554967

research-article2014

PMJ0010.1177/0269216314554967Palliative MedicineTaylor et al.

Original Article

Change in physiological variables in

the last 2 weeks of life: An observational
study of hospital in-patients with cancer

Palliative Medicine
2015, Vol. 29(2) 120127
The Author(s) 2014
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0269216314554967
pmj.sagepub.com

Paul Taylor1, Simon Crouch2, Debra A Howell2, Dawn W

Dowding3 and Miriam J Johnson4

Abstract
Background: Recognising dying remains a difficult clinical skill which has gained increasing importance in the United Kingdom
since the Neuberger review. Clinical and research methods exist to aid recognition of dying but do not exhibit the level of accuracy
required for such an important decision.
Aim: To explore change in key clinical parameters as cancer patients near the end of life.
Design: This is a retrospective cohort study of terminally ill patients. Data were collected from hospital case-notes. Case-note data
were analysed using multilevel modelling to explore absolute values and rates of change of given variables.
Setting/participants: Hospital in-patients who died from solid-tumour malignancies within a 3-month period in 2009 formed the
cohort. The setting was an acute hospital trust in the North of England.
Results: A total of 15,337 data points from the case-notes of 102 patients were analysed. There was a clinically and statistically
significant deterioration in respiratory function and renal function over the last 2weeks of life. Heart rate and serum sodium also
changed but did not vary greatly from normal limits. White cell parameters, haemoglobin and albumin showed evidence for change
over longer periods.
Conclusion: Results demonstrate statistically and clinically significant change in routinely measured respiratory and renal function
variables during the final 2weeks of life in people dying with cancer. Although useful in acute early warning scores, in a terminally ill
patient, relative haemodynamic stability should not be interpreted as reassuring. Further work is needed to understand how these
findings apply to the individual or inform other prognostic work.

Keywords
Palliative, terminal, cancer, dying, prognosis, observation

What is already known about this topic?

Diagnosing dying is a difficult clinical skill.
Diagnosing dying is of vital importance, both to the public and to the professionals.
Research into diagnosing dying faces ethical and methodological challenges.
What this paper adds?
Evidence of deranged respiratory and renal function as death approaches.
Evidence of relative stability of haemodynamic parameters as death approaches.
Methodological and statistical techniques for assessing measurable physiological change at the end of life.

1St

Benedicts Hospice, Sunderland, UK

of York, York, UK
3Columbia University School of Nursing and Visiting Nurse Service of
New York, New York, USA
4Hull and York Medical School, Hull, UK
2University

Corresponding author:
Paul Taylor, St Benedicts Hospice, St Benedicts Way, Ryhope,
Sunderland SR2 0NY, UK.
Email: drpaulmtaylor@gmail.com

121

Taylor et al.
Implications for practice, theory or policy?

Further research should explore change in respiratory function as a possible predictor of death.
Normal or near-normal blood pressure and pulse rate are not reassuring in terminally ill cancer patients nearing the
end of life.
Retrospective methods and multilevel modelling are valid techniques for minimising the research burden on the dying.

Introduction
Most health-care professionals will be involved in the care
of patients who are dying. The standard of care for the
dying has significant public importance,1 is recognised as
a national policy priority in the United Kingdom,2 for medicine as a profession,3 and is seen as a key aspect of palliative care provided by both specialists and non-specialists.4
Optimal care of dying patients requires that clinicians are
able to recognise this stage of an illness, a difficult clinical
skill.3 The diagnosis of dying involves a number of barriers,5,6 and errors receive significant attention in the public
domain.7,8
Traditionally, signs and symptoms of approaching death
have been based largely on experience and consensus9,10
rather than systematic study. An evidence base to guide the
clinician, applicable to the individual patient, would be
valuable but difficult to acquire due to the specific challenges of conducting research in such a vulnerable patient
population at such a time.11
Difficulties in the recognition of dying, and the subsequent action on this recognition, were highlighted in the
UK governments review into the Liverpool Care Pathway,
published last year,12 and addressing concerns around a
structured system for managing dying patients. The report
makes 44 recommendations and highlights areas warranting attention in both clinical and research domains. These
include funding research into the biology of dying, the use
and implementation of prognostic tools and the diagnosis
of dying.
Early warning13,14 and prognostic scoring systems1517
represent two potential approaches to the problem of recognising dying. They are typically developed from crosssectional survey data in relation to survival this
technique allows a strong measure of association between
variable value and time but assumes a common trajectory
among individuals. In addition, there is reason to suspect
that the physiology of the deteriorating terminally ill
patient exhibits significant changes to that of the healthy
or acutely unwell patient.18 A further approach is to assess
rate of change over time (trajectory), and evidence suggests that rate of change of some variables has greater
prognostic predictive power than absolute values.19,20
There is, however, little published data about the usefulness of change in physiological variables, measured in
daily clinical practice, over the last few days of life, in
the diagnosis of dying.

