A systematically structured review of biomarkers of dying in cancer patients in the last months of life; An exploration of the biology of dying

Background The Neuberger review made a number of recommendations to improve end of life care, including research into the biology of dying. An important aspect of the biology of dying is the identification of biomarkers as indices of disease processes. Biomarkers have the potential to inform the current, limited understanding of the dying process and assist clinicians in recognising dying, in particular how to distinguish dying from reversible acute deterioration. Objectives To critically appraise the literature on biological factors that may be used as prognostic indicators in advanced cancer patients and to identify candidate biomarkers of the dying process that can be measured serially in cancer patients’ bodily fluids. Methods A systematically structured review was conducted using three electronic databases. A hand search of six peer-reviewed journals and conference abstracts was also conducted. Studies reporting prognostic biomarkers in cancer patients with a median survival of ≤90 days and post-mortem studies were included. Final levels of evidence and recommendations were made using the Evidence Based Medicine modified GRADE system. Results 30 articles were included. Seven prognostic biological factors demonstrated Grade A evidence (lymphocyte count, white blood cell count, serum C-reactive protein, albumin, sodium, urea and alkaline phosphatase). An additional eleven prognostic factors were identified with Grade B evidence (platelet count, international normalised ratio, serum vitamin B12, prealbumin, bilirubin, cholesterol, aspartate aminotransferase, alanine transaminase, lactate dehydrogenase, pseudocholinesterase and urate). A number of biomarkers were specifically identified in the last two weeks of life but limitations exist. No post-mortem studies met the inclusion criteria. Conclusion The biology of dying is an important area for future research, with the evidence focused on signs, symptoms and prognostic factors. This review identifies a number of common themes shared amongst advanced cancer patients and highlights candidate biomarkers which may be indicative of a common biological process to dying.


Objectives
To critically appraise the literature on biological factors that may be used as prognostic indicators in advanced cancer patients and to identify candidate biomarkers of the dying process that can be measured serially in cancer patients' bodily fluids.

Methods
A systematically structured review was conducted using three electronic databases. A hand search of six peer-reviewed journals and conference abstracts was also conducted. Studies reporting prognostic biomarkers in cancer patients with a median survival of 90 days and post-mortem studies were included. Final levels of evidence and recommendations were made using the Evidence Based Medicine modified GRADE system.

Results
30 articles were included. Seven prognostic biological factors demonstrated Grade A evidence (lymphocyte count, white blood cell count, serum C-reactive protein, albumin, sodium, urea and alkaline phosphatase). An additional eleven prognostic factors were identified with Grade B evidence (platelet count, international normalised ratio, serum vitamin PLOS ONE | https://doi.org/10.1371/journal.pone.0175123 April 6, 2017 1 / 31 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 B12, prealbumin, bilirubin, cholesterol, aspartate aminotransferase, alanine transaminase, lactate dehydrogenase, pseudocholinesterase and urate). A number of biomarkers were specifically identified in the last two weeks of life but limitations exist. No post-mortem studies met the inclusion criteria.

Conclusion
The biology of dying is an important area for future research, with the evidence focused on signs, symptoms and prognostic factors. This review identifies a number of common themes shared amongst advanced cancer patients and highlights candidate biomarkers which may be indicative of a common biological process to dying.

