CONTINUING MEDICAL EDUCATION IN ACADEMIC EMERGENCY MEDICINE

CME Information: Accuracy of Signs and

Symptoms for the Diagnosis of Community-
Acquired Pneumonia: A Meta-analysis
CME Editor: Corey Heitz, MD
Authors: Mark H. Ebell, MD, MS, Hulme Chupp, MPH, Xinyan Cai, MPH, Michelle
Bentivegna, MPH, and Maggie Kearney, MPH
If you wish to receive credit for this activity, please refer to the website: www.wileyhealthlearn
ing.com/aem
Educational Objectives John Wiley and Sons, Inc. designates this journal-based
After reading the article, participants should be able to discuss CME activity for a maximum of 1.0 AMA PRA Category 1
the current evidence available regarding the use of chloroquine CreditTM. Physicians should only claim credit commensurate
and hydroxychloroquine to treat COVID-19. with the extent of their participation in the activity.
For information on applicability and acceptance of continu-
Activity Disclosures ing medical education credit for this activity, please consult
This activity received no commercial support. your professional licensing board.
CME Editor Corey Heitz discloses no relevant financial This activity is designed to be completed within 1 hour. To
relationships. successfully earn credit, participants must complete the activity
This activity underwent peer review in line with standards during the valid credit period, which is up to two years from
of editorial integrity and publication ethics. Conflicts of inter- initial publication. Additionally, up to 3 attempts and a score
est have been identified and resolved in accordance with John of 70% or better is needed to pass the post test.
Wiley and Sons, Inc.’s Policy on Activity Disclosure and
Conflict of Interest.

Accreditation
John Wiley and Sons, Inc. is accredited by the Accreditation
Council for Continuing Medical Education to provide continu-
ing medical education for physicians.
EVIDENCE-BASED DIAGNOSTICS

Accuracy of Signs and Symptoms for

the Diagnosis of Community-
acquired Pneumonia: A Meta-analysis
Mark H. Ebell, MD, MS1, Hulme Chupp, MPH1*, Xinyan Cai, MPH1*,
Michelle Bentivegna, MPH1*, and Maggie Kearney, MPH2

ABSTRACT
Background: Community-acquired pneumonia (CAP) is an important source of morbidity and mortality.
However, overtreatment of acute cough illness with antibiotics is an important problem, so improved diagnosis of
CAP could help reduce inappropriate antibiotic use.

Methods: This was a meta-analysis of prospective cohort studies of patients with clinically suspected
pneumonia or acute cough that used imaging as the reference standard. All studies were reviewed in parallel by
two researchers and quality was assessed using the QUADAS-2 criteria. Summary measures of accuracy
included sensitivity, specificity, likelihood ratios, the diagnostic odds ratio, and the area under the receiver
operating characteristic curve (AUROCC) and were calculated using bivariate meta-analysis.

Results: We identified 17 studies, of which 12 were judged to be at low risk of bias and the remainder at
moderate risk of bias. The prevalence of CAP was 10% in nine primary care studies and was 20% in seven
emergency department studies. The probability of CAP is increased most by an abnormal overall clinical
impression suggesting CAP (positive likelihood ratio [LR+] = 6.32, 95% CI = 3.58 to 10.5), egophony (LR+ = 6.17,
95% CI = 1.34 to 18.0), dullness to percussion (LR+ = 2.62, 95% CI = 1.14 to 5.30), and measured temperature
(LR+ = 2.52, 95% CI = 2.02 to 3.20), while it is decreased most by the absence of abnormal vital signs (LR =
0.25, 95% CI = 0.11 to 0.48). The overall clinical impression also had the highest AUROCC at 0.741.

Conclusions: While most individual signs and symptoms were unhelpful, selected signs and symptoms are of
value for diagnosing CAP. Teaching and performing these high value elements of the physical examination should
be prioritized, with the goal of better targeting chest radiographs and ultimately antibiotics.

A ntibiotic overuse for patients with acute lower res-

piratory tract infection (LRTI) is widespread in
the United States and around the world, in both pri-
Diseases Society of America do recommend an antibi-
otic for patients with community-acquired pneumonia
(CAP). Since it is not practical, safe, or cost-efficient to
mary care and emergency department (ED) set- obtain a 0 (CXR) in all patients with a LRTI accompa-
tings.1,2,3 It causes increasing rates of antibiotic nied by cough, it is important to understand which
resistance and increased costs and reinforces patient elements of the history and physical examination can
beliefs that every cough requires an antibiotic.4 How- be used to identify patients at risk for CAP who may
ever, recently updated guidelines from the Infectious be candidates for a CXR and who does not need one.

From the 1University of Georgia and 2; and Department of Epidemiology, Athens, GA.
Received January 16, 2020; revision received March 4, 2020; accepted March 8, 2020.
*Order of authorship among three participating doctoral students and their contribution should be considered equal.
The authors have no relevant financial information or potential conflicts to disclose.
Author contributions: ME conceived and designed the study, participated in data abstraction, performed the primary analysis, and wrote the initial
draft of the manuscript; CC, MB, XC, and MK all participated in abstraction of data for the meta-analysis; MB and CC duplicated the primary anal-
ysis by ME to confirm its accuracy; and all authors reviewed and approved the final manuscript.
Supervising Editor: Shahriar Zehtabchi, MD.
Address for correspondence and reprints: Mark H. Ebell, MD, MS; e-mail: [email protected].
ACADEMIC EMERGENCY MEDICINE 2020;27:542–553.

