Patients
From 1994 to 1999, eighty-six consecutive patients (with a mean age of
seventy-two years) who presented to our institution because of pain after a
total hip arthroplasty were referred to the Departments of Radiology and
Nuclear Medicine when both the history (pain in the hip during walking or
movement) and physical examination (an increase of pain during axial loading,
rotation, and walking) were suggestive of loosening of one or both of the
components. All patients were evaluated with plain radiography followed by a
one-day protocol that included digital subtraction arthrography, nuclear
arthrography, and bone scintigraphy. For this study, the diagnostic images
were retrospectively interpreted by four experts (two nuclear medicine
physicians and two radiologists) who were blinded with respect to the clinical
data and the patient outcomes. Thirteen patients were excluded from the study
because information from all four diagnostic modalities was not available for
retrospective analysis, leaving seventy-three patients (seventy-three
acetabular components) as the study group. The demographic characteristics of
the patients are presented in Table
I.
Plain Radiography
We routinely made anteroposterior radiographs of the hip and pelvis and a
true lateral radiograph of the affected hip. The anteroposterior radiograph of
the pelvis was made with the lower extremities internally rotated 15°.
The true lateral radiographs were made with the patient in the supine
position with the x-ray tube at a 45° cephalad angle. The table-to-film
distance was 100 cm; there was no correction for magnification. If it was
available, an early post-operative radiograph of the primary arthroplasty was
compared with the most recent plain radiograph. For assessment of the four
imaging techniques, the acetabular component was analyzed with use of the
method of DeLee and
Charnley21. The
acetabular component was considered to be loose if there was >2 mm of
bone-cement or prosthesis-bone radiolucency, migration of the prosthesis of
>3 mm, or a change in the amount of lateral tilt of
>5°15,22.
Digital Subtraction Arthrography
The contrast medium of Isovist 300 (iotrolan [0.641 g/mL]; Schering,
Berlin, Germany) mixed with 10 MBq indium-111-colloid was injected, under
sterile conditions, into the hip joint under radiographic guidance, with use
of an anterior approach and a 17-gauge spinal needle. Aspiration of hip fluid
was attempted and, if acquired, the material was sent for culture. If hip
fluid could not be aspirated, a small amount of sa-line solution was injected
and aspirated for culture. The contrast medium (10 to 35 mL) was then injected
until the patient complained of pain or until an increase in resistance was
felt in the syringe. Subtraction images were obtained at one frame per second,
with a maximum of twenty. Then, four representative images were copied to
radiographic film. Subtraction motion artifacts were compensated by means of
pixel shifting. The acetabular component was considered to be loose when
contrast was detected at the inferior aspect of the acetabular component (the
axial segment)8.
Nuclear Arthrography
After subtraction arthrography, the patient was asked to walk (about 300 m)
to increase intra-articular pressure in the hip joint and enhance tracer
transport along the prosthesis. For nuclear arthrography, anterior, posterior,
and lateral images (300,000 counts per view) were recorded with use of a 15%
window and 247 keV photopeak. The images were reviewed in a digital format
with use of superposition of the indium-111 subtraction images over the bone
scan, providing anatomical orientation. The acetabular component was
considered loose if tracer accumulated at the inferior aspect of the
acetabular component (the axial segment), if there was lateral or medial
tracking of contrast media, or
both8.
Bone Scintigraphy
Approximately three hours before the arthrographic procedure, 370 MBq of
technetium-99m hydroxymethylene diphosphonate was injected intravenously and
images were recorded with use of a dual-headed gamma camera (Genesys; ADAC
Laboratories, Milpitas, California) equipped with low-energy high-resolution
collimators. Dynamic images were obtained during the first sixty seconds by
scanning every eight seconds after injection, after which a blood pool image
of 120 seconds was recorded. Static images were recorded two hours after the
injection. Static anterior, posterior, and lateral images (300,000 counts per
view) were made with use of a 15% window and 140 keV photopeak
(technetium-99m) with an image matrix size of 256 × 256 pixels. The
static images were interpreted with use of the criteria for hip prosthesis
loosening described by Horoszowski et
al.23 and the
radiographic zones described by DeLee and
Charnley21. The
prosthesis was considered to be loose if there was a moderate increase in
tracer uptake in at least two zones or intense uptake in at least one
zone23.
Data Analysis
Surgical findings were used as the so-called gold standard. A diagnosis of
loosening was made if motion could be demonstrated between the acetabular
component and the pelvis with use of the revision instruments or by
hand7,14.
If the patients were managed nonoperatively, the prosthesis was considered to
be well fixed when the clinical symptoms subsided and subsequent radiographic
analysis showed no signs of loosening for at least one year.
Plain radiographs and subtraction arthrograms were evaluated by two
experienced radiologists, and nuclear arthrograms and bone scintigrams were
reviewed by two experienced nuclear medicine physicians. All four interpreters
were blinded with respect to the patient data, operative findings, and results
of clinical follow-up. They evaluated the components on each imaging
examination using a 5-point scale (with 1 indicating loose; 2, possibly loose;
3, equivocal; 4, possibly fixed; and 5, fixed). In case of disagreement
between the observers, a consensus was reached by discussion.
