Keywords: C-reactive protein - albumin - biomarkers - inflammation - terminal cancer - palliative
care
INTRODUCTION
In the terminal phase of cancer, accurate prognostication is important in aiding clinical
decision-making, for example, about systemic therapies, palliative procedures, or
artificial nutrition or hydration, and in planning and preparing for the time ahead[1 ]. There are significant barriers to consistently and accurately identifying the end-of-life
phase and no definitive diagnostic criteria exist[2 ]. Although several prognostic tools have been validated, they vary in their complexity,
subjectivity, and therefore their clinical utility[3 ]. It would therefore be useful to rationalize these subjective assessments into a
simpler format by carefully selecting and refining existing tools. Biological factors
have the potential to be used in such cases, helping to overcome the limited understanding
of the process of dying4 .
There are many biological prognostic factors that are associated with the terminal
disease process, but some of them may have limited applicability. Cancer-associated
inflammation leads to poor survival which means biomarkers of systemic inflammation
are objective criteria with the potential to assist clinicians in recognizing dying[5 ]. A systematic review reporting prognostic biomarkers in patients with cancer in
the last months of life described seven prognostic factors with a high level of evidence,
namely, lymphocyte count, white blood cell count, serum C-reactive protein (CRP),
albumin, sodium, urea, and alkaline phosphatase[4 ].
There is ample literature on hypoalbuminemia and elevated CRP as a prognostic biomarker
in patients with cancer[5 ]. In addition, indices based on albumin and CRP levels, such as the Glasgow Prognostic
Score (GPS)[5 ]
[6 ] or the CRP/albumin ratio (CAR)[5 ]
[7 ] have shown good prognostic predictability for several types of incurable cancer.
This rationale indicates that significantly deranged albumin and CRP values could
signal a relatively short remaining survival time in patients with terminal cancer[8 ]
[9 ].
Currently evidence about how inflammatory markers varies during the disease trajectory
is lacking, as most studies available in the literature are cross-sectional, and generally
assess biomarkers as a single threshold value at baseline, consequently revealing
little about the long-term linear changes of these biomarkers as the disease progresses.
In that regard, studies that assess longitudinal inflammatory biomarker levels and
their changes over time in relation to well-defined events such as death may help
improve prognostic information and plan supportive care.
Our main hypothesis is that albumin and CRP are inflammatory biomarkers that worsen
progressively with the approach of death and may express terminal cancer phase. To
the best of our knowledge, to date no longitudinal studies have explored changes in
these biomarkers' levels during the last months of life of patients with cancer. Thus,
in this study, we investigated the longitudinal changes in albumin and C-reactive
protein (CRP) levels, and CRP/albumin ratio (CAR) in patients with terminal cancer
receiving palliative care in the last three months of life.
METHODS
Patients and data collection
This was a secondary analysis of data from a prospective cohort study conducted in
the Palliative Care Unit (PCU) in Brazil. The institutional research ethics committee
approved the study protocol, and each patient provided informed consent before participating
in the research.
Patients included in the original cohort of the study were evaluated during their
first attendance at the PCU by trained researchers between July 27, 2016 and March
18, 2020. The patients all had metastatic or locally advanced malignancy and were
no longer subject to specific anticancer therapy with curative intent. The focus of
care in the PCU is symptom-oriented. It commences when anti-tumor treatment is discontinued
because of lack of effect and/or severe side-effects. The Karnofsky Performance Status
(KPS) score (ranging from 0 [death] to 100
[full function]) was assigned according to patient-reported daily physical function[10 ]. The study population has been described in more detail elsewhere[11 ]-[14 ].
From 2,153 patients eligible in the original dataset, 49 patients were initially excluded
because they were previously diagnosed with an infectious or autoimmune disease (human
papillomavirus n=1, paracoccidioidomycosis n= 1, tuberculosis n=1, psoriasis n=1,
lupus n=1, vitiligo n=1, rheumatoid arthritis n=4, pneumopathy n=15, and human immunodeficiency
virus infection n=22). Routine blood examination results of serum albumin and CRP
levels were retrospectively collected from the patients' electronic records. CAR values
were calculated using the same blood samples. The follow-up period was 90 days from
the date of death. This time point was chosen because it generally coincides with
the terminal cancer phase[15 ]. The time periods considered for analysis (in days) were: 0-15 (T1), 16-30 (T2),
31-45 (T3), 46-60 (T4), 61-75 (T5), and 76-90 (T6). When there was more than one measurement
in the same time interval, the one closest to the date of death was used. Only patients
with at least two study measurements (consecutive or not)
were included in the longitudinal analysis. Thus, a total of 467 patients were excluded
because they had only one or less data point for the CRP or albumin. A flow chart
of the study is presented in [Fig. 1 ]. There was no statistical difference between the sample analyzed and the excluded
patients in relation to age (p = 0.425), sex (p = 0.415), primary tumor site (p =
0.512), and KPS (p = 0.228).