Background
In a comprehensive evaluation of the challenges and actions required to provide the best care for dying patients, the Neuberger review identified the need for further research into the biology of dying as a priority [1]. The biology of dying is an umbrella topic that encompasses the physiological and biological changes attributed to the dying process, in addition to the aetiology of signs and symptoms commonly seen in the last days, weeks and months of life. There is significant uncertainty in consistently and accurately identifying the dying phase (last days of life) and no definitive diagnostic criteria exist [1,2]. Little is known about the process of dying [1] and in the final days of life, new symptoms or exacerbation or recurrence of previously well-controlled symptoms can occur [3,4]. A prospective cohort study of 343 doctors found only 20% of prognostic estimates in hospice patients were accurate and that, overall doctors overestimated survival by a factor of five [5].
It has been extensively documented that cancer patients experience a sharp functional decline in the last months of life [6]. The prevalence of common terminal symptoms amongst patients with advanced cancer is suggestive of a common terminal trajectory; historically termed the "terminal cancer syndrome" [7]. Although the concept of the "terminal cancer syndrome" has been largely superseded, there is significant evidence for the prognostic importance of dyspnoea (Grade B), cancer anorexia-cachexia syndrome (CACS, Grade B), delirium (Grade B) and low performance status (Grade A) in advanced cancer [8]. A systematic review by Kehl et al. demonstrated that dyspnoea (56.7%), pain (52.4%), respiratory tract secretions (51.4%) and confusion (50.1%) were the most prevalent signs and symptoms that occur in the last two weeks of life [9]. Further, Hui et al. identified 13 signs highly predictive of death within three days [10,11].
An important aspect of the biology of dying is the identification of biomarkers as indices of disease processes. Biomarkers have the potential to inform the current, limited understanding of the process of dying and assist clinicians in recognising dying, in particular how to distinguish dying from reversible acute deterioration. In 2005, Maltoni et al. published evidencebased recommendations for a number of prognostic biological factors in advanced cancer patients [8]. Subsequently, a number of prognostic models have been developed that incorporate prognostic biomarkers [12]. No systematic reviews have been conducted that summarise the evidence for prognostic biomarkers in advanced cancer patients.
In this systematically structured review, the authors summarise the evidence for prognostic biomarkers in advanced cancer patients in the last months of life and extrapolate which biological processes are affected.

Objectives
This systematically structured review was conducted to collate and critically appraise the literature on prognostic biomarkers in advanced cancer. The following questions formed the basis of this review: 1. What biological factors are prognostic in advanced cancer patients in the last days, weeks or months of life?
2. Can serial measurement of identified biomarkers detect the last days to weeks of life in advanced cancer patients?

Methods
Given that no randomised controlled trials have been conducted on this research topic, a systematically structured review was conducted to ensure a replicable and systematic synthesis of the evidence. The review was structured according to the PRISMA standards for conducting a systematic review [13].

Literature search methods
On 5 th February 2016, three electronic databases were searched (Medline, Scopus and Cochrane Database of Systematic Reviews) using combinations of the key words, described in Table 1. Limits were set to humans, adults (aged over 18  Institutional review board approval was not sought or required for this literature review. biolog* OR physiolog* OR pathophysiolog* OR patholog* OR biomarker* OR biologic marker* OR biological marker* OR biological factor* OR "biology of dying" OR prognostic* OR prognosis OR predict OR mortality OR "terminal cancer syndrome" OR "common terminal pathway" OR "common terminal trajectory" OR "final common pathway" OR "terminal syndrome"

Eligibility criteria
The primary objective was the identification of prognostic biomarkers in advanced cancer patients in the last days, weeks or months of life. Biomarkers were defined as objective, quantifiable characteristics of biological processes [14] quantifiable in bodily fluids and tissues. Given that most cancer patients seem to follow a common terminal trajectory, [6] only patients with cancer were included in this study. Diagnosing imminent death is often difficult and imprecise [1] and terminal physiological changes are often seen many weeks before death. Further, the definition of "advanced cancer" is often lacking in the literature. In line with the study by Maltoni et al. this review included only study populations with a median survival of 90 days [8] that, for the purpose of this study, defines the term "advanced cancer." This criterion enabled us to capture biological factors predictive of dying over days, weeks and months of life. Post-mortem studies of cancer patients specifically looking at biomarkers of the dying process were also included.
All types of peer-reviewed evidence were included and a 16-year timeframe was selected to ensure comprehensive yet current coverage of the literature, and build on the excellent review Articles were excluded if they described only signs, symptoms or physiological changes associated with imminent death. The following types of articles were also excluded: duplicates, non-English language, paediatric populations, editorials, commentaries, case reports, reviews and animal studies. The systematic review by Maltoni et al. was included as it provided an evidence-based summary of the literature up until 2005 [8]. Primary researchers were contacted by email, where necessary, to clarify information to ensure strict adherence to the inclusion criteria. Where survival data could not be confirmed, studies were excluded.

Study selection
A review protocol was developed by VLR in advance of the literature search (S2 File). VLR extracted the data from the studies and discussed the results with SC. Disagreements between reviewers were resolved by consensus. Reviewers were not blinded for authors, institutions, or journals of publication. The Liverpool Reviews and Implementation Group (LRiG) at University of Liverpool reviewed and agreed on the employed methodology.

