There is growing interest in not only intensive care unit (ICU) outcomes but also the resources required to deliver this care and its cost-effectiveness.1 The most available metric of resource utilisation is ICU length of stay, which is influenced by casemix, illness severity, and institutional characteristics, including delays in discharge. For instance, ICU length of stay is generally longer for more severely ill patients. Comparison of length of stay between units must therefore account for differences in baseline patient characteristics.
Recently, the Australian and New Zealand Intensive Care Society (ANZICS) Centre for Outcome and Resource Evaluation (CORE) developed a model to predict ICU length of stay2 based on baseline patient demographic information, illness severity and diagnosis. Actual length of stay can be compared with predicted values to identify ICUs with longer or shorter length of stay than expected. When this marker of resource use is combined with a measure of outcome such as in-hospital mortality, an assessment of ICU efficiency can be inferred.3
In 2018, "ICU efficiency plots” were introduced into routine ANZICS reporting.
Definition of an ICU efficiency plot
The ICU efficiency plot combines the standardised mortality ratio (SMR) — the ratio of observed to predicted deaths — plotted against the risk-adjusted length of stay ratio (LOSR). The risk-adjusted LOSR is a ratio of the geometric means of observed and predicted length of stay. The geometric mean can be considered the most typical length of stay for a patient group and is usually close to the median value.
Although there has been some controversy about which measures to use,4, 5 the ICU efficiency plot (using risk-adjusted ICU length of stay and risk- adjusted in-hospital mortality) is now routinely reported in multiple countries6, 7 and has been incorporated into ANZICS paediatric reporting since 2011.6
In this article, we provide a brief review of this performance metric for adult intensive care clinicians and tips on how these data may be interpreted. Clinicians can review their own ICUs' performance by logging into the ANZICS CORE portal.
Interpreting an ICU's position on the ICU efficiency plot
The SMR and risk-adjusted LOSR make up the two axes on the graph (Figure 1). Each ICU is displayed as a point estimate with 95% confidence intervals. Each unit falls within one of four quadrants, representing different outcome and resource use combinations. The "most efficient” ICUs are in the lower left quadrant, with both low SMR and a shorter ICU length of stay than predicted (the risk-adjusted LOSR is less than one). The "least efficient" ICUs are in the upper right quadrant, with both high SMR and a longer ICU length of stay than predicted (the risk-adjusted LOSR is greater than one).
Figure 1.
“Intensive care unit (ICU) efficiency plot”, with two examples highlighted to illustrate the most and least efficient ICUs
A risk-adjusted LOSR greater than one indicates a longer observed length of stay than predicted. Patients who deteriorate after admission would be expected to have a longer observed length of stay than predicted and would lead to a higher risk-adjusted LOSR for an ICU. However, individual patients with a very long length of stay generally do not affect the risk-adjusted LOSR because most ICUs have few of these atypical patients.8, 9
Causes of a longer observed length of stay than predicted
An analysis of 167 014 ICU admissions to 42 rural/regional, 32 metropolitan, 42 tertiary and 63 private ICUs in Australia and New Zealand between January and December 2018 showed statistically significant but clinically small differences between observed and predicted length of stay, typically less than 4 hours for most diagnoses and patient types. Exceptions included non-head injury trauma admissions, where observed length of stay was typically almost 10 hours shorter than predicted, and patients who required renal replacement therapy, in whom the observed length of stay was almost 2 days longer than predicted (Table 1).
Table 1.