Covariate Assessment
The covariates were recorded at baseline; that is, on the date of the patient's entry
to the cohort, by trained researchers. The demographic data (age and sex)
and clinical data (primary tumor site, tumor progression, previous antitumor treatment)
were collected from the patients' electronic records.
Fig. 1 Flow chart of the participant selection process. Note: n= number of observations;
CRP= C-reactive protein; Alb= albumin; CAR= CRP/albumin ratio. *There was no statistical
difference in the sample studied when compared to the all excluded (eligible and selected)
patients in relation to age (p = 0.425), sex (p = 0.415), primary tumor site (p = 0.512) and Karnofsky Performance Status (p = 0.228).
Weight (kg) was measured using a calibrated portable scale (Wiso®) with an accuracy
of 0.1 kg. For patients unable to stand, the Stryker® GoBed II in-bed weight system
was used (Stryker Medical, USA). Height (m) was measured using a tape stadiometer
on the wall. When this could not be used, height was estimated using the Chumlea et
al. formulas[16 ]
[17 ]. body mass index (BMI) was calculated by dividing weight by height squared and expressed
in kg/m2 . Muscle strength was assessed by handgrip strength (HGS, kg) using a Jamar® hydraulic
hand dynamometer (Baseline, Fabrication Enterprises Inc, Elmsord, USA). Three trials
were performed per hand, with a 1-minute rest interval between the trials of each
hand. Maximum strength was defined as the greatest of the six measurements and used
to represent HGS. Low muscle strength was defined when the HGS value was lower than
the 25th percentile[14 ]. All the patients completed the Portuguese validated version of the Patient-Generated
Subjective Global Assessment (PG-SGA, ©FD Ottery, 2015), available in pt-goblal.org[18 ]. The short form version (PG-SGA SF) consists of the first fourpart of the questionnaire
based on patient-reported weight, food intake, symptoms, and physical function. The
total score of the PG-SGA SF is the sum of the scores of these four parts and the
higher this score, the greater the nutritional risk.
Data analysis
The statistical analyses were conducted using Stata® 13.1. The Kolmogorov-Smirnov test was performed to assess the distribution of the
variables. Median and interquartile range (IQR) were used to describe the continuous
variables, and number of observations and frequencies were used for the categorical
variables. The descriptive statistics for the laboratory characteristics included
all the data points collected for each parameter. Significance was set at 5% for all
the statistical tests.
Changes in the trajectory of biomarkers until death were assessed using longitudinal
linear mixed-effects (LME) analysis. LME regression coefficients (slopes) provide
a combined estimate of the effect between and within the participants[19 ]
[20 ]. The models were fitted using the unstructured covariance matrix. Time until death
(in days) was included in all the LME models as both a random and a fixed effect to
adjust for the overall and individual variations in the biomarker concentrations over
time. All the other covariables were considered as fixed effects only. This analysis
technique provided a random coefficient and an intercept. The coefficient provided
a combined estimate of a “mean” trajectory for the patient cohort, describing the
relationship between the variable value and time to death (note that a value that
increases as death approaches will therefore have a negative coefficient). The intercept
predicts the mean values of the biomarkers at time is zero (death). It acts as a useful
“end point” for the predicted trajectory for clinical interpretation; i.e., 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 is statistically significant[21 ].
RESULTS
Summary of baseline information
A total of 1,637 patients were included in this analysis. The median age was 63 years
(IQR: 53-71 years), and 58.8% were female. The most common primary cancer sites were
gastrointestinal tract (29.9%) and gynecological (18.5%). KPS 50-60%
and distant metastatic disease were observed in 47.2% and 74.1% patients, respectively
([Table 1 ]).
