ORGINAL ARTICLE
Unravelling post-ICU mortality: predictors and causes of death Annemarije BraberM and Arthur R.H. van Zanten Background and objectives To study the characteristics of patients dying in the ICU, dying after ICU treatment during the same hospitalization period in general wards and post-ICU hospital survivors. In addition, causes of death and post-ICU mortality (PICUM) predictors were addressed. Methods The present study is a retrospective single centre cohort study in a mixed medical–surgical 12-bed ICU. Patients were divided into three groups: ICU deaths, post-ICU deaths and hospital survivors. Causes of death were determined by an independent review panel of three intensive care physicians. Daly’s mortality prediction model was applied in retrospect to evaluate risk of PICUM. Other predictors were also tested for predictive value. Results In total, 405 patients were included: 146 ICU deaths, 92 post-ICU deaths and 167 survivors (random computerized sample from 680 survivors). ICU mortality was 16.3% and PICUM 10.3%. Sepsis was the most common cause of death in both ICU deaths (48.3%) and post-ICU deaths (30.1%). Multivariate analysis identified age, comorbidities, length of stay in ICU, Acute Physiology and Chronic Health Evaluation II score
Introduction Critically ill patients are admitted to the ICU to reduce morbidity and prevent mortality related to acute illness, trauma or surgical procedures. However, owing to the severity of illness, death in the ICU may be inevitable in some patients. By definition, ICU mortality is the percentage of admitted patients who eventually die in the ICU. In addition to ICU mortality, post-ICU mortality (PICUM) during the same hospital stay can add to the mortality rate of patients admitted to the ICU. PICUM can range from 6.4 to 40% in different ICUs, depending on the severity of the illness.1,2 Little is known about the exact causes of death in general wards after ICU discharge. Previously, Ridley and Purdie3 in 1992 and Wallis et al.4 in 1997 concluded that respiratory failure was the main cause of post-ICU death. However, since these studies, new ICU treatment strategies have been developed and implemented. Therefore, we designed a retrospective study to analyse causes of death after ICU discharge. In addition, we were interested in knowing whether these deaths were predictable. At present, Annemarije Braber is a fellow in Intensive Care, University Medical
Centre Nijmegen St Radboud, Department of Intensive Care, The Netherlands. From the Department of Intensive Care, Gelderse Vallei Hospital, Ede, The Netherlands Correspondence to Arthur R.H. van Zanten, MD, Internist-intensivist, Department of Intensive Care, Gelderse Vallei Hospital, PO Box 9025, 6710 HN Ede, The Netherlands Tel: +31 318 434115; fax: +31 318 434116; e-mail:
[email protected] 0265-0215 ß 2010 Copyright European Society of Anaesthesiology
and a do-not-resuscitate code as independent predictors of PICUM. Based on Daly’s mortality prediction model, 63% of patients were discharged with a high risk of PICUM. Of these, 51% actually died. Specificity was low. Conclusion Causes of deaths were equally distributed among study groups, except for sepsis. Sepsis was more frequently encountered among ICU deaths. Five PICUM predictors were found: age, Acute Physiology and Chronic Health Evaluation II score, length of ICU stay, do-not-resuscitate code and comorbidities. A do-not-resuscitate code during the first 24 h after admission was the most important predictor of PICUM. Prospective research is warranted to evaluate the applicability of PICUM prediction models in individual ICU patients. Eur J Anaesthesiol 2010;27:486–490 Keywords: intensive care unit, mortality, outcome, post-intensive care unit mortality, prediction Received 17 April 2009 Revised 13 September 2009 Accepted 16 September 2009
Daly et al.5 developed a mortality prediction model for ICU patients at ICU discharge. They suggested that, in theory, a 39% mortality reduction after discharge could be achieved if patients with a high predicted mortality stayed another 48 h in the ICU. We retrospectively applied this model in patients discharged from the ICU and analysed other predictors for PICUM.
