Journal of Critical Care 29 (2014) 60–65
Contents lists available at ScienceDirect
Journal of Critical Care journal homepage: www.jccjournal.org
Admission factors associated with prolonged (N14 days) intensive care unit stay☆,☆☆ Fernando Godinho Zampieri, MD a, b,⁎, José Paulo Ladeira, MD a, b, Marcelo Park, MD, PhD b, Douglas Haib, MD a, Cintia Lovatto Pastore, ANP a, Cristiane M. Santoro, ANP a, Fernando Colombari, MD a a b
Intensive Care Unit, Hospital Alemão Oswaldo Cruz, São Paulo, Brazil Intensive Care Unit, Emergency Medicine Discipline, University of São Paulo, São Paulo, Brazil
a r t i c l e
i n f o
a b s t r a c t
Keywords: Prolonged ICU stay Critical illness Performance status
Purpose: To describe the admission factors associated with prolonged (N 14 days) intensive care unit (ICU) stay (PIS). Materials and Methods: Retrospective analysis of 3257 admissions during a 1.5-year period in a tertiary hospital. We tested the association between clinically relevant variables and PIS (N14 days) through binary logistic regression using the backward method. A Kaplan-Meier curve and the log-rank test were used to compare hospital outcomes for ICU survivors between patients with and without PIS. Results: In total, 6.6% of all admissions had a prolonged stay, consuming over 40% of all ICU bed-days. Illness severity; respiratory support at admission; performance status; readmission; admission from a ward, emergency room or other hospital; admission due to intracranial mass effect; severe chronic obstructive pulmonary disease; and the temperature at admission were all associated with PIS in a multivariate analysis. The created model had a good area under the curve (0.82) and was calibrated (Hosmer-Lemeshow test p = 0.431). Post hoc analysis on ICU survivors on in patients with at least two days of ICU stay yielded similar results. Hospital survival after ICU discharge was similar for patients with and without PIS (log-rank test p = 0.50). Conclusion: A small number of ICU admissions consume a great proportion of ICU bed-days. Illness severity, a need for support and performance status are important predictors of PIS. Patients who survive a PIS have similar hospital mortality to patients with a shorter stay. © 2014 Elsevier Inc. All rights reserved.
1. Introduction
Certain studies have focused on admission features and their impact on PIS [2,3], suggesting that illness severity, age and a need for organ support are all related to PIS [2,7]. However, none of the previous analyses has reported the impact of previous performance status (PS) on ICU LOS. Because PS has an important role in patient prognosis [8], its role in ICU LOS should be addressed. We performed a retrospective analysis of all patients admitted to a tertiary ICU in Brazil during a 1.5-year period to evaluate the admission features associated with PIS. All of the patients’ demographic data, need for organ support and clinical data (including PS) were recorded at admission. As a secondary end-point, hospital mortality after ICU discharge was also evaluated, categorizing patients as those with or without PIS.
Depending on the definition employed, 4% to 11% of patients admitted to the intensive care unit (ICU) will have a prolonged ICU stay (PIS) [1–3]. It has been suggested that up to 45% of all ICU days may be consumed by this apparently small percentage of patients [3]. Identifying patients at risk of a prolonged stay may help ICU management and avoid a shortage of ICU beds. Despite the increased use of resources, patients who eventually survive a PIS have acceptable long-term outcomes, including quality of life [2,4], even if specific subgroups, such as patients on prolonged mechanical ventilation [5] or the elderly [6], are considered. Therefore, a prolonged ICU length of stay (LOS) by itself should not be considered as a marker of poor outcome or a futile expenditure of resources [1].
2. Methods 2.1. Population
☆ Conflicts of Interest: All of the authors report that they have no conflicts of interest. ☆☆ Institution where the work was performed: Hospital Alemão Oswaldo Cruz. ⁎ Corresponding author. Intensive Care Unit, Hospital Alemão Oswaldo Cruz, Rua Dr. João Julião 331, São Paulo, Brazil. Zip code: 01323–903. Tel.: +55 1135490839. E-mail address: [email protected] (F.G. Zampieri). 0883-9441/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jcrc.2013.09.030
We retrospectively evaluated all critically ill patients admitted to a tertiary hospital in São Paulo, Brazil, from January 2011 to June 2012. The ICU is a mixed 34-bed unit with an in-house intensivist available 24/7. All data were collected in a database system during the ICU stay
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65
as part of the ICU routine (Epimed Monitor, Epimed®, Rio de Janeiro, Brazil). The collected data included demographic data, the reason for admission, previous PS, comorbidities, the LOS before ICU admission, a need for organ support and severity indexes. The local ethics committee approved the study protocol and waived the need for informed consent due to the study’s strictly retrospective and observational features. 2.2. Outcome definition PIS was defined as an ICU LOS of more than 14 consecutive days [2]. Readmissions were included in the analysis because ICU readmission was one of the factors hypothesized to be related to PIS. For mortality analysis after ICU discharge, however, only the last ICU admission was considered because including a previous admission with a known outcome would increase bias. ICU readmissions less than 48 hours before discharge were considered as discharge failure and grouped as a single ICU admission. 2.3. Variable selection A univariate analysis of factors that were preemptively associated with PIS due to clinical relevance was performed. Variables were tested for normality using the Kolmogorov-Smirnov test. Parametric data were compared using a t-test or analysis of variance. The MannWhitney or Kruskal-Wallis test was used for nonparametric data. Categorical variables were compared using the chi-square test. The variables included in the univariate analysis were age, body mass index, the LOS before ICU admission, a modified PS (see below), whether the admission was a readmission, the presence of major comorbidities (heart failure, dementia, chronic obstructive pulmonary disease (COPD) on ambulatory oxygen therapy, liver failure with Child-Pugh score B or C [9], chronic kidney disease on dialysis, a solid metastatic tumor or hematological malignancy), the site of origin (operating room, emergency room, ward or other hospital/nursing home; patients admitted from another hospital or a nursing home were included in the same group), the specific reasons for admission (sepsis or intracranial mass effect), the admission severity index (SAPS3 score [10]), the admission time (night or day), a need for organ support in the hour following ICU admission (vasoactive drugs, non-invasive ventilation, mechanical ventilation or dialysis) and other clinical data at admission (temperature, creatinine levels, platelet count, mean arterial pressure and heart rate).