Quality assessment
The primary authors (VLR and SC) used the UK National Institute for Health and Clinical Excellence (NICE) hierarchy of evidence to assign a quality rating to the potential articles for inclusion [15]. Given that a variety of research outputs were considered in this review, a critical appraisal tool described by Hawker et al. [16] was utilised to further evaluate the quality of studies. Briefly, a four-point scale from one (very poor) to four (good) was assigned to nine areas including: abstract and title, introduction and aims, methods and data, sampling, data analysis, ethics and bias, results, transferability/generalisability, and implications and usefulness with a total number between 9-36 assigned to each study [16]. The Maltoni et al. sevenpoint checklist of quality criteria for evaluation of studies on prognostic factors was also utilised, where one point is assigned to seven criteria including; prospective study design; welldefined cohort of patients assembled at a common point in the course of their disease; random patient selection, percentage of patients lost to follow-up 20%; ratio between the number of events (death) and the number of potential predictors !10; prognostic variables fully defined, accurately measured, and available for all or a high proportion of patients; and reliable measurement of outcome (date of death) [8]. A total score between 1-7 was assigned and high quality (or low probability of bias) was attributed to studies fulfilling at least five of the seven criteria [8]. VLR and SC assessed the quality of potential articles for inclusion independently and mean scores were assigned. Spearman's rank correlation coefficient was used to measure rank correlations of scores between assessors. Data was analyzed using Statistical Software Package for the Social Sciences 1 (SPSS 1 version 22.0; IBM SPSS Inc., Chicago, IL). Final levels of evidence and recommendations were made using the Evidence Based Medicine modified Grading of Recommendations Assessments, Development and Evaluation (GRADE) system (Grade A = high quality evidence with consistent results, to Grade D = very low quality evidence such as expert opinion) [17].

Study selection
The results of the literature search are summarised in Fig 1. Based on firm application of the inclusion criteria, 30 articles were included in this systematically structured review ( Table 2).  • In univariate analysis, PINI and its 4 components: CRP, AGP, albumin, prealbumin were predictors of length of survival.
• In multivariate analyses, CRP was the most important prognostic factor, and B12 was independent of CRP in predicting survival.
• There was a strong correlation between the presence of metastasis, hepatic dysfunction and an elevated serum B12 level (p<0.001).
• Serum urate concentration was measured consecutively for 3 weeks before death (n = 39) and was significantly increased between the first and the second week before death (p = 0.006).
Lam PT et al.