Observed and predicted length of stay of all admissions to Australian and New Zealand intensive care units (ICUs) in 2018
| Number of ICU admissions* | Observed ICU length of stay (hours) |
Predicted ICU length of stay (hours) |
|||
|---|---|---|---|---|---|
| Median (IQR) | Geometric mean (95% CI) | Geometric mean (95% CI) | Risk-adjusted LOSR (95%CI) | ||
| All ICU admissions | 167 014 | 40.7 (21.5–73.6) | 42.1 (41.9–42.3) | 42.2 (42.1–42.3) | 1.00 (0.99–1.01) |
| Cardiovascular diagnoses | |||||
| Cardiac surgery (CABG and valves) | 17 496 | 47.2 (25.6–73.1) | 48.8 (48.3–49.3) | 46.4 (46.2–46.6) | 1.05 (1.04–1.06) |
| Cardiac diagnoses | 9630 | 44.2 (21.5–87.3) | 42.5 (41.6–43.4) | 42.4 (42.0–42.7) | 1.00 (0.98–1.02) |
| Cardiac arrest | 3474 | 64.9 (26.8–122.3) | 52.3 (50.0–54.7) | 56.3 (55.5–57.2) | 0.93 (0.89–0.97) |
| Aortic aneurysms | 2325 | 41.5 (22.7–85.8) | 44.8 (43.0–46.7) | 43.7 (42.8–44.6) | 1.03 (0.99–1.06) |
| Other cardiovascular surgery | 15 127 | 24.4 (19.8–47.5) | 32.0 (31.6–32.4) | 31.7 (31.5–31.9) | 1.01 (1.00–1.02) |
| Respiratory diagnoses | |||||
| Chronic obstructive pulmonary disease | 2950 | 56.9 (31.6–98.1) | 54.7 (53.0–56.5) | 50.0 (49.3–50.6) | 1.09 (1.06–1.13) |
| Other respiratory diagnoses | 7410 | 48.9 (24.9–94.5) | 50.5 (49.3–51.6) | 51.9 (51.4–52.5) | 0.97 (0.95–0.99) |
| Admissions due to infection | |||||
| Pneumonia | 6076 | 74.6 (40.6–144.5) | 74.8 (73.0–76.7) | 76.4 (75.5–77.2) | 0.98 (0.96–1.00) |
| Other infections (including sepsis) | 14 114 | 62.2 (34.5–112.0) | 60.7 (59.8–61.7) | 58.2 (57.8–58.7) | 1.04 (1.03–1.06) |
| Other medical diagnoses | |||||
| Overdose | 5774 | 34.0 (18.8–55.8) | 33.2 (32.5–34.0) | 37.0 (36.6–37.3) | 0.90 (0.88–0.92) |
| Other medical diagnoses | 8236 | 44.6 (23.8–76.5) | 44.3 (43.4–45.2) | 44.8 (44.4–45.1) | 0.99 (0.97–1.01) |
| Gastrointestinal diagnoses | |||||
| Gastrointestinal (medical) | 5212 | 52.8 (27.0–99.4) | 52.2 (50.7–53.8) | 56.1 (55.3–56.9) | 0.93 (0.91–0.96) |
| Gastrointestinal surgery | 20 650 | 28.8 (20.3–65.0) | 36.9 (36.5–37.4) | 37.2 (36.9–37.4) | 0.99 (0.98–1.00) |
| Neurological and neurosurgical diagnoses | |||||
| Cerebrovascular accidents† | 2591 | 41.0 (21.6–78.6) | 40.7 (39.0–42.5) | 40.9 (40.2–41.7) | 0.99 (0.96–1.03) |
| Seizures | 2160 | 44.0 (23.7–83.3) | 45.6 (43.8–47.6) | 43.8 (43.0–44.7) | 1.04 (1.00–1.08) |
| Subarachnoid haemorrhage | 1708 | 65.9 (25.0–187.0) | 69.1 (65.2–73.2) | 70.2 (68.7–71.7) | 0.98 (0.94–1.04) |
| Other neurological diagnoses | 8273 | 26.7 (20.5–65.2) | 38.2 (37.4–39.0) | 38.1 (37.7–38.5) | 1.00 (0.98–1.02) |
| Other surgical diagnoses | |||||
| Orthopaedic surgery | 8733 | 23.5 (19.1–41.3) | 27.8 (27.4–28.2) | 25.5 (25.4–25.7) | 1.09 (1.07–1.10) |
| Spinal surgery | 5884 | 22.5 (18.5–32.8) | 26.2 (25.9–26.6) | 26.3 (26.2–26.5) | 1.00 (0.98–1.01) |
| Trauma (excluding head injuries) | 5300 | 46.6 (23.9–95.7) | 50.7 (49.3–52.2) | 60.4 (59.7–61.1) | 0.84 (0.82–0.86) |
| Head injury (+/- multitrauma) | 2414 | 70.8 (33.6–180.2) | 73.2 (69.6–76.8) | 76.2 (74.8–77.6) | 0.96 (0.92–1.00) |
| Other surgical diagnoses | 11 477 | 24.1 (19.0–45.8) | 30.4 (30.0–30.9) | 30.8 (30.6–31.1) | 0.99 (0.97–1.00) |
| Type of ICU admission | |||||
| Elective surgical admissions | 64 931 | 25.1 (20.4–48.3) | 32.9 (32.7–33.1) | 32 (31.9–32.1) | 1.03 (1.02–1.03) |
| Emergency admissions | 101 998 | 47.5 (23.9–93.6) | 49.2 (48.9–49.6) | 50.3 (50.1–50.4) | 0.98 (0.97–0.99) |
| Source of admission to ICU | |||||
| Operating theatre/recovery | 89 506 | 26.7 (20.5–55.4) | 35.8 (35.6–36.0) | 35.1 (35.0–35.2) | 1.02 (1.01–1.02) |
| Emergency department | 42 770 | 46.7 (24.2–88.6) | 46.5 (46.1–47.0) | 48.1 (47.9–48.3) | 0.98 (0.97–0.99) |
| Ward | 24 854 | 55.1 (27.2–109.3) | 53.4 (52.7–54.1) | 54.9 (54.6–55.2) | 0.99 (0.98–1.00) |
| Other hospital | 9255 | 61.7 (30.9–125.9) | 63.0 (61.7–64.4) | 66.4 (65.7–67.1) | 0.95 (0.93–0.97) |
| Therapies provided | |||||
| Not ventilated | 114 276 | 29.0 (20.2–62.6) | 35.3 (35.1–35.5) | 35.1 (35.0–35.2) | 1.00 (1.00–1.01) |
| Ventilated | 52 737 | 57.8 (30.6–1 17.0) | 61.7 (61.1–62.2) | 62.6 (62.3–62.8) | 0.99 (0.98–0.99) |
| Inotropes | 42 676 | 67.3 (38.9–124.1) | 68.4 (67.8–69.1) | 56.9 (56.7–57.2) | 1.20 (1.19–1.21) |
| Renal replacement therapy | 5245 | 123.2 (55.2–264.3) | 118.3 (114.6–122) | 72.9 (72.0–73.9) | 1.62 (1.58–1.67) |
CABG = coronary artery bypass graft; IQR = interquartile range; LOSR = length of stay ratio. * With available information. † It includes thrombotic cerebrovascular accidents and intra-cerebral haemorrhage.