This study aimed to explore objective physiological

change at the end of life in terms of rates and trajectories of
change. The aim of this study was to generate new hypotheses relating to trajectory of change in specific variables as
death approaches. Broad categories of illness (malignant
vs non-malignant) show distinct patterns of deterioration
as death approaches,21 although the extent to which all
patients within each group follow a common pathway is
disputed.22 Therefore, this report focuses on cancer patients
(solid-tumour malignancies). The work was conducted as
part of a larger mixed-methods project, the findings of
which will be reported elsewhere.

Methods
This was a retrospective cohort study. The study took place
at an acute National Health Service (NHS) trust (approximately 1300 beds) in the North of England, which included
a tertiary cancer service. Patients were admitted from a
variety of sources, including self-referral for specific
oncological emergencies, primary care team referral, A&E
and transfers from other specialties. During the study
period, there were 110 acute haematology and oncology
beds spread over five wards. One ward (24 beds) selected
palliative patients where possible, although not exclusively. The trust had links with a local 20-bedded independent hospice. Immediately prior to the period of the
study, a number of changes occurred regionally, including
the appointment of specialist palliative care physicians.
Patients were identified according to cause of death
using hospital coding and cause of death records for all
deaths occurring within the trust in AugustOctober 2009.
Adults (18years) dying as a result of a solid-tumour
malignancy, as identified from death certificate data
(where cancer was cited as 1a, 1b or 1c), were included.
The case-notes of patients whose deaths had been referred
to the coroner or were the subject of active litigation were
unavailable and thus excluded. Patients who died in an
intensive care unit (ICU) or high dependency unit (HDU)
environment were excluded as they potentially represented
a different dying process.
Study variables were identified from the prognostic literature in relation to poor survival in end-stage cancer
patients and are listed in the result tables. Variables were
included in the study if they met the following criteria:

122
Likely to be recorded routinely in medical
case-notes;
Had the potential to be subject to variation in the
final 2weeks of life;
Would be appropriate to be measured in terminally
ill patients.
The time period was the final 2weeks of life, chosen
to include both the last days of life and the preceding
week to provide context. Data relating to the study variables were extracted by one researcher (P.T.) from the
hospital records. Paper case-notes were used for observations, while computer records were used for investigations. In addition, data for haemoglobin and albumin
were extracted for 3months, as it was indicated from the
literature that these variables may also show change over
months or weeks. Each variable value was recorded
against the time (to the nearest hour for observations) or
date (for blood tests) the value was collected. Data were
based on all values that had been recorded for clinical
reasons and hence were included at all time points at
which they had been recorded in the case-notes; in some
cases, including the final hour of life.
Anonymised data were input directly into a bespoke
database, which was initially developed for a previous project by members of the research team.23 This relational
database was specifically designed to hold time and event
information, with each key event at any one time point
having the capacity to hold data on multiple specific subevents. A calendar-style interface allowed the researcher to
input each individuals data values according to time and
date recorded, with variables selected from a drop-down
box. When queried, the database generated parameter values against time recorded for each study ID. A paper linkage sheet was stored securely at the trust and destroyed
1year after completion of the study.
The research was reviewed and approved by several
bodies. Ethical, practical and legal implications were
assessed by a Research Governance Committee of the
sponsoring university, the local Research Ethics Committee
and the Trust Research and Development Department.
Information governance aspects were appraised by the
National Information Governance Board for Health and
Social Care.
An initial visual inspection of the data was performed
using scatter plots for the whole data set and individual
line graphs for each patient to provide a broad overview of
the data at the population level, as well as an impression of
variability between patients. Multilevel models were generated to explore the data in objective detail.
Multilevel modelling is an analysis technique, especially suited to longitudinal data such as these, in which
multiple regression analyses in sub-sets of a population are
combined to produce a regression model for the whole
population.24 The individual analyses, and the subsequent