Methods A retrospective cohort study from 1 January 2004 until 31 December 2005 in a mixed medical–surgical 12-bed ICU was performed. No neurosurgical or cardiothoracic surgical patients were treated. All patients older than 15 years admitted to the ICU were included. Patients were excluded if death occurred within 4 h after ICU admission, if ICU stay was shorter than 4 h, if a patient was transferred to an ICU in another hospital or if the medical record was lost. Following the guidelines of the Medical Ethics Committee of our institution, informed consent was waived. ICU patients were divided into three groups: hospital survivors, post-ICU in-hospital deaths and ICU deaths. Of the survivors, a random sample of 190 patients was selected by computer from 680 patients based on statistical calculations (Fig. 1). Information on patients’ characteristics such as age, sex, Acute Physiology and Chronic Health Evaluation II (APACHE-II) score and type of admission was extracted from the medical records DOI:10.1097/EJA.0b013e3283333aac
Copyright © European Society of Anaesthesiology. Unauthorized reproduction of this article is prohibited.
Unravelling post-ICU mortality: predictors and causes of death 487
Fig. 1
953 ICU admissions
169 ICU deaths
23 excluded
680 hospital survivors
104 general ward deaths
Random sample
12 excluded
146 ICU deaths
190 hospital survivors
92 ward deaths
23 excluded
167 hospital survivors
Study population (ICU admissions between 1 January 2004 and 31 December 2005). Readmissions not included.
of each patient. The APACHE-II score was used for severity of illness and outcome prediction. It is customary that these data are used to benchmark ICUs in the National Intensive Care Evaluation in the Netherlands. Data on comorbidities were collected using criteria proposed by van Gestel et al.6 (Addendum 1, http://links.lww. com/EJA/A10). For patients who stayed more than 2 consecutive days in the ICU, mortality prediction according to Daly’s mortality prediction model (DMPM) was calculated on the day of discharge to the general ward.5 Patients leaving the ICU to die in the general ward or codes that limited ICU readmission, except for a solitary do-not-resuscitate (DNR) code, were excluded from DMPM analysis. Daly’s logistic regression model consists of age, length of ICU stay, previous cardiothoracic surgery, chronic health points and acute physiology points according to the APACHE-II scoring system on the day of discharge to identify patients at risk of inappropriate early discharge: a cut-off point of 0.6 (60%) mortality risk demonstrated the best sensitivity (65.5%) and specificity (87.6%) in the original model; less than 0.6 representing a low probability and more than 0.6 a high probability of dying in the ward after ICU discharge. Of all the patients who died, relevant data from the period before death were extracted from 238 medical records and a short medical case report was made, providing information on five items: age, comorbidities, reason for ICU admission, ICU length of stay (LOS) and course of illness before death. Vital signs, relevant laboratory results, culture results and radiology test results were processed to obtain insight into the course of the illness before death. A reviewer panel of three independent ICU physicians individually scored the causes of death from these case reports using a predefined list of potential
causes based on 12 categories. Categories were pneumonia, pulmonary aspiration, irreversible cerebral damage, sepsis, malignancy, cardiovascular failure, end-stage renal failure, thromboembolic disease, end-stage liver failure, haemorrhagic shock, other and unknown. When two or all three reviewers scored the same cause of death reviewing a single case, that cause of death was stated. If no agreement was reached between the three reviewers, the case reports were individually evaluated once more. Statistics
From the total 680 survivors, a random computerized sample was taken of 190 patients. The size of this sample was sufficient to show significant differences between the three groups. Results are presented as mean SD. A x2 test and Wilcoxon test were used when appropriate. A P value less than 0.015 was accepted as statistically significant in comparing the three groups (overall comparison). A P value less than 0.05 was accepted as statistically significant in pairwise comparison of two groups. Clinically relevant and significant (P < 0.05) variables from univariate analysis were entered in a multivariate logistic regression model. Statistical analysis was performed using SAS software version 9.1.3.
