Introduction

Advances in oncology have led to a dramatic reduction in mortality rates in cancer patients over the past few decades [1]. However, this has come at the cost of using more complex and intensive cytotoxic and immunosuppressive regimens, in addition to undertaking major surgical procedures in patients who would previously received palliative care. As a result, there is an increasing demand for critical care input to support patients through these therapies or the complications that arise as a consequence.

Around one in seven patients admitted to general intensive care units (ICU) in Europe have a malignancy, the majority being solid tumours [2]. Multiple organ support can be offered but the burden of treatment can be significant for patients and therefore it is only undertaken when there is a reasonable expectation of survival [3]. Much of the controversy regarding ICU care of cancer patients surrounds expected survival in this group of patients. It may not always be appropriate to provide intensive care and organ support to patients with a reversible illness but an otherwise terminal process. In cancer patients, there are increasing questions about whether too much effort is made to prolong life at the cost of poorer quality in the last few months of life. For those patients approaching the end of their life, it has been recognised that care should be less medicalised and occur outside the hospital environment [4].

Due to these concerns, the presence of cancer is the second most common reason cited by intensive care physicians for refusal of ICU admission [5]. However, there may be a cohort of cancer patients who would benefit from an ICU admission but are refused on the basis of prejudices against this patient population. Cancer is one of many chronic comorbid conditions that are common in the ICU population. It is likely that the presence of chronic conditions has an influence on prognosis, and it has been recognised elsewhere that further work is required to determine the impact of cancer and other comorbidities on ICU outcomes [6].

The challenge facing ICU physicians is to identify which cancer patients are likely to benefit from ICU care and in which patients the expected outcome is unacceptable to justify the burdens of ICU. With an enhanced understanding of what happens to cancer patients with critical illness, clinicians would be in a better position to give prognostic information regarding expected outcomes. This would facilitate open discussion and better inform patients and relatives as well as physicians when making admission decisions pertaining to cancer patients.

While there have been a number of reports on outcomes of cancer patients after admission to ICU, to our knowledge there has been no attempt to systematically review their findings and assess the quality of evidence currently available. The aim of our study was to assess mortality among cancer patients admitted to ICU by carrying out a systematic review of published studies and to synthesise the results using both a narrative synthesis and the distribution of published mortality.

Methods

We carried out a systematic review of published literature and summarised the distribution of reported mortality using preferred reporting items for systematic reviews and meta-analyses guidelines [7].

Study identification and eligibility criteria

To be included in the review, the study had to report survival outcomes in patients with known malignancy (solid tumours) who had been admitted to a critical care area, that is, intensive care (or treatment) unit. We limited the study to solid cancers as survival in haematological cancers has been associated with consistently poorer ICU survival that has varied with time.

Two electronic databases (OVID MEDLINE and EMBASE to April 2014) were searched using a combination of medical subject headings (MeSH), title and abstract keywords. The MeSH terms “neoplasm” and “critical care”/“intensive care”/“ICU”/“critical illness” were exploded and articles containing both terms were combined. We then searched for the terms “death”, “mortality”, “surviv$”, “prognos$”, “hospital$” or “outcome”. Articles that featured in both searches were then limited to those human studies published in English. We excluded paediatric studies and review articles from the search.

Study selection and data extraction

From the initial search results, a title review was performed by two separate reviewers (K.P., D.S.M.). Editorials, letters or case reports were excluded. Studies were required to report survival outcomes in patients with solid tumours admitted to ICU/critical care. Where only one reviewer felt that the title may represent an article of interest, the abstract was sought. The abstracts were then scrutinised by both reviewers before obtaining the full text of all relevant articles. Where additional potential studies were identified by hand-searching or expert recommendation, they were reviewed for eligibility in the same way as studies identified from electronic databases.

Studies were excluded if the study population entirely pre-dated 1 January 2000, if outcomes were reported for solid tumours mixed with haematological tumours, or if the study reported outcomes from a study population that had already been included in another study. This date was chosen to ensure data were contemporaneous with current ICU management practices [8–10].

Data describing the study population were collected from each paper. Several studies only reported the average age and severity of illness scores for a mixed cohort of patients with solid and haematological tumours and did not differentiate patients with solid tumours. We did not place any restrictions on these studies. Reported survival data were collected, which was most commonly ICU mortality, hospital mortality, 6-month mortality and 1-year mortality. The authors were contacted for additional information when clarification was required.

STROBE criteria were used to examine what was reported in each study including study design, variables, data sources, participants, descriptive data and outcome data [11].

Distribution of mortality estimates

ICU and hospital mortality with exact 95 % confidence intervals (CIs) were calculated for each study. The distribution of mortality rates was examined and summarised using the Freeman–Tukey double arcsine transformation to estimate the average of the mortality distribution with 95 % CIs. Random-effects estimates were used to take account of heterogeneity between studies.