Palliative Medicine 29(2)

population analysis, are different levels. This study used
a random-slope and random-intercept model, which provides an association between variable value and time prior
to death, together with a measure of variation between
individual trajectories. This technique allows each individual to contribute a varied number of data points, as
found in real-time data recording found in clinical practice, making use of a detailed data set. All such analyses
were performed in Stata Version 11.0.25
In each model, there are two key values: co-efficient
and intercept. These values provide an estimate of a mean
trajectory for the patient cohort, describing the relationship
between the variable value and time prior to death (note
that a value that increases as death approaches will therefore have a negative co-efficient).
The intercept is a model prediction of the value where
time is zero. As this is the time of death, the value has no
true clinical meaning; rather, it acts as a useful end point
for the predicted trajectory, providing an indication of
predicted or expected values for that variable as death
approaches. This value is useful for clinical interpretation
of the model values. For example, if the intercept is
within or near the normal range of clinical values, this
suggests that the variable will have limited clinical significance in application, even where the model has statistical significance.
A final consideration in this approach was the impact of
short, limited or censored data sets. Reasons included clinical measurements being discontinued in the period prior
to death, cases where the patient had been admitted for less
than 2weeks prior to death and the fact that the frequency
of observations varied between patients, usually resulting
from clinical context. To assess for the presence of potential bias arising from these causes, a second (restricted)
data set was analysed for each variable, including only
those patients for whom data were retained on the final day
of life (or the final week of life for haemoglobin and albumin) and for those in whom at least 10 variable values
were recorded. Where this second set could be obtained,
both are provided in the results. This allows for any
observed association to be assessed for persistence in these
restricted circumstances.

Results
Data collected
Over the 3-month period, 189 potentially eligible patient
deaths were identified. These were requested through
existing trust channels for accessing patient notes, and the
process was repeated to maximise the number of notes
obtained. Of 189 records, 140 were available for evaluation. A final data set of 102 case-notes was retained for
analysis. Reasons for exclusions of retrieved case-notes
can be seen in Box A.

123

Taylor et al.
Box A. Reasons for exclusion of case-notes.
Non-cancer cause of death (16)
Cause of death not recorded in case-notes (5)
Insufficient information in notes to extract data (6)
Coroners cases (6)
ICU deaths (2)
Ongoing litigation (1)
Time of death not recorded in case-notes (2)

Table 1. Cancer diagnoses by primary site.

ICD code

Site

Number of cases
(% of total)

National
figures (%)

C15
C18C20
C32
C34
C50
C61
C67
C22
NA

Oesophagus
Colorectal
Larynx
Lung
Breast (female)
Prostate
Bladder
Cholangio/HCC
Others
primary known
Unknown
primary

4 (3.9)
8 (7.8)
4 (3.9)
37 (35.9)
8 (7.8)
5 (4.9)
6 (5.8)
4 (3.9)
17 (16.5)

4.8
10.0
0.5
21.8
7.7
6.8
3.3
Not given
Not given

10 (9.7)

Not given

103

NA

C77C80
Total

ICD: International Classification of Diseases; HCC: hepatocellular

carcinoma; ONS: Office for National Statistics.
National figures represent percentage of cancer deaths attributed to
that primary for England, 20072009 (ONS figures).26 Figures add to
103 as one patient had two primary cancers.

The analysis group consisted of 49 men and 53 women,

age range of 39102years (10th age centile: 52, median
age: 73years, 90th age centile: 85). Cancer diagnoses for
the commonest malignancies are provided in Table 1,
together with comparable national figures. Descriptive
statistics for the collected data are provided in Table 2,
which incorporates all data points collected for each
parameter. A summary of collected data on included variables is provided with the results (Tables 35). Other variables (performance status, weight, urine output, cognitive
impairment, nausea and vomiting) were identified from
the literature and thought to be relevant in daily clinical
practice but were insufficiently and inconsistently documented and were not studied further.

Multilevel models
Tables 35 show the results of multilevel modelling for
each variable. Where restricted data sets are available,
these are noted. Table 3 shows observations. Table 4 shows
blood tests collected for 2weeks prior to death. Table 5
shows blood tests collected for 3months prior to death.
Heart rate significantly increased during the last week
of life, although the model intercept value only reached
99bpm on the day of death. Changes were not observed in
blood pressure, and intercept values remained clinically
unremarkable. All respiratory variables measured changed
significantly with a rise in respiratory rate and reduction in
oxygen saturation.
Of blood results, serum urea and creatinine both showed
a statistically and clinically significant increase as death

Table 2. Descriptive statistics for all data collected.

