Representative phonocardiographic recordings from the V4 position
and electrocardiographic (ECG) recordings from the V6 position
from a patient without a diastolic heart sound (normal) and from a patient
with both a third and fourth heart sound (abnormal).
The box-and-whisker plots show the median and interquartile range (box),
the upper hinge (75th percentile plus 1.5 times the interquartile range) (upper
error bars), the lower hinge (25th percentile plus 1.5 times the interquartile
range) (lower error bars), and outlier values (circles) outside the hinges.
The P values represent comparisons with the group
without diastolic heart sounds. There were no other statistically significant
A normal left ventricular end-diastolic pressure (LVEDP) was defined
as 15 mm Hg or less and a normal left ventricular ejection fraction (LVEF)
was defined as 50% or more. Marked portions of the curves indicate the sensitivity
and specificity when using the given confidence cutoff for an S3 or
S4. AUROC indicates r statistic for the area under the receiver
operating characteristic curve; CI, confidence interval.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Marcus GM, Gerber IL, McKeown BH, et al. Association Between Phonocardiographic Third and Fourth Heart Sounds and Objective Measures of Left Ventricular Function. JAMA. 2005;293(18):2238–2244. doi:10.1001/jama.293.18.2238
Context The third (S3) and fourth (S4) heart sounds detected
by phonocardiography are considered to represent the criterion standards of
the gallop sounds, but their test characteristics have not been explored.
Objective To determine the diagnostic test characteristics of the S3 and
S4 for prediction of left ventricular dysfunction using a computerized
heart sound detection algorithm.
Design, Setting, and Participants Prospective study of 90 adult patients undergoing elective left-sided
heart catheterization at a single US teaching hospital between August 2003
and June 2004. The mean age was 62 (SD, 13) years (range, 24-90 years) and
61 (68%) were male. Within a 4-hour period, participants underwent computerized
heart sound phonocardiographic analysis, cardiac catheterization, transthoracic
echocardiography, and blood sampling for assessment of an S3/S4, left ventricular end-diastolic pressure (LVEDP), left ventricular
ejection fraction (LVEF), and B-type natriuretic peptide (BNP), respectively.
Main Outcome Measures Diagnostic test characteristics of the computerized phonocardiographic
S3 and S4 using markers of left ventricular function
as criterion standards.
Results Mean (SD) LVEDP was significantly elevated (18.4 [6.9] mm Hg vs 12.1
[7.3] mm Hg; P<.001), mean (SD) LVEF was reduced
(49.4% [20.2%] vs 63.6% [14.8%]; P<.001), and
median (interquartile range) BNP was elevated (330 [98-1155] pg/mL vs 86 [41-192]
pg/mL; P<.001) in those with an S3,
S4, or both compared with patients without a diastolic heart sound.
The sensitivities of these heart sounds to detect an elevated LVEDP, reduced
LVEF, or elevated BNP were 41%, 52%, and 32% for an S3, and 46%,
43%, and 40% for an S4, respectively. For abnormal levels of the
same markers of ventricular function, the specificities of the S3 were
92%, 87%, and 92%, while the specificities of the S4 were 80%,
72%, and 78%, respectively.
Conclusions Neither the phonocardiographic S3 nor the S4 is
a sensitive marker of left ventricular dysfunction. The phonocardiographic
S3 is specific for left ventricular dysfunction and appears to
be superior to the moderate specificity of the phonocardiographic S4.
