinical gestalt between those admitted and those declined, even within similar recorded tiers of acuity.

This article should be interpreted in the context of important limitations. The ICU database is limited to data at the time of ICU referral/admission and selected end-ICU outcomes, and does not contain information on longitudinal objective physiology and interventions during the ICU stay or post-ICU hospital course and outcomes. Chart review of mortuary records allowed recording hospital mortality and time of death, but does not provide information on hospital course or a reliable mechanism of death among decedents, nor on hospital discharge times (and therefore hospital length of stay) for survivors. Hospital mortality data were not available for patients never referred to the ICU, preventing an important ward comparison to the ICU-declined cohorts. Because final SCCM ICU Triage Priority is ultimately determined by the on-call consultant, in addition to the complex prognostic biases discussed, there may be additional bias toward assigning lower priority scores to some patients being declined due to capacity constraints, to avoid the moral injury associated with declining appropriate candidates (32). Lack of granular acuity data on declined ICU referrals, and the potential for bias toward both higher and lower acuity, preclude reliable quantification of this potential bias. Our data are from a late-pandemic era; these analyses should be repeated further into the post-pandemic era, as there is evidence of continued evolution of case-mix and critical care practice patterns, such as reduced use of mechanical ventilation, that may influence triage and outcomes (33, 34). This is a two-hospital, single-country study, and not immediately generalizable without other supporting data to other settings in and outside of South Africa.

CONCLUSIONS

In one sampling in the South African public healthcare system, many patients are referred but declined for ICU admission due to perceptions of being both “too well” or “too sick” to benefit and due to bed capacity, with a proportion of such patients experiencing potential clinical deviations from triage-expected trajectories. How best to optimize utilization of scarce ICU resources remains under active investigation and not completely understood.