61
[15]. A post hoc analysis of factors related to PIS in ICU survivors was also performed to evaluate whether the exclusion of patients who perished in the ICU would change our results. 2.6. Other analysis Hospital survival after ICU discharge was compared between ICU survivors in both groups through a Kaplan-Meier curve and the logrank test [16]. We also evaluated ICU LOS in quartiles of SAPS3 scores, categorizing patients as those with any degree of impairment (PS of 1 or 2) or without impairment (PS of 0). All analysis was performed using SPSS version 19.0 (IBM Corporation, Armonk, NY). P b .05 was considered significant. 3. Results In total, 3257 ICU admissions (2908 patients) were included in the analysis, of which 203 (6%) admissions were longer than 14 days. The number of patients who had an ICU LOS greater than 14 days was divided into 5-day intervals and is shown in Fig. 1. PIS consumed 42% of the 14 811 ICU days evaluated. The characteristics of the entire population and of the patients according to the presence or absence of PIS are shown in Table 1. Patients with PIS were older (70.3 ± 16.1 vs 65.8 ± 16.8 years; P b 0.001), had a worse PS, had a higher SAPS3 score at admission (58.8 ± 15.3 vs 42.5 ± 14.7; P b .001), had a longer hospital LOS before ICU admission (3 [0-12.5] vs 1 [0-3]; P b .001), were more frequently admitted from the ward, were more frequently admitted due to sepsis (30% vs 11%; P b .001) or intracranial mass effect (2.5% vs 0.2%; P b .001) and more often had comorbidities. Regarding the need for organ support at admission, patients with PIS required vasoactive drugs (23% vs 9%; P b .001), mechanical ventilation (24% vs 10%; P b .001) and noninvasive ventilation (28% vs 11%; P b .001) more frequently than patients without PIS. Patients with PIS had a higher heart rate and a higher temperature at admission (94.5 ± 22.5 vs 87.2 ± 21.7 bpm and 36.5 ± 0.92 vs 36.1 ± 0.86, respectively; both p b 0.01). The variables included in the model after univariate analysis were age; PS; SAPS3 score; the LOS before ICU admission; readmission, the site of origin and admission due to sepsis or intracranial mass effect; a previous diagnosis of heart failure, dementia, COPD on oxygen therapy, cirrhosis, a metastatic solid tumor or a hematological tumor; the use of organ support (noninvasive ventilation, mechanical ventilation or vasoactive support); and clinical admission features (the lowest mean arterial blood pressure, temperature and heart
2.4. Performance status PS was evaluated at ICU admission through a short patient or family/caregiver interview performed by the nurse who was responsible for data collection (CLP). PS is routinely evaluated at our institution for all admissions, and data are recorded in the same database. PS was categorized based on the need for help to perform self-care and categorized as 0, 1 or 2. A PS of 0 was defined as no help needed, corresponding to Eastern Corporative Oncologic Group performance status (ECOG) classes 0–2. A PS of 1 was defined as capable of limited self-care (ECOG 3), whereas a PS of 2 meant that the patient was unable to perform any self-care (ECOG 4) [11]. 2.5. Multivariate logistic regression All variables with p b 0.25 in the univariate analysis were included in a binary logistic regression using the backward method. Collinearity between variables was evaluated through calculation of the variance inflation factor [12]. An arbitrary threshold of 10 was used to define collinearity. The created model was tested through the creation of an receiver operating characteristic (ROC) curve [13,14]. The Hosmer-Lemeshow test was used to test the model’s calibration
Fig. 1. Number of admissions in each 5-day interval, from 15-60 days (total: 203 admissions).
62
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65
Table 1 Patient characteristics and outcomes
Table 2 Variables associated with LOS N14 days on logistic regression
All Number of admissions Number of patients General data Age
3257 2908
Platelets, ×103/mm3 Creatinine, mg/dL Temperature, °C Outcome LOS after ICU discharge, days ICU mortality, n (%) Hospital mortality
LOS N 14 days
P
3054 2742
203 166
-
70.35 (16.18) 55.2%
b.01
2334 (76%) 551 (18%)
129 (64%) 56 (27%)
b.001
169 (6%) 26.56 (5.31) 1143 (37%) 42.58 (14.79) 1 [0-2] 298 (9.8%)
18 (9%) 26.25 (5.81) 71 (35%) 58.80 (14.88) 3 [0-12.5] 55 (27.1%)
614 (20%) 822 (27%) 1481 (49%) 137 (4%)
100 (50%) 51 (25%) 32 (15%) 20 (10%)
b.001
330 (11%) 6 (0.2%)
60 (30%) 5 (2.5%)
b.01 b.001
218 (7%) 148 (5%) 93 (3%) 15 (0.5%) 89 (3%) 276 (9%) 94 (3%)
21 (10%) 23 (11%) 17 (9%) 4 (2%) 7 (3%) 23 (11%) 11 (5%)
.065 b.01 b.001 .027 .391 .165 .06
278 (9%) 331 (11%) 318 (10%) 29 (1%)
46 (23%) 57 (28%) 68 (24%) 2 (1%)
b.001 b.001 b.001 .585
66.10 (16.8) 65.82 (16.84) 53.8% 53.7%
Sex, % male PS Fully functional (0) 2463 (75%) Dependency for one daily 607 (19%) activity (1) Fully dependent (2) 187 (6%) Body mass index, kg/m4 26.54 (5.33) Nighttime admission 1214 (37%) SAPS3 43.59 (15.30) LOS before ICU admission 1 [0-3] Readmission 353 (10.8%) Place of origin Ward 714 (22%) Emergency department 873 (27%) Operating Room 1513 (47%) Other 157 (4%) Specific reasons for admission Sepsis 390 (12%) Intracranial mass effect 11 (0.3%) Comorbidities Heart failure 239 (7%) Dementia 171 (5%) COPD 110 (3%) Hepatic Failure 19 (1%) Dialysis 96 (3%) Metastatic tumor 299 (10%) Hematologic disease 105 (3%) Admission data Organ support Vasoactive drugs 324 (10%) Noninvasive ventilation 388 (12%) Mechanical ventilation 386 (12%) Dialysis 31 (1%) Clinical features MAP, mmHg 82.2 (17.89) HR, bpm
LOS ≤ 14 days
82.33 (17.78) 87.69 87.24 (21.89) (21.78) 195.66 195.31 (95.52) (94.92) 0.98 [0.760.98 [0.771.35] 1.34] 36.20 (0.87) 36.18 (0.86) †
4 [2-9] 255 (8.8%)⁎ 307 (10.6%)⁎
†
4 [2-8] 181 (6.6%)⁎ 226 (8.2%)⁎
.369
80.55 (19.52) 94.51 (22.54) 200.85 (104.26) 0.98 [0.711.54] 36.54 (0.92) †
16 [8-25] 74 (44%)⁎ 81 (48.8%)
.50 .484 b.01 b.001 b.001
.16 b.01 .47
Variable
OR
CI
P
SAPS 3 PS (vs PS 0) PS 1 PS 2 Site of origin (vs operating room) Ward Emergency room Other hospital/home care Readmission Intracranial mass effect COPD on oxygen therapy Need for support at admission Mechanical Ventilation Non-invasive Ventilation Maximum temperature at admission
1.03
1.02-1.04
b.001
1.70 1.91
1.18-2.44 1.09-3.35
.004 .024
2.17 1.77 4.27 1.79 6.07 1.87
1.23-3.82 1.04-3.00 2.25-8.10 1.19-2.69 1.56-23.51 1.02-3.42
.007 .038 b.001 .005 .009 .042
3.33 2.10 1.24
2.19-5.06 1.40-3.14 1.04-1.46
b.001 b.001 .015
We also evaluated the impact of PS on ICU LOS in each SAPS3 score quartile. Patients with a PS of 1 or 2 and with a SAPS3 score in the third quartile had a longer ICU LOS compared with patients with a PS of 0 (P = .001; Fig. 3). There was also a tendency toward a longer ICU LOS in the fourth quartile, although this trend was not statistically significant (P = .07). Patients with PIS had higher ICU and hospital mortality (36% vs 6% and 48% vs 10%, respectively; both P b .001). The LOS after ICU discharge was also longer for patients with PIS (22 [11.75-41] vs 5 [3-11]; P b .01). However, hospital mortality after ICU discharge was similar for patients in both groups (Fig. 4; log-rank test P = .50). 4. Discussion There are several important conclusions that can be drawn from the data presented in this manuscript. First, the study confirms previous reports that, albeit uncommon (6% in this analysis), prolonged admissions account for the occupancy of more than 40% of all ICU bed-days. Second, our analysis highlights that previous PS is associated with PIS. No comorbidity except COPD on domiciliary oxygen therapy was associated with PIS. Third, the study confirms that illness severity and a need for respiratory assistance are associated with PIS. Fourth, the increase in hospital mortality of patients with PIS is most likely due to their higher ICU mortality, as
.754 b.01 b.001 b.001 b.001
† LOS after last ICU discharge. ⁎ Percentage of mortality relative to number of patients.
rate). The variables independently associated with PIS in multivariate analysis are shown in Table 2. Excluding 50 patients in whom exclusive palliative care was adopted during the first 14 days of the ICU stay resulted in the removal of COPD on domiciliary oxygen therapy as a factor associated with PIS; the remaining variables continued in the model. The ROC curve of the model’s capability to predict PIS is shown in Fig. 2. The AUC of the model was 0.82 (95% confidence interval [CI] 0.80-0.85). The resulting model was accurate (Hosmer-Lemeshow test P = .431). The post-hoc analysis performed only on ICU survivors included 3002 admissions. Following this reanalysis, readmission and a previous diagnosis of COPD were no longer significantly associated with PIS for live ICU discharges. The remaining variables stayed in the model (Table 3).