ISSN 1553-2712 © 2020 by the Society for Academic Emergency Medicine

542 doi: 10.1111/acem.13965
ACADEMIC EMERGENCY MEDICINE • July 2020, Vol. 27, No. 7 • www.aemj.org 543

In previous meta-analyses, we have shown that Studies were excluded if they enrolled patients
normal vital signs and a normal lung examination because they had dyspnea or sepsis rather than sus-
effectively rule out CAP in patients with acute pected CAP. They were also excluded if patients were
cough,5 that the overall clinical impression is moder- in a specialized population such as only patients in
ately accurate for the diagnosis of CAP in adult,6 skilled nursing facilities, immunosuppressed patients,
and that C-reactive protein (CRP) is the preferred or patients with chronic lung disease. Studies of venti-
biomarker for the diagnosis of CAP in outpatients.7 lator or hospital-acquired pneumonia and studies of
However, there has been no recent meta-analysis of the diagnosis of a specific pathogen were excluded,
the accuracy of individual signs and symptoms for although studies limited to older adults were included.
the diagnosis of CAP in adults, with the most recent Studies were excluded if they used a case–control
published in 2007.8,9 In addition, previous meta-anal- design (i.e., recruited patients with known CAP and
yses did not use modern methods for the assessment healthy controls or matched patients with and without
of the quality of studies or to perform synthesis of radiographic pneumonia).
measures of test accuracy.10,11 We therefore set out
to perform an updated meta-analysis using modern Search Strategy
methods to answer the question: what is the accuracy This study is the second of three planned systematic
of signs and symptoms for the diagnosis of CAP in reviews (biomarkers to diagnose CAP, signs and symp-
adults. toms to diagnose CAP, and biomarkers for prognosis
in CAP) that used a single search strategy. The search
of the Medline database using the PubMed front-end
METHODS
was built around the concepts of “signs, symptoms,
This was a meta-analysis of previously published stud- and biomarkers”; “community-acquired pneumonia”;
ies of the accuracy of signs and symptoms for the diag- and “accuracy or prognosis” linked by Boolean AND
nosis of CAP. The study was registered with the joins and is shown in Data Supplement S1,
PROSPERO database (#CRD42018108036) and fol- Appendix S2. The limits “has abstract,” “human,”
lowed PRISMA guidance regarding conduct and and adult age ranges were applied to the search. In
reporting of a diagnostic meta-analysis (please see the addition, the reference lists of included studies were
Data Supplement S1, Appendix S1, available as sup- reviewed for additional articles, as were two older sys-
porting information in the online version of this tematic reviews identified by our search.8,9
paper, which is available at http://onlinelibrary.wiley.c
om/doi/10.1111/acem.13965/full, for the PRISMA Data Abstraction
checklist). All abstracts were reviewed for inclusion by the lead
author (MHE) and by one of four graduate students
Inclusion Criteria in epidemiology (CC, MK, MB, or XC). For any
Studies were included if they recruited a prospective abstract deemed potentially of interest, the full article
cohort of adolescents or adults presenting with symp- was obtained and reviewed by the lead author and
toms of respiratory infection or clinically suspected one other reviewer. Studies meeting inclusion and
pneumonia (including when it was based on the exclusion criteria were reviewed in parallel by the
physician ordering a CXR for respiratory symptoms) lead author and a graduate student who each
in the outpatient setting. The outpatient setting could abstracted variables describing study characteristics,
include primary care, urgent care, and the ED. Stud- study quality, and test accuracy data (true positives,
ies had to report sufficient information to calculate true negatives, false negatives, and false positives).
sensitivity and specificity for the diagnosis of CAP Discrepancies were resolved through consensus dis-
for at least one sign or symptom (including vital cussion.
signs). No limits were set for country, year, or lan-
guage. The reference standard had to be imaging (ra- Assessment of Study Quality
diography or CT) and had to have been performed The QUADAS-2 tool was adapted for our study and
in all participants or in all patients at high risk for definitions for low, unclear, and high risk of bias pre-
pneumonia and a random sample of low-risk specified for each domain.10 The full adapted tool is
patients, to avoid verification bias. shown in Data Supplement S1, Appendix S3.
544 Ebell et al. • SIGNS AND SYMPTOMS TO DIAGNOSE CAP

Analytic Strategy RESULTS

Similar signs and symptoms were grouped together.
A total of 792 abstracts were identified by our search,
For example, “purulent sputum,” “sputum (purulent),”
as well as eight from the review of reference lists and
and “mucopurulent sputum” were grouped into a sin-
one from the author’s files. A bridge search was per-
gle variable called “sputum (purulent).” Both the origi-
formed in August 2019 and identified 29 additional
nal sign or the symptom name and new groupings are
studies; one was reviewed in full but did not meet
shown in the full data table in the Data Supplement
inclusion criteria. Thus, a total of 830 abstracts were
S1, Appendix S3. For vital signs and other continuous
reviewed, of which 141 were reviewed in full and 16
variables, similar cutoffs to define an abnormal finding
met our inclusion and exclusion criteria and were
were combined where clinically reasonable, i.e., tem-
included in the final quantitative analysis. The search
perature greater than 37.7°C and temperature greater
process is summarized in Figure 1.
than 38.0°C. For studies reporting the overall clinical
Characteristics of included studies are summarized
impression in multiple categories, “higher than 75%”
in Table 1. The setting for data collection was primary
and “quite sure” or “very sure” were defined as overall
care for nine studies and the ED for seven studies,
clinical impression of pneumonia.12,13
and the number of patients studied ranged from 52 to
Data were imported into R (version 3.5.2) using the
2,820. The mean age of participants ranged from 32
R Studio framework (version 1.1.463). We performed
to 62 years, and 48% to 60% were female. Nine stud-
bivariate meta-analysis if there were three or more stud-
ies were set in Europe, five in the United States, and
ies of a sign or symptom using the mada package (ver-
one each in Israel and Chile. All studies used a new
sion 0.5.8) to calculate summary receiver operating
infiltrate or presence of radiographic diagnosis of
characteristic (ROC) curves and measures of accuracy
pneumonia on chest x-ray as the reference standard
with 95% confidence intervals (CIs).14 Where only a
test; we identified no studies that used CT as the refer-
single study described the accuracy of a test and cutoff,
ence standard. In one study, all patients where the
we used the diagti procedure in Stata version 15.1 (Stat-
clinician suspected pneumonia or with CRP> 50 mg/
Corp) to calculate measures of accuracy with 95% CIs.
L received a CXR, as well as a 25% random sample
Threshold effects occur when the cutoff for an
of all other patients (n = 97); none of the latter were
abnormal test varies, resulting in a tradeoff between
diagnosed with pneumonia.15 Four studies only
sensitivity and specificity. This can be an explicit varia-
included patients where the physician felt a CXR was
tion in cutoff (e.g., in a biomarker such as CRP) or
clinically indicated,16,17,18,19 while the remainder
implicit for a sign or symptom (e.g., any cough vs.
included patients presenting with respiratory symp-
only moderate or severe cough). When a threshold
toms, whether or not the physician ordered a CXR.
effect was observed based on inspection of the sum-
Because one study had somewhat different procedures
mary ROC curve, we presented the ROC curve but
for data collection in each of three states, in the assess-
did not necessarily report summary estimates of sensi-
ment of study quality and the analysis each state was
tivity and specificity. We reported the area under the
treated as a separate study.18 In the nine primary care
ROC curves (AUROCCs) as a measure of overall dis-
studies included in the current analysis, the prevalence
crimination where at least seven studies reported a
of CAP was 10% (556/5,579), while it was 20%
sign or symptom and the likelihood ratio differed sig-
(584/2,928) in the seven ED studies.
nificantly from 1.0 (this threshold was determined
Study quality was assessed using the QUADAS-2
post hoc after a review of the available studies). We
framework and was judged to be low risk of bias for
also calculated summary estimates of diagnostic accu-
12 studies and moderate risk of bias for five studies
racy (sensitivity, specificity, likelihood ratios, and the
(Table 2). A detailed description of the assessment of
diagnostic odds ratio) accompanied by 95% CIs. Posi-
study quality for each included study is shown in Data
tive and negative predictive values for typical preva-
Supplement S1, Appendix S3.
lences of CAP in the outpatient setting are selectively
The accuracy of studies is summarized in Table 3
reported for key signs and symptoms. Subgroup analy-
(individual study level data are provided in Data Sup-
ses were performed for selected signs and symptoms
plement S1, Appendix S6). The overall clinical impres-
by inclusion criteria (patients were recruited because
sion was more helpful for ruling in than for ruling
CXR was ordered vs. any patient with LRTI) and loca-
out CAP (positive likelihood ratio [LR+] = 6.32, 95%
tion (ED vs. primary care).
ACADEMIC EMERGENCY MEDICINE • July 2020, Vol. 27, No. 7 • www.aemj.org 545