2007, China
[27] • Prospective cohort study (explorative) • To identify potential prognostic factors affecting the survival in patients with advanced cancer who were treated in a palliative care unit at a regional hospital in Hong Kong.
• Hospital inpatients and outpatients at a palliative care unit at a regional teaching hospital in Hong Kong • Average serum LDH concentration was measured consecutively at 2 weeks and 1 week before death (n = 25) and was 5.04.04 +/-347.11IU/L and 630.40 +/-417.32IU/L respectively (p = 0.008).
(Continued) • PiPS-B scores can correctly order survival times for pairs of participants 67.5% of time.
• The PiPS prognostic models can predict whether patients will survive for "days," "weeks," or "months".
Cui • When the prognostic score of a patient was > 12, the prediction of survival appeared to be < 7 days.
• When the prognostic score of a patient was < 6, the prediction of survival appeared to be > 180 days (6 months • The overall accuracy between the PiPS-A and actual survival was 52.0%.
• The overall accuracy between the PiPS-B and actual survival was 49.5%.
• The overall accuracy between clinicians' estimates and actual survival was 46.5%, which were lower than those of the PiPS models.
• The "weeks" and "months" groups showed significantly prolonged survival rates than "days" group did in both PiPS-A and PiPS-B, by the Kaplan-Meier method.
• The sensitivity, specificity, PPV, and NPV of predictions of survival using the PiPS-A and the PiPS-B predictor models were variable.
• In the "days" and "weeks" groups, the PiPS-A/PiPS-B predictions and clinicians' estimates were consistent with actual median survival.
Amano • In this study, the feasibility of previously described predictor models: PaP Score, D-PaP Score, PPI, and modified PiPS models were assessed across all palliative care settings.
• The feasibility of PPI and modified PiPS-A was >90% in all groups, followed by PaP and D-PaP scores; modified PiPS-B had the lowest feasibility.
• The accuracy of prognostic scores was !69% in all groups and the difference was within 13%, while c-statistics were significantly lower with the PPI than PaP and D-PaP scores.
• All prognostic tools can predict short ( 13 days  • Changes were not observed in blood pressure, and intercept values remained clinically unremarkable • All respiratory variables measured changed significantly over the final 2 weeks of life with a rise in respiratory rate and reduction in oxygen saturation.
• Serum urea and creatinine showed a clinically and statistically significant increase.
• Serum sodium showed a statistically significant but clinically insignificant increase, while serum potassium showed no significant change over time.
• Evidence of leucocytosis and reduced lymphocyte counts, but there was no clinically significant change in the last 2 weeks of life. Recorded values were abnormal indicating that these variables are more likely to change slowly over weeks rather than days.
• Serum albumin showed a significant change over time and abnormal values as death approaches, showing prognostic power over both longer and shorter timescales.
• Other long-term variables including haemoglobin showed a significant change over 3 months of data collection, but results were masked by red cell transfusion. represents elevated CRP levels (>10 mg/L) and hypoalbuminaemia (<35 g/dL). The GPS is scored as 0, 1 or 2.
• The WPCBAL score was derived from the WBC, platelet count, BUN, AST, LDH, and CRP levels.
• The WPBAL score was derived from the WBC, platelet count, BUN, AST, and LDH levels.
• The WPCBAL and WPBAL scores showed a higher sensitivity, specificity, negative predictive value, and relative risk for the prognosis within 2 weeks of death compared to those of the GPS considering a less than three weeks prognosis.
• While the WPCBAL and WPBAL scores showed an equivalent positive predictive value to the GPS, both positive and negative likelihood ratio were superior to those of GPS.  -12(p70), TNF-α, IFN-γ, and MIP-1α values were below the lower limit of detection in the serum in more than 37 of the 43 analyzed patients, and these data were excluded from the analyses.
• Medium correlations were found between: Appetite and CRP, IL-6 and IL-1β Fatigue and IL-1ra Cognitive function and TGF-β1 Dyspnoea and IL-1β.
(Continued) • CAR, the CRP/albumin ratio is an inflammatory prognostic model that has been shown to predict survival in various malignancies.
• The mGPS combines serum albumin and CRP into a risk stratification score, and has been shown to predict survival in various malignancies. The mGPS divides patients into 3 subgroups: 0, 1 and 2.
• Neither of these predictive models has been used in palliative care settings.
• A significant survival difference was found between mGPS groups at 7(p = 0.01) and 180 (p = 0.02) but not 30 days.
• A lower CRP threshold did not improve prognostication.
Only A selected number of articles were excluded as patient populations had a median survival >90 days. An additional three articles were excluded due to lack of survival data.

Study characteristics
The characteristics of the included studies are summarised in Table 2. A formal meta-analysis was not conducted because of the heterogeneity of the published studies. A number of studies utilised univariate rather than multivariate analysis. Both types of studies were included to ensure a comprehensive summary of the literature. The use of the Evidence Based Medicine modified GRADE system has highlighted the limitations of univariate analysis for these types of studies [17]. Table 3 subdivides prognostic biomarkers by grades of evidence. Only five articles investigated changes in concentrations of identified biomarkers in the last days to weeks of life ( Table 3).
The methodological quality of the included studies ranged between 26-36 using the Hawker et al. appraisal tool [16]. Using the seven-point checklist of quality criteria, the selected studies ranged between 3-6 [8]. Spearman's rank correlation coefficient demonstrated a statistically significant rank correlation between assessors (rs 1⁄4 0.955 p = 0.000 and rs 1⁄4 0.921 p = 0.000, respectively). Frequently, sample size calculations were not conducted, and some studies lacked information about the study setting and patient characteristics. Further, the majority (97%) of studies had convenience rather than random sampling, a reliable method for identification of the date of death was often poor, and a large number of potential prognostic factors were analysed despite small sample sizes, which meant that the ratio between the number of deaths and the number of potential prognostic factors frequently scored less than ten.