Fig. 2. ROC curve for the model prediction of an ICU LOS greater than 14 days. AUC 0.82 (95% CI 0.80-0.85).
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65
63
Table 3 Variables associated with LOS N14 days on logistic regression performed only on ICU survivors Variable
OR
CI
P
SAPS 3 PS (vs PS 0) PS 1 PS 2 Site of origin (vs operating room) Ward Emergency room Other hospital/home care Intracranial mass effect Need for support at admission Mechanical ventilation Non-invasive ventilation Maximum temperature at admission
1.03
1.01-1.05
b.001
1.65 2.04
1.04-2.61 1.02-3.97
.031 .044
3.06 2.19 4.01 8.62
1.50-6.23 1.12-4.31 1.84-8.75 1.87-40.90
.007 .022 b.001 .007
6.00 2.03 1.55
3.60-10.02 1.23-3.34 1.24-1.94
b.001 .005 b.001
mortality after ICU discharge is similar for patients with and without PIS (Fig. 4). Arabi et al performed an analysis in a smaller sampler (947 admissions) and also found that patients with PIS consumed a greater proportion of ICU days (over 45%) [3]. This phenomenon emphasizes the importance of taking PIS risk factors at admission into account when planning resource allocation in the ICU and may influence the decision to increase the number of ICU beds or to implement an intermediary care unit (IMCU) [17]. However, the impact of an IMCU on hospital and ICU LOS is still unclear. It has been suggested that discharge to an IMCU may be a risk factor for ICU readmission [18]. Because readmission is independently associated with PIS, the effects of the implementation of an IMCU on ICU LOS may not be as straightforward as they seems and is most likely related to a balance between earlier ICU discharge and the possible increase in ICU readmission that may be modulated by the presence of an IMCU. Because there is no IMCU in our hospital, which is an important limitation to the generalization of the present data, there is no expected discharge bias involved in our analysis. Risk factors for PIS in previous analyses included illness severity, unplanned admission, readmission, admission due to respiratory reason or trauma, infection at admission, a need for organ support, the LOS before admission and age [3,7,19]. Similarly, our study suggests that illness severity and respiratory support (both invasive and noninvasive) are associated with PIS. However, the use of vasoactive drug support was not associated with PIS, most likely because patients who required vasopressors were sicker and had a
Fig. 3. ICU LOS for quartiles of SAPS3 scores in patients with good PS (0) or a need for assistance (PS of 1 or 2). The P values are non-significant, except for the third quartile.
Fig. 4. Kaplan-Meier curve for survival after ICU discharge for patients with and without PIS. Log-rank test P = 0.5. The data were censored at 30 days.
greater chance of dying early after ICU admission. Patients who were using vasopressors and died in the ICU died earlier (and therefore had a shorter stay) than patients who died in the ICU but were not using vasopressors at admission (3 [1-12.5] vs 10 [3.75-20] days; P b .001). The place of origin of the patient was also associated with PIS. Admissions from the ward or after an emergency surgical procedure were associated with PIS, corroborating the previous findings that unexpected admissions are related to an increased ICU LOS [3]. The same was true for admissions from other hospitals or from nursing homes. Flabouris et al suggested that patients admitted after interhospital transfer have higher mortality and a longer ICU LOS than similar patients admitted from the emergency department [20]. Admission from nursing homes has also been reported to be associated with an increased hospital LOS [21]. Sepsis was not associated with PIS in our analysis, contrary to the report from Arabi, in which the presence of infection was a risk factor for PIS [3]. The SOAP study also reported an increased ICU LOS for septic patients compared with non-septic patients [22]. This disparity may be explained due to differences in data collection, definitions and the time window. We included only whether the main reason for admission was sepsis, and not the presence of infection at admission [22,23]. Patients with sepsis have higher mortality compared with non-septic patients, so it is possible that when a 14-day cutoff is used, many patients with sepsis will already have perished, explaining the apparent nonexistence of an association between infection and PIS. The role of admission temperature in ICU LOS is intriguing. Kiekkas et al found that although fever was not associated with increased mortality, there was an association between increases in body temperature and mortality in the critically ill [24]. In a large series, Lee et al suggested that fever was markedly associated with mortality in non-septic patients, but not in patients with sepsis [25]. In a recent randomized trial, temperature control was associated with reduced mortality and the need for vasopressors in critically ill septic patients, which may be related to a reduction of oxygen consumption and, consequently, a better coupling of oxygen supply and demand [26]. It is uncertain whether temperature is a marker or cause of worse prognosis in critically ill patients, but our analysis suggests that this parameter may impact not only mortality but also LOS in a continuous fashion. To the best of our knowledge, no other study has evaluated the impact of previous PS on ICU LOS. PS is a key feature in the
64
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65
prognostication of critically ill patients with cancer [27,28] and is routinely used to define care goals for ambulatory cancer patients [29,30]. Our analysis shows that previous PS is associated with PIS, with an OR of 1.9 for patients with a poor previous status. The impact of PS on LOS appears to be more significant in the medium range of illness severity (Fig. 3). This finding suggests that for patients with reduced illness severity, the impact of PS on LOS is less important. Most of the patients in the first and second SAPS3 score quartiles were admitted from the operating room after elective surgery (90.2% and 60.2%, respectively) and had a very brief ICU LOS (Fig. 3); therefore, these groups most likely reflect routine ICU admissions. Although there was a trend toward a longer LOS in the fourth quartile of SAPS3 scores, this trend did not reach statistical significance (P = .07). This finding may be because the impact of PS on very ill patients is less important, as the severity of illness is most likely the greater determinant of prognosis. Except for COPD on domiciliary oxygen therapy, which is known to be associated with worse outcomes [31], no other comorbidity was associated with PIS. Therefore, we speculated that PS is more intimately associated with PIS than simply the presence or absence of comorbidity. Similar to a report by Laupland [2], we could not find any association between age and LOS before ICU admission and PIS. Although this finding may be a consequence of sampling issues, we cannot discard the possibility that LOS before ICU admission and age were surrogates for PS in previous analyses. As an important limitation, the evaluation of PS is subject to personal bias, especially when the family is asked about the previous PS of the relative. A more structured approach to PS should be adopted in further prospective studies on the subject. Two post hoc sensitivity analyses were performed. The first analysis was performed evaluating only ICU survivors. Excluding patients who perished in the ICU helped to reduce any eventual interpretation bias between PIS and mortality because ICU mortality competes with ICU LOS. In this analysis (Table 3), COPD on oxygen therapy and readmission were no longer associated with PIS. PS continued to be an important factor associated with PIS, suggesting that the association is robust and independent of ICU mortality. The second post-hoc analysis was performed excluding patients with a short LOS (≤ 2 days; data not shown). In this analysis, a PS of 2, admission from the emergency room and a need for noninvasive ventilation were no longer associated with PIS (although there was a trend in the association between MV and PIS; p = 0.06). The presence of severe liver disease was associated with PIS in this analysis (although the CI was wide: 4.7-20.7). These results are limited by the smaller number of patients with a PS of 2 in this subgroup (82 vs 187 patients in the main analysis). Nevertheless, a PS of 1 was still associated with PIS. The results of our analysis should be interpreted given the particular features of our hospital (including the absence of an IMCU). Our data do not support any change in current ICU admission practice but may aid attending physicians in considering that a recently admitted patient who presents several of the described risk factors is at increased risk of a prolonged stay. This consideration may help in answering families’ inquires and in planning ICU bed management. Additionally, our analysis confirms that in-hospital mortality for ICU survivors was similar for patients with and without PIS, suggesting that ICU LOS should not be taken into account during prognostication or decisions to withdraw life support [2]. Certain strengths of our analysis should be highlighted. First, our analysis adds new findings regarding risk factors for PIS, including the role of previous PS and the effect of admission temperature on ICU LOS. Second, albeit being a retrospective analysis, the large number of included patients makes this one of the largest analyses of risk factors for PIS. Third, a robust statistical analysis, including logistic regression, was applied to identify independent risk factors for PIS. The main results were unchanged even when post-hoc analysis was performed.