Parameter

Number of values
(patients)

Lower range

25th centile

50th centile

75th centile

Upper range

HR
SBP
DBP
RR
SpO2
White cells
Lymphocytes
Sodium
Potassium
Urea
Creatinine
Albumin
Hb

2003/97
2097/97
2097/97
1746/96
2090/97
225/83
212/79
254/85
246/83
255/85
253/85
724/98
706/100

45
65
16
3
60
0.5
0.18
108
2.1
1.7
33
<10a
5.7

80
108
61
17
93
8.75
0.57
133
3.8
5.7
69
20
9.5

94
123
70
20
95
12.95
0.87
136
4.5
10.8
111
25
10.8

106
138
78
26
97
18.1
1.22
140
5
16.7
177
30
12.1

180
206
141
50
100
59.5
3.15
166
7.1
61.9
635
46
16.2

HR: heart rate; SBP: systolic blood pressure; DBP: diastolic blood pressure; RR: respiratory rate; SpO2: oxygen saturation measured with finger
probe.
aLaboratory reports albumin below 10 as <10. Only three such values recorded in whole data set and treated as value of 10.

124

Palliative Medicine 29(2)

Table 3. Data collected and multilevel modelling for observations.

Variable

Data set

Intercept (95% CI)

Coefficient (95% CI)

Model p value

HR

SBP

DBP

RR

SpO2

Full
Restricted
Full
Restricted
Full
Restricted
Full
Restricted
Full
Restricted

98.7 (95.9, 101.4)

97.9 (94.1, 101.7)
122.0 (118.2, 125.7)
120.4 (115.3, 125.4)
69.9 (67.6, 72.2)
69.5 (66.2, 72.8)
21.1 (20.2, 21.9)
22.4 (21.1, 23.6)
92.7 (92.1, 93.4)
91.9 (91.0, 92.9)

0.028 (0.041, 0.016)

0.028 (0.047, 0.009)
0.013 (0.001, 0.027)
0.009 (0.002, 0.020)
0.002 (0.006, 0.011)
0.001 (0.012, 0.012)
0.010 (0.014, 0.006)
0.015 (0.021, 0.009)
0.010 (0.007, 0.013)
0.013 (0.007, 0.019)

<0.001
0.004
0.071
0.093
0.60
0.99
<0.001
<0.001
<0.001
<0.001

HR: heart rate; SBP: systolic blood pressure; DBP: diastolic blood pressure; RR: respiratory rate; SpO2: oxygen saturation measured with finger
probe; CI: confidence interval.
Data recorded for 2weeks prior to death. Coefficient is model predicted change in value per hour prior to death. Full data sets include all values.
Restricted data sets limited to those with 10 values in total, with at least one on the last day of life.

Table 4. Data collected and multilevel modelling for blood tests.

Variable

Intercept (95% CI)

Coefficient (95% CI)

Model p value

White cells
Lymphocytes
Sodium
Potassium
Urea
Creatinine

15.7 (13.3, 18.2)

0.87 (0.73, 1.01)
138.3 (136.5, 140.1)
4.3 (4.0, 4.5)
15.3 (12.6, 18.1)
140.3 (116.7, 163.9)

0.218 (0.427, 0.009)

0.013 (0.003, 0.028)
0.277 (0.459, 0.094)
0.009 (0.015, 0.032)
0.61 (0.876, 0.345)
3.493 (5.966, 1.020)

0.041
0.10
0.003
0.46
<0.001
0.006

CI: confidence interval.

Data collected over 2weeks. Coefficient is model predicted change in value per day prior to death. White cells and lymphocyte values in 109 per
litre; serum sodium, potassium and urea values in millimoles per litre; and creatinine values in micromoles per litre.

Table 5. Data collected and multilevel modelling for long-term variables.

