623292

research-article2015
PSH0010.1177/2010105815623292Proceedings of Singapore HealthcareZhang et al.

PROCEEDINGS
Original Article OF SINGAPORE HEALTHCARE

Proceedings of Singapore Healthcare

Prognostic value of Pneumonia Severity 2016, Vol. 25(3) 139­–147

© The Author(s) 2015
Reprints and permissions:
Index, CURB-65, CRB-65, and procalcitonin sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/2010105815623292

in community-acquired pneumonia in psh.sagepub.com

Singapore

Zoe Xiaozhu Zhang1, Weidong Zhang1, Ping Liu1, Yong Yang1,

Wan Cheng Tan2, Han Seong Ng3 and Kok Yong Fong4

Abstract
Objective: The purpose of this study was to evaluate the performance of three severity scoring tools and procalcitonin (PCT)
in severity stratification and mortality prediction among patients with community-acquired pneumonia (CAP) in Singapore.
Methods: The method used was a retrospective observational study of all the consecutive patients with CAP admitted
through the emergency department of Singapore General Hospital between 2012–2013.
Results: Among 1902 study subjects, the overall 30-day mortality was 15.7%. The mortality rates for Pneumonia Severity
Index (PSI) class I–III were 0, 0, and 3.7%, which were comparable to the original published data. CURB-65 and CRB-65 had
higher mortality rates in all severity levels. In three levels of risk stratification, the low risk group of PSI (class I–III) included
42.6% of the patients with mortality rate of 1.9%, whereas the low risk group defined by CURB-65 (score 0–1) and CRB-65
(score 0) included 52.0% and 24.4% of the patients with higher mortality rates (7.3% and 4.5% respectively). PSI was the most
sensitive in mortality prediction with area under receiver operating characteristic (ROC) curve of 0.82, higher than CURB-
65 (0.71), CRB-65 (0.67), and PCT (0.63) (p<0.001). The initial level of PCT was higher in non-survivors and intensive care
unit (ICU)-admitted patients compared to survivors (0.91 vs 0.36 ng/ml, p<0.001) and non-ICU patients (3.70 vs 0.38 ng/ml,
p<0.001). Incorporating PCT did not improve the discriminatory power of the scoring tools for mortality prediction.
Conclusions: PSI was a reliable tool for severity stratification and morality prediction among the patients with CAP in
Singapore.

Keywords
Community-acquired pneumonia, severity, mortality

Introduction
Community-acquired pneumonia (CAP) is one of the most
common infectious diseases worldwide. It is associated with
considerable mortality and morbidity, especially in elder
patients and patients with comorbidities.1,2 Pneumonia repre-
sents a significant health issue in Singapore. It is the fifth most
common cause of hospitalizations and the second principle
cause of death, leading to 13,000 admissions and over 3000
deaths annually, consisting of 18.5% of total deaths.3 The esti-
mated direct cost of a single CAP hospitalization ranges from
S$2000–16,000 (US$1400–12,000).
Pneumonia is the major respiratory disease seen in the
emergency department (ED). Timely and accurate severity
assessment is crucial to the initial management.4–6 To assist
clinicians in giving an objective severity evaluation,7–10 two
major clinical scoring tools, Pneumonia Severity Index (PSI)
and CURB-65, have been developed,11,12 PSI uses 20 clinical
variables to compute the severity score (Table 1).11 The
patients are then categorized into five severity classes with
increasing likelihood of death within 30 days. Class I–III are
1Epidemiology Department, Singapore General Hospital, Singapore
2UBC James Hogg Research Centre, St Paul’s Hospital, Canada
3CEO’s Office, Singapore General Hospital, Singapore
4Department of Rheumatology and Immunology, Singapore General
Hospital, Singapore
Corresponding author:
Zoe Xiaozhu Zhang, Epidemiology Department, Medical Board, Bowyer
Block A, Singapore General Hospital, Outram Road, Singapore 169608,
Singapore.
Email: zoe.zhang.x.z@sgh.com.sg

Creative Commons Non Commercial CC-BY-NC: This article is distributed under the terms of the Creative Commons Attribution-
NonCommercial 3.0 License (http://www.creativecommons.org/licenses/by-nc/3.0/) which permits non-commercial use,
reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open
Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage).
140 Proceedings of Singapore Healthcare 25(3)

Table 1.  Comparison of the components of Pneumonia Severity Index (PSI), CURB-65 and CRB-65.