Figure 1. PRISMA flow diagram describing the search process

CI 3.58-10.5; negative likelihood ratio [LR ] = 0.54, Regarding patient-reported symptoms, subjective fever
95% CI = 0.46 to 0.64). It had the highest LR + of and chills, the absence of coryza and rhinorrhea, dysp-
any finding and also had the highest AUROCC at nea, and chest pain significantly increased the likeli-
0.741 (Figure 2) with no clear pattern regarding accu- hood of CAP when present (LR+ = 1.21 to 1.47) and
racy for patients enrolled in studies because they pre- reduce the likelihood of CAP when absent (LR =
sented with acute RTI compared with those recruited 0.68 to 0.86). Cough had little discriminatory value,
because they had been referred for a CXR. but this is likely because cough was usually required
The only element of the medical history significantly as an entrance criterion for the studies.
associated with the likelihood of CAP based on the Egophony when present significantly increases the
likelihood ratios was chronic obstructive pulmonary likelihood of CAP when present (LR+ = 6.17, 95%
disease as a comorbidity (LR+ = 2.37, 95% CI = 1.21 CI = 1.34 to 18.0), although the sensitivity is quite
to 4.33; LR = 0.88, 95% CI = 0.78 to 0.97). low and the CI broad. Other signs significantly
Table 1
546

Characteristics of Included Studies, Stratified by Setting (Primary Care or Other Outpatient Versus ED)

Reference Standard (Prevalence

Author, Year Number Sex Inclusion Criteria Age CAP) Country Years
Primary care
Holm, 200728 364 51% female Consecutive adults 18 years or older with GP- Median 50 years CXR with new infiltrate (13.4%) Denmark 2002–2003
diagnosed LRTI; excluded if recent hospitalization,
severe illness requiring immediate hospitalization,
pregnancy, or already in study
Hopstaken, 200329 246 NR Consecutive adults 18 years or older with acute LRTI Mean 52 years CXR with new infiltrate (13.2%) Netherlands 1998–1999
defined as new or worsening cough and other
clinical characteristics; excluded if pregnant/
lactating, antibiotic allergy, other severe disease, or
recent hospitalization or antibiotics.
Lieberman, 200330 250 53% female Febrile adult 21 years or older with at least one of Mean 40 years CXR with new infiltrate that Israel 1999
cough, coryza, sore throat, or hoarseness resolved after treatment (7.6%)
Melbye, 198812 71 48% female Patients age 15 years or older with LRTI suspected Mean 48 years CXR with new infiltrate that Norway 1986
to be CAP resolved at 4 weeks (15.5%)
Melbye, 199215 581 58% female Patients age 18 years or older presenting with Mean 32 years CXR showing a density that Norway 1988–1989
symptoms suggestive of respiratory or throat resolved on follow-up (5.0%)
infection
Moberg, 201613 100 55% female Physician-suspected CAP, age 18+ years, and Mean 56 years CXR with new infiltrate (45.0%) Sweden 2011–2014
respiratory symptoms for at least 24 hours
Moore, 201716 720 49% female Acute cough as the main symptom, judged to be 45% were 60 years CXR with possible, probable, or United 2009–2013
infective by treating physician, 16+ years, and had or older definite pneumonia (16.0%) Kingdom
CXR ordered by treating physician due to suspicion
of pneumonia
van Vugt, 201322 2820 60% female Patients 18 years and older with acute cough or Mean 50 years CXR with new infiltrate (5.0%) 12 European 2007–2010
clinically suspected as having LRTI countries
Steurer, 201131 464 50% female Patients age 18 years or older with cough and Mean 47 years CXR with radiographic Switzerland 2006–2009
subjective or objectively measured fever. Excluded shadowing for which there is no
if known chronic lung disease other than chronic other explanation (20.5%)
bronchitis or immunosuppression.
ED
Wipf, 199932 52 0% female Adults with new or worsening cough accompanied Mean 62 years CXR with new infiltrate (46.2%) United States 1990–1993
by increased or darkening sputum production
Diehr, 198433 483 51% female Adults with cough less than 1 month duration; Mean 40 years CXR with radiographic United States NR
excluded if pregnant pneumonia (9.9%)
Gennis, 198919 308 43% female Patients 16 years or older where a CXR had been Mean 54 years CXR interpreted as positive or United States 1984–1985
ordered to evaluate for possible CAP; excluded if equivocal for pneumonia
pregnant (38.3%)
Gonzalez Ortiz, 199534 141 NR Patients 15 years or older presenting with fever NR CXR with findings suggestive of Spain NR
(>38°C) and respiratory symptoms; excluded if focal pneumonia (37.6%)
signs suggesting other infection such as meningitis

