Comments
Abstract
Rationale: ICU discharge delay occurs when a patient is considered ready to be discharged but remains in the ICU. The effect of discharge delay on patient outcomes is uncertain.
Objectives: To investigate the association between discharge delay and patient outcomes including hospital mortality, readmission to ICU, and length of hospital stay after ICU discharge.
Methods: Data were accessed from the Australian and New Zealand Intensive Care Society Adult Patient Database between 2011 and 2019. Descriptive analyses and hierarchical logistic and Cox proportional hazards regression were used to examine association between discharge delay and adjusted outcomes. Patients were stratified and analyzed by categories of mortality risk at ICU admission.
Measurements and Main Results: The study included 1,014,540 patients from 190 ICUs: 756,131 (75%) were discharged within 6 hours of being deemed ready, with 137,042 (13%) discharged in the next 6 hours; 17,656 (2%) were delayed 48–72 hours; 31,389 (3.1%) died in hospital; and 45,899 (4.5%) patients were readmitted to ICU. Risk-adjusted mortality declined with increasing discharge delay and was lowest at 48–72 hours (adjusted odds ratio, 0.87; 95% confidence interval, 0.79–0.94). The effect was seen in patients with predicted risk of death on admission to ICU of greater than 5% (adjusted odds ratio, 0.77; 95% confidence interval, 0.70–0.84). There was a progressive reduction in adjusted odds of readmission with increasing discharge delay.
Conclusions: Increasing discharge delay in ICUs is associated with reduced likelihood of mortality and ICU readmission in high-risk patients. Consideration should be given to delay the discharge of patients with high risk of death on ICU admission.
Discharge delays from ICUs are common around the world.
Increasing discharge delay in ICU is associated with reduced likelihood of mortality and readmission to ICU in patients who have a higher (>5%) risk of death on initial ICU admission.
ICUs frequently experience discharge delays, in which patients no longer require ICU-level support or intensive monitoring but remain in the ICU waiting for a ward bed. Small single-center studies from Australia, the United States, and Europe have found that 22–31% of ICU patients are discharge delayed (1–5). Discharge delay has been variably defined as between 4 and 8 hours in published literature (2, 6, 7).
The reasons for discharge delay are varied and include factors such as no bed is available on the appropriate ward, inadequate nursing staff skill mix on the receiving ward, and lack of a single room when infectious precautions exist (1, 5). When a patient is discharge delayed in the ICU, a ward-based team may assume that the ICU is leading the patient’s care. The ICU team may in turn assume that the ward team is taking control. This could lead to miscommunications and ultimately harmful gaps in the management of the patient. Alternatively, higher levels of monitoring within the ICU compared with a general ward may allow earlier detection and management of clinical deterioration.
Although the costly financial implications of discharge delay from the ICU have been examined (1, 2, 8), the effect on patient outcomes is less clear, with some studies reporting an association with increased length of hospital stay and higher incidence of adverse events such as delirium (7) and others suggesting little or no adverse effect (9). Discharge delay greater than 12 hours is considered an ICU key performance indicator in Australia and New Zealand.
We hypothesized that an increase in discharge delay would be associated with higher patient mortality and readmission rate to intensive care. Our aim was to study the prevalence and impact of discharge delay in Australian and New Zealand ICUs. Some of the results of these studies have been previously reported in the form of an abstract (10).
This was a retrospective study using data from the Adult Patient Database, one of four clinical quality registries run by The Australian and New Zealand Intensive Care Society (ANZICS) Centre for Outcome and Resource Evaluation. Ethics approval was granted by The Alfred Hospital (Human Research Ethics Committee number 713/18).
Deidentified individual patient data were accessed for all patients discharged alive to a ward from their first ICU admission at contributing hospitals between January 2011 and December 2019 inclusive.
“Time and date a patient is deemed ready for discharge from ICU” was introduced as a nonmandatory field into the Adult Patient Database in 2007 and made part of the minimum data set reported by all ICUs in mid-2009. Discharge delay was calculated as the time in hours between when a patient was deemed ready to leave the ICU and when the patient was discharged. This was categorized as “less than 6 hours,” “6–11.99 hours,” “12–23.99 hours,” “24–47.99 hours,” “48–71.99 hours,” and “≥72 hours.”