PSI Score CURB-65 Point CRB-65 Point

Characteristics Characteristics Characteristics

Demographic factors Confusion 1 Confusion 1
Age (in years) Blood urea nitrogen⩾20 1 Respiratory rate⩾30/min 1
mg/dl
Men Age Respiratory rate⩾30/min 1 Systolic BP<90 mm Hg or 1
diastolic BP⩽60 mm Hg
Women Age −10 Systolic BP<90 mm Hg or 1 Age⩾65 years 1
diastolic BP⩽60 mm Hg
Nursing home resident +10 Age⩾65 1  
Coexisting illnesses Total 5 Total 4
Neoplastic disease +30  
Liver disease +20  
Congestive heart failure +10  
Cerebrovascular disease +10  
Renal disease +10  
Findings on physical examination  
Altered mental status +20  
Respiratory rate⩾30/min +20  
Systolic blood pressure<90 mm Hg +20  
Temperature<35°C or ⩾40°C +15  
Pulse⩾125 beats/min +10  
Laboratory and radiographic  
findings
Arterial pH<7.35 +30  
Blood urea⩾30 mg/dl (11 mmol/l) +20  
Sodium<130 mmol/l +20  
Glucose⩾250 mg/dl (14 mmol/l) +10  
Hematocrit<30% +10  
Partial pressure of arterial oxygen<60 +10  
mm Hg or oxygen saturation<90%
Pleural effusion +10  

BP: blood pressure.

low risk groups with cumulative mortality rate <1%; whereas
class IV and V are intermediate and high risk groups with
mortality rates ranging from 9–30%. CURB-65 is a simple
severity calculator with use of only five criteria (confusion,
urea respiratory rate, blood pressure, age ⩾65).12–14 CRB-65
is a simplified version of CURB-65 obtained by omitting blood
urea measurement.15,16 PSI, CURB65, and CRB65 have been
validated extensively and proved to be powerful in severity
stratification and mortality prediction.11,12,16–20 In Europe and
North America PSI and CURB-65 are incorporated into clini-
cal guidelines as the standard of care of pneumonia.5,6,21
Procalcitonin (PCT), a systemic inflammatory protein, has
appeared to be a novel prognostic biomarker of pneumonia
in recent years. PCT helps to differentiate bacterial from viral
infections, and guides the decision on antibiotic therapy.22–25
PCT was associated with septic shock, organ failures, and
occurrence of complications in patients with pneumonia.26,27
Non-survivors and patients in the high severity class had
higher levels of PCT.18,27,28 Adding PCT to PSI or CURB-65
improved the prediction of adverse complications of CAP.18,27
Despite intensive investigations worldwide, the evidence
for the predictive efficacy of these severity scoring tools
among Asian patients is limited with various conclusions19, 29–32
(Table 2). Man et al. have reported that PSI, CURB-65, and
CRB-65 perform similarly in mortality prediction among CAP
patients in Hong Kong and recommended CURB-65 in daily
clinical practice because of its simplicity.19 A Korean study,
however, reported that for mortality prediction CURB-65
was more valid for less severe patients, but PSI performed
better among the severe cases.30 The only validation study
conducted in Singapore selectively included patients who
were 55 years or older and the study setting was not the hos-
pital ED.32 In view of the serious burden of pneumonia in ED
and in the Singapore healthcare system, a thorough validation
study is indispensable. Here we explicitly evaluated the effec-
tiveness of PSI, CURB-65, CRB-65, and PCT on severity strati-
fication and mortality prediction in a large cohort of patients
in the hospital emergency setting in Singapore.
Methods
Study setting and study subjects
The study was performed in Singapore General Hospital
(SGH), an adult urban tertiary hospital with more than 1500
beds. The ED has approximately 130,000 non-trauma
patients per year, and 10% seek medical care for respiratory
symptoms.
All the consecutive patients admitted to hospital through
the ED with a provisional diagnosis of pneumonia (International
Classification of Diseases, 9th revision, Clinical Modification,
Zhang et al. 141

Table 2.  Summary of Asian studies on pneumonia severity scoring scales.

Country/ Setting Severity scales Primary No. of Nature of Main findings Year of
region assessed outcomes patients the study publication
Hong Kong Hospital PSI, CURB-65, 30-Day mor- 1016 Prospective PSI, CURB-65 and CRB-65 per- 200719
emergency CRB-65 tality (⩾18 years) formed similarly in severity stratifica-
tion and mortality prediction.
Pakistan Hospital CURB-65, 30-Day mor- 137 Prospective CURB-65 and CRB65 performed 200929
CRB-65 tality (⩾18 years) similarly in predicting mortality.
Korea Hospitals PSI, CURB-65 30-Day mor- 883 Prospective CURB-65 was more accurate in pre- 201330
tality (⩾18 years) dicting mortality in low risk group,
whereas PSI was more accurate to
predict mortality among the high risk
group.
Taiwan Hospital PSI, CURB-65 30-Day mor- 987 Prospective PSI performed better than CURB- 201031
emergency tality (⩾18 years) 65 in mortality prediction among
young (18-64 years) and elderly
patients (65-84 years), but under-
performed among very old patients
(⩾85 years)
Singapore Hospital PSI, CURB-65 30-Day mor- 1052 Retrospec- The AUC of PSI and CURB-65 in 201232
tality (⩾55 years ) tive mortality prediction were 0.77 and
0.70 respectively among patients
over 55 years.

AUC: area under the curve; PSI: Pneumonia Severity Index.