Discussion
This is the first systematically structured review to evaluate prognostic biomarkers in a heterogeneous group of patients with advanced cancer. Seven prognostic biological factors had Grade A evidence: lymphocyte count, white blood cell (WBC) count, serum albumin, sodium, C-reactive protein (CRP), urea and alkaline phosphatase (ALP) ( Table 3). Few studies have specifically investigated changes in biomarkers in the last days to weeks of life. In the last two weeks of life a number of biomarkers were elevated in the blood including: WBC count, platelet count, serum CRP, urea, urate, alanine transaminase (ALT), lactate dehydrogenase (LDH), sodium and plasma interleukin-6 (IL-6). However limitations exist as only five studies specifically investigated serial measurements of candidate biomarkers in the last weeks of life.
A number of common themes emerged: systemic inflammation, haematological changes, CACS, hepatic dysfunction, renal dysfunction, and electrolyte changes. A number of biomarkers, such as serum albumin could be explained by multiple themes.

Systemic inflammation & haematological changes
There is consistent evidence that the presence of a systemic inflammatory response is associated with reduced survival in patients with a variety of solid organ tumors [48][49][50]. Elevated serum CRP levels are associated with poor prognosis independent of tumour stage [37,51]. IL-6, interleukin-1 (IL-1) and tumour necrosis factor alpha (TNFα) induce the hepatic synthesis of CRP. Further, elevated CRP and IL-6 are associated with CACS in advanced cancer [51]. Measurement of serum CRP is readily available and is an ideal biomarker of systemic inflammation. There is Grade A evidence that elevated serum CRP is an independent prognostic factor in advanced cancer [18,22,29,34,37,42]. Amano et al. demonstrated a clear dose-related  effect between elevated serum CRP and prognosis, and patients could be divided into four groups based on CRP concentrations [37] ( Table 2). Only two studies investigated the role of inflammatory cytokines at the end of life [25,46], this may be because most cytokines have a short plasma half-life and assays are expensive in relation to serum CRP. Iwase et al. measured changes in plasma levels of various cytokines in 28 terminally ill cancer patients with cachexia [25]. Only IL-6 was detected in all patients at a concentration greater than 10pg/mL [25]. The concentration of IL-6 was seen to gradually rise during the early stages of cachexia followed by a sharp rise in the week prior to death [25]. It was hypothesised that circulating macrophages and T lymphocytes produce IL-6 in response to tumour burden, or plasma IL-6 is produced by the tumour mass itself [25,52]. IL-6 is a pro-inflammatory cytokine that regulates immune reactions to tissue damage. There is a growing consensus that plasma IL-6 is a prognostic factor in solid organ malignancies [53,54], however changes in plasma IL-6 levels are affected by carcinoma type [55].
Paulsen et al. identified that a number of inflammatory cytokines were significantly elevated in the plasma in the last months of life [46] (Table 2). Interestingly, soluble tumour necrosis factor receptor-1 (sTNF-r1), IL-6, CRP and erythrocyte sedimentation rate (ESR) were highly correlated with quality of life, and interleukin-1β (IL-1β) was moderately correlated with breathlessness [46]. Further research is required to confirm these preliminary findings and assess their prognostic significance.
There is Grade A evidence that WBC count is an independent predictor of survival in advanced cancer [31,34,35]. In a UK multicentre cohort study, Gwilliam et al. demonstrated WBC count and platelet count independently predicted survival at two months and two weeks in a heterogeneous population of advanced cancer patients across a variety of palliative care settings [34].
There is Grade A evidence that lymphopenia is an independent predictor of survival in advanced cancer [23,31,33,34,41,43]. Gwilliam et al. confirmed that lymphocyte count was a significant predictor of two-month but not two-week survival [34]. The mechanisms driving lymphopenia in advanced cancer are unknown. Immunodeficiency in advanced cancer has been well documented and there is a significant trend of decreasing functional T cell populations (including CD4, CD8, CD4:CD8 ratio and naïve T cells) with cancer progression: this is associated with increased morbidity and mortality [56]. WBC differential rather than WBC counts may be a more useful prognostic marker for future studies.
The prognostic role of thrombocytopaenia (Grade B) [29,34] and haemoglobin count (Grade C) [33] has been indicated but further research is needed to confirm these findings.