Results In the 2-year study period, 895 patients were analysed after admission to the ICU (readmissions not included). Of these 895 ICU admissions, 146 patients died in the ICU (16.3%), 92 died post-ICU in general wards (10.3%) and 657 (73.4%) survived (Fig. 1). Total ICU mortality and PICUM was 238 of 953 (26.6%). Fifty-eight patients were excluded from this study for the following reasons: ICU stay was shorter than 4 h (n ¼ 19), European Journal of Anaesthesiology
2010, Vol 27 No 5
Copyright © European Society of Anaesthesiology. Unauthorized reproduction of this article is prohibited.
488 Braber and van Zanten
Table 1
Characteristics of ICU deaths, deaths after ICU discharge and hospital survivors
Patients; N (%) Age; mean SD (range; median) Male sex, N (%) APACHE-II score; mean SD (range; median) ICU length of stay; mean SD (range; median) Total comorbidity; mean SD (range; median) Comorbidity; N (%) DM CVA Malignancy Heart failure Respiratory failure Liver failure Renal failure DNR first 24 h; N (%) Mechanical ventilation; N (%) Acute admission; N (%) Surgical admission; N (%)
ICU deaths
Post-ICU deaths
Hospital survivors sample
146 (16.3) 69.2 13.5b (17.5–90.3; 72.9) 93 (64.1) 23.6 8.5b (2–45; 22.5) 9.3 12.5 (1–75; 4) 0.88 0.78 (0–3; 1)b
92 (10.3) 73.9 10.5c (48.8–99; 74.9) 54 (58.5) 21 7.1c (6–40; 20) 9.8 12.2 (1–79; 6)c 1.05 0.81 (0–3; 1)c
167a 60.1 18.6 (15.6–89.4; 65.3) 96 (57.5) 15.3 5.7 (3–29; 15) 5.9 8.1 (1–54; 2) 0.54 0.68 (0–3; 0)
26 13 36 24 13 9 7 57 130 143 34
(17.9) (8.9) (24.8) (16.6)b (9)b (6.2)d,b (4.8) (40.4)b (90.3)d,b (98.6)d,b (23.5)d,b
26 9 28 17 11 0 7 36 72 82 36
(28) (9.7) (30.1) (18.3)c (11.8)c (0) (7.5) (39.1)c (78.3)c (88.2) (38.7)
27 7 31 12 3 1 9 13 105 132 68
(16.2) (4.2) (18.6) (7.2) (1.8) (0.6) (5.4) (7.8) (62.9) (79.5) (41)
APACHE-II, Acute Physiology and Chronic Health Evaluation II; CVA, cerebrovascular accident; DM, diabetes mellitus; DNR, do not resuscitate. a Sample of 680 survivors (74.1%). b Significant ICU deaths versus survivors. c Significant post-ICU deaths versus survivors. d Significant ICU deaths versus post-ICU deaths.
death occurred within 4 h (n ¼ 18), the patient was transferred to an ICU outside our hospital (n ¼ 17) and if the medical records were lost (n ¼ 4). Comparative data of the groups with respect to patients’ demographics, comorbidities and other data are shown in Table 1. Causes of ICU deaths and post-ICU deaths are listed in Table 2. No differences in sepsis sources between ICU deaths and post-ICU deaths were found. An abdominal source was found in 51.4 versus 59.3%, respectively, and a pulmonary source in 32.9 and 25.9%, respectively. After the initial scoring of causes of death by the independent reviewers, agreement was complete in 62.7% of the cases; in 37.3%, agreement was found in two out of the three physicians. From the group of patients who eventually died in general wards after successful ICU discharge, in 23.6% of these post-ICU death was expected based on the medical condition at the time of discharge.
discharge in the ICU survivors and the post-ICU deaths using the mortality prediction model with a cut-off value of 60%. In our study population, the sensitivity of this model was 91% and the specificity was 52%. Forward stepwise multivariate analysis was performed and identified five independent predictors of PICUM (Table 3). The DNR code given in the first 24 h after ICU admission is the strongest PICUM predictor followed by the cumulative comorbidities.
