A longitudinal study of the role of patient-reported outcomes on survival prediction of palliative cancer inpatients in Taiwan.

Abstract

This study explores the significance of patient-reported outcomes for predicting length of survival of palliative cancer patients.
Patients were recruited upon admission to the inpatient palliative care unit. Weekly assessment of 180 terminal cancer patients was carried out throughout their survival time using the Medical Outcome Study 36-Item Short-Form Health Survey, the Taiwanese version of the M.D. Anderson Symptom Inventory (MDASI-T), the Karnofsky Performance Status (KPS), the Brief Pain Inventory, and the Brief Fatigue Inventory. Generalized estimating equations (GEE) were utilized to analyze whether the patient-reported outcomes predicted survival time.
Of all patients, 64 had one assessment, 51 had two, 25 had three, and 40 had four or more assessments, up to a maximum of eight. The univariate analysis showed that gender (P < 0.01), KPS (P < 0.01), the physical component summary score (P = 0.02), the MDASI-T total score (P < 0.01), composite fatigue severity (P < 0.01), and composite pain severity (P < 0.01) were significantly associated with length of survival. The multivariate analysis showed that gender (P < 0.01), KPS (P < 0.01), and the MDASI-T total score (P = 0.01) were significant predictors of survival time.
This is the first study to explore the significance of patient-related outcomes for predicting length of survival of palliative cancer patients using the GEE method. This study confirms that overall symptom severity is a significant factor in assessing the length of survival of palliative cancer patients.

Full text

Page 1

ORIGINAL ARTICLE
A longitudinal study of the role of patient-reported outcomes
on survival prediction of palliative cancer inpatients
in Taiwan
Jing-An Chang & Chia-Chin Lin
Received: 27 September 2008 /Accepted: 12 January 2009 /Published online: 12 February 2009
# Springer-Verlag 2009
Abstract
Goals of work This study explores the significance of
patient-reported outcomes for predicting length of survival
of palliative cancer patients.
Patients and methods Patients were recruited upon admis-
sion to the inpatient palliative care unit. Weekly assessment
of 180 terminal cancer patients was carried out throughout
their survival time using the Medical Outcome Study 36-
Item Short-Form Health Survey, the Taiwanese version of
the M.D. Anderson Symptom Inventory (MDASI-T), the
Karnofsky Performance Status (KPS), the Brief Pain
Inventory, and the Brief Fatigue Inventory. Generalized
estimating equations (GEE) were utilized to analyze whether
the patient-reported outcomes predicted survival time.
Main results Of all patients, 64 had one assessment, 51 had
two, 25 had three, and 40 had four or more assessments, up
to a maximum of eight. The univariate analysis showed that
gender (P<0.01), KPS (P<0.01), the physical component
summary score (P=0.02), the MDASI-T total score (P<
0.01), composite fatigue severity (P<0.01), and composite
pain severity (P<0.01) were significantly associated with
length of survival. The multivariate analysis showed that
gender (P<0.01), KPS (P<0.01), and the MDASI-T total
score (P=0.01) were significant predictors of survival time.
Conclusions This is the first study to explore the signifi-
cance of patient-related outcomes for predicting length of
survival of palliative cancer patients using the GEE method.
This study confirms that overall symptom severity is a
significant factor in assessing the length of survival of
palliative cancer patients.
Keywords Prognosticfactor.Qualityoflife.Symptom
severity.Survival.Palliativepatients
Introduction
Patient-reported experiences are highly valued in palliative
care. Patient-reported outcomes (PROs) are defined as
“measurement of any aspect of a patient’s health status
that comes directly from the patient (i.e., without the
interpretation of the patient’s responses by a physician or
anyone else,” including disease symptoms, patient func-
tioning, and quality of life (QoL) [61]. In a recent review of
the impact of PROs on patient survival after a cancer
diagnosis, it was concluded that the most commonly
assessed PRO was QoL [19]. Improvement of a patient’s
QoL is widely agreed to be the ultimate goal of palliative
care, making QoL an important end point in palliative care
research; however, several recent studies have suggested
that QoL may also have an independent prognostic value
for assessing the length of survival of terminal patients [44,
56, 57, 67]. The predictive value of QoL for cancer survival
has also been noted in descriptive reviews [23, 48, 58]. The
World Health Organization’s project to develop a quality-
of-life measure defines quality of life as “individuals’
perception of their position in life in the context of the
culture and value systems in which they live and in relation
to their goals, expectations, standards, and concerns” [59].
Clinical predictions of the length of survival of patients in
the end stages of cancer have therapeutic and psychological
Support Care Cancer (2009) 17:1285–1294
DOI 10.1007/s00520-009-0583-9
J.-A. Chang
Taipei Veterans General Hospital, Taipei Medical University,
Taipei, Taiwan
C.-C. Lin (*)
School of Nursing, Taipei Medical University,
250 Wu-Hsing Street,
Taipei 110, Taiwan
e-mail: clin@tmu.edu.tw