Our study has several limitations. First, this is a retrospective analysis of one center, which may limit the findings’ external validity. Nevertheless, we obtained similar results as other analyses performed in different facilities [3]. Second, we were not able to evaluate long-term prognosis (including 6-month or 1-year outcomes) after ICU stay and were limited to hospital mortality as the mortality outcome. It should be highlighted that hospital mortality is most likely inadequate as a robust outcome, especially because we are unaware of the quality of life at discharge and the destination facility (a nursing home, for example). An analysis using relevant quality-of-life outcomes, including quality-adjusted life-years after discharge, would improve current knowledge on this subject. Third, we used a limited scale to evaluate PS, and we have no information regarding PS or quality of life after hospital discharge. Fourth, the inclusion of readmissions in the analysis may be controversial because it is conceivable that the readmission of patients with other risk factors for PIS (PS, for example) could potentiate an association or create bias. However, readmission was defined a priori as a risk factor for PIS and therefore included in the analysis. We lastly were unable to perform a cost analysis. 5. Conclusion Albeit uncommon, patients with PIS consume an important proportion of all available ICU bed-days. Illness severity, a need for respiratory support, patient origin, temperature at admission, admission due to intracranial mass effect, COPD on domiciliary oxygen therapy, reduced PS and previous ICU admission are associated with PIS. Prognosis after ICU discharge appears to be independent of the presence of PIS. Further prospective studies should focus on the impact of previous PS on outcomes. Acknowledgments None. References [1] Heyland DK, Konopad E, Noseworthy TW, et al. Is it 'worthwhile' to continue treating patients with a prolonged stay (N14 days) in the ICU? An economic evaluation. Chest 1998;114:192–8. [2] Laupland KB, Kirkpatrick AW, Kortbeek JB, et al. Long-term mortality outcome associated with prolonged admission to the ICU. Chest 2006;129:954–9. [3] Arabi Y, Venkatesh S, Haddad S, et al. A prospective study of prolonged stay in the intensive care unit: predictors and impact on resource utilization. Int J Qual Health Care 2002;14:403–10. [4] Rimachi R, Vincent JL, Brimioulle S. Survival and quality of life after prolonged intensive care unit stay. Anaesth Intensive Care 2007;35:62–7. [5] Combes A, Costa MA, Trouillet JL, et al. Morbidity, mortality, and quality-of-life outcomes of patients requiring N or = 14 days of mechanical ventilation. Crit Care Med 2003;31:1373–81. [6] Montuclard L, Garrouste-Orgeas M, Timsit JF, et al. Outcome, functional autonomy, and quality of life of elderly patients with a long-term intensive care unit stay. Crit Care Med 2000;28:3389–95. [7] Higgins TL, McGee WT, Steingrub JS, et al. Early indicators of prolonged intensive care unit stay: impact of illness severity, physician staffing, and pre-intensive care unit length of stay. Crit Care Med 2003;31:45–51. [8] Christodoulou C, Rizos M, Galani E, et al. Performance status (PS): a simple predictor of short-term outcome of cancer patients with solid tumors admitted to the intensive care unit (ICU). Anticancer Res 2007;27:2945–8. [9] Pugh RN, Murray-Lyon IM, Dawson JL, et al. Transection of the oesophagus for bleeding oesophageal varices. Br J Surg 1973;60:646–9. [10] Moreno RP, Metnitz PG, Almeida E, et al. SAPS 3–From evaluation of the patient to evaluation of the intensive care unit. Part 2: development of a prognostic model for hospital mortality at ICU admission. Intensive Care Med 2005;31: 1345–55. [11] Oken MM, Creech RH, Tormey DC, et al. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol 1982;5:649–55. [12] Hsieh FY, Lavori PW, Cohen HJ, et al. An overview of variance inflation factors for sample-size calculation. Eval Health Prof 2003;26:239–57. [13] Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation 2007;115:928–35. [14] Heagerty PJ, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics 2005;61:92–105. [15] Hosmer DW, Lemeshow S. Applied logistic regression. 2nd ed. Wiley; 2000.
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65 [16] Bland JM, Altman DG. The logrank test. BMJ 2004;328:1073. [17] Peelen L, de Keizer NF, Peek N, et al. The influence of volume and intensive care unit organization on hospital mortality in patients admitted with severe sepsis: a retrospective multicentre cohort study. Crit Care 2007;11:R40. [18] Renton J, Pilcher DV, Santamaria JD, et al. Factors associated with increased risk of readmission to intensive care in Australia. Intensive Care Med 2011;37: 1800–8. [19] Estenssoro E, González F, Laffaire E, et al. Shock on admission day is the best predictor of prolonged mechanical ventilation in the ICU. Chest 2005;127: 598–603. [20] Flabouris A, Hart GK, George C. Outcomes of patients admitted to tertiary intensive care units after interhospital transfer: comparison with patients admitted from emergency departments. Crit Care Resusc 2008;10:97–105. [21] Suter-Widmer I, Christ-Crain M, Zimmerli W, et al. Predictors for length of hospital stay in patients with community-acquired pneumonia: results from a Swiss Multicenter study. BMC Pulm Med 2012;12:21. [22] Vincent JL, Sakr Y, Sprung CL, et al. Sepsis in European intensive care units: results of the SOAP study. Crit Care Med 2006;34:344–53. [23] Suarez JI, Zaidat OO, Suri MF, et al. Length of stay and mortality in neurocritically ill patients: impact of a specialized neurocritical care team. Crit Care Med 2004;32: 2311–7.
65
[24] Kiekkas P, Velissaris D, Karanikolas M, et al. Peak body temperature predicts mortality in critically ill patients without cerebral damage. Heart Lung 2010;39:208–16. [25] Lee BH, Inui D, Suh GY, et al. Association of body temperature and antipyretic treatments with mortality of critically ill patients with and without sepsis: multicentered prospective observational study. Crit Care 2012;16:R33. [26] Schortgen F, Clabault K, Katsahian S, et al. Fever control using external cooling in septic shock: a randomized controlled trial. Am J Respir Crit Care Med 2012;185: 1088–95. [27] Soares M, Salluh JI, Toscano L, et al. Outcomes and prognostic factors in patients with head and neck cancer and severe acute illnesses. Intensive Care Med 2007;33:2009–13. [28] Soares M, Caruso P, Silva E, et al. Characteristics and outcomes of patients with cancer requiring admission to intensive care units: a prospective multicenter study. Crit Care Med 2010;38:9–15. [29] Kalser MH, Barkin J, MacIntyre JM. Pancreatic cancer. Assessment of prognosis by clinical presentation. Cancer 1985;56:397–402. [30] Buccheri G, Ferrigno D, Tamburini M. Karnofsky and ECOG performance status scoring in lung cancer: a prospective, longitudinal study of 536 patients from a single institution. Eur J Cancer 1996;32A:1135–41. [31] Funk GC, Bauer P, Burghuber OC, et al. Prevalence and prognosis of COPD in critically ill patients between 1998 and 2008. Eur Respir J 2013;41:792–9.