Variable

Data set

Intercept (95% CI)

Coefficient (95% CI)

Model p value

Albumin

Hb

Full
Restricted
Full
Restricted

22.5 (21.2, 23.7)

18.8 (16.2, 21.5)
10.8 (10.4, 11.2)
10.0 (9.3, 10.7)

0.145 (0.119, 0.171)

0.173 (0.131, 0.216)
0.016 (0.008, 0.024)
0.017 (0.002, 0.035)

<0.001
<0.001
<0.001
0.075

Hb: haemoglobin; CI: confidence interval.

Data collected over 3months. Coefficient is model predicted change in value per day prior to death. Full data sets include all values. Restricted data
sets limited to those with 10 values in total, with at least one in the last week of life. Haemoglobin values in gram per decilitre and albumin values
in gram per litre.

approached. Serum sodium showed a statistically significant but clinically insignificant increase while potassium
showed no significant change over time. White cell parameters showed a tendency to a leukocytosis with reduced
lymphocytes but without clinically significant change
throughout the study period. Of the long-term variables,
both decrease as death approaches but over a timescale of
weeks rather than days.

Discussion
The purpose of this study was to explore evidence for
physiological change in the final 2weeks of life, which

may be used to inform and improve further research. These

results provide evidence for specific physiological changes
in this time period in people with cancer.
The most striking results are those relating to respiratory function. These show evidence both for deterioration
over the final 2weeks of life (respiratory rate increasing
and SpO2 decreasing) and for abnormal values as death
approaches.
Heart rate also shows an increase as death approaches,
but the intercept value only represents a mild tachycardia.
Therefore, while a change in heart rate exhibits a statistically significant relationship with time to death, clinical
significance may be missed. Systolic and diastolic blood

125

Taylor et al.
Table 6. RCP National Early Warning Score table for physiological parameters.
Parameter

RR
SpO2 (%)
Supplemental oxygen
Temp
SBP
HR
LOC

8
91
35.0
90
40

2
9293
Yes
91100

0
911
9495

35.136.0
101110
4150

1220
96
No
36.138.0
111219
5190
Alert

2
2124

38.139.0
91110

39.0
111130

3
25

220
130
V, P, U

Source: adapted from original.28

RCP: Royal College of Physicians; RR: respiratory rate; SpO2: oxygen saturations measured with finger probe; Temp: temperature; SBP: systolic
blood pressure (mmHg); HR: heart rate (beats/min); LOC: level of consciousness; V, P, U: responds to voice, responds to pain, unresponsive.

pressures show neither strong associations with time to

death, nor abnormal intercept values. This observation has
been a source of conflicting findings in prior research,
with some studies supporting the role of blood pressure as
a predictor,27 while others outline its limits.14
These findings provide an interesting point of comparison with early warning scores. Such scoring systems are
designed to identify acutely unwell patients who may benefit from escalation of care and are at risk of dying without
such escalation.13,28 A recent publication by a Royal
College of Physicians (RCP) working party28 proposed a
standardised National Early Warning Score (NEWS,
Table 6), which can be used as a comparator for this study.
In the NEWS, parameter values generate a weighted score,
which is used to assign a risk of death within a given time
period; higher scores represent higher risk.
Comparison between this study and NEWS shows that
respiratory dysfunction is also associated with poor outcomes in acutely unwell medical patients. Conversely,
haemodynamic dysfunction (as demonstrated by tachycardia and hypotension) which shows little association in this
study, also scores highly in NEWS. These observations
may highlight a distinction between physiological deterioration leading to death in the acutely unwell patient and the
deterioration leading to death in people dying from cancer
and highlights the need to revalidate any such scoring systems in any population to which they will be applied.
The blood results show that markers of renal function
worsen as death approaches; these markers show both
increasing abnormality and consistent change. Serum
sodium increases slightly as death approaches, but the values recorded are not abnormal, again limiting clinical usefulness. Serum potassium shows neither significant change
nor abnormal values.
White cell variables do not appear to change in the final
weeks of life, but the recorded values are abnormal implying they change more slowly and hence may be more applicable to long-term prognostication. Haemoglobin does
show some change over 3
months of data collection,
although inspection of individual trajectories suggests that
this change may have been masked by red cell transfusions.