The following exclusion criteria were applied: patients less than 16 years old, readmission episodes to ICU, patients who died during their first ICU admission, patients originally admitted for palliative care or organ donation, those discharged to a nonward location (e.g., another hospital), those with an unknown mortality outcome, and those who could not have discharge delay calculated owing to missing data. Sensitivity analyses examined subgroups of patients stratified into three prespecified groups (approximate tertiles) based on predicted risk of death at ICU admission (low risk, <1%; intermediate risk, 1–4.9%; high risk, ≥5%). To account for changes in case mix and incidence of discharge delay over the studied period, data on high-risk patients were also stratified and compared between the times of 2011–2014 and 2015–2019. Finally, further sensitivity analyses were conducted, which included patients who died in the ICU after being listed as ready for ICU discharge.
The primary outcome was in-hospital mortality after discharge from ICU. Secondary outcomes included readmission to ICU and length of stay in hospital after discharge from ICU.
Results are reported as counts (with proportions), means (with SD), or medians (with interquartile ranges). Comparisons used χ2 tests, two-sample t tests, and Wilcoxon rank-sum or Kruskal-Wallis tests as appropriate depending on type and distribution of data. Hierarchical logistic regression was used to examine the association between discharge delay and mortality or ICU readmission, with patients clustered by site and site treated as a random effect. Models considered discharge delay both as a categorical variable and separately using restricted cubic splines to account for nonlinear relationships between discharge delay and outcomes. Length of stay in hospital after discharge from ICU was examined using Cox proportional hazards regression with censoring for death. All models accounted for illness severity at admission to ICU, changes over time, regional variation, hospital type, year, month and day of the week of discharge, time in ICU before discharge ready decision, time of discharge decision, and time of actual discharge from ICU. Illness severity was assessed using the Australian and New Zealand Risk of Death, a locally derived mortality prediction model incorporating diagnosis, age, treatment limitations on admission to ICU, and subscores from the Acute Physiological and Chronic Health Evaluation III scoring system. The Australian and New Zealand Risk of Death is highly discriminatory with an area under the receiver operating characteristic of >0.9 within the overall Australian and New Zealand ICU population (10, 11). Results are presented as odds ratios (ORs) or hazard ratios, both with 95% confidence intervals (CIs). All statistical analyses were performed using Stata version 16 (StataCorp).
Data for 1,350,083 ICU admissions reported to the ANZICS Adult Patient Database between 2011 and 2019 were extracted. After exclusions, there were 1,145,089 eligible patients, of whom 130,549 patients had either unknown mortality outcomes (4,254 patients) or missing or incorrect “ready to go” times (126,295 patients). The final study cohort consisted of 1,014,540 patients (Figure 1) who were discharged alive from 190 ICUs (43 rural or regional, 40 metropolitan, 40 tertiary, 67 private) to a ward. Comparisons of patients with missing information with the final study cohort are provided in Table E1 in the online supplement.
Figure 1. Flow chart displaying patients screened for study from ANZICS Adult Patient Database and excluded patients. ANZICS = Australian and New Zealand Intensive Care Society.
[More] [Minimize]Overall, 756,131 (75%) of patients left the ICU within 6 hours of being deemed ready for discharge. Nine percent of all patients were delayed more than 24 hours (Table 1). The majority of patients were deemed ready (time of day when the decision was made) for discharge between 9 and 11 a.m. (Figure 2). Discharges from ICU have increased between 2011 and 2019, and longer discharge delays have become more frequent (Figure E1).