Contents lists available at ScienceDirect
Journal of Critical Care journal homepage: www.jccjournal.org
Admission factors associated with prolonged (N14 days) intensive care unit stay☆,☆☆ Fernando Godinho Zampieri, MD a, b,⁎, José Paulo Ladeira, MD a, b, Marcelo Park, MD, PhD b, Douglas Haib, MD a, Cintia Lovatto Pastore, ANP a, Cristiane M. Santoro, ANP a, Fernando Colombari, MD a a b
Intensive Care Unit, Hospital Alemão Oswaldo Cruz, São Paulo, Brazil Intensive Care Unit, Emergency Medicine Discipline, University of São Paulo, São Paulo, Brazil
a r t i c l e
i n f o
a b s t r a c t
Keywords: Prolonged ICU stay Critical illness Performance status
Purpose: To describe the admission factors associated with prolonged (N 14 days) intensive care unit (ICU) stay (PIS). Materials and Methods: Retrospective analysis of 3257 admissions during a 1.5-year period in a tertiary hospital. We tested the association between clinically relevant variables and PIS (N14 days) through binary logistic regression using the backward method. A Kaplan-Meier curve and the log-rank test were used to compare hospital outcomes for ICU survivors between patients with and without PIS. Results: In total, 6.6% of all admissions had a prolonged stay, consuming over 40% of all ICU bed-days. Illness severity; respiratory support at admission; performance status; readmission; admission from a ward, emergency room or other hospital; admission due to intracranial mass effect; severe chronic obstructive pulmonary disease; and the temperature at admission were all associated with PIS in a multivariate analysis. The created model had a good area under the curve (0.82) and was calibrated (Hosmer-Lemeshow test p = 0.431). Post hoc analysis on ICU survivors on in patients with at least two days of ICU stay yielded similar results. Hospital survival after ICU discharge was similar for patients with and without PIS (log-rank test p = 0.50). Conclusion: A small number of ICU admissions consume a great proportion of ICU bed-days. Illness severity, a need for support and performance status are important predictors of PIS. Patients who survive a PIS have similar hospital mortality to patients with a shorter stay. © 2014 Elsevier Inc. All rights reserved.
1. Introduction
Certain studies have focused on admission features and their impact on PIS [2,3], suggesting that illness severity, age and a need for organ support are all related to PIS [2,7]. However, none of the previous analyses has reported the impact of previous performance status (PS) on ICU LOS. Because PS has an important role in patient prognosis [8], its role in ICU LOS should be addressed. We performed a retrospective analysis of all patients admitted to a tertiary ICU in Brazil during a 1.5-year period to evaluate the admission features associated with PIS. All of the patients’ demographic data, need for organ support and clinical data (including PS) were recorded at admission. As a secondary end-point, hospital mortality after ICU discharge was also evaluated, categorizing patients as those with or without PIS.
Depending on the definition employed, 4% to 11% of patients admitted to the intensive care unit (ICU) will have a prolonged ICU stay (PIS) [1–3]. It has been suggested that up to 45% of all ICU days may be consumed by this apparently small percentage of patients [3]. Identifying patients at risk of a prolonged stay may help ICU management and avoid a shortage of ICU beds. Despite the increased use of resources, patients who eventually survive a PIS have acceptable long-term outcomes, including quality of life [2,4], even if specific subgroups, such as patients on prolonged mechanical ventilation [5] or the elderly [6], are considered. Therefore, a prolonged ICU length of stay (LOS) by itself should not be considered as a marker of poor outcome or a futile expenditure of resources [1].
2. Methods 2.1. Population
☆ Conflicts of Interest: All of the authors report that they have no conflicts of interest. ☆☆ Institution where the work was performed: Hospital Alemão Oswaldo Cruz. ⁎ Corresponding author. Intensive Care Unit, Hospital Alemão Oswaldo Cruz, Rua Dr. João Julião 331, São Paulo, Brazil. Zip code: 01323–903. Tel.: +55 1135490839. E-mail address: [email protected] (F.G. Zampieri). 0883-9441/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jcrc.2013.09.030
We retrospectively evaluated all critically ill patients admitted to a tertiary hospital in São Paulo, Brazil, from January 2011 to June 2012. The ICU is a mixed 34-bed unit with an in-house intensivist available 24/7. All data were collected in a database system during the ICU stay
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65
as part of the ICU routine (Epimed Monitor, Epimed®, Rio de Janeiro, Brazil). The collected data included demographic data, the reason for admission, previous PS, comorbidities, the LOS before ICU admission, a need for organ support and severity indexes. The local ethics committee approved the study protocol and waived the need for informed consent due to the study’s strictly retrospective and observational features. 2.2. Outcome definition PIS was defined as an ICU LOS of more than 14 consecutive days [2]. Readmissions were included in the analysis because ICU readmission was one of the factors hypothesized to be related to PIS. For mortality analysis after ICU discharge, however, only the last ICU admission was considered because including a previous admission with a known outcome would increase bias. ICU readmissions less than 48 hours before discharge were considered as discharge failure and grouped as a single ICU admission. 2.3. Variable selection A univariate analysis of factors that were preemptively associated with PIS due to clinical relevance was performed. Variables were tested for normality using the Kolmogorov-Smirnov test. Parametric data were compared using a t-test or analysis of variance. The MannWhitney or Kruskal-Wallis test was used for nonparametric data. Categorical variables were compared using the chi-square test. The variables included in the univariate analysis were age, body mass index, the LOS before ICU admission, a modified PS (see below), whether the admission was a readmission, the presence of major comorbidities (heart failure, dementia, chronic obstructive pulmonary disease (COPD) on ambulatory oxygen therapy, liver failure with Child-Pugh score B or C [9], chronic kidney disease on dialysis, a solid metastatic tumor or hematological malignancy), the site of origin (operating room, emergency room, ward or other hospital/nursing home; patients admitted from another hospital or a nursing home were included in the same group), the specific reasons for admission (sepsis or intracranial mass effect), the admission severity index (SAPS3 score [10]), the admission time (night or day), a need for organ support in the hour following ICU admission (vasoactive drugs, non-invasive ventilation, mechanical ventilation or dialysis) and other clinical data at admission (temperature, creatinine levels, platelet count, mean arterial pressure and heart rate).