Consistency between study outcomes was assessed using the Chi square test for homogeneity. The extent of variation between study outcomes was measured using the I 2 statistic. This indicates the proportion of the total variation across study outcomes attributable to heterogeneity. Values greater than 50 % suggest substantial heterogeneity.

Regression and stratified analyses were used to examine the role of selected study-level characteristics (for example, location, setting and severity) in contributing to differences between study outcomes. There was insufficient study information to generate informative results from multivariable meta-regression and only univariable analyses were performed. All statistical analyses were performed using Stata v.13.1 (Stata, College Station, TX, USA).

Results

Electronic database searches identified 668 references to April 2014 of which 48 papers were included in the final selection. The studies broadly fell into two groups: those reporting on a mixture of solid cancers together, and those reporting on specific tumour types. Details of the selection processes are described in Fig. 1.

Fig. 1
figure 1

Flowchart of study selection

Full size image

Study characteristics

We identified 48 studies that reported mortality (including short-term ICU or in-hospital mortality) for patients with solid tumours after admission to ICU (Table 1) [2, 12–58]. While the inclusion criteria required that admissions from 2000 onward were reported, the studies spanned admissions between 1997 and 2011. Where cancer site-specific outcomes were reported, the commonest (20 studies) were for lung cancers. Four papers each reported head and neck and colorectal cancers; oesophageal, breast and pancreatic cancers were each reported in three papers; stomach and gynaecological cancers each had two papers; and single papers were identified for upper gastrointestinal, all gastrointestinal, urological, prostate and melanomas. Soares et al. has published several outcome studies on a cohort of patients from May 2000 variously to December 2004 [59], to December 2005 [60], and to January 2004 [47, 61], and we chose the single most comprehensive of these papers [47].

Table 1 Characteristics of studies reporting short term mortality for patients with solid cancers admitted to ICU
Full size table

The mean age of patients, where given, ranged from 47 to 75 years and, unless tumours were sex-specific, patients comprised a mixture of men and women. Where the presence of metastatic disease was reported, this ranged from 0–100 %. Twenty-two studies were from Europe, ten were from Brazil and nine from North America. Slatore’s study was the largest [46], with 49,373 patients whilst Azoulay et al., Darmon et al., and Song et al. reported on all cancer patients of which solid tumours made up 10, 12 and 13 patients, respectively [16] [25] [52]. The majority of the studies (36/48, 75 %) included a measure of physiological status, and there was wide variation between them; 20 studies provided mean SAPS II scores, ranging from 25.1 to 63.4, and 15 studies gave mean APACHE II scores, ranging from 10.3 to 25.6.

We considered bias in each of the studies, using information reported under PRISMA criteria. With respect to information or measurement biases, it is possible that studies misclassified whether or not patients had cancer, or misclassified the site of the primary tumour. However, because no additional sources of information were available to validate the classification, we were unable to evaluate the extent, if any, of such biases. As all studies reported total and not cause-specific mortality, misclassification of outcomes is unlikely, although the timing of death may have been incorrect. Mortality could be misrepresented in those studies that excluded routine postoperative patients, patients with ICU stays of less than 24 h, readmissions or only inclusion of patients requiring specific interventions, such as renal replacement therapy or mechanical ventilation.

Overall mortality up to 5 years after ICU admission

Mortality was reported for ICU, hospital and up to 5 years after ICU admission [Table 1, Electronic Supplementary Material (ESM)]. ICU mortality was reported in 35 studies and ranged widely from 4.5 to 85 %. Hospital mortality was reported in 31 studies ranging from 4.6 to 76.8 %. Mortality at 1 year was reported in five studies and ranged from 35.9 to 88.0 %, and a single study by Bissell reported 62.8 % survival at 5 years [17]. ICU and hospital mortality are plotted showing the average of the distribution of rates in Fig. 2. The average mortality summarises only those studies using independent samples, so that where several papers reported on the same group of patients, only one representative paper was used. The overall average mortality within ICU was 31.2 % (95 % CI 24.0–39.0 %) based on a total sample of 25,339 patients with a solid tumour, of which Bos’s study comprised nearly half of the patients. There was substantial variation between studies in the proportion of patients who died in ICU (I 2 = 99.1 %, χ 2 = 3,483, df = 31, p < 0.001). The average hospital mortality was 38.2 % (33.8–42.7 %) among a total sample of 74,061 patients with solid tumours, of which Slatore’s study comprised two-thirds of the patients. There was substantial variation in hospital mortality between studies (I 2 = 98.8 %, χ 2 = 1,829, p < 0.001). ESM Fig. 1 provides summary average mortality estimates at all follow-up periods.