Time-honored bedside auscultation with the stethoscope is an important
clinical tool to help characterize heart sounds and identify abnormalities
associated with cardiac dysfunction. Potain1 first
described the gallop rhythm in 1880, referring to the cadence produced by
presence of abnormal diastolic cardiac sounds. These diastolic sounds refer
to the presence of a third heard sound (S3) and/or fourth heart
sound (S4). The auscultated S3 and S4 have
long been used as clinical signs of heart disease, with diagnostic and prognostic
the value of these physical findings has been diminished by reports of poor
accuracy and interobserver reliability.8,9
As an objective instrument that supplies measurable data, the phonocardiogram
has traditionally been the criterion (gold) standard tool for the detection
of ventricular gallop sounds. Phonocardiography has been used to understand
the mechanisms and associated clinical characteristics of diastolic sounds,10-13 and
results of phonocardiography have been used to determine the accuracy of physician
auscultation.9 However, the test characteristics
of the phonocardiogram to detect abnormal left ventricular function have not
We undertook a prospective study to correlate the presence of S3 and S4 heart sounds detected by computerized phonocardiography
with invasive and noninvasive objective markers of left ventricular systolic
and diastolic function in patients undergoing diagnostic cardiac catheterization.
Adult patients referred for nonemergent left-sided heart catheterization
at the University of California San Francisco (UCSF) Medical Center were eligible
for enrollment between August 2003 and June 2004. Exclusion criteria included
age younger than 18 years, systolic blood pressure less than 90 mm Hg, intravenous
vasopressor, inotropic, or vasodilator pharmacotherapy, cardiac rhythm other
than a sinus or paced atrial rhythm, severe mitral regurgitation or stenosis,
constrictive pericarditis, serum creatinine level of 4.0 mg/dL (354 μmol/L)
or higher, severe pulmonary hypertension, and mechanical ventilation.
From systematic review of the clinical chart, patients’ primary
diagnoses and significant comorbidities were recorded, including coronary
artery disease (defined as ≥1 coronary artery with ≥70% diameter stenosis),
systemic hypertension, clinical heart failure, aortic stenosis, mitral regurgitation,
chronic renal insufficiency, hypertrophic obstructive cardiomyopathy, and
chronic obstructive pulmonary disease. Study enrollment was prospectively
set at 100 participants using cross-sectional sampling. All patients gave
written informed consent prior to enrollment and the protocol was approved
by the UCSF Committee on Human Research.
Blood was drawn from the arterial sheath for measurement of B-type natriuretic
peptide (BNP) level using a membrane immunofluorescence assay (Biosite Inc,
San Diego, Calif). Within 4 hours, cardiac catheterization, transthoracic
echocardiography, and computerized heart sound phonocardiographic analysis
Prior to angiography, left ventricular pressure was recorded. A BNP
level higher than 100 pg/mL was prospectively specified as abnormal.14
Patients underwent recording of left ventricular end-diastolic pressure
(LVEDP) using a 6F pigtail catheter and a fluid-filled pressure transducer.
Pressure was recorded using a 50–mm Hg scale at 50 mm/s paper speed.
A physician, blinded to all clinical and diagnostic testing data, measured
the post–A wave pressure. A minimum of 5 consecutive cardiac cycles
were used to measure mean LVEDP. An LVEDP higher than 15 mm Hg was prospectively
specified as abnormal.15,16
Transthoracic echocardiographic data were obtained by an experienced
echocardiographer (Acuson Sequoia, Mountain View, Calif, or SONOS 5500, Philips
Medical Systems, Andover, Mass). Echocardiographic contrast (Optison, Amersham,
Little Chalfont, England; 0.3-0.5 mL injected into a peripheral vein) was
administered when required to improve endocardial border detection and enhance
Doppler signals. Echocardiographic data were stored on magneto-optical disks
and analyzed off-line by a single experienced reader blinded to any clinical
or study data. The average of 3 measurements was used for the analysis. End-diastolic
and end-systolic volumes were calculated using the biplane method of discs17 and were then indexed to body surface area. These
volumes were used to calculate left ventricular ejection fraction (LVEF).
An LVEF less than 50% was prospectively defined as abnormal.
A 3-minute audioelectrocardiographic tracing (Audicor, Inovise Medical
Inc, Portland, Ore) was obtained (Figure 1).
