Introduction

The development of life-sustaining treatments over the last half century has resulted in some patients remaining dependent on life support until death [1]. In these patients, continued curative treatment is rarely the best option [2]. Prolonging non-beneficial treatments robs patients of their dignity and families of an opportunity to prepare for bereavement [3]. Intensivists have, therefore, limited the use of life-sustaining treatments in these situations. Presently, most deaths in the intensive care unit (ICU) occur after decisions to forgo life-sustaining treatment (DFLSTs) [4–6], and the incidence of decisions to forgo LSTs may be increasing [7]. Making DFLST, which may consist in withholding and/or withdrawing life support, marks a shift from curative care to comfort care. Patients with DFLSTs are closely monitored and given palliative care as needed to ensure optimal comfort.

DFLSTs must be ethically appropriate. Perceptions of what is ethical, however, may vary. Substantial variability in the decision making process has been documented in previous research. These variations concern the incidence of decisions to forgo LSTs, the characteristics of patients who receive these and the procedure that is followed for making decisions to forgo LSTs [4, 5, 8–12]. Variations were also identified in responses to ethical scenarios [6, 13]. There is widespread agreement that there is no ethical difference between withholding and withdrawal [6], although withdrawal has been described as more difficult for intensivists, and is not used in some countries [10]. A single large ICU study recorded practices in 37 ICUs from 17 European countries [10]. The results show considerable variability in decisions to forgo LSTs in Europe. However, no large study across widely disparate geographic areas has been reported to date. The objectives of this study were to collect data on decisions to forgo LSTs in 14,488 patients admitted to 282 ICUs in seven different regions, and to identify factors associated with decisions to forgo LSTs in ICUs.

Patients and methods

We used the prospective international cohort created for the SAPS 3 study [14, 15]. The organization of the project, data collection, and study cohort have been described in detail elsewhere [14, 15]. This is a pre-planed analysis of the SAPS 3 study. Participating countries can be seen from Table E10 of the ESM of the SAPS 3 cohort description [14, 15]. Definitions of major therapeutic limitation during ICU stay were collected at ICU discharge. The questions asked to researchers evaluates if major therapeutic limitations were used during the ICU stay. Only those limitations expected to have had a relevant impact on patient’s morbidity and/or mortality were registered. Date where withholding or withdrawing therapy was first used was registered.

Database

The SAPS 3 hospital outcome cohort comprises 16,784 patients from 303 ICUs. We excluded the 2,296 patients for whom no data were available regarding decisions to forgo LSTs. This left 14,488 (86.3%) patients for the study. Among them, 1,239 (8.6%) received decisions to forgo LSTs.

Data quality

The database was evaluated for completeness and reliability. Independent raters rescored the data, and kappa coefficients and intra-class correlation coefficients were computed, as appropriate [16]. Data quality proved excellent, as shown by the detailed results reported in the ESM file of the SAPS 3 primary report [14, 15].

Statistical analysis

Statistical analysis was performed using the SAS system, version 9.1 (SAS Institute Inc., Cary, NC). All P values smaller than 0.05 were considered significant. Unless otherwise specified, results are expressed as median and quartiles. The chi-square test was used for categorical data. For continuous variables, ANOVA was used. Univariate logistic regression analyses were performed to identify patient- and ICU-related factors that might predict decisions to forgo LSTs. Factors that were significant in the univariate analyses were entered into a multivariate stepwise logistic regression analysis. These have been added in the footnote of Table 3. If regions were introduced into the model, significant differences could have been found. However, since participating countries were not representative samples in each country, we did not introduced this variable in the model.

Results

Figure 1 shows the patient flow chart of 14,488 patients admitted to 282 ICUs in seven geographic areas. ICU organizational and managerial characteristics were available for 271 ICUs (Table 1). Overall the median (quartile) number of admissions per year was 441.5 (267–723) patients per ICU.

Fig. 1
figure 1

Patient flow chart showing incidence and outcome of decisions to forgo life-sustaining therapies in patients included in the SAPS3 database

Full size image
Table 1 ICU characteristics
Full size table

As shown in Fig. 1, decisions to forgo LSTs were implemented in 1,239 (8.6%) patients, including 677 (54.6%) patients who received withholding decisions and 562 (45.4%) who received withdrawal decisions. Hospital mortality was 21% (3,050/14,488). Among the deaths, 1,105 (36.2%) occurred after decisions to forgo LSTs. Hospital mortality was 86.4% in patients with withholding decisions and 92.5% in patients with withdrawal decisions.

Table 1 shows that decisions to forgo LSTs were more common in hospitals without emergency departments, in smaller ICUs, and in ICUs with lower nurse-to-patient ratios and larger numbers of physicians per ICU bed. DFLSTs were also more common when intensivists were present only during weekdays (compared to ICUs where intensivists were present during weekdays and weekends), when multidisciplinary meetings were held, and when nurses and intensivists performed clinical rounds together. Conversely, DFLSTs were less common in ICUs that had at least one full time intensivist and in those with intensivists available at night and over weekends.