ICD-9CM code 480.x–487.x) between 1 January 2012–31
December 2013 were included. Ethical approval and waiver of
informed consent was obtained from the Centralized
Institutional Review Boards of the SingHealth Authority of
Singapore (2014/226/A). This study was kindly funded by
SingHealth Foundation Research Grant SHF/FG590S/2013.
Data collection
The patients’ case notes were reviewed by experienced and
trained research coordinators with a nursing or medical back-
ground. The 20 variables required for PSI calculation were
extracted, including patients’ demographics, comorbidities, ini-
tial vital signs, laboratory test results, and chest X-ray reports.11
All the collected data was the first reading in ED. Information
on six additional chronic conditions, including ischemia heart
disease, chronic obstructive pulmonary disease (COPD), dia-
betes mellitus, hypertension, dementia, and Parkinson’s dis-
ease were collected as well. The level of PCT measured within
the first 24 h of admission was taken as the initial level of PCT.
Pneumonia definition and exclusion criteria
Pneumonia was defined as an acute infection of the lung
parenchyma characterized by symptoms of acute respiratory
infection and the presence of an acute pulmonary infiltrate
on chest X-ray or abnormal auscultatory findings.33 Patients
with human immunodeficiency virus (HIV) infection, pulmo-
nary tuberculosis, cystic fibrosis, or were on long term immu-
nosuppressant or steroid treatment were excluded.
Severity score calculation and stratification
The scores of PSI, CURB-65, and CRB65 were calculated
according to original studies.11,12 Patients were subsequently
stratified into three levels of risk groups:11,12,19 (a) PSI: low risk
(classes I-III); intermediate risk (class IV); and high risk (class V);
(b) CURB-65: low risk (scores 0–1); intermediate risk (score
2); and high risk (scores 3-5); (c) CRB65: low risk (score 0);
intermediate risk (score 1–2); and high risk (score 3–4).
The primary outcome, all-cause mortality at 30 days after
hospital admission were compared. Length of stay (LOS) in
hospital was analyzed as well.
Statistical analysis
Categorical variables were expressed as counts (percent-
ages) and continuous variables as mean±standard deviation
(SE) or median with 25–75th quartile range) if the data were
not normally distributed. Frequency comparison was done by
chi-square test. For non-parametric continuous variables,
Mann-Whitney U test or Kruskal-Wallis test was used respec-
tively. Receiver-operating characteristic (ROC) curve analysis
was performed to evaluate the discriminatory power of the
risk scores and PCT on mortality prediction. Area under the
curve (AUC) and 95% confidence interval was computed.
Standard sensitivity, specificity, positive, and negative predictive
values were calculated at various cut-offs. Adopted from pre-
vious studies,18, 27 the initial level of PCT was stratified into the
4 ranges, which were <0.12, ⩽0.12<0.25; ⩽0.25<0.5, and
⩾0.5 ng/ml (0.12 ng/ml was the lowest detection limit of the
PCT assay). To test whether incorporating PCT improved the
predictive performance of the risk scores, joint logistic regres-
sion and ROC curve analysis were conducted as other studies
described.18,27 All statistical tests were two-tailed. Values of
p<0.05 were considered significant. All calculations were per-
formed using SPSS, version 20 for Windows.
Results
In the two-year study period, a total of 3224 patients in ED
received ICD-9-CM code 480 (pneumonia viral), 482
142 Proceedings of Singapore Healthcare 25(3)
Figure 1.  Study design. ICD: international classification of diseases; ED: emergency department.
(bacterial pneumonia), 485 (bronchopneumonia), and 486
(pneumonia unspecified) as the primary diagnosis (Figure 1).
Among them, 347 patients were discharged from ED right
after the treatment. Of 2841 admitted patients, 1902 patients
satisfied the study criteria and comprised the study cohort;
939 patients were excluded either because they were not
pneumonia or they fulfilled the exclusion criteria. The 30-day
mortality data was kindly provided by National Registry of
Disease Office, Health Promotion Board Singapore.
Patients’ demographic information and baseline clinical
features were summarized in Table 3. All the cut-off values in
physical examinations and laboratory tests were recom-
mended by Fine et al. in their original paper for PSI calcula-
tion.11 The overall 30-day mortality rate was 15.7%. Intensive
care unit (ICU) admission rate was 5.8%. The median length
of hospital stay was four days with 25th–75th interquartile
range (IQR) of 2–8 days.
The 30-day mortality rates for PSI class I–III were 0, 0, and
3.7%, which were comparable to the original data, whereas
PSI class IV and V had significantly higher mortality rates
(Table 4). For CURB-65 and CRB-65, the mortality rates
were significantly higher for all severity levels than the original
study. In three-level risk stratification, the low risk group of
PSI (class I–III) included 42.6% of the subjects with the mor-
tality rate of 1.9% (Table 5); whereas CURB-65 (score 0–1)
included more patients (52.0%) with significantly higher mor-
tality rate (7.3%, p<0.001); CRB-65 (score 0) included sub-
stantially fewer patients (24.4%), whereas the mortality rate
was much higher (4.5%). The median LOS increased signifi-
cantly with increasing risk levels (p<0.001).
The level of PCT within 24 h of admission was available in
1410 patients (Table 6). The median level of PCT significantly
increased with increasing risk levels in all tested scoring scales
(p<0.001). Non-survivors had significantly higher levels of
PCT than the survivors (0.91 µg/l vs 0.36 µg/l, p<0.001); so did
the patients who were admitted to ICU when compared with
those who had ICU admission (3.7 ug/l vs 0.38 ug/l, p<0.001).
Sensitivity, specificity, and positive and negative predictive
value (PPV and NPV) for prediction of 30-day mortality at
different cut-offs for each severity score and PCT are shown
in Table 7. PSI gave the most sensitive prediction of mortality
at all cut-offs (100%, 100%, 100%, 95.0%, and 47.8%) with
the highest NPV (NA, 100%, 100%, 98.1%, and 90.0%). Initial
level of PCT appeared to be the poorest in prediction of
30-day mortality.
The ROC curves for prediction of 30-day mortality for
each scoring tool were shown in Figure 2. AUC of PSI was
0.82 (95% CI: 0.80–0.84), which was significantly higher than
CURB-65 (AUC 0.71, 95% CI: 0.68–0.74) and CRB-65 (AUC
0.67, 95% CI: 0.63–0.70). The AUC of PCT was 0.63 (95%
CI: 0.59–0.67) for mortality prediction, which was the lowest
among the tested scoring tools (Figure 3). Incorporating PCT
did not further improve the discriminatory power of the
scoring tools on mortality prediction (p>0.05).
Discussion
This was the first large scale validation study to evaluate the
performance of the three severity scoring tools and PCT in
severity stratification and mortality prediction in hospital
emergency department setting in Singapore. In comparison
to CURB-65 and CRB-65, PSI was more accurate in defining
patients in low risk classes and more sensitive in mortality
prediction. PSI classified reasonable proportion of patients
into the low risk group with comparable mortality rate to the
original published data and other subsequent studies.11,17,19,34
Conversely, CURB-65 and CRB-65 underestimated the
severity and miscategorized some patients with high risk of
death into the low risk classes. Our finding was consistent
with many previous studies.17,35,36 A systematic review and
meta-analysis which screened 402 articles and included the
retrieved data from 23 studies also concluded that PSI was
superior to CURB-65 and CRB-65 in mortality prediction
and more accurate in identifying non-severe patients with low
risk of death.37
PSI makes use of 20 variables ranging from clinical presen-
tations, laboratory test results, and radiological findings to
compute the severity score. The estimation is generally more
accurate and sensitive than CURB-65 and CRB-65 although
the process of information collection and score computation
Zhang et al. 143