Audioelectrocardiographic leads were attached to the V3 and V4 positions and connected to a Marquette MAC 5000 (General Electric
Healthcare Technologies, Waukesha, Wis). The audioelectrocardiographic data
were stored electronically to a compact disc. A 10-second segment free of
artifact was selected off-line by a technician, blinded to all clinical and
diagnostic testing data, for a computer-generated report regarding presence
of an S3/S4 heart sound. The Audicor software develops
a confidence score between 0 and 1.0 for each heart sound based on the intensity,
persistence, and frequency content, with a value of at least 0.5 indicating
the presence of a diastolic heart sound.
Data are presented as mean values and standard deviations for normally
distributed continuous variables. Because BNP was highly right-skewed, this
variable is presented as median and interquartile range (IQR) and analyses
were performed on log-transformed values. Categorical data are presented as
exact numbers and proportions. Sensitivity, specificity, and receiver operating
characteristic (ROC) curves were calculated for S3 and S4 confidence scores as the predictors of elevated LVEDP, reduced LVEF,
and elevated BNP using the predefined cutoffs. Continuous variables were compared
using t tests and analysis of variance where appropriate.
Categorical variables were compared using the Fisher exact test. All analyses
were performed using STATA, version 8.2 (Stata Corp, College Station, Tex).
A 2-tailed P<.05 was considered significant.
One hundred patients were enrolled. Eight patients were excluded because
of poor phonocardiographic sound quality. Because the phonocardiographic software
cannot assess paced rhythms, 2 additional patients were excluded, leaving
90 patients for analysis.
The mean age was 62 (13) years (range, 24-90 years), and 61 (68%) were
male. Twenty-six (29%) had diabetes, 72 (80%) had systemic hypertension, 32
(36%) had a clinical diagnosis of heart failure, and 16 (18%) were hospitalized
for an acute coronary syndrome. Sixty-four patients (71%) had angiographic
evidence of coronary artery disease, 40 (44%) had a prior percutaneous coronary
intervention, and 17 (19%) had prior coronary artery bypass graft surgery.
Five (6%) had moderate to severe calcific aortic stenosis and 2 (2%) had severe
hypertrophic obstructive cardiomyopathy. The mean (SD) body surface area was
1.91 (0.26) m2, the mean (SD) body mass index (calculated as weight
in kilograms divided by the square of height in meters) was 29.0 (8.1), and
the mean (SD) creatinine level was 1.47 (1.32) mg/dL (130  μmol/L)
(>1.5 mg/dL [>133 μmol/L] in 14 [16%]).
Computerized heart sound analysis detected no extra heart sound in 49
patients (54%), an S3 only in 12 patients (13%), an S4 only
in 20 patients (22%), and both an S3 and an S4 in 9
patients (10%), so that there was an S3 and/or an S4 in
Mean (SD) heart rate was 69/min (12/min). Mean (SD) central aortic pressure
was 131 (26) mm Hg for systolic pressure and 66 (13) mm Hg for diastolic pressure.
Mean (SD) LVEDP was 15.0 (7.8) mm Hg (range, 1-31 mm Hg). Forty-one patients
(46%) had an abnormal LVEDP (>15 mm Hg).
Seventy-nine patients had an adequate echocardiographic assessment of
LVEF. Mean (SD) LVEF was 57% (19%) (range, 7%-85%). Twenty-three patients
(28%) had an abnormal LVEF (<50%).
The median BNP was 133 pg/mL (IQR, 59-394 pg/mL; range, 5-4490 pg/mL;
n = 89). Fifty-two patients (58%) had an abnormal BNP (>100 pg/mL).
There was no difference with respect to age, sex, or diabetes regarding
the presence of diastolic heart sounds (Table
1). Patients with an isolated S4 tended to have a higher
prevalence of coronary artery disease compared with those without a diastolic
heart sound or those with an S3 alone. One third of patients with
both an S3 and an S4 had an acute coronary syndrome,
while only 6% of patients without a diastolic heart sound were undergoing
cardiac catheterization for an acute coronary syndrome. The prevalence of
systemic hypertension was highest in patients with an S3 or an
S4 compared with patients without a diastolic heart sound. A clinical
diagnosis of heart failure was significantly higher in patients with an S3 alone and those with both an S3 and an S4 compared
with those with neither.