Finally, serum albumin shows both significant change over

time and abnormal values as death approaches, suggesting
prognostic power over longer and shorter timescales.
Finally, this study has shown how multilevel modelling
and epidemiological methods may be used to explore complex retrospective data in detail. These methods may be of
use in future research where limiting the burden on dying
patients remains a priority.
These findings are of particular importance given the
recent findings of the Neuberger review, which outlined
difficulties in recognising dying, the importance of this
recognition for end of life care and called for further
research into the biology of dying. The findings of this
study therefore contribute vital information to this evidence base, highlighting both promising parameters for
further study and a research approach that makes detailed
use of retrospective cohort data. Two key areas which
future studies may explore include a detailed analysis of
deteriorating respiratory function in the final days/weeks
of life and a similar analysis of deteriorating renal
function.
While we have outlined a number of areas of interest
for future research, caution must be taken when interpreting these findings directly. The variables showing most
significant change over the final days of life are vulnerable
to confounding through a number of clinical, demographic
and treatment-related factors. For example, probe-measured SpO2 may be affected by poor peripheral circulation,
inspired oxygen and pre-existing respiratory disease.
Similar observations may be made of renal dysfunction
(diuretic use, patient age, pre-existing renal disease, concomitant dehydration and reduced oral intake as death
approaches). The data here are therefore presented as sufficient to warrant a more detailed prospective study, rather
than sufficient for a change in clinical practice alone.

Limitations and strengths

As can be seen from Table 1, the distribution of cancer
primary sites at the study centre broadly mirrors that of
national mortality figures29 but with respiratory tract

126
malignancies over-represented. This is reflected in local
figures,30 in which mortality from such malignancies is 1.8
times the national average. This observation provides a
degree of caution against direct clinical application of the
study findings to different populations, but the results
remain useful to researchers aiming to explore physiological change as death approaches and provide a useful basis
for such studies. In keeping with the stated aims, by using
multilevel modelling, all available, contemporaneously
documented, clinically relevant data points have been used
in the analysis to successfully generate a hypothesis for
future work.
In retrospect, given the variables showing significant
change in the study, it would have been valuable to have
obtained data on intravenous (IV) fluid use, clinical
dehydration, oxygen use, presence of respiratory illness
and similar parameters related to the outcomes of interest. Nevertheless, these should be readily obtained in any
future study electing to explore these variables in further
detail.
A final limitation is the reliance of death certificate data
to identify participants. The risk of inaccuracy of death
certificate data is well documented;26 nevertheless, this
occurs less frequently when the cause of death is attributed
to a malignant disease.26

Conclusion
These data demonstrate statistically and clinically significant change in routinely measured respiratory and renal
function variables during the final 2weeks of life in people
with cancer. The importance of change in variable rather
than an absolute value at a single time point is highlighted
as a crucial component of assessment.
Blood pressure, serum sodium and serum potassium are
unlikely to be useful in this area, while change in heart rate
may have a limited role in this situation. This appears at
odds with their accepted role in early warning scores but
may indicate that dying in terminal illness is not always
the same process as the physiological deterioration in
acute illness. White cell count and, independently, lymphocyte count may warrant further study over longer
periods.
Although further work is needed in order to understand
how these findings may inform other early warning or
prognostic work, and application to the individual, these
variables are likely to be useful for prospective study of
change over the last few days of life.
Declaration of conflicting interests
The authors declare that there is no conflict of interest.

Funding
This work was part of the lead authors PhD fellowship, which was
funded and supervised through Hull and York Medical School

Palliative Medicine 29(2)

(HYMS), a collaboration between the Universities of York and
Hull. This research received no specific grant from any funding
agency in the public, commercial or not-for-profit sectors.