Table 3.  Baseline characteristics and outcomes of study subjects. with less chronic conditions (data not shown). Should some
of these patients be cared for in an outpatient setting, the
Characteristics Total (n=1902)
admission rate, and the usage of healthcare resources would
Demographics be markedly decreased. We also noticed that patients in high
Age (mean±SD years) 70±17 severity classes had much higher mortality rates compared to
Male 1055 (55.6) the original studies where the severity scores were derived
Race   and validated.11,12 The factors leading to this discrepancy
Chinese 1474 (77.7) could be complicated. Low ICU admission rate could be one
Malay 193 (10.1) of the reasons. In this study approximately 85% of the deaths
Indian 153 (8.0) occurred before hospital discharge (data not shown). For
Others 77 (4.0) patients in intermediate and high severity classes, the fre-
Nursing home residents 49 (2.6)
quencies of ICU admission were significantly lower than the
Coexisting illnesses and medical history
mortality rates (Table 4) which meant many severe patients
Neoplastic disease 348 (18.3)
died with no intensive care received. As is known that many
Cerebrovascular disease 301 (15.8)
factors play a crucial role in the decision on ICU admission,
Renal dysfunction 280 (14.7)
Congestive heart failure 153 (8.0)
patient’s age, coexisting diseases, long-term prognosis etc. are
Liver disease 21 (1.1) often the determinant factors besides the severity of disease.
Hypertension 1122 (59.0) Elderly patients with multiple comorbid diseases, especially
Diabetes mellitus 580 (30.5) the diseases with extremely poor prognosis would have less
Ischemia and heart disease 478 (25.1) opportunity of being admitted to ICU because they may not
Parkinson's disease and dementia 183 (9.6) have long-term benefit from ICU care.38 Moreover, patients’
COPD 112 (5.9) social and economical status and support have a big influence
Bronchiectasis 61 (3.2) on patients’ clinical management and outcome.
Asthma 137 (7.2) In addition to the severity scoring tools, we evaluated the
TB history 114 (6.0) prognostic value of PCT. Here PCT showed significant asso-
Physical examination findings ciation with the severity of pneumonia which was consistent
Altered mental status 65 (3.4) with the published data, that patients with high severity score
Pulse⩾125/min 206 (10.8) had higher initial PCT level.20,27,39,40 However, PCT was not a
Respiratory rate⩾30/min 74 (3.9) good predictor for mortality in our patient cohort. Huang
Systolic blood pressure<90 mm Hg 84 (4.4) et al. and Schuetz et al. have also reported that initial PCT had
Temperature<35°C or ⩾40°C 88 (4.6) significantly lower discriminatory power in mortality predic-
Laboratory and radiologic findings tion compared to PSI and CURB-65.18,27 Adding PCT could
Urea level⩾11 mmol/l 426 (22.4) not improve the performance of the scoring tools in predic-
Glucose⩾14 mmol/l 177 (9.3) tion of mortality.18,27
Sodium<130 mmol/l 378 (19.9)
The study had limitations. Firstly, the study was not novel.
Hematocrit<30% 344 (18.1)
PSI and CURB-65 have been developed for more than a dec-
PaO2<60 mm Hg or O2 saturation<90% 239 (12.6)
ade. Numerous validation studies have been performed
Arterial pH<7.35 119 (6.3)
worldwide, especially among western nations. In Singapore,
Pleural effusion 672 (35.3)
Outcome parameters
however, this was the first large validation study conducted in
ICU admission 111 (5.8) hospital emergency setting among adult patients. The clear
30-day mortality 299 (15.7) findings may bring significant clinical impact, especially on low-
Hospital LOS (median, 25–75th IQR 4 (2-8) ering the admission rate and medical cost of pneumonia.
days ) Secondly, the study used the retrospective design. All the data
were extracted from case-note review, which was not as accu-
COPD: chronic obstructive pulmonary disease; ICU: intensive care unit; rate as information directly collected through face-to-face
IQR: interquartile range; LOS: length of stay; SD: standard deviation; TB:
tuberculosis.
interview or questionnaire. However, a retrospective study is
Data are presented as n (%) unless other stated. All the cut-offs in physical quick, economical, and easy to conduct, a good method for a
examination and laboratory findings are recommended by Fine et al. for PSI pilot study. Thirdly, there was no stringent exclusion of health-
calculation.11 care associated pneumonia (HAP) because of the lack of data
on patients’ hospital admission history prior to the episode we
is complicated and time-consuming.37 However, the perfor- studied. Fourthly, not all study subjects had PCT value for the
mance of scoring tools varies from population to population. analysis of the mortality predictive value, and the time of
Different clinical characteristics and racial composition of the blood sample taken for PCT measurement varied although all
patients could result in various ranking. Therefore, thorough the reported levels of PCT in this study were measured within
validation study among a particular population is critical. 24 h of admission. Serum PCT changes dynamically over time.
Patients in PSI class I and II can be safely treated as outpa- Ideally all the blood samples should be taken at the same time.
tients.11,36 In this study more than 20% of the study subjects However, it is very difficult to achieve this in a retrospective
were in these two severity classes. They were much younger study. Another limitation we would like to mention was the
144 Proceedings of Singapore Healthcare 25(3)