Descriptive statistics including all the data points collected are summarized in
[Table 2 ]. Median albumin was 3.00 g/dL (IQR: 2.50-3.60) across the whole time-period analyzed
and decreased with the approach of death (3.40 to 2.60 g/dL). In contrast, median
CRP was 9.31 mg/L (IQR: 4.42-17.30) and median CAR was 3.22 (IQR: 1.42-6.68), and
both increased before death (CRP: 5.82 to 14.19 mg/L and CAR: 1.62 to 5.94).
Table 1
Baseline demographic and clinical characteristics of the patients included in this
study (n= 1,637)
Variables
n (%)
Age (years)[a ]
Sex
63 (53; 71)
Female Primary tumor site
962 (58.8)
GI tract[b ]
490 (29.9)
Gynecological[c ]
303 (18.5)
Breast
201 (12.3)
Head and neck[d ]
201 (12.3)
Lung
165 (10.1)
Skin
73 (4.5)
Bones and soft tissues
55 (3.3)
Kidney and urinary tract
41 (2.5)
Otherse
Cancer stage
108 (6.6)
Locally advanced
424 (25.9)
Metastatic Current medical status
1213 (74.1)
Inpatient
393 (24.0)
Outpatient Previous treatment
1244 (76.0)
Quimiotherapy
1108 (67.7)
Radiotherapy
779 (47.6)
Surgery
670 (40.9)
KPS (%)
≥70
227 (13.9)
50-60
773 (47.2)
30-40
637 (38.9)
Note: n= number of observations; %= frequency; GI= Gastrointestinal; KPS= Karnofsky Performance
Status.
a Median (interquartile range).
b Upper and lower GI tract.
c Cervix, uterus, endometrium, ovary and vulva.
d Oral and nasal cavity, pharynx, larynx, salivary glands, paranasal sinuses, thyroid
and eyes.
e Central nervous system, hematologic, male genital organs, peritoneum, mediastinum
and unrecognized site.
Association of change in inflammatory biomarkers with death
A significant negative correlation between CRP and albumin levels was observed ([Fig. 2 ]); CRP and CAR increased ([Fig. 2B ] and 2C), while albumin levels showed significant reduction during the last 90 days
of life ([Fig. 2A ]). The LME analysis revealed that serum albumin (p <0.001), CRP (p <0.001), and CAR
(p <0.001) showed a significant linear dose-response relationship with time to death
(slope significantly different from 0). In other words, the results of the longitudinal
analysis showed a significant increase in CRP (slope: -0.10 to -0.13) and CAR (slope:
-0.05 to
-0.07) and a decrease in albumin (slope: all 0.01) ([Table 3 ]).
Fig. 2 Changes in (A) albumin, (B) C-reactive protein and (C) C-reactive protein/albumin
ratio until the days before death according to longitudinal linear mixed-effects analysis
in patients with terminal cancer. Note: CRP= C-reactive protein; CAR= CRP/albumin
ratio. *p -value refers to the statistical model included all patients with ≥2 measures, without
adjusting for variables (crude).
Table 2
Descriptive statistics for all data collected of the patients with terminal cancer
Time to death
Number of observations/ Patients
5th centile
25th centile
50th centile
75th centile
95th centile
CRP (mg/L)
T1
2705/1052 738
0.78 1.82
4.42 7.20
9.31 14.19
17.30 24.31
32.56 36.90
T2
538
0.97
4.86
9.41
17.32
32.42
T3
445
0.78
3.87
7.64
14.00
26.65
T4
392
0.58
3.86
7.64
14.24
27.00
T5
324
0.47
3.27
8.24
14.45
25.80
T6
268
0.43
2.51
5.82
11.10
28.10
Alb
(g/dL)
T1
4522/1637 1320
1.90 1.60
2.50 2.20
3.00 2.60
3.60 3.10
4.20 3.80
T2
858
1.90
2.50
3.00
3.50
4.10
T3
736
1.90
2.60
3.10
3.60
4.20
T4
626
2.20
2.80
3.30
3.70
4.40
T5
526
2.30
3.00
3.40
3.80
4.20
T6
456
2.30
3.00
3.40
3.90
4.50
CAR
2014/836
0.23
1.42
3.22
6.68
14.58
T1
597
0.59
2.89
5.94
10.33
18.44
T2
409
0.32
1.59
3.12
6.49
13.64
T3
321
0.23
1.12
2.63
4.87
11.47
T4
266
0.18
1.11
2.38
4.86
10.67
T5
228
0.12
0.83
2.32
5.21
11.10
T6
193
0.11
0.68
1.62
3.28
9.65
Note: CRP= C-reactive protein; Alb= albumin; CAR= CRP /albumin ratio.
a Time before death (in days): T1= 0-15; T2= 16-30; T3= 31-45; T4= 46-60; T5= 61-75;
and T6= 76-90.