References

1. http://dyingmatters.org/overview/about-us (accessed 16
January 2014).
2. End of Life Care Team, Department of Health. End of life
care strategy: 4th annual report. London: End of Life Care
Team, Department of Health, 2012.
3. Sleeman KE and Collis E. Caring for a dying patient in hospital. BMJ 2012; 346: f2174.
4. Ellershaw J and Wilkinson S. Care of the dying: a pathway to excellence. 2nd ed. Oxford: Oxford University Press,
2011.
5. Coackley A and Ellershaw J. The terminal phase. Medicine
2007; 36: 105108.
6. Taylor PM and Johnson MJ. Recognizing dying in terminal
illness. Br J Hosp Med 2011; 72: 446450.
7. Devlin K. Sentenced to death on the NHS. The Daily
Telegraph, 2 September 2009, http://www.telegraph.co.uk/
health/healthnews/6127514/Sentenced-to-death-on-theNHS.html
8. Donnelley L. Leading doctors fears about Liverpool Care
Pathway. The Daily Telegraph, 3 March 2013, http://www.
telegraph.co.uk/health/healthnews/9904650/Leadingdoctors-fears-about-Liverpool-Care-Pathway.html
9. Higgs R. The diagnosis of dying. J R Coll Physicians Lond
1999; 33: 110112.
10. Ellershaw J, Sutcliffe JM and Saunders CM. Dehydration
and the dying patient. J Pain Symptom Manage 1995; 10:
192197.
11. Stone PC, Gwilliam B, Keeley V, et al. Factors affecting
recruitment to an observational multicentre palliative care
study. BMJ Support Palliat Care 2013; 3: 318323.
12. Neuberger J, Guthrie C, Aaronovitch D, et al. More care,
less pathway: a review of the Liverpool Care Pathway.
Department of Health, Crown Copyright, 2013.
13. Prytherch DR, Smith GB, Schmidt PE, et al. ViEWS
towards a national early warning score for detecting adult
inpatient deterioration. Resuscitation 2010; 81: 932937.
14. Cooksley T, Kitlowski E and Haji-Michael P. Effectiveness
of Modified Early Warning Score in predicting outcomes in
oncology patients. QJM 2012; 105: 10831088.
15. Morita T, Tsunoda J, Inoue S, et al. The Palliative

Prognostic Index: a scoring system for survival prediction of terminally ill cancer patients. Support Care Cancer
1999; 7: 128133.
16. Maltoni M, Caraceni A, Brunelli C, et al. Prognostic factors in advanced cancer patients: evidence-based clinical
recommendations a study by the Steering Committee of
the European Association for Palliative Care. J Clin Oncol
2005; 23: 62406248.
17. Gwilliam B, Keeley V, Todd C, et al. Development of
Prognosis in Palliative care Study (PiPS) predictor models
to improve prognostication in advanced cancer: prospective
cohort study. BMJ 2011; 343: d4920.
18. Lasheen W and Walsh D. The cancer anorexia-cachexia
syndrome: myth or reality? Support Care Cancer 2010; 18:
265272.

Taylor et al.
19. Chan EY, Wu HY and Chan YH. Revisiting the Palliative
Performance Scale: change in scores during disease trajectory predicts survival. Palliat Med 2013; 27: 367374.
20. Bennett M and Ryall N. Using the modified Barthel index
to estimate survival in cancer patients in hospice: observational study. BMJ 2000; 321: 13811382.
21. Murray SA, Kendall M, Boyd K, et al. Illness trajectories
and palliative care. BMJ 2005; 330: 10071011.
22. Gott M, Barnes S, Parker C, et al. Dying trajectories in heart
failure. Palliat Med 2007; 21: 9599.
23. Howell DA, Wang HI, Roman E, et al. Variations in specialist palliative care referrals: findings from a populationbased patient cohort of acute myeloid leukaemia, diffuse
large B-cell lymphoma and myeloma. BMJ Support Palliat
Care. Epub ahead of print 19 February 2014. DOI:10.1136/
bmjspcare-2013-000578.
24. Rabe-Hesketh S and Skrondal A. Multilevel and longitudinal
modeling using Stata. College Station, TX: Stata Press, 2008.

127
25. StataCorp. Stata statistical software: release 11. College
Station, TX: StataCorp LP, 2011.
26. Maclaughlan J and Wells C. Death certification reform: a
case study on the potential impact on mortality statistics,
England and Wales. Office for National Statistics, 2012,
http://www.ons.gov.uk/ons/dcp171778_288141.pdf
27. Ramchandran KJ, Shega JW, Von Roenn J, et al. A predictive model to identify hospitalized cancer patients at risk for
30-day mortality based on admission criteria via the electronic medical record. Cancer 2013; 119: 20742080.
28. https://www.rcplondon.ac.uk/sites/default/files/documents/
national-early-warning-score-standardising-assessmentacute-illness-severity-nhs.pdf (accessed 21 July 2014).
29. http://www.ons.gov.uk/ons/publications/re-referencetables.html?edition=tcm%3A77-251462 (accessed 31 July
2014).
30. http://www.cancerresearchuk.org/cancer-info/cancerstats/
local-cancer-statistics/ (accessed 31 July 2014).

Copyright of Palliative Medicine is the property of Sage Publications, Ltd. and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for
individual use.