Table 4.  30-Day mortality rates and intensive care unit (ICU) admission rates in each severity level of Pneumonia Severity Index (PSI),
CURB-65 and CRB-65.

No. of subjectsa 30-Day mortality Expected morality ICU admission rateb

(n=1902) rates# (n=299) rates, %11,12 (n=111)
PSI  
I 173 (9.1) 0 0.1 2 (1.2)
II (⩽70) 227 (11.9) 0 0.6 3 (1.3)
III (71–90) 409 (21.5) 15 (3.7) 2.8 7 (1.7)
IV (91–130) 754 (39.6) 141 (18.7) 8.2 44 (5.8)
V (>130) 339 (17.8) 143 (42.2) 29.2 55 (16.2)
p Value <0.001  
CURB-65  
0 382 (20.1) 11 (2.9) 0.7 9 (2.4)
1 607 (31.9) 61 (10.0) 2.1 34 (5.6)
2 600 (31.5) 127 (21.2) 9.2 35 (5.8)
3 241 (12.7) 70 (29.0) 14.5 26 (10.8)
4 63 (3.3) 23 (36.5) 40 6 (9.5)
5 9 (0.5) 7 (77.8) 14 1 (11.1)
p Value <0.001  
CRB-65  
0 464 (24.4) 21 (4.5) 1.2 18 (3.9)
1 902 (47.4) 138(15.3) 5.3 52 (5.8)
2 438 (23.0) 99 (22.6) 12.2 31 (7.1)
3 88 (4.6) 34 (38.6) 32.9 9 (10.2)
4 10 (0.5) 7 (70.0) 18.2 1(10.0)
p Value <0.001  
aData are numbers and percentages of the total study subjects (n=1903).
bData are numbers and percentages of the subjects under each severity level.

Table 5.  30-Day mortality and length of stay (LOS) in three-level risk categories of Pneumonia Severity Index (PSI), CURB-65 and CRB-65.