Table 3
Longitudinal mixed effects models for inflammatory biomarkers in the last three months
before death in patients with terminal cancer
Variables
Models
Number of observations/ patients
Intercept (95% CI)
Slope[a ] (95% CI)
P value[b ]
CRP (mg/L)
1 2
2705 / 1052 2704 / 960
44.94 (41.28; 48.92) 44.73
(41.10; 48.68)
-0.11 (-0.12; -0.10) -0.11
(-0.14; -0.08)
<0.001 <0.001
3
1258 / 629
46.53 (41.66; 51.96)
-0.13 (-0.14; -0.11)
<0.001
4
1448 / 423
44.29 (39.79; 49.30)
-0.10 (-0.12; -0.08)
<0.001
Alb (g/dL)
1 2
4522 / 1637 4210 / 1519
0.15 (0.14; 0.16) 0.15 (0.14;
0.16)
0.01 (0.01; 0.01) 0.01 (0.01;
0.01)
<0.001 <0.001
3
1572 / 786
0.16 (0.14; 0.19)
0.01 (0.01; 0.01)
<0.001
4
2944 / 850
0.14 (0.13; 0.15)
0.01 (0.01; 0.01)
<0.001
CAR
1
2014 / 836
8.78 (7.90; 9.76)
-0.06 (-0.07; -0.05)
<0.001
2
1868 / 775
8.00 (7.18; 8.93)
-0.05 (-0.07; -0.04)
<0.001
3
1159 / 581
8.13 (6.96; 9.50)
-0.07 (-0.08; -0.06)
<0.001
4
848 / 254
9.22 (8.00; 10.63)
-0.06 (-0.07; -0.05)
<0.001
Note: CI= confidence interval; CRP= C-reactive protein; Alb= albumin; CAR= CRP/albumin
ratio.
a Slope is a model predicted change in value per day prior to death.
b
P value refers to longitudinal linear mixed-effects.
Model 1: Included ≥ 2 data points without adjust for confounder variables.
Model 2: Included ≥ 2 data points with adjust for confounder variables[* ].
Model 3: Included only 2 data points without adjust for confounder variables.
Model 4: Included ≥ 3 data points without adjust for confounder variables.
* The variables with P < 0.25 in univariate analysis: age (years), sex (female), Karnofsky
Performance Status (%), primary tumor site (gastrointestinal tract/others), distant
metastasis (yes/no), systemic treatment (yes/no), body mass index (kg/m2 ), weight loss in 6 months (%), handgrip strength (<25th percentile), and score of Patient-Generated Subjective Global Assessment (total points).
DISCUSSION
In this study, we evaluated inflammatory biomarkers at baseline and within three months
until death of patients who received palliative care for terminal cancer. Our findings
provide novel evidence that there was a significant longitudinal linear relationship
between change in CRP, CAR and albumin and death. CRP and CAR levels increased while
albumin decreased during the last three months of life.
Systemic inflammation is a recognized as a hallmark of cancer and reflects the body's
defense to mitotic processes and develops through the action of various proinflammatory
mediators, including cytokines, such as interleukins (IL) 1, IL 6 and tumor necrosis
factor alpha (TNF-α) leading to accelerated tumor progression[22 ]-[24 ]. According to our results, terminal cancer is related to a progressive increase
in serum CRP/CAR levels, and a decrease in serum albumin levels with the approach
of death. This can be explained by the fact that cytokines play a role in inflammation,
including induction of acute phase reactants and down-regulation of albumin production,
therefore CRP synthesis by the liver increases as the disease progresses, while albumin
synthesis could be significantly decreased[25 ].