| Characteristics | Discharge Delay Category | ||||||
|---|---|---|---|---|---|---|---|
| <6 h | 6–12 h | 12–24 h | 24–48 h | 48–72 h | ≥72 h | All Patients | |
| Number of patients, n (%) | 756,131 (75) | 137,042 (13) | 33,578 (3) | 57,792 (6) | 17,656 (2) | 12,341 (1) | 1,014,540 |
| Discharge delay, median (IQR), h | 1.5 (0.0–3.5) | 7.4 (6.5–8.9) | 16.5 (13.5–20.8) | 29.0 (26.8–31.8) | 53.3 (51.0–56.3) | 88.3 (77.1–110.7) | 2.9 (0.0–6.0) |
| Age, mean (SD), yr | 61.9 (17.9) | 61.6 (17.9) | 60.7 (18.2) | 61.3 (17.8) | 62.2 (17.4) | 63.4 (16.9) | 61.8 (17.9) |
| Sex, M, n (%) | 427,019 (56) | 76,941 (56) | 19,144 (57) | 33,373 (58) | 10,271 (58) | 7,183 (58) | 573,931 (57) |
| Invasive ventilation, n (%) | 50,409 (34) | 13,922 (36) | 4,328 (36) | 6,365 (38) | 1,822 (37) | 1,312 (40) | 78,158 (35) |
| Treatment limitation at ICU admission, n (%) | 27,144 (4) | 7,495 (6) | 2,210 (7) | 3,625 (6) | 1,227 (7) | 1,021 (8) | 42,722 (4) |
| APACHE II score, mean (SD) | 14.1 (6.6) | 15.6 (7.0) | 15.9 (7.1) | 16.5 (7.1) | 17.1 (7.1) | 18.0 (7.2) | 14.6 (6.8) |
| APACHE III score, mean (SD) | 47.8 (21.3) | 52.4 (22.6) | 53.2 (23.0) | 54.8 (22.9) | 56.6 (23.1) | 59.8 (23.4) | 49.3 (21.8) |
| ANZROD %, median (IQR) | 1.2 (0.4–4.2) | 2.1 (0.7–7.2) | 2.5 (0.8–8.3) | 2.8 (0.9–9.1) | 3.9 (1.1–10.5) | 4.9 (1.6–13.8) | 1.4 (0.5–5.1) |
| ≥1 chronic comorbidity, n (%) | 170,608 (22.6) | 35,571 (26.0) | 8,944 (26.6) | 16,447 (28.5) | 5,366 (30.4) | 3,974 (32.2) | 240,910 (23.7) |
| Hours in ICU before decision, median (IQR) | 31.8 (18.5–64.2) | 37.0 (17.2–70.7) | 37.4 (18.3–74.8) | 41.6 (19.8–85.5) | 44.4 (21.0–92.0) | 54.8 (24.3–111.4) | 35.0 (18.5–66.7) |
| Total length of ICU stay, h, median (IQR) | 33.6 (20.4–66.2) | 45.6 (24.0–79.2) | 52.8 (36.0–93.6) | 72.0 (50.4–115.2) | 98.4 (76.8–146.4) | 158.4 (117.6–228.0) | 40.8 (21.6–74.4) |
| Cause of admission to ICU | |||||||
| Medical diagnoses, n (%) | 290,336 (38.4) | 67,559 (49.3) | 19,473 (58.0) | 33,043 (57.2) | 10,520 (59.6) | 7,688 (62.3) | 428,619 (42.2) |
| Infection and sepsis, n (%) | 67,597 (8.9) | 17,910 (13.1) | 5,090 (15.2) | 9,017 (15.6) | 3,088 (17.5) | 2,387 (19.3) | 105,089 (10.4) |
| Trauma, n (%) | 28,346 (3.7) | 7,764 (5.7) | 2,457 (7.3) | 4,050 (7.0) | 1,282 (7.3) | 919 (7.4) | 44,818 (4.4) |
| All postoperative diagnoses, n (%) | 465,795 (61.6) | 69,483 (50.7) | 14,105 (42.0) | 24,749 (42.8) | 7,136 (40.4) | 4,653 (37.7) | 585,921 (57.8) |
| Elective surgery, n (%) | 386,806 (51.4) | 50,302 (37.0) | 8,953 (26.8) | 15,506 (27.2) | 4,276 (24.6) | 2,525 (20.8) | 468,368 (46.4) |
| Cardiac surgery, n (%)* | 102,425 (13.5) | 12,736 (9.3) | 2,677 (8.0) | 5,252 (9.1) | 1,384 (7.8) | 654 (5.3) | 125,128 (12.3) |
Patients in the longer discharge delay groups had higher illness severity scores on admission to ICU, had more comorbidities, were more commonly admitted because of a medical condition or infection, had spent longer in the ICU, were less commonly elective postoperative or cardiac surgical patients, and more commonly had treatment limitations in place on admission to ICU (Table 1).