The ICU LOS was less than 2 consecutive days in 39% of the hospital survivor group, compared with 17.3% in the post-ICU death group. Owing to the short ICU LOS, Daly’s model was not applicable in these patients. Figure 2 shows the probability for PICUM after ICU
In agreement with other studies, the APACHE-II score in the first 24 h was significantly higher in nonsurvivors and was an independent risk factor for PICUM.5,8 Cumulative comorbidity scores were higher in those who died after ICU discharge and were the second strongest
Table 2
Discussion PICUM was 10.9%, which was comparable to percentages found in other studies.1,2 Of the three groups analysed, post-ICU deaths were in the oldest age groups. ICU survivors were the youngest. We found that age was an independent risk factor for PICUM, as was reported before by Azoulay et al.7
Causes of death in the ICU and general wards after ICU discharge
Cause of death, N (%) Sepsis Irreversible cerebral damage Congestive heart failure/cardiac arrest Malignancy Pneumonia Renal failurec Thromboembolic diseasec Aspirationc Hepatic failurec Bleedingc Unknownc Otherc,d
ICU deaths, 145 (100) 70 21 20 6 7 2 2 4 3 7 0 3
(48.3)b (14.5) (13.8) (4.1) (4.8)b (1.4) (1.4) (2.8) (2.1) (4.8) (0) (2.1)
Post-ICU deaths, 93 (100) 28 19 13 6 12 1 3 2 0 1 4 4
(30.1) (20.4) (14) (6.5) (12.9) (1.1) (3.2) (2.1) (0) (1.1) (4.3) (4.3)
Discharged with maximum treatment limitations, 22 (100)a 5 10 2 2 1 0 0 0 0 1 1 0
(22.6) (45.5) (9.1) (9.1) (4.5) (0) (0) (0) (0) (4.5) (4.5) (0)
a Subgroup of the post-ICU deaths (23.6%). b Significant. c Groups too small for statistical analysis. d Diabetic ketoacidosis, pancreatitis, end-stage Duchenne muscular dystrophy, voluntarily starvation and tension pneumothorax.
European Journal of Anaesthesiology 2010, Vol 27 No 5
Copyright © European Society of Anaesthesiology. Unauthorized reproduction of this article is prohibited.
Unravelling post-ICU mortality: predictors and causes of death 489
Fig. 2
Number of patients
60 50 40 Low mortality risk
30
High mortality risk
20 10 0 Post ICU deaths
Survivors
Distribution of post-ICU mortality risk on the day of ICU discharge using Daly’s model representing a low mortality risk less than 60% and high mortality risk more than 60% in survivors (n ¼ 102) and post-ICU deaths (n ¼ 56).
predictor for PICUM. Norena et al.9 demonstrated that in-hospital mortality after ICU discharge was associated with a comorbidity score similar to our results. Apart from respiratory failure, cardiac failure was also found to be significantly more frequent in deceased patients than in survivors, as has also recently been shown in a large study.10 Apart from age, APACHE-II score and cumulative comorbidities, length of ICU stay and DNR code in the first 24 h were also proven to be independent risk factors for PICUM. A DNR code was the strongest predictor, however, with large confidence intervals, suggesting that the result should be interpreted with caution. This is remarkable as a DNR coding is based on the subjective opinions of physicians. It is impossible to determine whether this association is caused by the optimal selection of patients at high risk for mortality or the assumption that a DNR code could lead to suboptimal care and poor outcome. Thus, potentially, this could be a reflection of a self-fulfilling prophecy. Some studies indeed showed that withdrawal of support in patients is likely to result in a poor outcome that may bias predictive models and may indeed lead to self-fulfilling prophecies.11–13 Interestingly, the causes of death in both groups of deceased patients are comparable except for sepsis. Sepsis is the most frequently found primary cause of Table 3 Multivariate analysis (independent predictors of post-ICU mortality)
Age APACHE-II score Total comorbidity score Length of stay in ICU DNR first 24 h after ICU admission
Odds ratio
95% CI
P
1.04 1.13 1.86 1.04 3.50
1.01–1.07 1.07–1.20 1.20–2.89 1.01–1.08 1.53–8.06
0.005