(Continued)
Ebell et al. • SIGNS AND SYMPTOMS TO DIAGNOSE CAP
ACADEMIC EMERGENCY MEDICINE • July 2020, Vol. 27, No. 7 • www.aemj.org 547

associated with an increased risk of CAP include dull-

United States 1986–1987

United States 1987–1988

ness to percussion, confusion, crackles, decreased

Years
breath sounds, any abnormal lung founds, the pres-

2005
ence of rhonchi, and toxic or ill-appearance (LR+ =
1.46 to 2.62). As with symptoms, the absence of a
Country

finding had less effect on reducing the likelihood of

CAP (LR = 0.61 to 0.96) than its presence did on
Chile

increasing it.
Abnormal vital signs were also associated with an
Reference Standard (Prevalence

equivocal, or possible infiltrate

CXR interpreted as probable or

definite pneumonia (13.9%) increased risk of CAP, including measured tempera-
CXR interpreted as positive,

ture ≥ 37.7 to 38.0°C, O2 saturation < 95%, heart

CXR with radiographic

rate> 100 beats/min, and respiratory rate> 20 to 25/

pneumonia (34.5%)
CAP)

min (LR+ = 2.02 to 2.52). The absence of any abnor-

mal vital signs had a negative likelihood ratio of 0.25,
providing good evidence against the presence of CAP.
(15.7%)

Only the presence of “any abnormal vital sign” had

a sensitivity above 80% (93%), while clinical character-
istics with a specificity greater than 90% were history
Illinois, 47.6 years

41 years Virginia

of chronic obstructive pulmonary disease, egophony,

Mean 45.4 years

Nebraska, and
Mean 54 years
Mean 53 years

dullness to percussion, and confusion. Clinical charac-

Age

teristics with a positive likelihood ratio greater than

CAP = community-acquired pneumonia; CXR = chest radiograph; GP = general practitioner; NR = not reported.

2.5 include egophony, dullness to percussion, confu-

sion, and measured temperature greater than 37.7 to
38.0°C. Only the absence of any abnormal vital sign
58% female Physician thought pneumonia was a possibility or at
59% female Patients 15 years or older with fever or respiratory

58% female Patients 18 years or older where a CXR had been

dyspnea, wheeze, or altered mentation and CXR

least two of fever, cough, sputum, pleuritic pain,

had a negative likelihood ratio less than 0.5, at 0.25.

immunocompromised or chronic lung disease.
symptoms presenting to the ED. Excluded if

The highest AUROCCs were found for the overall

clinical impression (0.741), any abnormal lung finding
ordered to evaluate for possible CAP

(0.669), and measured temperature greater than 37.7

Inclusion Criteria

to 38.0°C (0.637).
Another measure of overall diagnostic accuracy or
discrimination is the diagnostic odds ratio (DOR).
Findings with the highest DOR for the diagnosis of
had been ordered

CAP were the overall clinical impression (11.5, 95%

CI = 6.7-18.5), egophony (6.5, 95% CI = 1.4-18.9),
and any abnormal vital sign (6.0, 95% CI = 3.0-
10.6).
Comparing studies of patients recruited because
they had a CXR ordered versus studies that recruited
all patients with acute RTI, there was no clear pattern
Sex

for the accuracy of abnormal lung examination, chest

pain, crackles, dullness to percussion, dyspnea, or
Number

1364
325

255

tachycardia when stratified by these variables (Data

Supplement S1, Appendix S4). We also compared
studies set in the ED with those set in primary care
for selected signs and symptoms where there were an
Table 1 (continued)

Sald
ıas, 200735

adequate number of studies in each setting. There was

Singal, 198917

Tape, 199118

no clear pattern for overall clinical impression or

Author, Year

crackles on examination. For the symptom of dyspnea,

all five studies in the primary care setting were more
sensitive than the ED studies, while all four ED
548 Ebell et al. • SIGNS AND SYMPTOMS TO DIAGNOSE CAP

Table 2
Overview of Study Quality

Patient Reference Flow and L = 0, M = 1, and H = 2 + With High

Selection Index Test Standard Timing Likelihood of Bias
Diehr, 198433 L L L L L
Gonzalez Ortiz, 199534 L L L L L
28
Holm, 2007 L L L L L
Hopstaken, 200329 L L L L L
Lieberman, 200330 L L L L L
Melbye, 198812 L L L L L
16
Moore, 2017 L L L L L
Sald
ıas, 200735 L L L L L
Steurer, 201131 L L L L L
van Vugt, 201322 L L L L L
Wipf, 199932 L L L L L
18
Tape (Nebraska, Illinois), 1991 * L L L L L
Singal, 198917 H L L L M
Moberg, 201613 L L H L M
Melbye, 199215 L L L H M
Gennis, 198919 H L L L M
Tape (Virginia), 199118* L L H L M

*Because this study used different data collection procedures in Virginia, quality is assessed separately for that state.

studies were more specific than the primary care stud- any abnormal lung founds, the presence of rhonchi,
ies (Data Supplement S1, Appendix S5). and toxic or ill appearance were significantly associated
with the presence or absence of CAP. However, the
likelihood ratios were all between 0.5 and 2.0 for
DISCUSSION
these findings, so they have little impact on the diag-
The history and physical examination is a critical com- nostic likelihood of CAP and were especially unhelpful
ponent of the evaluation of patients with acute cough. when negative. On the other hand, physician integra-
However, many individual signs and symptoms have tion of individual signs and symptoms has much
limited value (especially when absent), and knowledge higher diagnostic accuracy. This has been previously
of the signs and symptoms most predictive of CAP shown—that the overall clinical impression can
can help physicians focus their evaluation. Based on approximate the accuracy of a clinical prediction
the DOR, a measure of overall discrimination, the fol- rule.20
lowing elements of the clinical examination are most The summary ROC curve for overall clinical
useful: the overall clinical impression, the presence of impression shows data consistent with a threshold
egophony, any abnormal vital sign, any abnormal lung effect for studies including any patient with acute RTI.
finding, tachypnea, and the presence of measured For those who were only included if a CXR was
fever. Based on the comparison of subjective with ordered, specificity was consistent but sensitivity var-
objective temperature, one concludes that the absence ied, perhaps due to differences in the threshold for
of subjective fever helps rule out CAP, while the pres- ordering the test. Review of the summary ROC curve
ence of measured fever tends to rule it in (Figure 2). for fever (Figure 3) revealed that subjective fever was
However, the converse (absence of measured fever or more sensitive (63% vs. 34%) but less specific (55%
presence of subjective fever) is less helpful diagnosti- vs. 87%) than measured temperature> 37.7 to 38.0°C
cally. for the diagnosis of CAP.
The LR + and LR– for a number of signs and The symptom of dyspnea showed a pattern of diag-
symptoms that are often acquired as part of the history nostic accuracy by setting, being more sensitive in pri-
and physical examination such as dullness to percus- mary care and more specific in the ED. This may
sion, confusion, crackles, decreased breath sounds, reflect different implicit cutoffs, with primary care
ACADEMIC EMERGENCY MEDICINE • July 2020, Vol. 27, No. 7 • www.aemj.org 549