Patients with an S3 or S4 had a significantly
higher LVEDP compared with patients without a diastolic heart sound
(Table 2). For patients with both an S3 and
an S4, the mean (SD) LVEDP was significantly elevated at 19.8 (5.1)
mm Hg (Figure 2A). The differences in
LVEDP between patients with an S3 alone, an S4 alone,
or both were not statistically significant.
Patients with an LVEDP of more than 15 mm Hg had a higher prevalence
of an S3 (41% vs 8%; P<.001) and an
S4 (46% vs 20%; P = .01), a
lower mean (SD) LVEF (50.4% [21.4%] vs 61.8% [14.6%]; P = .006), and a higher BNP (median, 329 pg/mL [IQR, 172-1002
pg/mL] vs 75 pg/mL [IQR, 39-133 pg/mL]; P<.001)
than those with a normal LVEDP.
The LVEF was significantly lower for patients with an S3 alone,
an S4 alone, and those with both an S3 and an S4 compared with those without a diastolic heart sound (Table 2 and Figure 2B). Patients
with an LVEF less than 50% had a higher prevalence of an S3 (52%
vs 13%; P<.001) and an S4 (43% vs 28%; P = .20), a higher mean (SD) LVEDP (19.1 [7.5]
mm Hg vs 13.7 [7.3] mm Hg; P = .003), and
a higher BNP (median, 391 pg/mL [IQR, 151-1350 pg/mL] vs 96 pg/mL [IQR, 44-271
pg/mL]; P = .01) than those with a normal
LVEF. The BNP level tended to progressively increase comparing patients without
a diastolic heart sound, those with an S4 alone, those with an
S3 alone, and those with both an S3 and an S4 (Table 2).
In patients with an abnormal LVEDP, defined as more than 15 mm Hg, the
presence of a very soft S3 (defined as a confidence score of 0-0.25)
had a sensitivity of 54% and a specificity of 82%, while the clinically used
confidence score of 0.50 had a sensitivity of 41% and a specificity of 94%
(Figure 3 and Table 3). The S4 yielded a sensitivity of 46% with a
specificity of 80%. The area under the ROC curve (AUROC) for the S3 confidence
score was 0.76 (95% confidence interval [CI], 0.63-0.88; Figure 3A). The test performance for the S4 confidence
score was slightly lower at 0.68 (95% CI, 0.57-0.79).
The test characteristics were similar based on an abnormal LVEF, defined
as less than 50% (Table 3): both the
S3 and the S4 had low sensitivities (52% and 43%, respectively),
and the specificity of the S3 (87%) was superior to that of the
S4 (72%). The AUROC for the S3 confidence score was
0.73 (95% CI, 0.62-0.84) and for the S4 confidence score was 0.62
(95% CI, 0.46-0.78) (Figure 3B).
The test characteristics of the phonocardiographic S3 and
S4 using a BNP level of more than 100 pg/mL as the criterion standard
for ventricular dysfunction also demonstrated low sensitivities (32% and 40%,
respectively) and high specificities (92% and 78%, respectively; Table 3). The AUROCs for the S3 and
S4 confidence scores as predictors of an elevated BNP were 0.65
(95% CI, 0.54-0.76) and 0.62 (95% CI, 0.50-0.73), respectively. The test characteristics
for any diastolic heart sound (S3, S4, or both) had
higher sensitivities but lower specificities, with no significant improvement
in overall diagnostic accuracy (Table 3).