Among patients who died, the proportion with DFLSTs ranged from 26 to 63.5% according to the region where the patient was admitted. Moreover, the proportion of hospital survivors with withdrawal decisions ranged from 2.4 to 30.3% and the proportion with withholding decisions ranged from 4 to 40% according to the region. Table E1 and figure E1 describes significant differences across the participating regions. As shown in Table 2, overall patients with DFLSTs were older, and a larger proportion of them exhibited severe co-morbid conditions and immunosuppression. Admission from a ward and life-sustaining treatment before ICU admission were more common among patients with than without decisions to forgo LSTs. SAPS 3 and SOFA scores at ICU admission were 45 (36–57) and 3 (2–5) in patients without decisions to forgo LSTs compared to 67 (58–77) and 6 (4–9) in patients with decisions to forgo LSTs (P < 0.0001 for both scores), respectively.

Table 2 Patient characteristics
Full size table

Table 3 reports independent predictors of DFLST implementation identified by stepwise logistic regression. The following variables were associated with increased incidence of DFLST: higher age, hospital location before ICU admission, unplanned ICU admission, documented infection at admission, non-surgical status or emergency surgery, higher SOFA score at ICU admission, comorbidities such as NYHA-IV chronic heart failure, hematological malignancies and solid tumors; ICU admission for shock, ICU admission for neurological cause such as cerebrovascular accident, intracranial tumor or post-anoxic coma; pancreatitis and other digestive causes (excluding cholecistitis). The need for vasoactive agents and longer length of ICU stay were determinants of DFLST. Among ICU-related variables, a higher number of nurses per patient was associated with increased incidence of DFLST [odds ratio of 1.03 (1.005–1.058)/nurse per bed].

Table 3 Results of multivariate stepwise logistic regression on decision to forgo life-sustaining therapies
Full size table

Seven variables were independently associated with a decreased incidence of DLST, namely, ICU admission for diabetic complication, rhythm disturbances, acute lung injury, or cholecistitis. Among ICU and hospital-related variables, availability of an emergency department in the same hospital, presence of a full time ICU-specialist and doctors presence during nights and week-ends were also associated with a decreased incidence of DFLST.

Discussion

Decisions to forgo LSTs in adult ICU patients have been a focus of increasing research over the last two decades. Descriptive studies were performed at local, national, [4, 5, 8–10] and multinational levels [5, 8, 10]. Using the SAPS 3 database of 14,488 patients in 282 ICUs, we found that in addition to previously identified predictors (case-mix, severity, co-morbidities and nature of the acute medical disease), organizational variables were independently associated with the incidence of DFLSTs. Namely, the number of nurses, availability of an ED in the same hospital, the presence of full time intensivist including doctors who make the rounds during the week-end days were independently associated with incidence of DFLSTS.

We decided not to study the impact of geographic area on the incidence of DFLSTs. Indeed, center participation to the SAPS 3 database included criteria to minimize heterogeneity in terms of outcome in homogeneous groups of patients.

A major strength of this study is the large sample of patients. In addition, we collected information on ICU characteristics. Very little is known about the potential impact of ICU characteristics and critical-care organization on end-of-life practices. The status of the institution, e.g., private versus public and teaching versus non-teaching, has been reported to affect end-of-life practices [17–19]. As expected, DFLSTs were made in the sickest ICU patients [4, 5, 8–10]. However, the impact of organizational factors on the incidence of DFLSTs suggests that these substantially influence the end-of-life decision-making procedure. Indeed, these results suggest that in ED-patients who were admitted from another hospital, DFLSTs were more likely to occur. Along this line, presence of a full time ICU-specialist and availability of doctors making rounds during weekend days is associated with decrease in incidence of DFLST. These findings must be integrated in a strategy to better understand factors that influence end-of-life care.

This finding of significant impact of organizational factors on DFLSTs invite qualitative studies into factors that determine the incidence, pattern, and outcomes of DFLSTs. Studies have shown variations in decisions to forgo LSTs with personal physician characteristics, experience [20], gender [21], specialty [19] or time working in ICUs [22]. Religious beliefs and cultural background play a role [6, 10, 21, 23]. The current study suggests that, in addition, ICU resources, case-mix, co-morbidities, patterns of ICU may also influence decisions to forgo LSTs. Along this line, the fact that the presence of full time intensivist was associated with lower risk of decisions to forgo LST may be ascribed to a less opened ICU admission policy when each single admission is discussed with the senior intensivist rather than commanded by the primary physician.

Our study has several limitations. First, the database used for the study was not designed for an investigation of decisions to forgo LSTs. Nevertheless, a sub-study on decisions to forgo LSTs was planned early in the designing of the SAPS 3 study, so that investigators were aware of the need to collect accurate data on treatment withholding and with withdrawal decisions. Second, we did study whether the country or the region were potential determinants of DFLSTs. Beyond the lack of representative sample of each country or region, we also may hypothesize that the variability across geographic areas demonstrated in our study may mask variability within each country and within each ICU, as previously reported [13]. Last, information on DFLSTs was missing for about 15% of the patients in the database. The patients did not differ from the rest of the cohort in terms of severity or mortality, suggesting that missing data did not indicate absence of decisions to forgo LSTs but instead reflected failure to record information on decisions to forgo LSTs.

In summary, this multicenter international study documents variables that influence significantly the procedure of end-of-life decisions. The finding that organizational factors may have significant impact on incidence of DFLST raises crucial questions about the determinants of DFLSTs and the definition of optimal DFLST practice.

In the future, guidelines may help spread excellence in end-of-life care. However, certain types of cultural variations are permissible and should not be perceived as incorrect practices. In addition, organizational factors should be recognized as factors potentially influencing ICU end-of-life care.