Fig. 2
figure 2

ICU and hospital mortality among ICU patients with solid tumours

Full size image

Predictors of survival

When the effects of risk factors associated with survival in multivariable analyses were reported these have been described in Table 2. Severity of illness scores were generally associated with greater risks of mortality with each unit increase; the relatively larger effect size reported by Song et al. [52] reflects the comparison between patients with SOFA scores of ≤10 and >10. The relationship between higher severity of illness score and increased mortality was reported up to 90 days after admission and was of a similar magnitude to its effect on shorter-term mortality. This may suggest that it imposes little additional risk after the initial ICU admission period.

Table 2 Multivariable prognostic factors associated with survival among ICU cancer patients
Full size table

Cancer patients who are admitted as medical, as opposed to surgical admissions had increased risk of ICU mortality of between two- and fourfold, and increased risks of in-hospital mortality of between six- and eightfold. WHO Performance Status, which is used to quantify cancer patients’ general health and functioning, was associated with survival. Generally, papers compared patients with scores of 3 and 4 (from capable of only limited self-care to completely disabled and confined to bed or chair) with those of 0, 1 and 2; where alternative thresholds were used, these have been indicated in Table  2. Poorer performance status was associated with increased ICU mortality of between four- and sevenfold, and increased mortality by two- to threefold at 90 days after ICU admission. Sepsis increased ICU mortality by fivefold and in-hospital mortality between two- and fivefold. Mechanical ventilation increased ICU mortality by around sixfold in most studies; Mourad’s much higher figure of nearly 17-fold increased ICU mortality included haematological patients and the confidence interval around the estimate was wide [37]. Use of vasopressors increased the risks of both ICU and in-hospital mortality. The increased mortality at 30 days reported by Darmon is of a similar magnitude to that reported in ICU by several authors, and again this may suggest that its effects are not sustained beyond ICU [25].

The impact of cancer stage has been inconsistently associated with mortality. Mendoza’s study of 147 patients with mixed solid tumours found metastatic disease to be independently predictive of hospital mortality on multivariate analysis [35]. Soares’ study of patients with head and neck cancer identified that those with stage IV disease had a higher risk of in-hospital death with OR 3.8 (95 % CI 1.28–11.28) [62]. A further study by Soares of cancer patients receiving >24 h of invasive mechanical ventilation found that hospital mortality was associated with recurrence/progression of cancer status (OR 3.43, 95 % CI 1.81–6.53 compared with controlled cancer) and airway/pulmonary involvement by the tumour as the reason for ventilation (OR 5.73 CI 1.92–17.08) [61]. However, an additional three papers did not find an association between metastatic disease and short-term outcomes such as ICU or hospital mortality [2, 24, 33]. The impact of leukopenia and neutropenia has been assessed in many of the studies and was not found to be associated with short-term mortality [2, 25, 29, 32, 54].

Studies varied widely in their mean acute illness scores (Table 1) and reported mortality (ESM Table 1). Generally, studies with higher acute illness score reported higher ICU mortality (ESM Fig.2). However, there was some variation in mortality between studies with similar acute illness scores, and these reflect differences in other patient characteristics, particularly cancer site.

ICU and hospital mortality for specific tumour sites

Figure 3 summarises short-term mortality where this has been described for specific tumour sites, indicating the site-specific average and overall average (additional data are provided in ESM Table 2). It is apparent from Fig. 3 that survival varies between studies on the same cancer site and between different cancer sites. Precision of estimates varied considerably between studies. Generally, mortality was highest for patients with lung cancers and lowest for patients with gynaecological cancers (average ICU mortality 40.1 %, 95 % CI 28.6–52.2 and 12.0 %, 95 % CI 7.4–10.4, respectively). Other individual tumours were less frequently reported with four papers reporting outcomes for colorectal cancer (average hospital mortality 35.0 %), four papers in head and neck tumours (average hospital mortality 54.7 %), three papers for oesophageal malignancy (average hospital mortality 19.9 %) and four papers for breast cancer (average hospital mortality 57.7 %).

Fig. 3
figure 3

ICU and hospital mortality among ICU patients with specific tumours

Full size image

Discussion

This systematic literature review identified a set of studies that describe contemporary outcomes in non-haematological cancer patients after admission to ICU. The 48 papers that reported on solid cancers were characterised by a wide range of case-mix variables both in terms of critical illness and in tumour types. The broad range of observed outcomes, with ICU mortality ranging from 4.5 to 85 %, reflects the heterogeneity of patients described in the available literature. However, average mortality across all included studies with ICU mortality of 31.2 % and hospital mortality of 38.2 % is considerably lower than that which has been historically reported. Cancer patients should therefore not be excluded from ICU exclusively based on having an underlying malignancy, and instead those factors known to be associated with survival should be taken into account.