In our prospective study of 90 patients referred for elective cardiac
catheterization, diastolic heart sounds as assessed by computerized phonocardiography
had poor sensitivity for detection of left ventricular dysfunction. Although
patients with either diastolic sound detected by computerized phonocardiographic
heart sound analysis did have significantly higher LVEDP and BNP levels and
lower LVEF measurements compared with those without a diastolic heart sound,
overall diagnostic accuracy was modest, as evidenced by low AUROCs. The low
sensitivities and AUROCs we observed indicate that diastolic heart sounds
are not high-quality diagnostic tests for left ventricular dysfunction. Finally,
the specificities of each sound to detect abnormalities in the objective markers
of left ventricular function were consistently high, with the S3 proving
to be superior to the S4 in separating patients based on abnormal
left ventricular hemodynamics.
The auscultated third and fourth heart sounds
are known to be associated with abnormal left ventricular hemodynamics.4,18,19 An auscultated gallop
sound has been shown to be associated with a worse prognosis in patients undergoing
noncardiac surgery,20,21 in those
with asymptomatic left ventricular dysfunction,6 and
in those with overt heart failure.6,7
Several studies have reported test characteristics of the auscultated
S3 or S4 to detect abnormal markers of left ventricular
function, including elevated LVEDP,16 echocardiographic
measurements of left atrial filling,19 LVEF,4 and BNP levels.22 One
small, blinded study examined the test characteristics of the auscultated
S4, yielding an 84% sensitivity and 75% specificity for an elevated
atrial filling fraction in 41 patients.19 An
important limitation to the S3 studies involves the auscultators’
lack of blinding to the patients’ clinical conditions; nonetheless,
they demonstrated low sensitivities (31%-51%) and high specificities (90%-97%)
in detecting an abnormal level of the chosen hemodynamic marker.4,16,22 As
a caveat to all of these studies, the generalizability and applicability of
the true test characteristics are difficult to determine given substantial
interobserver variability and poor accuracy of physician auscultation.8,9,23
As an objective unbiased instrument that supplies measurable data, the
phonocardiogram has been viewed as the criterion standard for detection of
S3 and S4. Phonocardiography has been used to elucidate
the hemodynamic mechanisms of the S3 and S42,10-13,18,24 and
has served to provide objective evidence of the S3 and S4 in studies on auscultation.9,23 However,
to the best of our knowledge, the test characteristics of the phonocardiographic
S3 and S4 have not been previously reported.
In the present study, we compared the phonocardiographic S3 and
S4 with 3 clinically relevant objective markers of left ventricular
function: LVEDP, LVEF, and BNP levels. Clinically, an S3 alone
was associated with hypertension and heart failure, an S4 alone
was associated with coronary disease and hypertension, and the combination
of the sounds tended to have a higher prevalence in those with acute coronary
syndromes or heart failure. Sex and presence of diabetes were not related
to the prevalence of these diastolic heart sounds.
The level of each marker of left ventricular function was significantly
more abnormal in patients with an S3 alone compared with those
without an S3. Patients with an S3 alone tended to have
a lower LVEF and higher BNP compared with those with an S4 alone.
Those with an S4 alone had a significantly higher LVEDP and BNP
compared with those without an S4. Finally, patients exhibiting
both diastolic sounds had the highest LVEDP and BNP levels.
Prespecified and conventionally accepted levels of each marker of ventricular
function (LVEDP >15 mm Hg, LVEF <50%, and BNP >100 pg/mL) were prospectively
defined as abnormal, and these abnormal levels served as criterion standards
to determine the test characteristics of the phonocardiographic S3 and
S4. Both sounds demonstrated consistently low sensitivities to
detect an abnormal level of any of the markers (ranging from 32% to 52%).
This suggests that neither sound should be used to screen for the presence
of an abnormal LVEDP, LVEF, or BNP level. More generally, absence of the sounds
does not appear to rule out ventricular dysfunction.
The specificities of the S3 were consistently high, ranging
from 87% to 92%, suggesting that, if present, an S3 can be useful
to rule in a diagnosis of ventricular dysfunction. The specificities of the
S4 to detect an abnormal marker of ventricular function were consistently
lower than those of the S3 but remained moderately high (72%-80%).