Table 3
Diagnostic accuracy for individual elements of the medical history and physical examination

Diagnostic
Studies Sensitivity Specificity odds ratio
Sign or symptom (patients) (95% CI) (95% CI) LR+ (95% CI) LR (95% CI) (95% CI) AUROCC
Overall clinical 7 (5081) 0.50 (0.39-0.61) 0.92 (0.84-0.96) 6.32 (3.58-10.5) 0.54 (0.46-0.64) 11.5 (6.7-18.5) 0.741
impression
Medical history
Chronic obstructive 3 (748) 0.19 (0.13-0.27) 0.91 (0.86-0.95) 2.37 (1.21-4.33) 0.88 (0.78-0.97) 2.74 (1.24-5.51)
pulmonary disease
Previous pneumonia 3 (1245) 0.13 (0.02-0.47) 0.90 (0.63-0.98) 1.32 (0.81-2.00) 0.96 (0.81-1.02) 1.39 (0.79-2.21)
Any comorbidity 3 (3904) 0.44 (0.33-0.55) 0.63 (0.50-0.75) 1.19 (0.99-1.48) 0.90 (0.80-1.01) 1.34 (0.98-1.80)
Alcohol use disorder 3 (988) 0.06 (0.02-0.23) 0.96 (0.93-0.98) NC NC NC
Smoking (current) 4 (3425) 0.32 (0.13-0.59) 0.69 (0.54-0.81) 1.06 (0.53-1.78) 0.97 (0.66-1.22) 1.18 (0.44-2.73)
Male sex 4 (3539) 0.46 (0.39-0.54) 0.57 (0.52-0.61) 1.08 (0.93-1.23) 0.94 (0.83-1.06) 1.15 (0.88-1.47)
Smoking (ever) 3 (1434) 0.50 (0.30-0.69) 0.52 (0.36-0.67) 1.03 (0.78-1.28) 0.97 (0.75-1.18) 1.09 (0.66-1.70)
Symptoms
Pleuritic chest pain 3 (1245) 0.32 (0.26-0.39) 0.87 (0.65-0.96) 2.76 (0.97-7.133) 0.81 (0.70-1.02) 3.56 (0.95-9.77)
Fever (subjective) 8 (4907) 0.63 (0.50-0.74) 0.55 (0.38-0.71) 1.47 (1.26-1.71) 0.68 (0.58-0.80) 2.10 (1.48-2.87) 0.623
Chills 7 (2453) 0.55 (0.43-0.67) 0.62 (0.50-0.72) 1.44 (1.26-1.65) 0.73 (0.63-0.83) 2.00 (1.58-2.49) 0.610
Coryza and 4 (1106) 0.60 (0.40-0.77) 0.57 (0.22-0.66) 1.43 (1.11-2.00) 0.71 (0.56-0.86) 2.07 (1.31-3.13)
rhinorrhea absent
Sputum (bloody) 4 (1582) 0.13 (0.06-0.27) 0.90 (0.84-0.94) 1.33 (0.80-2.06) 0.96 (0.84-1.02) 1.41 (0.78-2.47)
Dyspnea 10 (5626) 0.63 (0.48-0.75) 0.51 (0.31-0.71) 1.30 (1.07-1.65) 0.75 (0.66-0.85) 1.75 (1.28-2.34) 0.598
Sore throat absent 3 (782) 0.60 (0.49-0.70) 0.52 (0.28-0.75) 1.29 (0.75-1.77) 0.81 (0.57-1.34) 1.78 (0.65-3.83)
Chest pain 8 (5031) 0.51 (0.33-0.69) 0.58 (0.37-0.76) 1.21 (1.05-1.42) 0.86 (0.78-0.94) 1.41 (1.13-1.74) 0.549
Headache 3 (1188) 0.65 (0.46-0.81) 0.42 (0.21-0.65) 1.19 (0.93-1.49) 0.85 (0.67-1.08) 1.35 (0.90-1.94)
Sputum (any) 6 (4441) 0.71 (0.60-0.81) 0.35 (0.21-0.51) 1.11 (0.96-1.32) 0.84 (0.63-1.11) 1.37 (0.87-2.07)
Myalgias 3 (1424) 0.49 (0.41-0.56) 0.57 (0.45-0.68) 1.10 (0.91-1.45) 0.92 (0.77-1.10) 1.26 (0.82-1.86)
Sputum (purulent) 3 (1365) 0.52 (0.35-0.70) 0.52 (0.39-0.65) 1.09 (0.90-1.26) 0.92 (0.73-1.08) 1.21 (0.83-1.71)
Cough 7 (1866) 0.88 (0.82-0.93) 0.16 (0.07-0.34) 1.07 (0.97-1.27) 0.77 (0.41-1.37) 1.57 (0.71-3.01)
Signs
Egophony 3 (1116) 0.05 (0.03-0.10) 0.99 (0.95-0.99) 6.17 (1.34-18.0) 0.96 (0.93-0.99) 6.46 (1.36-18.9)
Dullness to 7 (1932) 0.14 (0.10-0.19) 0.94 (0.88-0.97) 2.62 (1.14-5.30) 0.92 (0.87-0.98) 2.89 (1.17-5.90) NC
percussion
Confusion 4 (1596) 0.11 (0.08-0.15) 0.95 (0.92-0.97) 2.15 (1.36-3.34) 0.94 (0.90-0.98) 2.29 (1.39-3.63)
Crackles 12 (5898) 0.42 (0.32-0.52) 0.79 (0.68-0.86) 2.00 (1.54-2.58) 0.74 (0.66-0.82 2.70 (1.95-3.63) 0.611
Decreased 6 (4322) 0.25 (0.20-0.32) 0.87 (0.78-0.92) 1.96 (1.23-3.02) 0.87 (0.79-0.95) 2.29 (1.31-3.73)
breath sounds
Abnormal lung 8 (2875) 0.60 (0.40-0.78) 0.67 (0.42-0.85) 1.90 (1.26-2.91) 0.61 (0.47-0.75) 3.18 (1.83-2.08) 0.669
exam (any finding)
Rhonchi 5 (2375) 0.23 (0.16-0.32) 0.87 (0.78-0.92) 1.76 (1.26-2.41) 0.89 (0.83-0.95) 1.99 (1.35-2.81)
Toxic or ill 5 (4162) 0.42 (0.22-0.65) 0.70 (0.43-0.88) 1.46 (1.08-2.15) 0.83 (0.71-0.94) 1.77 (1.17-2.64)
appearance
Pleural rub 5 (1885) 0.07 (0.04-0.11) 0.97 (0.91-0.992) 3.02 (0.74-8.02) 0.96 (0.91-1.02) 3.20 (0.72-8.81)
Wheeze (any) 8 (2519) 0.25 (0.19-0.32) 0.75 (0.68-0.92) 1.00 (0.82-1.22) 1.00 (0.94-1.07) 1.00 (0.77-1.30)
Vital signs
Temp>=37.7-38.0 10 (5490) 0.34 (0.25-0.56) 0.87 (0.79-0.92) 2.52 (2.02-3.20) 0.77 (0.70-0.83) 3.30 (2.60-4.16) 0.637
O2 saturation < 95% 3 (1089) 0.36 (0.22-0.53) 0.83 (0.78-0.87) 2.12 (1.47-2.71) 0.77 (0.61-0.92) 2.83 (1.61-4.39)
Heart rate> 100 bpm 8 (5172) 0.33 (0.23-0.44) 0.84 (0.74-0.90) 2.04 (1.59-2.62) 0.80 (0.73-0.86) 2.55 (1.93-3.31) 0.606
Respiratory 3 (3638) 0.53 (0.25-0.79) 0.84 (0.44-0.91) 2.02 (1.34-3.02) 0.65 (0.45-0.84) 3.14 (2.08-4.51)
rate> 20-25 bpm
Any abnormal 3 (604) 0.93 (0.74-0.98) 0.30 (0.12-0.59) 1.37 (1.10-1.84) 0.25 (0.11-0.48) 6.01 (3.03-10.6)
vital sign