No. of patientsa (n=1902) 30-Day mortalityb (n=299) LOSc

PSI
Low (I–III) 809 (42.6) 15 (1.9) 3 (2–5)
Intermediate (IV) 754 (39.6) 141 (18.7) 5 (3–9)
High (V) 339(17.8) 143 (42.2) 7 (3–14)
p Value < 0.001 <0.001
CURB-65
Low (score 0–1) 989 (52.0) 72 (7.3) 3 (2–7)
Intermediate (score 2) 600 (31.5) 127 (21.2) 5 (3–9)
High (score 3–5) 313 (16.4) 100 (31.9) 6 (3–12.5)
p Value < 0.001 <0.001
CRB-65
Low (score 0) 464 (24.4) 21 (4.5) 3 (2–6)
Intermediate (score 1–2) 1340 (70.4) 237 (17.7) 5 (3–9)
High (score 3–4) 98 (5.1) 41 (41.8) 7 (3–14.3)
p Value <0.001 <0.001
aThenumber and percentage of the total study subjects (n=1977).
bThe number and percentage of the subjects under each risk class, and it was tested by chi2 test.
cLOS in hospital was presented as median (25th–75th interquartile range) as it was not normally distributed, and was tested by Kruskal-Wallis test.

exclusion of non-hospitalized patients with compatible ICD-9
code (383 out of 3224 patients) because in the pilot study we
found that majority of these patients were short of positive
radiological support for the diagnosis of pneumonia or they
did not have the necessary laboratory test results for comput-
ing PSI and CURB-65 score (data not shown). Unfortunately
we still missed a small proportion of patients who had mild
pneumonia and were treated as outpatients. This could pro-
duce a selection bias since hospitalized patients were generally
more severe. However, PSI in this study was able to accurately
identify the mild cases with low risk of death among the hos-
pitalized patients. If un-hospitalized mild cases were included
into the analysis, the mortality rate of the low risk group could
be even lower. Therefore, the conclusion of study would not
Zhang et al. 145

Table 6.  Procalcitonin (PCT) levels in different severity classes and major clinical outcome groups.

No. of subjects PCT median (25th–75th IQR) p Value

(n=1410) (ng/ml)
PSI class
Low (I–III) 546 0.20 (0.12–0.72) <0.001
Intermediate (IV) 586 0.57 (0.18–2.78)
High (V) 278 1.40 (0.38–9.83)
CURB-65
Low (score 0–1) 704 0.26 (0.12–1.10) < 0.001
Intermediate (score 2) 459 0.59 (0.19–2.60)
High (score 3–5) 247 1.30 (0.37–9.00)
CRB-65
Low (score 0) 318 0.22 (0.12–1.20) <0.001
Intermediate (score 1–2) 1012 0.51 (0.17–2.48)
High (score 3–4) 80 3.45 (0.43–14.05)
30-Day mortality
Survived 1166 0.36 (0.13–1.73) <0.001
Dead 244 0.91 (0.28–7.15)
ICU admission
ICU admitted 107 3.70 (0.87–13.8) <0.001
Non-ICU admitted 1303 0.38 (0.14–1.80)

ICU: intensive care unit; IQR: interquartile range; PSI: Pneumonia Severity Index.
The level of initial PCT is presented as median (25th–75th IQR). The difference in PCT levels between severity levels were compared using non-parametric
Kruskal-Willis test.

Table 7.  Sensitivity, specificity, positive predictive values (PPVs)

and negative predictive values (NPVs) for 30-day mortality
prediction at different cut-offs for each severity scoring tool and
procalcitonin (PCT).