Page 2

implications for the patient and especially for the family
[36]. However, accurate survival time estimates for patients
with terminal cancer continue to be a challenge for
palliative care clinicians.
Some studies have explored the prognostic indicators of
length of survival of patients with advanced or terminal
cancer [8–10, 19, 46, 49]. To date, several factors have
been thought to influence the duration of the terminal phase
for cancer patients, but conclusive findings have remained
elusive. Identification of such factors could improve
prognostic accuracy and facilitate health care providers’
decision making in selecting the most appropriate care plan
[14]. Prior studies have shown the following to have
predictive value for assessing the length of survival of
terminal cancer patients: performance status [8–10, 14, 40],
pain [40], fatigue [8, 10, 39], clinical symptoms [8, 40], and
quality of life [12, 36, 64]. Some studies of cancer patients,
including patients not necessarily in a terminal phase, have
identified quality of life as having an important prognostic
factor [6, 16, 20]. In summary, of the patient-reported
outcomes previously identified in the literature, perfor-
mance status, clinical symptoms, and quality of life are of
great value in predicting length of survival of terminal
cancer patients. The ability to accurately estimate a patient’s
length of survival may improve the decision-making
process for palliative care and allow patients and their
families to better prepare for death.
To determine prognostic factors, most prior studies
collected data on only one occasion, compromising results.
Moreover, most prior studies focused on individual symp-
toms, such as pain or fatigue, meaning that the role of overall
symptom severity in predicting survival has not been
extensively investigated. Therefore, this prospective and
longitudinal study aimed to evaluate the significance of
overall symptom severity, quality of life, pain, fatigue, and
performance status data in survival prediction from a sample
of cancer patients admitted to a palliative care unit and
collected on a weekly basis throughout their survival time.
Patients and methods
Participants and settings
A longitudinal design was used in this study, which
recruited a convenience sample from the palliative care
inpatient units at three medical centers in northern Taiwan.
Patients were required to meet the following criteria to be
included in this study: (1) have a pathological diagnosis of
cancer and be admitted to a palliative care unit, (2) be age
18 years or older, and (3) have the ability to communicate
in Mandarin or Taiwanese. Patients were not included if
they were cognitively impaired (by physicians’ judge-
ments), did not want to participate, or could not understand
the intent of the study. The final sample consisted of 180
terminal cancer patients enrolled consecutively from the
palliative care units. The Institutional Review Boards
approved the study.
Instruments
A six-part survey was used to collect data. The survey
included the following: (1) the Medical Outcome Study 36-
Item Short-Form Health Survey (SF-36), (2) the Taiwanese
version of the M.D. Anderson Symptom Inventory
(MDASI-T), (3) the Karnofsky Performance Status (KPS),
(4) the Brief Pain Inventory-Chinese Version (BPI-C), (5)
the Brief Fatigue Inventory-Taiwanese Version (BFI-T),
and (6) a demographic questionnaire.
The Medical Outcomes Study Short-Form-36, Taiwanese
version
The SF-36 was used to assess quality of life in this study
because this generic quality-of-life instrument has been used
widely across different disease populations, which provides an
opportunity to compare results form this study with other
studies or disease populations. The SF-36 has been used in
cancerpatients[21, 26, 42, 52, 62] and has been examined for
its significance in predicting survival in cancer patients [26,
52]. The SF-36 measures health-related quality of life,
including physical functioning (ten items); role limitations
due to physical health problems (RP, four items); bodily pain
(BP, two items); general health (GH, five items); vitality (VT,
four items); social functioning (SF, two items); role limi-
tations due to emotional problems (RE, three items); and
mental health (MH, five items). A physical component
summary (PCS) score was computed by averaging the scores
on PF, RP, BP, and GH. A mental component summary
(MCS) score was computed by averaging the scores on VT,
SF, RE, and MH [66]. The Taiwanese version of SF-36 was
validated in a healthy adult sample [38, 60].
The Taiwanese version of the M. D. Anderson Symptom
Inventory
The MDASI-T was used to assess multiple symptoms. The
original MDASI was developed to measure 13 symptoms in
the previous 24-h period of patients with cancer, including
symptom intensity and subsequent interference with life
activities. The first part of the MDASI consists of 13 single-
item measures of symptom intensity, including fatigue,
sleep disturbance, pain, drowsiness, poor appetite, nausea,
vomiting, shortness of breath, numbness, difficulty remem-
bering, dry mouth, distress, and sadness. Each symptom
1286 Support Care Cancer (2009) 17:1285–1294

