CARDIOVASCULAR JOURNAL OF AFRICA • Vol 21, No 5, September/October 2010
AFRICA
275
Two previous publications
10,11
investigated estimation of left
atrial size by measuring the carinal angle and comparing that
to echo findings. In the first study, 35 patients with enlarged
left atria (
>
4.5 cm), and 35 paired, age-matched patients with
normal atria (
<
4.0 cm) were selected as determined by echo.
Interbronchial angle on chest radiographs (standard and supine
portable films) was measured by a blinded observer using a goni-
ometer. Left atrial size could be accurately predicted to be larger
than 5.0 cm in diameter if the carinal angle was greater than 100°
(
r
=
0.746 with
p
<
0.001), but the sample size of 63 patients may
have been too small.
In the second study, the postero-anterior chest radiographs
and echocardiographs of 108 clinically stable patients (53 men
and 55 women) were respectively reviewed. Left atrial size
correlated poorly with both interbronchial angle (
r
=
0.33) and
sub-carinal angle (
r
=
0.25). An interbronchial angle of 76.4° and
a sub-carinal angle of 65.4° were the best discriminators between
patients with normal and those with enlarged left atrial dimen-
sions (sensitivities: 63 and 51%, specificities: 63 and 66%, for
interbronchial angle and sub-carinal angle, respectively) but no
intra- or inter-observer variation were tested.
The aim of this study was to investigate the precision and
accuracy of chest radiographs to predict left atrial enlargement in
comparison to measurement by echocardiography.
Methods
The study was conducted at Steve Biko Academic Hospital, a
regional tertiary referral centre. The study was a cross-sectional
retrospective analysis.
Patients older than 18 years admitted to the hospital between
January 2000 and December 2003 who had had both echocar-
diography and chest radiography performed during the same
admission were included in the analysis. The sampling frame
was determined by accessing records at the ultrasound depart-
ment and finding all those patients who had had echocardiogra-
phy during the specified period. All radiographs of the identified
patients were retrieved and assessed according to quality criteria,
and only those patients with adequate radiographs were included
in the final analysis.
Demographic data were collected on all patients. Left atrial
size was determined with two-dimensional targeted M-mode
echocardiography using a Vivid 3 General Electric ultrasound
machine. Left atrial size was determined according to the
American Society of Echocardiography (ASE) criteria.
12
In determining left atrial size, the maximal dimension was
measured from the parasternal long-axis view between the lead-
ing edge of the posterior aortic wall and the leading edge of the
posterior wall of the left atria at end-systole. Left atrial size may
be underestimated in the parasternal long-axis view because this
chamber may enlarge longitudinally.
13
Therefore, left atrial size
was measured from two optical orthogonal views (four-chamber
and two-chamber) as well, from the tip of the mitral valve to the
posterior wall of the left atrium at end-systole, and the larger of
the two values was used in the analysis.
Two independent observers assessed chest radiographs twice.
Duplicate radiographs were removed from the analysis so that
each cardiac ultrasound was matched with one radiograph (either
supine or erect). Radiographs were read on a radiographic view-
ing box. The angle of divergence (
α
) of the first few centimetres
of the inferior main-stem bronchial borders was measured using
a protractor. The sub-angle distance (SAD)
x
(in mm) on the side
opposite to the sub-carinal angle (SCA)
α
(alpha, in degrees) was
measured 20 mm from the sub-carinal angle along the medial
borders of the bronchi (
y
) using a tape measure.
The sample size calculation was performed using the program
nQuery Advisor
®
. With
α
and power set at 5% and 90%, respec-
tively,
δ
(expected proportion of subjects with enlarged left
atrial size) estimated at 70%, K
0
(hypothetical perfect agree-
ment between two methods) chosen as 90%, and K
1
(expected
agreement between two methods) as 75%, the sample size was
estimated at 106. Allowing for 10% of patients with echocar-
diograms not having traceable chest radiographs, and expecting
20% of carinal angles not to be clearly visible on chest radio-
graphs, an initial sample of 150 was determined.
Statistical analysis
Intra- and inter-observer agreement for the SCA and SAD were
performed using Lin’s concordance correlation coefficient
14
limits of agreement (Bland and Altman methodology
15
and Bland
and Altman plots). Logistic regression was used to determine the
predictive value of the SCA and SAD for both erect and supine
radiographs. First, a linear model was determined. This was
followed by assessment of the fit of the model and its perform-
ance characteristics. Finally, to assess whether the dependent
variable was linear in the logit, three methods, as proposed by
Hosmer and Lemeshow,
16
were used: lowess (locally weighted
least squares) smoothing curves, design variables and fractional
polynomials.
Cut-off values for the SCA or SAD that resulted in the best
compromise between sensitivity and specificity were determined
using receiver operating characteristic (ROC) curves.
17
P
-values
<
0.05 were regarded as statistically significant for all compari-
sons. All calculations were performed using Intercooled Stata
version 8.2.
Results
Echocardiography and radiography data were available on 159
patients, 74 males and 85 females. The mean age of the study
sample was 59 years (range 18–88 years). Five echocardiograms
were incomplete as left atrial size was not measured, and there-
fore only 154 echocardiograms were included in the logistic
regression analysis. As several patients had more than one chest
radiograph taken, 178 chest radiographs were available for
determination of intra- and inter-observer variability. Table 1
describes the radiographic quality of the X-rays used to deter-
mine left atrial size.
Agreement of both SCA and SAD using observer 1 (intra-
observer agreement) was near perfect, with Lin’s correlation
coefficient ranging from 0.98 to 0.99 (
p
=
0.000) across all
categories of radiographs (erect and supine, good, fair and poor
quality). Agreement of both SCA and SAD using observer 2
TABLE 1. RADIOGRAPH CHARACTERISTICS
Chest radiographs
Good
Fair
Poor
Total
Supine
43
1
3
47
Erect
115
10
6
131
Total
158
11
9
178