Sensitivity Specificity PPV NPV

PSI risk class  
⩾1 100.0 0.0 15.7 NA
⩾II 100.0 10.8 17.3 100.0
⩾III 100.0 25.0 19.9 100.0
⩾VI 95.0 49.5 26.0 98.1
V 47.8 87.8 42.2 90.0
CURB-65 score  
⩾0 100.0 0.0 15.7 NA
⩾1 96.3 23.5 19.0 97.2
⩾2 75.9 57.2 24.9 92.7
⩾3 33.4 86.7 31.9 87.5
⩾4 10.0 97.4 41.7 85.3
⩾5 2.3 99.9 77.8 84.6
CRB-65 score  
Figure 2.  Receiver-operating characteristic (ROC) curves of
⩾0 102.0 0.0 15.4 NA
Pneumonia Severity Index (PSI), CURB-65, and CRB-65 with
⩾1 93.0 27.6 19.3 95.5
respect to prediction of 30-day mortality. AUC: area under the
⩾2 46.8 75.3 26.1 88.4 curve; CI: confidence interval.
⩾3 13.7 96.4 41.8 85.7
4 2.3 99.8 70.0 84.6
Initial PCT levels   many published data.37 Meanwhile, we retrieved patient clini-
<0.12 11.5 76.8 9.4 80.6 cal data through case-note review, and the findings would be
<0.25 21.3 58.6 9.7 78.1 more accurate and informative than those data obtained
<0.5 37.3 44.6 12.4 77.3 merely by matching ICD code.
⩾0.5 62.7 55.4 22.7 87.6 In conclusion, PSI was a valid and reliable scoring system
for severity stratification and mortality prediction in patients
NA: not available; PSI: Pneumonia Severity Index.
with pneumonia in Singapore despite its complexity and cum-
bersomeness. Implementation of PSI into daily clinical prac-
be affected. The main strength of the study lay in the size of tice would help to guide the management of pneumonia,
the study cohort, which was comparable or even larger than reduce unnecessary admission and improve the efficiency of
146 Proceedings of Singapore Healthcare 25(3)
Figure 3.  Receiver-operating characteristic (ROC) curves of
procalcitonin (PCT) and combined PCT with Pneumonia Severity
Index (PSI), CURB-65, and CRB-65 respectively with respect to
prediction of 30-day mortality. AUC: area under the curve; CI:
confidence interval.
medical resource usage. Further prospective validation study
is still required.
Acknowledgements
The authors wish to thank Shen Lian in the Biostatistics Unit, Yong
Loo Lin of the School of Medicine, National University Health
System for their assistance in biostatistical analysis. The authors are
also grateful to the National Registry of Diseases Office, Singapore
Health Promotion Board for providing 30-day mortality data.
Conflict of interest
The authors declare that there is no conflict of interest
Funding
This research received no specific grant from any funding agency in
the public, commercial, or not-for-profit sectors..
References
1. Mandell LA, Wunderink RG, Anzueto A, et  al. Infectious
Diseases Society of America/American Thoracic Society con-
sensus guidelines on the management of community-acquired
pneumonia in adults. Clin Infect Dis 2007; 44: S27–S72.
2. Murphy SL, Xu J and Kochanek KD. Deaths: final data for 2010.
National vital statistics reports : from the Centers for Disease
Control and Prevention, National Center for Health Statistics,
National Vital Statistics System. 2013; 61: 1–117.
3. Singapore Ministry of Health. Singapore Health Facts: Principle
Causes of Death 2013 (https://HYPERLINK “http://www.
moh.gov.sg/content/moh_web/home/statistics/Health_Facts_
Singapore/Principal_Causes_of_Death.html)”www.moh.
gov.sg/content/moh_web/home/statistics/Health_Facts_
Singapore/Principal_Causes_of_Death.html). Accessed on
June 2015.
4. Brar NK and Niederman MS. Management of community-
acquired pneumonia: a review and update. Ther Adv Resp Dis
2011; 5: 61–78.
5. Bartlett JG, Dowell SF, Mandell LA, et al. Practice guidelines for
the management of community-acquired pneumonia in adults.
Infectious Diseases Society of America. Clin Infect Dis 2000; 31:
347–382.
6. Mandell LA, Marrie TJ, Grossman RF, et al. Canadian guidelines
for the initial management of community-acquired pneumonia:
an evidence-based update by the Canadian Infectious Diseases
Society and the Canadian Thoracic Society. The Canadian
Community-Acquired Pneumonia Working Group. Clin Infect
Dis 2000; 31: 383–421.
7. Neill AM, Martin IR, Weir R, et al. Community acquired pneu-
monia: aetiology and usefulness of severity criteria on admis-
sion. Thorax 1996; 51: 1010–1016.
8. Woodhead MA, Macfarlane JT, McCracken JS, et al. Prospective
study of the aetiology and outcome of pneumonia in the com-
munity. Lancet 1987; 1: 671–674.
9. Almirall J, Bolibar I, Vidal J, et al. Epidemiology of community-
acquired pneumonia in adults: a population-based study. Eur
Resp J 2000; 15: 757–763.
10. Angus DC, Marrie TJ, Obrosky DS, et al. Severe community-
acquired pneumonia: use of intensive care services and evalu-
ation of American and British Thoracic Society Diagnostic
criteria. Am J Resp Crit Care Med 2002; 166: 717–723.
11. Fine MJ, Auble TE, Yealy DM, et al. A prediction rule to identify
low-risk patients with community-acquired pneumonia. N Engl
J Med 1997; 336: 243–250.
12. Lim WS, van der Eerden MM, Laing R, et al. Defining commu-
nity acquired pneumonia severity on presentation to hospital:
an international derivation and validation study. Thorax 2003;
58: 377–382.
13. Lim WS, Macfarlane JT, Boswell TC, et al. Study of community
acquired pneumonia aetiology (SCAPA) in adults admitted to
hospital: implications for management guidelines. Thorax 2001;
56: 296–301.
14. Macfarlane JT and Boldy D. 2004 update of BTS pneumonia