Cancer anorexia-cachexia syndrome
CACS is a significant prognostic factor in advanced cancer [8] and the biological mechanisms have been extensively studied [57,58]. There is Grade A evidence for the independent prognostic role of hypoalbuminaemia in advanced cancer [19,27,33,34,41]. It is hypothesised that hypoalbuminaemia is caused by a combination of hepatic dysfunction and CACS. Interestingly, Gwilliam et al. demonstrated hypoalbuminaemia was a predictor of two-month but not twoweek survival, suggesting serum albumin is a predictor of dying over a longer timeframe [34]. This is contrary to the findings by Taylor et al. who demonstrated a significant change over time and as death approached [41]. These findings highlight the need for further prospective studies that analyse serial measurements of serum albumin at the end of life.
One article demonstrated low serum prealbumin as a significant predictor of survival in terminally ill cancer patients by multivariate analysis (Grade B) [24]. Albumin and prealbumin levels reflect the visceral protein pool [24]. Given that prealbumin has a shorter half-life than albumin; prealbumin may be a more sensitive marker of nutritional status [24].

Hepatic dysfunction
A number of biomarkers of liver dysfunction have been indicated as prognostic in advanced cancer including: serum vitamin B12, albumin, bilirubin, aspartate aminotransferase (AST), ALT, ALP, gamma-glutamyl transferase (GGT), international normalised ratio (INR), LDH and cholesterol. The aetiology of hepatic dysfunction at the end of life is unclear. Certainly, studies have demonstrated mixed results on the prognostic significance of liver metastasis in advanced cancer [27,29,33,34,43].
There is Grade B evidence for the independent prognostic significance of elevated serum vitamin B12 in advanced cancer [18]. Geissbühler et al. demonstrated an inverse relationship between survival and serum vitamin B12 levels [18]. There was also a strong correlation between the presence of metastasis or hepatic dysfunction and an elevated serum vitamin B12 [18]. Importantly, there was no difference in serum vitamin B12 levels in patients with haematological malignancies (a potential confounding factor) compared to other cancers [18]. To our knowledge no additional studies have been published that investigate vitamin B12 as a prognostic factor.
Studies assessing serum bilirubin and other liver function tests have found mixed results. There is Grade B evidence that hyperbilirubinaemia is an independent prognostic factor [26,29,33,35]. Laboratory parameters ALP (Grade A), AST (Grade B) and ALT (Grade B) proved to be prognostically significant in at least one multivariate analysis [24,34,35]. Gwilliam et al. also found ALT predictive of two-week, but not two-month survival [34]. There are conflicting results whether GGT is prognostic (Grade C) [33,35]. These contrasting results may be explained by the geographical differences in patient groups, the type of patients included and differences in the methods of biochemical analysis of liver function tests. In contrast to serum AST, elevations in serum ALT are rarely seen in conditions other than liver parenchymal damage and may be a more sensitive marker for liver disease [24].
Prolonged INR was an independent prognostic factor by multivariate analysis in two prospective studies of terminally ill cancer patients (Grade B) [26,29]. To our knowledge, no studies have been published that measure serial changes in INR and liver function tests at the end of life.
Serum LDH is an independent prognostic factor in advanced cancer [26,29,33] (Grade B). Suh et al. demonstrated a significant increase in LDH concentrations in the last two weeks of life [29]. This retrospective study was limited however by sample size and was restricted to hospitalised patients [29]. Prospective multicentre studies across a variety of palliative care settings are required to confirm these preliminary findings. It has been extensively published that elevated serum LDH in cancer reflects tumour burden and aggressiveness [29,59]. Serum concentrations can be further elevated due to hepatic necrosis caused by hepatic dysfunction and/or metastasis in advanced cancer [29,60].
There is Grade B evidence for the prognostic role of hypocholesterolaemia in advanced cancer [26,29,33]. A number of risk factors for hypocholesterolaemia have been identified including: cancer, hepatic dysfunction, CACS, haematological disease and elderly populations [61].

Renal dysfunction
A number of potential prognostic biomarkers of renal dysfunction have been indicated in advanced cancer. Maltoni et al. demonstrated insufficient evidence for the prognostic role of proteinuria in advanced cancer [8]. Proteinuria is prognostic in a number of solid organ tumours [62,63] however, to our knowledge, there have been no recent published studies that investigate proteinuria as a predictor of survival in advanced cancer.
Elevated serum urea was demonstrated as a significant predictor of survival in three studies by multivariate analysis (Grade A) [24,34,41]. Gwilliam et al. identified serum urea as an independent predictor of two-week and two-month survival in advanced cancer [34]. Taylor et al. also found that serum urea and creatinine showed a statistically and clinically significant increase in the last two weeks of life [41]. There is Grade C evidence that serum creatinine is prognostic in advanced cancer [26,33].
There is Grade B evidence that serum urate is a significant prognostic factor in advanced cancer [26,29,33] and serum urate levels were significantly increased in the last two weeks of life [26]. A number of potential explanations for raised serum urate concentrations have been hypothesised including: renal dysfunction and cellular injury caused by hypoxia and/or inflammation [26].
The mechanisms for renal dysfunction at the end of life are unclear. Oliguria was identified as a significant prognostic factor in two univariate analyses [26,29]. It is hypothesised that glomerular filtration rate is reduced at the end of life. Although hypotension is a prognostic factor, there are conflicting results whether blood pressure falls during the dying phase [41,64].