The overall mortality was 3.1%, with the lowest unadjusted mortality seen in those who were discharged within 6 hours of a decision that the patient was ready to leave (Table 2). After adjusting for confounders (including the time the decision was made and the hour, day, and month the patient left the ICU, also independently associated with mortality), discharge delay between 24 and 72 hours was associated with a lower odds of mortality. This was lowest between 48 and 72 hours (OR, 0.87; 95% CI, 0.79–0.94) (Tables 2 and E2 and Figure 3). Adjusted in-hospital mortality was highest when the decision for ICU discharge occurred in the evening (Table E2) (OR, 1.26; 95% CI, 1.19–1.34).
| Primary Outcome: Hospital Mortality | |||
|---|---|---|---|
| Discharge Delay Category | Deaths [n (%)] | Adjusted Odds Ratio (95% CI) | P Value |
| All patients | 31,389 (3.1) | — | — |
| <6 h | 21,135 (2.8) | 1.00 (reference value) | — |
| 6–12 h | 5,381 (3.9) | 0.98 (0.94–1.02) | 0.31 |
| 12–24 h | 1,380 (4.1) | 0.96 (0.90–1.02) | 0.21 |
| 24–48 h | 2,210 (3.8) | 0.94 (0.90–0.99) | 0.025 |
| 48–72 h | 692 (3.9) | 0.87 (0.79–0.94) | 0.001 |
| ≥72 h | 591 (4.8) | 0.92 (0.84–1.01) | 0.09 |
| Secondary Outcome: ICU Readmission | |||
| Discharge Delay Category | Readmissions [n (%)] | Adjusted Odds Ratio (95% CI) | P Value |
| All patients | 45,899 (4.5) | — | — |
| <6 h | 33,274 (4.4) | 1.00 (reference value) | — |
| 6–12 h | 7,318 (5.3) | 0.98 (0.95–1.02) | 0.34 |
| 12–24 h | 1,424 (4.2) | 0.88 (0.83–0.93) | <0.001 |
| 24–48 h | 2,599 (4.5) | 0.86 (0.82–0.90) | <0.001 |
| 48–72 h | 727 (4.1) | 0.75 (0.69–0.81) | <0.001 |
| ≥72 h | 557 (4.5) | 0.78 (0.71–0.85) | <0.001 |
Figure 3. Adjusted odds ratios (95% confidence interval) for mortality versus discharge delay (modeled using restricted cubic splines).
[More] [Minimize]There was a statistically significant interaction between categories of baseline risk and discharge delay (P < 0.001). There was no relationship between mortality and discharge delay for patients who were in the lowest (<1%) and intermediate (1–5%) risk subgroups. Patients in the highest (>5%) risk of death subgroup on admission to ICU had a reduction in adjusted mortality as discharge delay increased. Adjusted OR for mortality was lowest at 48–72 hours of discharge delay (OR, 0.77; 95% CI, 0.70–0.84) (Table 3 and Figure 4). This general relationship was preserved for the high-risk patients regardless of whether they were admitted before or after 2015 (Table E3).
| Raw Outcomes by Admission Risk Subgroup | Low Risk (<1%) | Intermediate Risk (1–5%) | High Risk (>5%) | |||
|---|---|---|---|---|---|---|
| Number of patients | 417,645 | 336,586 | 257,192 | |||
| Discharge delay, h | 2.3 (0.0–5.0) | 3.0 (0.0–6.2) | 4.0 (0.0–8.0) | |||
| Mortality, n (%) | 835 (0.2) | 4,375 (1.3) | 25,976 (10.1) | |||
| Readmission to ICU, n (%) | 10,441 (2.5) | 16,493 (4.9) | 18,775 (7.3) | |||
| Days in hospital after “ready to go” decision* | 4.2 (2.1–6.5) | 5.6 (3.1–10.3) | 8.1 (4.2–16.1) | |||
| Days in hospital after ICU* | 4.0 (2.0–6.2) | 5.3 (2.8–10.0) | 7.8 (3.9–15.5) | |||
| Multivariable Logistic Regression Analysis for Mortality | ||||||
| Discharge Delay Category | Odds Ratio (95% CI) | P Value | Odds Ratio (95% CI) | P Value | Odds Ratio (95% CI) | P Value |
| <6 h | 1 (reference) | — | 1 (reference) | — | 1 (reference) | — |
| 6–12 h | 0.99 (0.76–1.30) | 0.97 | 1.02 (0.92–1.14) | 0.68 | 0.94 (0.90–0.99) | 0.012 |
| 12–24 h | 0.92 (0.58–1.44) | 0.70 | 0.94 (0.79–1.12) | 0.49 | 0.89 (0.83–0.95) | 0.001 |
| 24–48 h | 1.22 (0.89–1.68) | 0.22 | 0.97 (0.85–1.10) | 0.62 | 0.85 (0.80–0.89) | <0.001 |
| 48–72 h | 0.92 (0.47–1.81) | 0.82 | 0.93 (0.75–1.16) | 0.53 | 0.77 (0.70–0.84) | <0.001 |
| ≥72 h | 1.10 (0.49–2.50) | 0.82 | 1.11 (0.87–1.42) | 0.40 | 0.78 (0.71–0.86) | <0.001 |
Figure 4. Adjusted odds ratios (95% confidence interval) for mortality versus discharge delay (modeled using restricted cubic splines) for different baseline risk groups.