61
[15]. A post hoc analysis of factors related to PIS in ICU survivors was also performed to evaluate whether the exclusion of patients who perished in the ICU would change our results. 2.6. Other analysis Hospital survival after ICU discharge was compared between ICU survivors in both groups through a Kaplan-Meier curve and the logrank test [16]. We also evaluated ICU LOS in quartiles of SAPS3 scores, categorizing patients as those with any degree of impairment (PS of 1 or 2) or without impairment (PS of 0). All analysis was performed using SPSS version 19.0 (IBM Corporation, Armonk, NY). P b .05 was considered significant. 3. Results In total, 3257 ICU admissions (2908 patients) were included in the analysis, of which 203 (6%) admissions were longer than 14 days. The number of patients who had an ICU LOS greater than 14 days was divided into 5-day intervals and is shown in Fig. 1. PIS consumed 42% of the 14 811 ICU days evaluated. The characteristics of the entire population and of the patients according to the presence or absence of PIS are shown in Table 1. Patients with PIS were older (70.3 ± 16.1 vs 65.8 ± 16.8 years; P b 0.001), had a worse PS, had a higher SAPS3 score at admission (58.8 ± 15.3 vs 42.5 ± 14.7; P b .001), had a longer hospital LOS before ICU admission (3 [0-12.5] vs 1 [0-3]; P b .001), were more frequently admitted from the ward, were more frequently admitted due to sepsis (30% vs 11%; P b .001) or intracranial mass effect (2.5% vs 0.2%; P b .001) and more often had comorbidities. Regarding the need for organ support at admission, patients with PIS required vasoactive drugs (23% vs 9%; P b .001), mechanical ventilation (24% vs 10%; P b .001) and noninvasive ventilation (28% vs 11%; P b .001) more frequently than patients without PIS. Patients with PIS had a higher heart rate and a higher temperature at admission (94.5 ± 22.5 vs 87.2 ± 21.7 bpm and 36.5 ± 0.92 vs 36.1 ± 0.86, respectively; both p b 0.01). The variables included in the model after univariate analysis were age; PS; SAPS3 score; the LOS before ICU admission; readmission, the site of origin and admission due to sepsis or intracranial mass effect; a previous diagnosis of heart failure, dementia, COPD on oxygen therapy, cirrhosis, a metastatic solid tumor or a hematological tumor; the use of organ support (noninvasive ventilation, mechanical ventilation or vasoactive support); and clinical admission features (the lowest mean arterial blood pressure, temperature and heart
2.4. Performance status PS was evaluated at ICU admission through a short patient or family/caregiver interview performed by the nurse who was responsible for data collection (CLP). PS is routinely evaluated at our institution for all admissions, and data are recorded in the same database. PS was categorized based on the need for help to perform self-care and categorized as 0, 1 or 2. A PS of 0 was defined as no help needed, corresponding to Eastern Corporative Oncologic Group performance status (ECOG) classes 0–2. A PS of 1 was defined as capable of limited self-care (ECOG 3), whereas a PS of 2 meant that the patient was unable to perform any self-care (ECOG 4) [11]. 2.5. Multivariate logistic regression All variables with p b 0.25 in the univariate analysis were included in a binary logistic regression using the backward method. Collinearity between variables was evaluated through calculation of the variance inflation factor [12]. An arbitrary threshold of 10 was used to define collinearity. The created model was tested through the creation of an receiver operating characteristic (ROC) curve [13,14]. The Hosmer-Lemeshow test was used to test the model’s calibration
Fig. 1. Number of admissions in each 5-day interval, from 15-60 days (total: 203 admissions).
62
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65
Table 1 Patient characteristics and outcomes
Table 2 Variables associated with LOS N14 days on logistic regression
All Number of admissions Number of patients General data Age
3257 2908
Platelets, ×103/mm3 Creatinine, mg/dL Temperature, °C Outcome LOS after ICU discharge, days ICU mortality, n (%) Hospital mortality
LOS N 14 days
P
3054 2742
203 166
-
70.35 (16.18) 55.2%
b.01
2334 (76%) 551 (18%)
129 (64%) 56 (27%)
b.001
169 (6%) 26.56 (5.31) 1143 (37%) 42.58 (14.79) 1 [0-2] 298 (9.8%)
18 (9%) 26.25 (5.81) 71 (35%) 58.80 (14.88) 3 [0-12.5] 55 (27.1%)
614 (20%) 822 (27%) 1481 (49%) 137 (4%)
100 (50%) 51 (25%) 32 (15%) 20 (10%)
b.001
330 (11%) 6 (0.2%)
60 (30%) 5 (2.5%)
b.01 b.001
218 (7%) 148 (5%) 93 (3%) 15 (0.5%) 89 (3%) 276 (9%) 94 (3%)
21 (10%) 23 (11%) 17 (9%) 4 (2%) 7 (3%) 23 (11%) 11 (5%)
.065 b.01 b.001 .027 .391 .165 .06
278 (9%) 331 (11%) 318 (10%) 29 (1%)
46 (23%) 57 (28%) 68 (24%) 2 (1%)
b.001 b.001 b.001 .585
66.10 (16.8) 65.82 (16.84) 53.8% 53.7%
Sex, % male PS Fully functional (0) 2463 (75%) Dependency for one daily 607 (19%) activity (1) Fully dependent (2) 187 (6%) Body mass index, kg/m4 26.54 (5.33) Nighttime admission 1214 (37%) SAPS3 43.59 (15.30) LOS before ICU admission 1 [0-3] Readmission 353 (10.8%) Place of origin Ward 714 (22%) Emergency department 873 (27%) Operating Room 1513 (47%) Other 157 (4%) Specific reasons for admission Sepsis 390 (12%) Intracranial mass effect 11 (0.3%) Comorbidities Heart failure 239 (7%) Dementia 171 (5%) COPD 110 (3%) Hepatic Failure 19 (1%) Dialysis 96 (3%) Metastatic tumor 299 (10%) Hematologic disease 105 (3%) Admission data Organ support Vasoactive drugs 324 (10%) Noninvasive ventilation 388 (12%) Mechanical ventilation 386 (12%) Dialysis 31 (1%) Clinical features MAP, mmHg 82.2 (17.89) HR, bpm
LOS ≤ 14 days
82.33 (17.78) 87.69 87.24 (21.89) (21.78) 195.66 195.31 (95.52) (94.92) 0.98 [0.760.98 [0.771.35] 1.34] 36.20 (0.87) 36.18 (0.86) †
4 [2-9] 255 (8.8%)⁎ 307 (10.6%)⁎
†
4 [2-8] 181 (6.6%)⁎ 226 (8.2%)⁎
.369
80.55 (19.52) 94.51 (22.54) 200.85 (104.26) 0.98 [0.711.54] 36.54 (0.92) †
16 [8-25] 74 (44%)⁎ 81 (48.8%)
.50 .484 b.01 b.001 b.001
.16 b.01 .47
Variable
OR
CI
P
SAPS 3 PS (vs PS 0) PS 1 PS 2 Site of origin (vs operating room) Ward Emergency room Other hospital/home care Readmission Intracranial mass effect COPD on oxygen therapy Need for support at admission Mechanical Ventilation Non-invasive Ventilation Maximum temperature at admission
1.03
1.02-1.04
b.001
1.70 1.91
1.18-2.44 1.09-3.35
.004 .024
2.17 1.77 4.27 1.79 6.07 1.87
1.23-3.82 1.04-3.00 2.25-8.10 1.19-2.69 1.56-23.51 1.02-3.42
.007 .038 b.001 .005 .009 .042
3.33 2.10 1.24
2.19-5.06 1.40-3.14 1.04-1.46
b.001 b.001 .015
We also evaluated the impact of PS on ICU LOS in each SAPS3 score quartile. Patients with a PS of 1 or 2 and with a SAPS3 score in the third quartile had a longer ICU LOS compared with patients with a PS of 0 (P = .001; Fig. 3). There was also a tendency toward a longer ICU LOS in the fourth quartile, although this trend was not statistically significant (P = .07). Patients with PIS had higher ICU and hospital mortality (36% vs 6% and 48% vs 10%, respectively; both P b .001). The LOS after ICU discharge was also longer for patients with PIS (22 [11.75-41] vs 5 [3-11]; P b .01). However, hospital mortality after ICU discharge was similar for patients in both groups (Fig. 4; log-rank test P = .50). 4. Discussion There are several important conclusions that can be drawn from the data presented in this manuscript. First, the study confirms previous reports that, albeit uncommon (6% in this analysis), prolonged admissions account for the occupancy of more than 40% of all ICU bed-days. Second, our analysis highlights that previous PS is associated with PIS. No comorbidity except COPD on domiciliary oxygen therapy was associated with PIS. Third, the study confirms that illness severity and a need for respiratory assistance are associated with PIS. Fourth, the increase in hospital mortality of patients with PIS is most likely due to their higher ICU mortality, as
.754 b.01 b.001 b.001 b.001
† LOS after last ICU discharge. ⁎ Percentage of mortality relative to number of patients.
rate). The variables independently associated with PIS in multivariate analysis are shown in Table 2. Excluding 50 patients in whom exclusive palliative care was adopted during the first 14 days of the ICU stay resulted in the removal of COPD on domiciliary oxygen therapy as a factor associated with PIS; the remaining variables continued in the model. The ROC curve of the model’s capability to predict PIS is shown in Fig. 2. The AUC of the model was 0.82 (95% confidence interval [CI] 0.80-0.85). The resulting model was accurate (Hosmer-Lemeshow test P = .431). The post-hoc analysis performed only on ICU survivors included 3002 admissions. Following this reanalysis, readmission and a previous diagnosis of COPD were no longer significantly associated with PIS for live ICU discharges. The remaining variables stayed in the model (Table 3).