The test characteristics of any diastolic heart sound (S3 and/or
S4) showed an increase in sensitivity (57%-74%) but a reduction
in specificity (64%-73%) compared with using either heart sound alone.
The test characteristics of the phonocardiogram may vary depending on
the cutoff used to designate a positive S3 or S4. Our
particular phonocardiogram was an audioelectrocardiographic tracing that created
a computer-generated confidence score for each heart sound between 0 and 1.0;
a value of at least 0.50 was prospectively specified as indicative of a diastolic
heart sound. As demonstrated by the AUROCs (Figure
3), this cutoff generally yielded the highest accuracy. Reducing
the cutoff to 0.25 or higher, which likely includes the detection of some
sounds not audible to the human ear, provided somewhat higher sensitivities
for the presence of elevated LVEDP (54% for the S3 and 56% for
the S4) but resulted in a modest reduction in specificity (82%
for the S3 and 76% for the S4).
The findings of this study should not be extrapolated to patients with
conditions known to have a high prevalence of these diastolic sounds that
would have been excluded from the study, such as severe mitral regurgitation
or constrictive pericarditis. Because an S3 can be physiologic
in younger people (particularly those <40 years old),25 our
findings should not be used to interpret the meaning of an S3 in
younger patients. A significant limitation of the study likely represents
a limitation of phonocardiography in general or perhaps any instrument that
requires high-quality data for accurate interpretation—8 patients were
excluded in our study because of poor quality of the phonocardiographic tracings.
A final limitation is that there is no single universally accepted tool for
phonocardiography; however, we did use prospectively defined cutoffs for both
the phonocardiographic tracings and hemodynamic parameters.
In conclusion, the absence of a phonocardiographic S3 or
S4 is not sufficient to exclude ventricular dysfunction. If present,
the phonocardiographic S3 and S4 are specific for an
elevated LVEDP, a depressed LVEF, and an elevated BNP level. The S3 appears
to have superior test characteristics compared with the S4 in identifying
patients with abnormal left ventricular function.
Corresponding Author: Andrew D. Michaels,
MD, Division of Cardiology, University of California San Francisco Medical
Center, 505 Parnassus Ave, Box 0124, San Francisco, CA 94143-0124 (email@example.com).
Author Contributions: Dr Michaels had full
access to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Marcus, Gerber, Vessey,
Acquisition of data: Gerber, McKeown, Jordan,
Huddleston, Foster, Michaels.
Analysis and interpretation of data: Marcus,
Gerber, McKeown, McCulloch, Michaels.
Drafting of the manuscript: Marcus, Michaels.
Critical revision of the manuscript for important
intellectual content: Marcus, Gerber, McKeown, Vessey, Jordan, Huddleston,
McCulloch, Foster, Chatterjee, Michaels.
Statistical analysis: McCulloch, Michaels.
Administrative, technical, or material support:
Study supervision: Foster, Chatterjee, Michaels.
Financial Disclosures: Dr Michaels has received
an unrestricted educational grant from Inovise Medical Inc, Portland, Ore.
None of the other authors reported disclosures.
Funding/Support: Dr Michaels is supported by
National Institutes of Health Mentored Patient-Oriented Research Career Development
Award K23 RR018319-01 A1. Training, shipment of equipment, supplies, use of
the phonocardiographic equipment, and interpretation of the tracings was provided
free of charge by Inovise Medical Inc.
Role of the Sponsor: Inovise Medical Inc was
not involved in the design of the study, data management, data analysis, or
manuscript preparation or authorship. Inovise Medical Inc was allowed to review
the manuscript. However, any decisions regarding manuscript revision were
made by the authors.
Acknowledgment: We acknowledge the patients
who participated in this study; the staff in the UCSF Cardiac Catheterization
Laboratory for their technical assistance; and Patti Arand, Nancy Forman,
and Robert Warner of Inovise Medical Inc for training and technical support.