[More] [Minimize]Subgroup analyses showed there was a reduction in mortality risk for ventilated, nonventilated, medical, infected, respiratory, and trauma subsets of patients between 24 and 72 hours of discharge delay. This mortality benefit between 24 and 72 hours of delay was also evident when looking at the subset of 137 sites with less than 5% missing data for discharge delay and when considering patients only in public hospitals. Surgical patients (excluding cardiac and trauma) and cardiac surgical patients derived no mortality benefit if discharge delayed. Delay 72 hours or more was associated with harm for cardiac surgical patients. There was no outcome benefit associated with discharge delay for patients in private hospitals. The case mix in private hospitals comprised predominantly low-risk elective surgical patients (Tables E4 and E5).
Overall readmission to ICU was 4.5%, with the lowest unadjusted readmission rate when patients waited 48–72 hours to leave the ICU after a discharge decision was made (Table 2). After adjustment for confounders, odds of readmission to the ICU fell with progressively longer discharge delay, reaching a nadir at 48–72 hours (adjusted odds of readmission, 0.75; 95% CI, 0.69–0.81) (Tables 2 and E6 and Figures E2 and E3).
Median length of stay in hospital after discharge from ICU was 5 days (interquartile range, 2.6–9.4). Before adjustment, it appeared that all degrees of discharge delay were associated with a longer time in hospital after discharge from ICU. However, after adjusting for confounders, patients delayed between 24 and 72 hours had no difference in length of stay in hospital compared with those with a delay of less than 6 hours (Table 4). Refer to Table E7 and Figures E4 and E5 for full multivariable analysis and heat maps.
| Discharge Delay Category | Days in Hospital after “Ready to Go” Decision [Median (IQR)] | Days in Hospital after Discharge from ICU | ||||
|---|---|---|---|---|---|---|
| Median (IQR) | Unadjusted Hazard Ratio (95% CI) | P Value | Adjusted Hazard Ratio (95% CI)* | P Value | ||
| All patients | 5.2 (3.0–9.7) | 5.0 (2.6–9.4) | — | — | — | — |
| <6 h | 5.1 (2.6–9.0) | 5.0 (2.5–8.9) | 1 (reference value) | — | 1 (reference value) | — |
| 6–12 h | 6.0 (3.2–11.2) | 5.7 (2.9–10.9) | 0.87 (0.87–0.88) | <0.001 | 0.97 (0.97–0.98) | <0.001 |
| 12–24 h | 5.8 (3.1–11.4) | 5.0 (2.4–10.7) | 0.90 (0.89–0.92) | <0.001 | 0.99 (0.98–1) | 0.05 |
| 24–48 h | 6.4 (3.6–12.4) | 5.2 (2.4–11.2) | 0.86 (0.85–0.86) | <0.001 | 1.00 (0.99–1.01) | 1.00 |
| 48–72 h | 7.6 (4.5–14.3) | 5.3 (2.2–12.0) | 0.81 (0.8–0.82) | <0.001 | 0.99 (0.97–1.01) | 0.21 |
| ≥72 h | 10.6 (6.4–19.5) | 6.5 (2.4–14.9) | 0.69 (0.67–0.7) | <0.001 | 0.87 (0.86–0.89) | <0.001 |
A sensitivity analysis was conducted including all 3,409 patients who died after being listed as ready for ICU discharge. Compared with those with discharge delay less than 6 hours, there was an increased risk of death at 12–24 hours, and no association with mortality after this. Risk of readmission was reduced for delays of more than 12 hours. In the subgroup of high-baseline-risk patients, there was a reduced mortality with all discharge delays greater than 24 hours (Table E8).To address the possibility that the discharge decision time had been used to indicate time of brain death declaration in potential organ donors, a further sensitivity analysis was conducted including the 2,209 patients who died in the ICU and had admission diagnoses unrelated to trauma, neurological conditions, or cardiac arrest. The protective effect of discharge delay was again seen in high-risk patients with more than 24 hours of discharge delay, whereas risk of readmission was reduced in all patients for delays of more than 12 hours (Table E9).