Where the positive likelihood ratio (LR+), negative likelihood ratio (LR ) or diagnostic odds ratio differed significantly from 1.0, the value is
shown in bold face.
NC, not calculable from data; AUROCC = area under the receiver operating characteristic curve.
550 Ebell et al. • SIGNS AND SYMPTOMS TO DIAGNOSE CAP

Overall clinical impression Any fever

1.0
1.0

0.8
0.8

0.6
0.6

Sensitivity
Sensitivity

0.4
0.4

0.2
0.2

Acute RTI Measured fever > 37.7−38.0 C

Referred for CXR Subjective fever

0.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

False Positive Rate False Positive Rate

Figure 2. Summary receiver operating characteristic curve for the Figure 3. Summary receiver operating characteristic curve for sub-
overall clinical impression. CXR = chest radiograph; RTI = respira- jective fever versus measured temperature> 37.7 to 38.0°C
tory tract infection

physicians using a lower threshold for diagnosing CAP 95% CI = 1.1 to 1.8; LR = 0.25, 95% CI = 0.11
to achieve a higher sensitivity, so appropriate patients to 0.48). Thus, normal vital signs provide reassurance
can be referred to the ED for further evaluation. that CAP is less likely, and we showed in a previous
While guidelines recommend a CXR to confirm the systematic review that the combination of normal vital
diagnosis of CAP before initiating therapy,21 not all signs and normal lung examination has a negative
patients with acute cough should receive a CXR. Also, likelihood ratio of 0.1 for CAP.5 The combination of
CXRs may not be readily available outside of the ED normal vital signs and a normal lung examination
setting or in low resource settings. The presence of an would reduce the likelihood of CAP to approximately
overall clinical impression suggesting CAP (LR+ = 0.5% given a prevalence of 5%, 1% given a prevalence
6.3, 95% CI = 3.6 to 10.5), egophony (LR+ = 6.2, of 10%, and 2% given a prevalence of 20%, obviating
95% CI = 1.3 to 18.0), dullness to percussion the need for a chest x-ray in most patients. On the
(LR+ = 2.6, 95% CI = 1.1 to 5.3), and measured other hand, an abnormal overall clinical impression or
fever (LR+ = 2.5, 95% CI = 2.0 to 3.2) were all mod- the presence of egophony would increase the likeli-
erately useful for increasing the likelihood of CAP and hood of CAP to 25% given a prevalence of 5%, 36%
could prompt a clinician to order a radiograph in a given a prevalence of 10%, and 56% given a probabil-
patient with acute cough. Only a single clinical finding ity of 20%, situations in which a chest x-ray (and pos-
had a LR– less than 0.5 (absence of any abnormal sibly empiric therapy) would be appropriate for most
vital sign), while the past medical history and comor- patients. The excellent test characteristics of the overall
bidities were of relatively little diagnostic value. clinical impression mean that experienced ED and pri-
Combinations of symptoms were not generally stud- mary care physicians can trust their overall judgment
ied, other than any abnormal lung finding (LR+ = of the likelihood of pneumonia and value it is a diag-
1.9, 95% CI = 1.3 to 2.9; LR = 0.61, 95% CI = nostic test. It is also an important message for physi-
0.47 to 0.75) and any abnormal vital sign (LR+ = 1.4, cians that we should not rely too much on the
ACADEMIC EMERGENCY MEDICINE • July 2020, Vol. 27, No. 7 • www.aemj.org 551