Fig. 2. ROC curve for the model prediction of an ICU LOS greater than 14 days. AUC 0.82 (95% CI 0.80-0.85).
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65
63
Table 3 Variables associated with LOS N14 days on logistic regression performed only on ICU survivors Variable
OR
CI
P
SAPS 3 PS (vs PS 0) PS 1 PS 2 Site of origin (vs operating room) Ward Emergency room Other hospital/home care Intracranial mass effect Need for support at admission Mechanical ventilation Non-invasive ventilation Maximum temperature at admission
1.03
1.01-1.05
b.001
1.65 2.04
1.04-2.61 1.02-3.97
.031 .044
3.06 2.19 4.01 8.62
1.50-6.23 1.12-4.31 1.84-8.75 1.87-40.90
.007 .022 b.001 .007
6.00 2.03 1.55
3.60-10.02 1.23-3.34 1.24-1.94
b.001 .005 b.001
mortality after ICU discharge is similar for patients with and without PIS (Fig. 4). Arabi et al performed an analysis in a smaller sampler (947 admissions) and also found that patients with PIS consumed a greater proportion of ICU days (over 45%) [3]. This phenomenon emphasizes the importance of taking PIS risk factors at admission into account when planning resource allocation in the ICU and may influence the decision to increase the number of ICU beds or to implement an intermediary care unit (IMCU) [17]. However, the impact of an IMCU on hospital and ICU LOS is still unclear. It has been suggested that discharge to an IMCU may be a risk factor for ICU readmission [18]. Because readmission is independently associated with PIS, the effects of the implementation of an IMCU on ICU LOS may not be as straightforward as they seems and is most likely related to a balance between earlier ICU discharge and the possible increase in ICU readmission that may be modulated by the presence of an IMCU. Because there is no IMCU in our hospital, which is an important limitation to the generalization of the present data, there is no expected discharge bias involved in our analysis. Risk factors for PIS in previous analyses included illness severity, unplanned admission, readmission, admission due to respiratory reason or trauma, infection at admission, a need for organ support, the LOS before admission and age [3,7,19]. Similarly, our study suggests that illness severity and respiratory support (both invasive and noninvasive) are associated with PIS. However, the use of vasoactive drug support was not associated with PIS, most likely because patients who required vasopressors were sicker and had a
Fig. 3. ICU LOS for quartiles of SAPS3 scores in patients with good PS (0) or a need for assistance (PS of 1 or 2). The P values are non-significant, except for the third quartile.
Fig. 4. Kaplan-Meier curve for survival after ICU discharge for patients with and without PIS. Log-rank test P = 0.5. The data were censored at 30 days.
greater chance of dying early after ICU admission. Patients who were using vasopressors and died in the ICU died earlier (and therefore had a shorter stay) than patients who died in the ICU but were not using vasopressors at admission (3 [1-12.5] vs 10 [3.75-20] days; P b .001). The place of origin of the patient was also associated with PIS. Admissions from the ward or after an emergency surgical procedure were associated with PIS, corroborating the previous findings that unexpected admissions are related to an increased ICU LOS [3]. The same was true for admissions from other hospitals or from nursing homes. Flabouris et al suggested that patients admitted after interhospital transfer have higher mortality and a longer ICU LOS than similar patients admitted from the emergency department [20]. Admission from nursing homes has also been reported to be associated with an increased hospital LOS [21]. Sepsis was not associated with PIS in our analysis, contrary to the report from Arabi, in which the presence of infection was a risk factor for PIS [3]. The SOAP study also reported an increased ICU LOS for septic patients compared with non-septic patients [22]. This disparity may be explained due to differences in data collection, definitions and the time window. We included only whether the main reason for admission was sepsis, and not the presence of infection at admission [22,23]. Patients with sepsis have higher mortality compared with non-septic patients, so it is possible that when a 14-day cutoff is used, many patients with sepsis will already have perished, explaining the apparent nonexistence of an association between infection and PIS. The role of admission temperature in ICU LOS is intriguing. Kiekkas et al found that although fever was not associated with increased mortality, there was an association between increases in body temperature and mortality in the critically ill [24]. In a large series, Lee et al suggested that fever was markedly associated with mortality in non-septic patients, but not in patients with sepsis [25]. In a recent randomized trial, temperature control was associated with reduced mortality and the need for vasopressors in critically ill septic patients, which may be related to a reduction of oxygen consumption and, consequently, a better coupling of oxygen supply and demand [26]. It is uncertain whether temperature is a marker or cause of worse prognosis in critically ill patients, but our analysis suggests that this parameter may impact not only mortality but also LOS in a continuous fashion. To the best of our knowledge, no other study has evaluated the impact of previous PS on ICU LOS. PS is a key feature in the
64
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65
prognostication of critically ill patients with cancer [27,28] and is routinely used to define care goals for ambulatory cancer patients [29,30]. Our analysis shows that previous PS is associated with PIS, with an OR of 1.9 for patients with a poor previous status. The impact of PS on LOS appears to be more significant in the medium range of illness severity (Fig. 3). This finding suggests that for patients with reduced illness severity, the impact of PS on LOS is less important. Most of the patients in the first and second SAPS3 score quartiles were admitted from the operating room after elective surgery (90.2% and 60.2%, respectively) and had a very brief ICU LOS (Fig. 3); therefore, these groups most likely reflect routine ICU admissions. Although there was a trend toward a longer LOS in the fourth quartile of SAPS3 scores, this trend did not reach statistical significance (P = .07). This finding may be because the impact of PS on very ill patients is less important, as the severity of illness is most likely the greater determinant of prognosis. Except for COPD on domiciliary oxygen therapy, which is known to be associated with worse outcomes [31], no other comorbidity was associated with PIS. Therefore, we speculated that PS is more intimately associated with PIS than simply the presence or absence of comorbidity. Similar to a report by Laupland [2], we could not find any association between age and LOS before ICU admission and PIS. Although this finding may be a consequence of sampling issues, we cannot discard the possibility that LOS before ICU admission and age were surrogates for PS in previous analyses. As an important limitation, the evaluation of PS is subject to personal bias, especially when the family is asked about the previous PS of the relative. A more structured approach to PS should be adopted in further prospective studies on the subject. Two post hoc sensitivity analyses were performed. The first analysis was performed evaluating only ICU survivors. Excluding patients who perished in the ICU helped to reduce any eventual interpretation bias between PIS and mortality because ICU mortality competes with ICU LOS. In this analysis (Table 3), COPD on oxygen therapy and readmission were no longer associated with PIS. PS continued to be an important factor associated with PIS, suggesting that the association is robust and independent of ICU mortality. The second post-hoc analysis was performed excluding patients with a short LOS (≤ 2 days; data not shown). In this analysis, a PS of 2, admission from the emergency room and a need for noninvasive ventilation were no longer associated with PIS (although there was a trend in the association between MV and PIS; p = 0.06). The presence of severe liver disease was associated with PIS in this analysis (although the CI was wide: 4.7-20.7). These results are limited by the smaller number of patients with a PS of 2 in this subgroup (82 vs 187 patients in the main analysis). Nevertheless, a PS of 1 was still associated with PIS. The results of our analysis should be interpreted given the particular features of our hospital (including the absence of an IMCU). Our data do not support any change in current ICU admission practice but may aid attending physicians in considering that a recently admitted patient who presents several of the described risk factors is at increased risk of a prolonged stay. This consideration may help in answering families’ inquires and in planning ICU bed management. Additionally, our analysis confirms that in-hospital mortality for ICU survivors was similar for patients with and without PIS, suggesting that ICU LOS should not be taken into account during prognostication or decisions to withdraw life support [2]. Certain strengths of our analysis should be highlighted. First, our analysis adds new findings regarding risk factors for PIS, including the role of previous PS and the effect of admission temperature on ICU LOS. Second, albeit being a retrospective analysis, the large number of included patients makes this one of the largest analyses of risk factors for PIS. Third, a robust statistical analysis, including logistic regression, was applied to identify independent risk factors for PIS. The main results were unchanged even when post-hoc analysis was performed.