This study of more than 1 million patients discharged from ICUs in Australia and New Zealand over an 8-year period demonstrated that 121,367 (12%) had their discharge from ICU delayed by more than 12 hours. After adjusting confounders, discharge delay appeared to be protective and was associated with a reduction in the odds of hospital mortality and readmission to ICU. The effect was mainly accounted for by patients who had a predicted risk of death greater than 5% on admission to ICU.
Discharge delay has been reported to be associated with increased hospital length of stay (7). The costly financial implications of delayed discharge have been examined by small single-center studies in the United States and Europe (1, 2). A review by Hicks and colleagues in 2019 demonstrated that the mean cost per patient bed-day in Australian ICUs is $4,375 AUD (8). However, in 2013, Garland and colleagues found in a prospective single-center study that 30-day mortality decreased as discharge delay increased (12). This relationship held until 20 hours of discharge delay, at which point odds of 30-day mortality steadily increased but was still 0.93 at 72 hours, similar to our nonsignificant point estimate of 0.92 (95% CI, 0.84–1.01). A single-center study from 2018 by Bose and colleagues did not find any correlation between discharge delay and mortality (9). Our finding that patients with higher risk of death on ICU admission benefit most is consistent with the trend reported by Gilligan in a brief report of approximately 500,000 intensive care patients in the United Kingdom. This audit found that “sicker” critical care patients in the United Kingdom had a reduction in their expected mortality if ICU discharge delay was longer than 24 hours (6).
The reason why discharge delay appears to have a protective effect is unclear. A period of delay may allow patients to declare whether they are truly stable off ICU-level supports that have recently been ceased. Doctors and nurses in the ICU can offer better care than on the ward, where staffing and training levels are lower. Discharge-delayed ICU patients may have central vascular access that allows for prompt intravenous therapy and for daily monitoring blood tests, as opposed to the ward environment where lack of intravenous access can delay the detection and treatment of problems. It may also be that some patients were discharged too early and had worse outcomes—that is, the decision for discharge was in retrospect incorrect.
Although there is a risk of overinterpreting findings in the multiple subgroups examined, those who experienced the greatest mortality benefit with discharge delay, such as medical, infected, respiratory, trauma, and ventilated patients, also tended to be those with higher illness severity. The patient groups where discharge delay had little or no effect were those with low-risk elective surgery. In only one group (cardiac surgery patients) were discharge delays associated with harm.
Our study also found that readmission rates to ICU declined as discharge delay increased. Even though there have been previous studies that have examined predictors of readmission to ICU such as severity of disease and presence of comorbidities (13–16), none of these studies have examined discharge delay as one of the possible factors for readmission to the ICU. Making efforts to avoid readmission to ICU is important as this is associated with increased mortality and hospital length of stay (17–19). The effect of discharge delays should be examined as one of the confounding factors in future studies.
Our primary analysis excluded all patients (n = 3,409) who died while discharge delayed in ICU. This may have introduced bias. It was not possible to tell how many were patients initially considered “ready to go” who then unexpectedly died and how many were dead patients awaiting transfer from the ICU (e.g., brain dead waiting for organ donation surgery), where owing to local coding practices, the “ready to go” time was used to represent the time of death declaration. During the study period, there were 2,933 brain-dead donors (data from The Australian and New Zealand Organ Donation Registry) across Australia and New Zealand (20). It is likely that they represent a proportion of the 3,409 ICU deaths as this study includes all major ICUs in Australia and New Zealand. Our sensitivity analyses included initially all 3,409 patients who died in the ICU after being listed as ready for ICU discharge and were then repeated including only the 2,209 patient deaths admitted with diagnoses unrelated to trauma, neurological conditions, or cardiac arrest (i.e., those most likely to be potential organ donors). These had consistent findings with our primary analysis and showed a protective effect of discharge delay among patients with high baseline mortality risk at ICU admission. Furthermore, it is possible that some patients who died within the ICU were patients deemed “palliative” (21) and were awaiting transfer to the ward to continue end-of-life care. This is potentially supported by the fact that 25% of 3,386 deaths in the ICU (where treatment goals were listed) had a treatment limitation in place on admission to the ICU compared with 4% of 979,023 ICU survivors.