Electrolyte changes
There is Grade A evidence that serum sodium is a significant predictor of survival in advanced cancer [30,35]. These studies involved hospitalised patients; and acute illness may be a confounding factor. One multivariate analysis identified hyponatraemia as an independent predictor of survival in advanced cancer [42]. The most common aetiology of hyponatraemia in cancer is syndrome of inappropriate secretion of antidiuretic hormone (SIADH) [42,65]. Conversely, hypernatraemia at the end life is commonly caused by dehydration [30] and is associated with shorter overall survival in hospitalised patients receiving palliative care [30]. Although, Taylor et al. demonstrated a statistically significant elevation in serum sodium in the last two weeks of life, results were not clinically significant [41]. Serum sodium levels rather than serum sodium may be a more useful prognostic factor for future studies.
The prognostic significance of hypercalcaemia has been described in a number of solid organ and haematological malignancies including: lung [66], prostate [67], renal [68], head and neck [69,70] and the aerodigestive tract [71]. There is conflicting evidence whether hypercalcaemia predicts survival in advanced cancer (Grade C

Non-invasive research methodologies
Coyle et al. recently presented preliminary results of a statistically significant increase in the number of volatile organic compounds (VOCs) in the urine in the last weeks of life using Gas Chromatography Mass Spectrometry (GC-MS) [44]. Interestingly, the steepest rise in significant VOCs was seen in the last week of life [44].

Prognostic models
Based on the outcome of these studies a number of prognostic models have been developed to assist clinicians. Simmons et al. recently reviewed the role of prognostic models in advanced cancer [12]. They concluded that 'various prognostic tools have been validated, but vary in their complexity, subjectivity and therefore clinical utility' [12].
What makes this study unique?
The Neuberger review made a number of recommendations to improve end of life care, including research into the biology of dying [1]. This is the first study to specifically investigate biomarkers in advanced cancer patients in the last months of life and attempts to extrapolate which biological processes are affected. A number of common themes emerged including: systemic inflammation, organ dysfunction and CACS.
What is the significance of the findings of this analysis?
This review demonstrates that there are many biological prognostic factors that have an association with the dying process. Although this review is unable to provide evidence of causation it is important for healthcare professionals to be aware of these prognostic factors and their incorporation into existing prognostic models [12], which may be useful in clinical practice. However, further research is needed to understand the application of biomarkers in prognostication at the end of life.
Identification of biomarkers of dying is an important area for future research that will lead to both improved clinical tools for managing patients, and shed light on fundamental processes such as the answer to the question: why do patients die from cancer?
The "terminal cancer syndrome" theory is characterised by common terminal symptoms including dry mouth, dyspnoea, malnutrition and susceptibility to infection, which are shared amongst heterogeneous groups patients with advanced cancer [7]. In the last three days of life, Bruera et al. demonstrated that blood pressure and oxygen saturations decrease significantly in cancer patients [64]. Heart rate variability [72] and autonomic dysfunction [73] have also been described. Thus, changes in vital signs and biomarkers suggest end organ dysfunction.
In non-cancer animal models, McDonald et al. demonstrated a rapid loss of body weight and disintegration of the circadian rhythmicity in deep body temperature several days before death in senescent but not presenescent rats [74]. Tankersley et al. also demonstrated a predictable sequence of pathophysiological events associated with dying including a fall in daily mean heart rate and a loss of circadian pattern in deep body temperature 3-4 weeks before death [75]. Further, increased lung permeability, reduced lung volume and compliance were seen during a period of terminal senescence [76].
At post-mortem, Kadhim et al. demonstrated the over-expression of interleukin-2 (IL-2) in situ in brainstem neuronal centres implicated in autonomic control of vital homeostatic functions in adults and children who died from severe illness [77,78]. It was hypothesised that biological stressors trigger over-expression of IL-2 in brainstem neuronal centres, inducing a neurochemical cascade that results in disturbed homeostatic control of cardiorespiratory responses, and eventual death; however, populations were confined to non-cancer diagnoses [78]. Further, Perry et al. hypothesised that a neurochemicals including glutamate decarboxylase, tissue pH and tryptophan could reflect novel biomarkers of agonal status [79] and hence the dying phase.
Prognostic factors and post-mortem therefore studies suggest a common biological process to dying. Measurable parameters in the blood suggest a systemic inflammatory response including: elevated CRP, IL-6 and other pro-inflammatory cytokines, hypoalbuminaemia, leucocytosis and neutrophilia. Interestingly, post-mortem studies demonstrated evidence of silent pneumonia in advanced cancer [80,81]. Although it is not possible to assume causation, these findings are of clinical and research interest. For example, the presence of silent pneumonia, may to some extent explain why 24% of patients with terminal cancer have breathlessness despite a lack of risk factors [7]. Further, Kontoyiannis et al. demonstrated pulmonary candidiasis in 21% of patients with evidence of pneumonia at post-mortem and 42% of these patients had disseminated candidiasis [82]. What is the role of immunodeficiency towards the end of life? Is there an overarching aetiology for common terminal symptoms? Does over-expression of IL-2 in brainstem neuronal centres together with silent pneumonia, contribute to breathing changes seen during the dying phase? When is the "point of no return" when the administration of antibiotics is futile? The current lack of research in these areas, together with our desire to improve patient care, highlights the importance for further research into the biology of dying. Biomarkers are important adjuncts in prognostication and the recognition of dying, but equally increase our understanding of the dying process.
Key areas for future research include: serial blood measurements of candidate biomarkers including inflammatory cytokines during the dying phase, intracerebral measurement of cytokines; and correlation with physiological parameters observed during the dying phase. Interestingly, Coyle et al. recently presented a feasibility study for taking serial urine samples from hospice inpatients towards the end of life [83] and demonstrated elevated levels of a number of volatile VOCs in the urine during the dying phase [44]. Non-invasive research methodologies are an important area for future research.