Page 3

item is rated on a scale of 0 (not at all) to 10 (as bad as you
can imagine). The second part of the MDASI assesses the
extent to which symptoms interfere with the following:
general activities, mood, normal work, relations with other
people, walking, and enjoyment of life. Each of the six
interference items is rated on a scale of 0 (does not
interfere) to 10 (completely interferes). A multisymptom
severity composite score (an average of the 13 symptom
items) and an interference composite score (an average of
the six interference items) were computed. The validity and
reliability of the MDASI has been established [11].
The MDASI-T was developed using a translation and
back-translation process. The MDASI was first translated
from English into Taiwanese by a bilingual translator. A
second bilingual translator who had not seen the original
English version back-translated each item from Taiwanese
into English. The back-translated items were compared
with the original English items for congruency. This
process was repeated until the back-translated items and
the original items were in agreement. The MDASI-T has
good validity and reliability [35].
Brief Pain Inventory-Chinese version
The BPI-C was used to assess pain intensity and its
resulting interference with life activities [47]. Each item
on the BPI is rated on a scale of 0 (no pain) to 10 (the worst
pain I can imagine). The first part of the BPI consists of the
following four single-item measures of pain intensity: (1)
worst pain (please rate your pain by circling the number
that best describes your pain at its worst in the last 24 h),
(2) least pain (please rate your pain by circling the number
that best describes your pain at its least in the last 24 h), (3)
average pain (please rate your pain by circling the number
that best describes your pain on average), and (4) pain now
(please rate your pain by circling the number that tells how
much pain you have right now). The second part of the BPI
consists of seven items (each item again rated on a scale of
0 to 10) which assess the extent of pain interference with
the following: general activities, mood, walking, working,
relations with others, sleeping, and enjoyment of life. An
interference score (the average of the seven items) was
computed. The reliability and validity of the BPI have been
established [7, 32, 65].
Brief Fatigue Inventory-Taiwanese version
Fatigue was assessed with the BFI-T, which was originally
developed at the University of Texas M. D. Anderson Cancer
Center to measure fatigue in cancer patients. The BFI uses a
scale of 0 to 10 to measure fatigue severity and interference
with life activities in the previous 24 h. The first part of the
BFI measures worst fatigue during the past 24 h, usual fatigue
during past24 h, and fatigue now, with each item rated from 0
(no fatigue) to 10 (fatigue as bad as you can imagine). A
composite fatigue severity score is the average of the three
severity items. The second part of the BFI assesses the extent
to which fatigue interferes with the following: general
activities, mood, walking, normal work, relations with other
people, and enjoyment of life. Each of the previous items is
rated on a scale of 0 (does not interfere) to 10 (completely
interferes). A composite fatigue interference score is the
average of the six interference items. The validity and
reliability of the BFI has been established [43]. The BFI-T
was developed using a translation and back-translation
process. The psychometric properties of the BFI-T have
been established [34].
Karnofsky Performance Status
The KPS was used to assess patients’ performance status.
The KPS is rated on a scale of 1–100, in steps of ten and
has been documented to have good predictive validity [5].
Questionnaire for demographic and disease information
A demographic information sheet covered basic patient
information, including age, gender, education, marital status,
religious beliefs, and occupation. A disease information
sheet recorded each patient’s diagnosis, medications, and
treatment status, as well as whether metastasis had occurred.
Procedure
Approvalfor thisstudywas obtainedfromthe HumanSubject
Committeeofthehospital.Theresearcherapproachedpatients
individually to describe the study and obtain informed
consent. Upon admission to the palliative care unit and after
their informed consent was obtained, the patients were asked
to complete all self-administered questionnaires. The re-
searcher completed the KPS. Assessment with the SF-36,
MDASI-T, BFI-C, BFI-T, and KPS was repeated every week
for the remainder of the patient’s life. The mean (SD) number
of evaluations performed was 2.43 (1.56). Specifically, 64
(36%) patients had one assessment, 51 (28%) had two, 25
(14%) had three, and 40 (22%) had four or more assessments,
up to a maximum of eight.
Statistical analysis
Data on demographic and disease information and on the
KPS, SF-36, MDASI-T, BPI-C, and BPI-T were analyzed
by descriptive statistics. To take into account the depen-
dence of repeated measurements, a statistical method called
generalized estimating equations (GEE) [30, 31, 68] was
applied in this study. The GEE method, an extension of the
Support Care Cancer (2009) 17:1285–1294 1287