The current findings are of clinical importance given that these biomarkers are commonly
available as part of the standard routine management of patients with cancer. In addition,
assessing the change in a variable rather than an absolute value at a single point
in time is crucial because albumin and CRP alterations over time indicate an ongoing
increase in inflammation. Furthermore, there is lesser susceptibility to biases related
to an acute elevation of the biomarker. Another point to be mentioned is that prognostic
scoring systems[3 ]
[6 ]
[26 ]
[27 ] are typically developed from cross sectional data in relation to survival, and this
technique allows a measure of association between a variable cut-off value and time
but assumes a common trajectory among individuals[21 ]. Our findings demonstrated that the assessment of the rate of change over time of
the biomarkers (trajectory) demonstrated prognostic predictive power. These suggest
that CRP, albumin and CAR could be used to predict death in patients with incurable
cancer referred for exclusive palliative care regardless of a specific cut-off point
evaluated by a single measurement at baseline.
The median concentrations of inflammatory biomarkers in our study were worse than
those reported in a longitudinal study in patients with acute myeloid leukaemia[28 ] which may be explained by the fact that our study dealt with patients at the most
advanced stage of the disease. To our knowledge, no studies have specifically studied
longitudinal albumin, CRP, and CAR in an exclusive palliative setting. Concerning
cross-sectional studies that involve patients with cancer in palliative care[4 ]
[8 ]
[9 ]
[29 ] median and cut-off values vary according to the cancer population and period analyzed
which makes it difficult to compare our results with those observed in other previous
studies.
At the present time there is no consensus on the best cutoff points for CRP, albumin
and CAR, with different studies using different values as reference[5 ]
[7 ]. The meta-analysis conducted by Dolan et al.[5 ] showed variation in the cutoff points for albumin and CRP in the included studies.
The most widely used cut-off point in the 63 articles that assessed the prognostic
power of CRP was >10 mg/L (n= 19; 30.1%), followed by >5 mg/L (n= 5; 7.9%). As for
the 33 that related albumin to overall survival, the most common cutoff points studied
were <3.5 g/dL (n= 13; 39.4%) and <3.0 g/dL (n= 5; 15.1%). According to the results
of a meta-analysis by Li et al.[7 ]
different CAR cutoff points, ranging from 0.25 to 6.7, are used to describe the association
between CAR and overall survival[29 ].
Regarding the intercept values, our results provided a combined estimate of the associations
between participants and within participants over time as a useful “end point” indication
of predicted or expected values for that variable as death approaches. The clinical
interpretation consists in observing that these values were above the normal reference
limits or thresholds values of these biomarkers[21 ]. In addition, it is worth highlighting that according to our crude and adjusted
LME models, changes in inflammatory biomarkers over 90 days were associated with death
regardless of age, sex, KPS, primary tumor site, distant metastasis, previous systemic
treatment, BMI, percentage weight loss in six months, HGS, and score of PG-SGA SF.
In practical terms, for example each 0.11mg/L increase in CRP above the mean CRP at
baseline was associated with one-day survival reduction, or considering the mean baseline
value is expected a 0.9g/dL decrease in serum albumin levels and an increase of 7.2
to 12.6 mg/L in CRP and 3.6 to 6.3 in CAR levels until patient's death.
Another important point was that the inclusion of more than two measures in the multivariate
model did not alter our results. It is interesting to note that two routine measurements
of blood-based biomarkers were enough to observe such dynamics in relation to death.
Although the current study presents clinically relevant data, there are strengths
and weaknesses that should be considered when interpreting our findings. The strength
is the longitudinal design of the study, which allowed a better understanding of the
relationship between the biomarkers evaluated and the patients' prognosis. A potential
weakness was the high number of excluded patients. However, no differences were observed
in relation to demographic and clinical variables when we compared the excluded and
the analyzed patients. To the best of our knowledge, this is the only study that has
described changes in albumin, CRP, and CAR values in patients with terminal cancer
receiving exclusive palliative care using an LME model. Thus, comparisons between
our data and the available literature are limited. Finally, this was a single-center
study without an external validation cohort, which limits generalizability of the
study results.
Identification of biomarkers of dying is an important area for future research, which
could contribute to improved clinical practices for patient management. However, further
investigations in multicenter studies are required to understand how these laboratory
measures and proposed cutoffs should be used to better employ biomarkers in prognostication
of terminal cancer.
CONCLUSION
This study of longitudinal measures allowed the exploration of the inflammatory biomarker's
response throughout the end-of-life and demonstrated that decreased albumin levels
and increased CRP and CAR values are significantly related to the terminal illness
process in the last three months of life of patients with cancer. Appraisal of these
biomarkers may be useful in clinical practice to predict survival in patients with
terminal cancer in palliative care.
Bibliographical Record