The most common factor associated with a high risk-adjusted LOSR was discharge delay (ie, a prolonged time in the ICU after being deemed ready for discharge), which is dependent on both ICU and hospital-wide practices (Figure 2).
Figure 2.
Number of hours that predicted length of stay is longer than observed length of stay plotted against discharge delay in hours (time between decision to discharge and exit from the intensive care unit) (P < 0.001) for increasing difference between predicted and observed length of stay, with increasing discharge delay in hours
Implications
The ICU efficiency plot is an innovative display of overall ICU performance. It provides the opportunity to benchmark institutional resource utilisation against mortality. It updates quarterly as ANZICS data are submitted but will require scrutiny to determine overall accuracy of predictions.10 In the future, it may also facilitate monitoring of interventions to improve overall ICU performance. This information will hopefully stimulate review of hospital-wide processes that affect ICU length of stay, patient disposition, and workflow.
Competing interests
No relevant disclosures.
References
- 1.Rothen H.U., Stricker K., Einfalt J., et al. Variability in outcome and resource use in intensive care units. Intensive Care Med. 2007;33:1329–1336. doi: 10.1007/s00134-007-0690-3. [DOI] [PubMed] [Google Scholar]
- 2.Straney L.D., Udy A.A., Burrell A., et al. Modelling risk-adjusted variation in length of stay among Australian and New Zealand ICUs. PLoS ONE. 2017;12 doi: 10.1371/journal.pone.0176570. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Rapoport J., Teres D., Zhao Y., Lemeshow S. Length of stay data as a guide to hospital economic performance for ICU patients. Med Care. 2003;41:386–397. doi: 10.1097/01.MLR.0000053021.93198.96. [DOI] [PubMed] [Google Scholar]
- 4.Teres D., Higgins T., Steingrub J., et al. Defining a high performance ICU system for the 21st century: a position paper. J Intensive Care Med. 1998;13:195–205. [Google Scholar]
- 5.Rapoport J., Teres D., Lemeshow S., Gehlbach S. A method for assessing the clinical performance and cost-effectiveness of intensive care units: a multicenter inception cohort study. Crit Care Med. 1994;22:1385–1391. doi: 10.1097/00003246-199409000-00006. [DOI] [PubMed] [Google Scholar]
- 6.Straney L.D., Clements A., Alexander J., Slater A. Measuring efficiency in Australian and New Zealand paediatric intensive care units. Intensive Care Med. 2010;36:1410–1416. doi: 10.1007/s00134-010-1916-3. [DOI] [PubMed] [Google Scholar]
- 7.Niskanen M., Reinikainen M., Pettilä V. Case-mix-adjusted length of stay and mortality in 23 Finnish ICUs. Intensive Care Med. 2009;35:1060–1067. doi: 10.1007/s00134-008-1377-0. [DOI] [PubMed] [Google Scholar]
- 8.Crozier T.M., Pilcher D.V., Bailey M.J., et al. Long-stay patients in Australian and New Zealand intensive care units: demographics and outcomes. Crit Care Resusc. 2007;9(327):33. [PubMed] [Google Scholar]
- 9.Iwashyna T.J., Hodgson C.L., Pilcher D., et al. Timing of onset and burden of persistent critical illness in Australia and New Zealand: a retrospective, population-based, observational study. Lancet Respir Med. 2016;4(566):73. doi: 10.1016/S2213-2600(16)30098-4. [DOI] [PubMed] [Google Scholar]
- 10.Kramer A.A. Are ICU length of stay predictions worthwhile? Crit Care Med. 2017;45:379–380. doi: 10.1097/CCM.0000000000002111. [DOI] [PubMed] [Google Scholar]