Page 4

quasi-likelihood approach, is being increasingly used to
analyze longitudinal and other correlated data, especially
when they are binary or in the form of counts [22]. The
method of GEE is a generalization of generalized linear
model that takes into account this within-group correlation
[30, 31, 68]. The GEE method was used in this study to
analyze whether survival time was predicted by the KPS,
SF-36, MDASI-T, BPI-C, and BPI-T measured weekly.
Log transformation was used with survival time in order to
achieve a normal distribution. Therefore, survival time with
a log transformation serves as an outcome variable in the
GEE model. All P values were set at 0.05.
Results
Participant and disease-related characteristics
Demographic and disease-related characteristics of patients
are presented in Table 1. The participants were diagnosed
with various types of cancer, including, but not limited to,
lung, colorectal, gastric, liver, and head and neck cancer.
Sixty-eight percent of participants were male and 76% were
retired. The average age was 67.27 years. The mean survival
duration from admission to death was 18.91 days and
median survival time was 14 days. Figure 1 shows the
global survival curve of the sample, which highlights
unequivocally that the study was performed on a group of
patients in the terminal phase. The means (SD) of major
outcome variables and severity for 13 symptoms in the
patients’ last 4 weeks of life are summarized in Table 2,
showing that all variables were deteriorating for patients
toward the end of their lives. The four most severe symptoms
experienced by patients in the last week of their lives were
fatigue, sleep disturbance, drowsiness, and lack of appetite.
Factors for predicting length of survival of terminal cancer
patients
The GEE method was used to analyze predictors of survival
time, as well as to control for the baseline heterogeneity and
time effects (i.e., changes in outcome variables resulting from
the passage of time). After controlling for the baseline
heterogeneity and time effects, univariate analysis of survival
time (after log transformation) highlighted prognostic signif-
icance of several variables and symptoms investigated
(Table 3). Gender, KPS, the PCS score, the MDASI-T total
score, composite fatigue severity, and composite pain
severity were significantly associated with survival time in
the univariate analysis. In addition to the MDASI-T total
score, most items in the MDASI-T, including pain, fatigue,
sleep disturbance, distress, shortness of breath, difficulty
remembering, poor appetite, drowsiness, dry mouth, sadness,
Days
6040 200
Cum Survival
1.0
0.8
0.6
0.4
0.2
0.0
Fig. 1 Overall survival of 180 patients
Table 1 Demographic and disease-related characteristics (N=180)
Characteristics MeanSD Number Percentage
Age (years)
Education (years)
KPS
Days from admission
to death
Sex
Male
Female
Marital status
Married
Other
Religious affiliation
Buddhist and Taoist
Christian
None
Other
Retired
Yes
No
Cancer diagnosis
Lung
Colorectal
Gastric
Liver
Head and neck
Pancreatic and bile track
Breast
Urology
Other
Days from admission to death
1–7 days
8–14 days
15–21 days
22–30 days
31–67 days
67.27
9.99
22.61
18.91
13.10
4.77
7.05
16.71
123
57
68
32
128
52
71
29
114
28
22
16
63
16
12
9
137
43
76
24
39
29
21
19
17
14
8
2
30
22
16
12
11
10
8
4
1
16
56
36
32
19
37
31
20
18
11
20
1288Support Care Cancer (2009) 17:1285–1294