In addition to the potential “immortal time bias” described above, this study has further limitations. It is retrospective and observational. The basis of ICU clinician decisions for discharge were unknown, as was the accuracy of recording of the decision time. We do not know how often patients deteriorated after initially listed as “ready to go” and had their time of decision rescinded and then reentered later. We do not know the actual time a bed on the ward was requested or if this was different from the time the patient was considered ready. Therapies provided during ICU stay and at the time of discharge to the ward, were unknown. Severity of illness, requirements for ongoing treatment, and presence of treatment limitations at the time of discharge from ICU were also unknown. Data about use of ICU outreach liaison services to detect early signs of deterioration and need for intervention on the ward were unavailable. It is also unknown whether the exclusion of the 126,295 patients with missing data (11.4% of the 1,145,089 eligible patients) influenced our results. Lastly, we have not assessed the effect of ICU or ward “strain.” The availability of ward beds, or pressure on ICUs under strain to provide space for new admissions (22) when they had limited capacity (23) or the availability of a ward bed, may have influenced how and when clinicians declared a patient ready for discharge. However, hour of discharge from the ICU, which may be an indirect marker of strain, was included in our models.
Strengths of our study include the large number of patients, including all diagnostic categories (medical, surgical, cardiac surgery, trauma), and the large number of ICUs from all regions of Australia and New Zealand and all hospital types. Findings are thus likely generalizable. Accurate clinical data allowed robust risk adjustment for available confounders that might have influenced associations with outcome.
Future research should investigate reasons for the benefits of discharge delay to better understand its protective effect and how it should influence clinical practice and should also investigate options such as step-down units (24).
Adjusted mortality and readmission rates to ICU declined with increasing discharge delay, in patients with high severity of illness on admission to ICU. For such patients, consideration should be given to keeping them in the ICU after when initially thought ready for discharge, even if they no longer require ICU-level supports.
The authors thank the ANZICS Clinical Outcomes and Resource Evaluation Centre for providing the data used in the current study. The authors and the ANZICS Clinical Outcomes and Resource Evaluation management committee also thank clinicians, data collectors, and researchers at the contributing sites listed in Table E10.
| 1. | Johnson DW, Schmidt UH, Bittner EA, Christensen B, Levi R, Pino RM. Delay of transfer from the intensive care unit: a prospective observational study of incidence, causes, and financial impact. Crit Care 2013;17:R128. |
| 2. | Edenharter G, Gartner D, Heim M, Martin J, Pfeiffer U, Vogt F, et al. Delay of transfer from the intensive care unit: a prospective observational analysis on economic effects of delayed in-house transfer. Eur J Med Res 2019;24:30. |
| 3. | Anstey MH, Thompson K, Seppelt I. Exit block in the intensive care unit. Med J Aust 2017;207:224. |
| 4. | Williams TA, Leslie GD, Brearley L, Leen T, O’Brien K. Discharge delay, room for improvement? Aust Crit Care 2010;23:141–149. |
| 5. | Williams T, Leslie G. Delayed discharges from an adult intensive care unit. Aust Health Rev 2004;28:87–96. |
| 6. | Gilligan S. Critical care delayed discharge: good or bad? J Intensive Care Soc 2017;18:146–148. |
| 7. | Tiruvoipati R, Botha J, Fletcher J, Gangopadhyay H, Majumdar M, Vij S, et al.; Australia and New Zealand Intensive Care Society (ANZICS) Clinical Trials Group. Intensive care discharge delay is associated with increased hospital length of stay: a multicentre prospective observational study. PLoS One 2017;12:e0181827. |