Our study has several limitations. First, this is a retrospective analysis of one center, which may limit the findings’ external validity. Nevertheless, we obtained similar results as other analyses performed in different facilities [3]. Second, we were not able to evaluate long-term prognosis (including 6-month or 1-year outcomes) after ICU stay and were limited to hospital mortality as the mortality outcome. It should be highlighted that hospital mortality is most likely inadequate as a robust outcome, especially because we are unaware of the quality of life at discharge and the destination facility (a nursing home, for example). An analysis using relevant quality-of-life outcomes, including quality-adjusted life-years after discharge, would improve current knowledge on this subject. Third, we used a limited scale to evaluate PS, and we have no information regarding PS or quality of life after hospital discharge. Fourth, the inclusion of readmissions in the analysis may be controversial because it is conceivable that the readmission of patients with other risk factors for PIS (PS, for example) could potentiate an association or create bias. However, readmission was defined a priori as a risk factor for PIS and therefore included in the analysis. We lastly were unable to perform a cost analysis. 5. Conclusion Albeit uncommon, patients with PIS consume an important proportion of all available ICU bed-days. Illness severity, a need for respiratory support, patient origin, temperature at admission, admission due to intracranial mass effect, COPD on domiciliary oxygen therapy, reduced PS and previous ICU admission are associated with PIS. Prognosis after ICU discharge appears to be independent of the presence of PIS. Further prospective studies should focus on the impact of previous PS on outcomes. Acknowledgments None. References [1] Heyland DK, Konopad E, Noseworthy TW, et al. Is it 'worthwhile' to continue treating patients with a prolonged stay (N14 days) in the ICU? An economic evaluation. Chest 1998;114:192–8. [2] Laupland KB, Kirkpatrick AW, Kortbeek JB, et al. Long-term mortality outcome associated with prolonged admission to the ICU. Chest 2006;129:954–9. [3] Arabi Y, Venkatesh S, Haddad S, et al. A prospective study of prolonged stay in the intensive care unit: predictors and impact on resource utilization. Int J Qual Health Care 2002;14:403–10. [4] Rimachi R, Vincent JL, Brimioulle S. Survival and quality of life after prolonged intensive care unit stay. Anaesth Intensive Care 2007;35:62–7. [5] Combes A, Costa MA, Trouillet JL, et al. Morbidity, mortality, and quality-of-life outcomes of patients requiring N or = 14 days of mechanical ventilation. Crit Care Med 2003;31:1373–81. [6] Montuclard L, Garrouste-Orgeas M, Timsit JF, et al. Outcome, functional autonomy, and quality of life of elderly patients with a long-term intensive care unit stay. Crit Care Med 2000;28:3389–95. [7] Higgins TL, McGee WT, Steingrub JS, et al. Early indicators of prolonged intensive care unit stay: impact of illness severity, physician staffing, and pre-intensive care unit length of stay. Crit Care Med 2003;31:45–51. [8] Christodoulou C, Rizos M, Galani E, et al. Performance status (PS): a simple predictor of short-term outcome of cancer patients with solid tumors admitted to the intensive care unit (ICU). Anticancer Res 2007;27:2945–8. [9] Pugh RN, Murray-Lyon IM, Dawson JL, et al. Transection of the oesophagus for bleeding oesophageal varices. Br J Surg 1973;60:646–9. [10] Moreno RP, Metnitz PG, Almeida E, et al. SAPS 3–From evaluation of the patient to evaluation of the intensive care unit. Part 2: development of a prognostic model for hospital mortality at ICU admission. Intensive Care Med 2005;31: 1345–55. [11] Oken MM, Creech RH, Tormey DC, et al. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol 1982;5:649–55. [12] Hsieh FY, Lavori PW, Cohen HJ, et al. An overview of variance inflation factors for sample-size calculation. Eval Health Prof 2003;26:239–57. [13] Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation 2007;115:928–35. [14] Heagerty PJ, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics 2005;61:92–105. [15] Hosmer DW, Lemeshow S. Applied logistic regression. 2nd ed. Wiley; 2000.
F.G. Zampieri et al. / Journal of Critical Care 29 (2014) 60–65 [16] Bland JM, Altman DG. The logrank test. BMJ 2004;328:1073. [17] Peelen L, de Keizer NF, Peek N, et al. The influence of volume and intensive care unit organization on hospital mortality in patients admitted with severe sepsis: a retrospective multicentre cohort study. Crit Care 2007;11:R40. [18] Renton J, Pilcher DV, Santamaria JD, et al. Factors associated with increased risk of readmission to intensive care in Australia. Intensive Care Med 2011;37: 1800–8. [19] Estenssoro E, González F, Laffaire E, et al. Shock on admission day is the best predictor of prolonged mechanical ventilation in the ICU. Chest 2005;127: 598–603. [20] Flabouris A, Hart GK, George C. Outcomes of patients admitted to tertiary intensive care units after interhospital transfer: comparison with patients admitted from emergency departments. Crit Care Resusc 2008;10:97–105. [21] Suter-Widmer I, Christ-Crain M, Zimmerli W, et al. Predictors for length of hospital stay in patients with community-acquired pneumonia: results from a Swiss Multicenter study. BMC Pulm Med 2012;12:21. [22] Vincent JL, Sakr Y, Sprung CL, et al. Sepsis in European intensive care units: results of the SOAP study. Crit Care Med 2006;34:344–53. [23] Suarez JI, Zaidat OO, Suri MF, et al. Length of stay and mortality in neurocritically ill patients: impact of a specialized neurocritical care team. Crit Care Med 2004;32: 2311–7.
65
[24] Kiekkas P, Velissaris D, Karanikolas M, et al. Peak body temperature predicts mortality in critically ill patients without cerebral damage. Heart Lung 2010;39:208–16. [25] Lee BH, Inui D, Suh GY, et al. Association of body temperature and antipyretic treatments with mortality of critically ill patients with and without sepsis: multicentered prospective observational study. Crit Care 2012;16:R33. [26] Schortgen F, Clabault K, Katsahian S, et al. Fever control using external cooling in septic shock: a randomized controlled trial. Am J Respir Crit Care Med 2012;185: 1088–95. [27] Soares M, Salluh JI, Toscano L, et al. Outcomes and prognostic factors in patients with head and neck cancer and severe acute illnesses. Intensive Care Med 2007;33:2009–13. [28] Soares M, Caruso P, Silva E, et al. Characteristics and outcomes of patients with cancer requiring admission to intensive care units: a prospective multicenter study. Crit Care Med 2010;38:9–15. [29] Kalser MH, Barkin J, MacIntyre JM. Pancreatic cancer. Assessment of prognosis by clinical presentation. Cancer 1985;56:397–402. [30] Buccheri G, Ferrigno D, Tamburini M. Karnofsky and ECOG performance status scoring in lung cancer: a prospective, longitudinal study of 536 patients from a single institution. Eur J Cancer 1996;32A:1135–41. [31] Funk GC, Bauer P, Burghuber OC, et al. Prevalence and prognosis of COPD in critically ill patients between 1998 and 2008. Eur Respir J 2013;41:792–9.