Page 5

and numbness were also related with survival time in the
univariate analysis (see Table 4).
A multivariate analysis allows one to consider several
variables simultaneously and was therefore adopted to
evaluate the joint effect of the variables measured at each
time point. The variables included in the final model were
gender, age, KPS, PCS, MCS, the MDASI-T total score,
fatigue composite score, and pain composite score. In the
final model, results revealed that gender, KPS, and the
MDASI-T total score were significant prognostic indicators
of survival time for terminal care patients. Being male,
having a lower performance status, and reporting higher
levels of overall symptom severity were associated with a
shorter length of survival time (Table 5). In other words,
overall symptom severity and functional status remain
significant prognostic indicators of survival for terminal
cancer patients after controlling for demographic variables,
quality of life, pain, and fatigue.
Discussion
Patients with terminal cancer and their families frequently
request estimations of length of survival in order to plan for
Table 2 Mean (SD) major outcome variables and severity for 13 symptoms during the 4 weeks prior to death
Weeks before death<1 week >1–<2 weeks>2–<3 weeks>3–<4 weeks
(n=180)(n=116)(n=65)(n=40)
M (SD)M (SD)M (SD)M (SD)
KPS
PCS score
MCS score
MDASI-T
Fatigue severity
Pain severity
MDASI-T items
Pain
Fatigue
Nausea
Sleep disturbance
Sadness
Shortness of breath
Difficulty remembering
Poor appetite
Drowsiness
Dry mouth
Distress
Vomiting
Numbness
15.33
21.06
30.05
7.32
8.96
6.26
(6.38)
(3.05)
(6.36)
(1.07)
(1.01)
(1.96)
22.50
21.39
31.09
6.86
8.19
5.84
(6.17)
(3.21)
(6.71)
(1.23)
(1.14)
(1.76)
24.00
21.91
31.50
6.70
8.15
5.69
(5.81)
(3.51)
(6.71)
(1.40)
(1.20)
(1.93)
25.00
22.09
31.20
6.67
8.12
5.58
(5.55)
(3.49)
(7.27)
(1.26)
(1.15)
(1.95)
7.67
9.23
4.69
8.88
7.87
7.84
7.60
8.59
8.74
7.68
7.62
2.26
6.42
(2.16)
(1.08)
(3.76)
(1.24)
(1.61)
(1.99)
(1.71)
(1.79)
(1.29)
(1.77)
(1.55)
(3.19)
(2.32)
7.07
8.55
4.70
8.23
7.48
7.05
6.96
7.83
8.00
7.27
7.00
2.77
6.22
(2.28)
(1.03)
(3.64)
(1.34)
(1.76)
(1.71)
(1.74)
(1.89)
(1.43)
(1.72)
(1.96)
(3.39)
(2.46)
6.75
8.26
4.80
7.89
7.43
6.83
6.86
7.65
7.62
7.29
7.08
2.54
6.08
(2.49)
(1.20)
(3.65)
(1.72)
(1.70)
(2.06)
(1.88)
(2.17)
(1.83)
(2.15)
(1.64)
(3.22)
(2.25)
6.75
7.90
5.03
8.03
7.30
6.45
6.73
7.98
7.78
7.18
6.95
2.93
5.70
(2.68)
(1.34)
(3.58)
(1.72)
(1.79)
(1.81)
(1.52)
(1.29)
(1.25)
(2.06)
(1.81)
(3.96)
(2.14)
PCS physical component summary, MCS mental component summary, MDASI-T average of the 13 symptom items, fatigue severity average of
three severity items, pain severity average of four severity items
Table 3 Univariate analysis of survival by GEE method (N=180)
β
SE95% CIZP
Lower Upper
Gender (control: male)
Age
KPS
PCS
MCS
MDASI-T
Composite fatigue severity
Composite pain severity
−0.15
0.00
0.03
0.02
0.01
−0.14
−0.13
−0.05
0.05
0.00
0.00
0.01
0.00
0.02
0.02
0.01
−0.25
−0.01
0.02
0.00
0.00
−0.17
−0.17
−0.06
−0.05
0.00
0.03
0.03
0.01
−0.11
−0.10
−0.03
9.31
0.92
0.00**
0.34
0.00**
0.02*
0.09
0.00**
0.00**
0.00**
180.15
5.08
2.89
76.61
49.37
20.65
*P<0.05; **P<0.01
Support Care Cancer (2009) 17:1285–12941289