guidelines: what’s new? Thorax 2004; 59: 364–366.
15. Bauer TT, Ewig S, Marre R, et al. CRB-65 predicts death from
community-acquired pneumonia. J Int Med 2006; 260: 93–101.
16. Capelastegui A, Espana PP, Quintana JM, et al. Validation of a
predictive rule for the management of community-acquired
pneumonia. Eur Resp J 2006; 27: 151–157.
17. Aujesky D, Auble TE, Yealy DM, et al. Prospective comparison
of three validated prediction rules for prognosis in community-
acquired pneumonia. Am J Med 2005; 118: 384–392.
18. Huang DT, Weissfeld LA, Kellum JA, et al. Risk prediction with
procalcitonin and clinical rules in community-acquired pneu-
monia. Ann Emerg Med 2008; 52: 48–58.
19. Man SY, Lee N, Ip M, et al. Prospective comparison of three
predictive rules for assessing severity of community-acquired
pneumonia in Hong Kong. Thorax 2007; 62: 348–353.
20. Menendez R, Martinez R, Reyes S, et al. Biomarkers improve
mortality prediction by prognostic scales in community-
acquired pneumonia. Thorax 2009; 64: 587–591.
21. Mandell LA, Wunderink RG, Anzueto A, et  al. Infectious
Diseases Society of America/American Thoracic Society con-
sensus guidelines on the management of community-acquired
pneumonia in adults. Clin Infect Dis 2007; 44: S27–S72.
22. Masia M, Gutierrez F, Shum C, et al. Usefulness of procalcitonin
levels in community-acquired pneumonia according to the
patients outcome research team pneumonia severity index.
Chest. 2005; 128: 2223–9.
Zhang et al. 147
23. Christ-Crain M, Stolz D, Bingisser R, et al. Procalcitonin guid-
ance of antibiotic therapy in community-acquired pneumo-
nia: a randomized trial. Am J Resp Crit Care Med 2006; 174:
84–93.
24. Christ-Crain M, Jaccard-Stolz D, Bingisser R, et al. Effect of pro-
calcitonin-guided treatment on antibiotic use and outcome in
lower respiratory tract infections: cluster-randomised, single-
blinded intervention trial. Lancet. 2004; 363: 600–607.
25. Muller F, Christ-Crain M, Bregenzer T, et al. Procalcitonin levels
predict bacteremia in patients with community-acquired pneu-
monia: a prospective cohort trial. Chest 2010; 138: 121–129.
26. Boussekey N, Leroy O, Georges H, et al. Diagnostic and prog-
nostic values of admission procalcitonin levels in community-
acquired pneumonia in an intensive care unit. Infection 2005;
33: 257–263.
27. Schuetz P, Suter-Widmer I, Chaudri A, et al. Prognostic value
of procalcitonin in community-acquired pneumonia. Eur Resp J
2011; 37: 384–392.
28. Bloos F, Marshall JC, Dellinger RP, et al. Multinational, observa-
tional study of procalcitonin in ICU patients with pneumonia
requiring mechanical ventilation: a multicenter observational
study. Crit Care 2011; 15: R88, pp. 1–9.
29. Zuberi FF and Khan JA. Prospective comparison of prediction
rules of mortality risk for CAP in a developing country. The
international journal of tuberculosis and lung disease : the official
journal of the International Union against Tuberculosis and Lung
Disease. 2008; 12: 447–52.
30. Kim HI, Kim SW, Chang HH, et al. Mortality of community-
acquired pneumonia in Korea: assessed with the pneumonia
severity index and the CURB-65 score. Journal of Korean medi-
cal science. 2013; 28: 1276–82.
31. Chen JH, Chang SS, Liu JJ, et al. Comparison of clinical char-
acteristics and performance of pneumonia severity score and
CURB-65 among younger adults, elderly and very old subjects.
Thorax. 2010; 65: 971–7.
32. Abisheganaden J, Ding YY, Chong WF, Heng BH and Lim
TK. Predicting mortality among older adults hospitalized for
community-acquired pneumonia: an enhanced confusion, urea,
respiratory rate and blood pressure score compared with
pneumonia severity index. Respirology. 2012; 17: 969-75.
33. Bartlett JG, Dowell SF, Mandell LA, et al. Practice guidelines for
the management of community-acquired pneumonia in adults.
Infectious Diseases Society of America. Clin Infect Dis 2000; 31:
347–382.
34. Flanders WD, Tucker G, Krishnadasan A, et al. Validation of the
pneumonia severity index. Importance of study-specific recali-
bration. J Gen Int Med 1999; 14: 333–340.
35. Ananda-Rajah MR, Charles PG, Melvani S, et al. Comparing the
pneumonia severity index with CURB-65 in patients admitted
with community acquired pneumonia. Scand J infect Dis 2008;
40: 293–300.
36. Renaud B, Coma E, Labarere J, et  al. Routine use of the
Pneumonia Severity Index for guiding the site-of-treatment
decision of patients with pneumonia in the emergency depart-
ment: a multicenter, prospective, observational, controlled
cohort study. Clin Infect Dis 2007; 44: 41–49.
37. Loke YK, Kwok CS, Niruban A, et al. Value of severity scales
in predicting mortality from community-acquired pneumonia:
systematic review and meta-analysis. Thorax. 2010; 65: 884–890.
38. Guidelines for intensive care unit admission, discharge, and
triage. Task Force of the American College of Critical Care
Medicine, Society of Critical Care Medicine. Critical care medi-
cine. 1999; 27: 633–8.
39. Kruger S, Ewig S, Marre R, et al. Procalcitonin predicts patients
at low risk of death from community-acquired pneumonia
across all CRB-65 classes. The European respiratory journal.
2008; 31: 349–55.
40. Menendez R, Cavalcanti M, Reyes S, et al. Markers of treat-
ment failure in hospitalised community acquired pneumonia.
Thorax. 2008; 63: 447–52.