| 8. | Hicks P, Huckson S, Fenney E, Leggett I, Pilcher D, Litton E. The financial cost of intensive care in Australia: a multicentre registry study. Med J Aust 2019;211:324–325. |
| 9. | Bose S, Johnson AEW, Moskowitz A, Celi LA, Raffa JD. Impact of intensive care unit discharge delays on patient outcomes: a retrospective cohort study. J Intensive Care Med 2019;34:924–929. |
| 10. | Forster GM, Bihari S, Tiruvoipati R, Pilcher D. The association between discharge delay from intensive care and patient outcomes [abstract]. Am J Respir Crit Care Med 2020;201:A6215. |
| 11. | Paul E, Bailey M, Kasza J, Pilcher DV. Assessing contemporary intensive care unit outcome: development and validation of the Australian and New Zealand Risk of Death admission model. Anaesth Intensive Care 2017;45:326–343. |
| 12. | Paul E, Bailey M, Pilcher D. Risk prediction of hospital mortality for adult patients admitted to Australian and New Zealand intensive care units: development and validation of the Australian and New Zealand Risk of Death model. J Crit Care 2013;28:935–941. |
| 13. | Garland A, Connors AF Jr. Optimal timing of transfer out of the intensive care unit. Am J Crit Care 2013;22:390–397. |
| 14. | Ho KM, Dobb GJ, Lee KY, Finn J, Knuiman M, Webb SA. The effect of comorbidities on risk of intensive care readmission during the same hospitalization: a linked data cohort study. J Crit Care 2009;24:101–107. |
| 15. | Frost SA, Alexandrou E, Bogdanovski T, Salamonson Y, Davidson PM, Parr MJ, et al. Severity of illness and risk of readmission to intensive care: a meta-analysis. Resuscitation 2009;80:505–510. |
| 16. | Woldhek AL, Rijkenberg S, Bosman RJ, van der Voort PH. Readmission of ICU patients: a quality indicator? J Crit Care 2017;38:328–334. |
| 17. | Lee H, Lim CW, Hong HP, Ju JW, Jeon YT, Hwang JW, et al. Efficacy of the APACHE II score at ICU discharge in predicting post-ICU mortality and ICU readmission in critically ill surgical patients. Anaesth Intensive Care 2015;43:175–186. |
| 18. | Grochla M, Saucha W, Borkowski J, Knapik P. [Readmission to the intensive care unit - epidemiology, prediction and clinical consequences] [in Polish]. Wiad Lek 2019;72:1387–1396. |
| 19. | Ponzoni CR, Corrêa TD, Filho RR, Serpa Neto A, Assunção MSC, Pardini A, et al. Readmission to the intensive care unit: incidence, risk factors, Resource use, and outcomes: a retrospective cohort study. Ann Am Thorac Soc 2017;14:1312–1319. |
| 20. | ANZOD. Australia and New Zealand Organ Donation Registry annual reports. [accessed 2020 May 5]. Available from: https://www.anzdata.org.au/anzod/publications-2/annual-reports/. |
| 21. | Curtis JR, Treece PD, Nielsen EL, Downey L, Shannon SE, Braungardt T, et al. Integrating palliative and critical care: evaluation of a quality-improvement intervention. Am J Respir Crit Care Med 2008;178:269–275. |
| 22. | Anesi GL, Liu VX, Gabler NB, Delgado MK, Kohn R, Weissman GE, et al. Associations of intensive care unit capacity strain with disposition and outcomes of patients with sepsis presenting to the emergency department. Ann Am Thorac Soc 2018;15:1328–1335. |
| 23. | Wallace DJ, Angus DC, Seymour CW, Barnato AE, Kahn JM. Critical care bed growth in the United States: a comparison of regional and national trends. Am J Respir Crit Care Med 2015;191:410–416. |
| 24. | Prin M, Wunsch H. The role of stepdown beds in hospital care. Am J Respir Crit Care Med 2014;190:1210–1216. |
Author Contributions: Concept: S.B., R.T., and D.P. Design: D.P. Analysis: M.B. and D.P. Interpretation: G.M.F., S.B., and R.T. Manuscript preparation: G.M.F., S.B., and D.P. Manuscript revision and approval: G.M.F., S.B., R.T., M.B., and D.P.
This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.
Originally Published in Press as DOI: 10.1164/rccm.201912-2418OC on July 10, 2020
Author disclosures are available with the text of this article at www.atsjournals.org.