Limitations
There are several limitations with this review. This review only included published studies from year 2000, in order to conduct an in-depth analysis of current studies since the Maltoni et al. paper. The authors recognise that a systematic review of randomised controlled trials is the gold standard in research synthesis, however within the constraints of current available evidence, the PRISMA standards for reporting evidence were applied to ensure a systematic structure to the search, selection and review of the literature [13]. However, reviewers were not blinded to the authors, institutions, or journals of publication, which could have introduced selection bias.
This review excluded non-cancer conditions and was limited to patient populations with a median survival of 90 days, which meant that a selected number of articles were excluded. This was important given that "advanced cancer" is poorly defined in the literature; exclusion of these articles did not change the overall findings of this review.
Many of the studies included in Table 2 were heterogeneous in nature, small and underpowered, and the quality of studies varied considerably. Few studies have specifically investigated changes in the last weeks of life. We screened some studies that may have contained information specific to the objectives of this review, but data was not presented in a way that was specific to advanced cancer. There are considerable ethical implications of conducting research in patients in the last few weeks of life, in particular when this involves invasive procedures. We recognise that this may have limited the number of studies for inclusion in this review. Certainly, a number of included studies were limited by sample size and lack of protocols for collection of blood samples.
It was not possible to provide a synthesis of the evidence due to the heterogeneity of the sample. Despite this, the findings of this review are of particular research interest. This review makes an important contribution to the evidence base on the biology of dying and highlights potential areas for research funding and analysis. Attention has been given to the ethical and methodological implications of research into the biology of dying and future strategies are described.

Conclusion
The biology of dying is an important area for future research interest. The evidence to date is largely focused on signs, symptoms and prognostic factors. Despite appearances, the extent to which cancer patients follow a common terminal trajectory is uncertain and high quality research should be conducted to explore this concept further. This review identifies a number common themes shared amongst advanced cancer patients and highlights candidate biomarkers which may be indicative of a common biological process to dying. Attention should be placed on understanding the physiological process of dying and identification of candidate biomarkers of imminent death. This will increase clinicians' confidence in identifying the dying process, inform decision making surrounding end of life care, and ensure the best possible care for patients and their families.