A number of studies looked at factors associated with survival. Poorer physiological status was associated with poorer ICU and hospital survival, but the range of different scoring systems employed, and the absence of any measure from some studies, makes summary estimates difficult in this patient group. All types of organ failure were associated with a poorer outcome; however, of these, requirement for invasive mechanical ventilation had the strongest effect on mortality. In addition, poor chronic health or limited functional status have been demonstrated to be associated with poorer survival. These findings are consistent with those seen in the general ICU population and are not specific to cancer patients. Whilst tumour type has not been demonstrated to be associated with mortality in any of the individual studies, the average mortality by tumour types described in this systematic review suggests that there is variation. The impact of disease stage and presence of metastasis has been inconsistently associated with mortality and the exact role these factors play is uncertain. Thus, despite around 15 % of ICU admissions being for patients with known malignancies [2], current literature is unable to predict likely survival of the individual cancer patient after ICU admission; however, it can guide prognostication.

Our study has strengths and weaknesses. It is the first systematic review on outcomes among cancer patients admitted to ICU. We have used established systematic review procedures, PRISMA guidelines [7], and STROBE criteria for assessing observational studies [11]. Unlike narrative reviews, our study has a specific, explicit and therefore reproducible methodological strategy for identifying and synthesising relevant papers and is not based on the subjectivity of the authors. Our analysis adds to current literature by consolidating and summarising a large amount of information from primary published sources which many busy practicing ICU clinicians would not have time to identify or read. As such, its strength lies in pooling this information in an accessible form while highlighting the paucity and heterogeneity of information available on specific cancer outcomes.

The principal weakness of our study is that we have included all studies within this field whether the focus was on the general ICU cancer population or a subgroup such as those with sepsis, a specific organ failure or an individual tumour type. As a result, there are large differences between the study populations of the included papers which is reflected in the distribution of mortality rates that they report. The differences in terms of patient population, type of cancer, or type of patients (surgical vs. non-surgical) are confounding biases and contribute towards the heterogeneity we report. There are vast differences in the outcomes associated with surgical admissions versus medical admissions to ICU, and this effect seems to be exaggerated within the oncological population. For example, the three studies with the lowest mortality [30, 41, 43] includes mainly elective surgical patients. In contrast, the studies with higher mortality [20, 31, 33, 54] includes only a few surgical patients, and instead selected patients based on their severity (mechanical ventilation plus at least one organ failure) [31]: only patients requiring renal replacement therapy [33], the extent of the underlying disease [20] or both. The difference in reported mortality is therefore to be expected.

Bos et al. [19] noted that many studies which report on prognostic factors mainly involve specialised oncological ICUs, making it difficult to extrapolate to general ICUs where the mixture of tumour types and expertise is different. They suggested that their own study was most appropriately compared to papers by Taconne et al. and Soares et al. [2, 50]. But even in this limited comparison, there were important differences in terms of study size, duration, and discrimination of emergency admissions, medical and surgical patients. Given the paucity and variability of the published literature, we felt that it would have been even less informative to have further restricted our inclusion criteria as this would have lost more than half of the published data with no increase in precision. This is important because, although we have summarised the distribution of reported mortality using the average mortality with 95 % confidence intervals, the mortality reference range of published studies is justifiably very wide.

As the incident number of cancers will continue to increase globally [63], driven largely by aging populations but in some countries by increased exposures to risk factors, demand for ICU support for cancer patients and for information to guide admission policies will also rise. Our work adds to current literature because it draws together different strands of international research and highlights the ways in which these studies of cancer patients in ICU differ. We feel that current evidence on outcomes of cancer patients after admission to ICU would be improved with additional clinical details such as treatment intent, cancer staging/histology or pre-ICU clinical pathways. Targets for future research should build upon the work that has already been published by reporting outcomes in clearly defined groups and producing further prognostic information on those factors that are associated with poorer survival.

Whilst oncological ICUs provide a large patient population in which to study these patients, they may not be representative of the general ICU cancer population and therefore additional studies require to be performed on both groups. Where patients are admitted to general ICUs, there is a value in reporting comparative outcomes for non-cancer patients to allow calculation of the additional risk posed by having cancer. In addition to describing survival over the short, medium and longer terms, research is needed to describe whether cancer patients are likely to receive planned anti-cancer therapy after discharge from ICU, as admissions are currently assumed to be suitable for ongoing treatment following successful discharge. Because there are international variations in the definition of ICUs and in the case-mix of patients they treat, it will be important for future research to clearly define staffing levels, interventions provided, and patient characteristics, so that results can be usefully generalised [64].