absence of individual physical findings such as egoph- clinicians rather than trainees. How to best teach this
ony, dullness to percussion, crackles, decreased breath skill of “clinical gestalt,” how many exposures to
sounds, or rhonchi, which all have summary estimates patients with acute respiratory illness are needed to
of the LR– between 0.74 and 0.96. develop it, and how to best integrate it with other
Knowing how to best use signs and symptoms can information remain to be determined. The same ques-
help physicians avoid inappropriate antibiotic use. tions apply to egophony, which had the highest
Those with normal vitals and a normal lung examina- LR + but which not all physicians may be comfortable
tion (in the absence of other bacterial infections such eliciting. We also concluded that patients with normal
as streptococcal pharyngitis or acute otitis media, lung findings and normal vital signs are very unlikely
which are easily ruled out) should not receive antibi- to have CAP (LR = 0.1).5 Finally, CRP is moder-
otics. Knowing that the likelihood of CAP is extremely ately accurate for the diagnosis of CAP (AUROCC =
low can bolster the confidence of physicians not to 0.82; M.H. Ebell, submitted for publication) and has
prescribe an antibiotic. For those at increased risk of also been shown to be a tool that can reduce inappro-
CAP based on the overall clinical impression or the priate antibiotic use.24–26
presence of one or more signs, a negative CXR can
again provide confidence not to prescribe antibiotics.
STRENGTHS AND LIMITATIONS
By targeting CXR, we also avoid its overuse.
An open question is the degree of statistical inde- Strengths of the current study include a comprehen-
pendence of individual signs, symptom, vital signs, sive literature search, the generally good methodologic
and the CRP. In a previous study by van Vugt and quality of included studies, and the use of contempo-
colleagues,22 the presence of crackles, diminished rary bivariate meta-analysis. Limitations include hetero-
vesicular breathing, tachycardia, fever, the absence of geneity in clinical settings and countries, differences in
rhinorrhea, and elevated CRP were all independent the inclusion criteria, and a failure to define what is
predictors of CAP, suggesting that CRP provides diag- abnormal for a sign or symptom. In addition, some
nostic value in addition to that of the physical exami- signs and symptoms such as the overall clinical
nation. impression had likelihood ratios with relatively wide
Van Vugt et al.22 also identified several clinical pre- CIs (LR+ = 6.3, 95% CI = 3.6 to 10.5). While there
diction rules for the diagnosis of CAP, but none per- was a fairly broad range of prevalence of CAP in the
formed particularly well as measured by the included studies, this should not impact sensitivity,
AUROCC either in the study population in van Vugt specificity, or LRs, which are characteristics of the test.
et al. or in a small validation study by Graffelman Finally, all studies used chest radiography as the refer-
et al.23 Van Vugt and colleagues22 have proposed their ence standard, which is imperfect. In one study of
own clinical prediction rule based on the largest study 2,251 patients who received both CXR and CT, 97%
to date, but it has yet to be prospectively validated. It of patients had pneumonia diagnosed on both studies,
would also be worth exploring novel modeling strate- and only 3% had pneumonia only seen on CT.27
gies such as artificial neural networks or fast and fru-
gal trees, as well as a two-stage process for clinical
CONCLUSION
diagnosis. For example, those with normal vital signs
and normal examination can be excluded at stage 1, In conclusion, while the history and physical examina-
in stage 2 a clinical prediction rule used to identify tion is important, only a few key signs and symptoms
those at high risk for CAP who should all undergo significantly change the underlying likelihood of com-
CXR, and in a moderate-risk group where clinicians munity-acquired pneumonia. The probability of com-
would use their judgement and other sources of infor- munity-acquired pneumonia is appreciably increased
mation. by an overall clinical impression suggesting commu-
In the past 2 years, our group has now performed nity-acquired pneumonia, egophony, dullness to per-
a set of four related systematic reviews on the diagno- cussion, and measured temperature, while it is
sis of CAP. We conclude that the overall clinical significantly decreased by the absence of abnormal
impression is a valuable diagnostic tool, with accuracy vital signs or (from a previous study) the combination
similar to that of clinical prediction rules.6 Most of of abnormal vital signs and a normal lung examina-
the studies in that review included experienced tion.5 Clinical education should focus on teaching
552 Ebell et al. • SIGNS AND SYMPTOMS TO DIAGNOSE CAP

high-value elements of the examination such as egoph- 11. Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt
ony or dullness on percussion and on providing suffi- PM, Zwinderman AH. Bivariate analysis of sensitivity
cient clinical examples of acute cough to hone the and specificity produces informative summary mea-
overall clinical impression. Future research should be sures in diagnostic reviews. J Clin Epidemiol
2005;58:982–90.
performed to validate promising clinical prediction
12. Melbye H, Straume B, Aasebo U, Brox J. The diagnosis
rules and to integrate signs, symptoms, and point-of-
of adult pneumonia in general practice. The diagnostic
care tests such as C-reactive protein and to explore
value of history, physical examination and some blood
novel approaches to the development and validation tests. Scand J Prim Heal Care 1988;6:111–7.
of these rules. 13. Moberg AB, Taleus U, Garvin P, Fransson SG, Falk M.
Community-acquired pneumonia in primary care: clinical
assessment and the usability of chest radiography. Scand J
References
Prim Heal Care 2016;34:21–7.
1. Antimicrobial Resistance Posing Growing Health Threat. 14. Doebler P. mada: Meta-Analysis of Diagnostic Accuracy. R
Available at: http://www.cdc.gov/media/releases/2011/ package version 0.5.8. 2017. Available at: https://cran.r-
p0407_antimicrobialresistance.html. Accessed Mar 30, project.org/package=mada. Accessed Mar 30, 2020.
2020. 15. Melbye H, Straume B, Aasebøs U, Dale K. Diagnosis of
2. Fleming-Dutra KE, Hersh AL, Shapiro DJ, et al. Preva- pneumonia in adults in general practice relative impor-
lence of inappropriate antibiotic prescriptions among US tance of typical symptoms and abnormal chest signs evalu-
ambulatory care visits, 2010–2011. JAMA ated against a radiographic reference standard. Scand J
2016;315:1864–73. Prim Health Care 1992;10:226–33.
3. Joint Commission. Antimicrobial Stewardship Tools. 16. Moore M, Stuart B, Little P, et al. Predictors of pneumo-
2015. Available at: http://www.jointcommission.org/topic nia in lower respiratory tract infections: 3C prospective
s/hai_portal_external_resources_antimicrobial_stewa cough complication cohort study. Eur Respir J 2017;50:
rdship.aspx. Accessed Mar 30, 2020. pii:1700434.
4. Little P, Rumsby K, Kelly J, et al. Information leaflet and 17. Singal BM, Hedges JR, Radack KL. Decision rules and
antibiotic prescribing strategies for acute lower respiratory clinical prediction of pneumonia: evaluation of low-yield
tract infection: a randomized controlled trial. J Am Med criteria. Ann Emerg Med 1989;18:13–20.
Assoc 2005;293:3029–35. 18. Tape TG, Heckerling PS, Ornato JP, Wigton RS. Use of
5. Marchello CS, Ebell MH, Dale AP, Harvill ET, Shen Y, clinical judgment analysis to explain regional variations in
Whalen CC. Signs and symptoms that rule out commu- physicians’ accuracies in diagnosing pneumonia. Med
nity-acquired pneumonia in outpatient adults: a systematic Decis Mak 1991;11:189–97.
review and meta-analysis. J Am Board Fam Med 19. Gennis P, Gallagher J, Falvo C, Baker S, Than W. Clini-
2019;32:234–47. cal criteria for the detection of pneumonia in adults:
6. Dale AP, Marchello C, Ebell MH. The accuracy of the guidelines for ordering chest roentgenograms in the emer-
overall clinical impression for the diagnosis of pharyngitis, gency department. J Emerg Med 1989;7:263–8.
sinusitis, and pneumonia: a meta-analysis. Br J Gen Pract 20. Dale AP, Marchello C, Ebell MH. Clinical gestalt to diag-
2019;69:e444–53. nose pneumonia, sinusitis, and pharyngitis: a meta-analy-
7. Ebell MH, Bentivegna M, Cai X, Hulme C, Kearney M. sis. Br J Gen Pract 2019;69:e444–53.
Accuracy of biomarkers for the diagnosis of adult commu- 21. Metlay JP, Waterer GW, Long AC, et al. Diagnosis and
nityacquired pneumonia: a meta-analysis. Acad Emerg treatment of adults with community-acquired pneumonia.
Med 2020; 27:195–206. An official clinical practice guideline of the American Tho-
8. Metlay JP, Kapoor WN, Fine MJ. Does this patient have racic Society and infectious diseases Society of America.
community-acquired pneumonia? Diagnosing pneumonia Am J Respir Crit Care Med. 2019;200:e45–67.
by history and physical examination. JAMA 1997;278: 22. van Vugt SF, Broekhuizen BD, Lammens C, et al. Use of
1440–5. serum C reactive protein and procalcitonin concentrations
9. M€ uller B, Harbarth S, Stolz D, et al. Diagnostic and prog- in addition to symptoms and signs to predict pneumonia
nostic accuracy of clinical and laboratory parameters in in patients presenting to primary care with acute cough:
community-acquired pneumonia. BMC Infect Dis diagnostic study. BMJ 2013; 346:f2450.
2007;7:10. 23. Graffelman AW, Le Cessie S, Knuistingh Neven A, Will-
10. Whiting PF, Rutjes AW, Westwood ME, et al. QUADAS- emssen FE, Zonderland HM, Van Den Broek PJ. Can
2: a revised tool for the quality assessment of diagnostic history and exam alone reliably predict pneumonia? J Fam
accuracy studies. Ann Intern Med 2011;155:529–36. Pract 2007;56:465–70.
ACADEMIC EMERGENCY MEDICINE • July 2020, Vol. 27, No. 7 • www.aemj.org 553

24. Aabenhus R, Jensen JU, Jorgensen KJ, Hrobjartsson A, 31. Steurer J, Held U, Spaar A, et al. A decision aid to rule
Bjerrum L. Biomarkers as point-of-care tests to guide pre- out pneumonia and reduce unnecessary prescriptions of
scription of antibiotics in patients with acute respiratory antibiotics in primary care patients with cough and fever.
infections in primary care. Cochrane Database Syst Rev BMC Med 2011;9:56.
2014;(6):CD010130. 32. Wipf JE, Lipsky BA, Hirschmann JV, et al. Diagnosing
25. Cals JW, Ebell MH. C-reactive protein: guiding antibiotic pneumonia by physical examination: relevant or relic?
prescribing decisions at the point of care. Br J Gen Pract Arch Intern Med 1999;159:1082–7.
2018;68:112–3. 33. Diehr P, Wood RW, Bushyhead J, Krueger L, Wolcott B,
26. Cals JW, Schot MJ, de Jong SA, Dinant GJ, Hopstaken Tompkins RK. Prediction of pneumonia in outpatients
RM. Point-of-care C-reactive protein testing and antibiotic with acute cough-A statistical approach. J Chronic Dis
prescribing for respiratory tract infections: a randomized 1984;37:215–25.
controlled trial. Ann Fam Med 2010;8:124–33. 34. Gonzalez Ortiz MA, Carnicero Bujarrabal M, Varela
27. Upchurch CP, Grijalva CG, Wunderink RG, et al. Com- Entrecanales M. Prediction of the presence of pneumonia
munity-acquired pneumonia visualized on CT scans but in adults with fever. Med Clin (Barc) 1995;105:521–4.
not chest radiographs: pathogens, severity, and clinical out- 35. Sald
ıas PF, D
ıaz FA, Cabrera TD, Gederlini GA, De
comes. Chest 2018;153:601–10. Solminihac LI, Hern
andez AP. Predictive value of history
28. Holm A, Nexoe J, Bistrup LA, et al. Aetiology and predic- and physical examination for the diagnosis of community-
tion of pneumonia in lower respiratory tract infection in acquired pneumonia in adults. Rev Med Chil
primary care. Br J Gen Pract 2007;57:547–54. 2007;135:143–52.
29. Hopstaken RM, Muris JW, Knottnerus JA, Kester AD,
Rinkens PE, Dinant GJ. Contributions of symptoms,
signs, erythrocyte sedimentation rate, and C-reactive pro- Supporting Information
tein to a diagnosis of pneumonia in acute lower respira- The following supporting information is available in
tory tract infection. Br J Gen Pract 2003;53:358–64. the online version of this paper available at http://
30. Lieberman D, Shvartzman P, Korsonsky I, Lieberman D.
onlinelibrary.wiley.com/doi/10.1111/acem.13965/full
Diagnosis of ambulatory community-acquired pneumonia.
Data Supplement S1. Supplementary material.
Comparison of clinical assessment versus chest X-ray.
Scand J Prim Heal Care 2003;21:57–60.