Statistics
Our statistical analysis included the construction of receiver operating
characteristic curves. A receiver operating characteristic curve is the
graphical representation of the trade-off between false-negative and
false-positive rates for every cutoff. By tradition, the plot shows the
sensitivity on the y-axis and 1-specificity on the x-axis. The more of the
curve that is in the left upper quadrant, the better the diagnostic accuracy
of a particular imaging technique. The closer the curve is to a diagonal line,
the poorer the accuracy of the technique. The cutoff of a test is chosen
according to the relative "costs" associated with false-positive
and false-negative results. The best cutoff is that which maximizes the sum of
the sensitivity and specificity, which is the point nearest to the top
left-hand corner24.
When the sensitivity, specificity, and accuracy of each imaging modality were
calculated, scores of 1, 2, and 3 were considered to reflect positive results
(a loose prosthesis) and scores of 4 and 5 were considered to reflect negative
results (a fixed prosthesis). Positive predictive values and negative
predictive values were calculated with the Bayes theorem.
The predictive values of the diagnostic techniques in relation to loosening
of the acetabular component were compared by means of multivariate regression
models. A p value of <0.05 was considered significant (Wald test, two
tailed).
The intraclass correlation coefficient was used to assess interobserver
variability. Intraclass correlation coefficient values can be interpreted as
weighted Kappa values where the weights are quadratic. Values range from 0,
reflecting no interobserver agreement, to 1, reflecting perfect interobserver
agreement.
Patients
Of the seventy-three patients (seventy-three hips) who underwent all four
imaging examinations, fifty-two underwent revision surgery and twenty-one were
managed nonoperatively. Of those who underwent revision arthroplasty,
forty-one were found to have a loose acetabular component. Patients who were
managed nonoperatively were monitored clinically and radiographically every
six months. Five of the patients who were managed nonoperatively had an
acetabular component that was considered to be clinically and radiographically
loose, but the physical condition of the patients precluded surgery. The other
sixteen patients who were treated nonoperatively were considered to have a
well-fixed acetabular component both clinically and radiographically
(Fig. 1). All cultures acquired
during arthrography were negative for aerobic and anaerobic bacteria.
Cemented Acetabular Component
Fifty-one of the seventy-three patients had a cemented acetabular
component. Thirty-six of them underwent revision surgery, and the acetabular
component was found to be loose in twenty-eight (78%). Of the fifteen patients
who were monitored clinically for one year, five were considered to have a
loose acetabular component and ten were considered to have a solid
component.
When the retrospective independent imaging evaluations were compared with
the results found at surgery and with the presumed diagnosis based on clinical
evaluations in the patients who did not undergo revision surgery, plain
radiography had a sensitivity of 85% (95% confidence interval, 68 to 95) and a
specificity of 89% (95% confidence interval, 65 to 99). The positive
predictive value was 93% (95% confidence interval, 78 to 99), and the negative
predictive value was 76% (95% confidence interval, 53 to 92)
(Table II). Thus, twenty-eight
patients were diagnosed correctly as having a loose acetabular component, five
patients were incorrectly diagnosed as having a well-fixed component (a
false-negative diagnosis), and two patients were incorrectly diagnosed as
having a loose component (a false-positive diagnosis).
Subtraction arthrography had a sensitivity of 76% (95% confidence interval,
58 to 89), a specificity of 72% (95% confidence interval, 47 to 90), a
positive predictive value of 83% (95% confidence interval, 65 to 94), and a
negative predictive value of 62% (95% confidence interval, 38 to 82).
Twenty-five loose and thirteen fixed cups were diagnosed correctly. There were
five false-positive and eight false-negative diagnoses.
Nuclear arthrography had a sensitivity of 67% (95% confidence interval, 48
to 82) and a specificity of 72% (95% confidence interval, 47 to 90). We found
a positive predictive value of 82% (95% confidence interval, 62 to 94) and a
negative predictive value of 54% (95% confidence interval, 33 to 74).
Twenty-two loose components were correctly identified. There were five
false-positive and eleven false-negative diagnoses.
Bone scintigraphy had a sensitivity of 85% (95% confidence interval, 68 to
95) and a specificity of 61% (95% confidence interval, 36 to 83), which was
lower than that of the other imaging modalities. Similarly, the positive
predictive value was 80% (95% confidence interval, 63 to 92), although the
negative predictive value was 69% (95% confidence interval, 41 to 89).
Twenty-eight patients had a loose acetabular component correctly diagnosed;
seven patients had a false-positive diagnosis, and five patients had a
false-negative diagnosis. Visual inspection of the receiver operating
characteristic curves showed that plain radiography was superior to the other
imaging techniques for detecting loosening of the acetabular component
(Fig. 2-A).
Uncemented Acetabular Component
Twenty-two patients had uncemented acetabular components. Sixteen underwent
revision surgery, and six were followed clinically. At the time of surgery,
thirteen acetabular components were found to be loose. It was determined that
the acetabular component was not loose in any of the six patients who were
followed clinically for one year. When the retrospective independent imaging
evaluations were compared with the results found at surgery and with the
presumed diagnosis based on clinical evaluations in those patients who did not
undergo revision surgery, plain radiography proved to be the best imaging
modality to detect loosening in patients with uncemented acetabular
components. The sensitivity of plain radiography was 85% (95% confidence
interval, 55 to 98), the specificity was 78% (95% confidence interval, 40 to
97), the positive predictive value was 85% (95% confidence interval, 55 to
98), and the negative predictive value was 78% (95% confidence interval, 40 to
97). There were two false-negative and two false-positive diagnoses; eleven
patients were correctly diagnosed as having a loose cup.
Subtraction arthrography had a sensitivity of 62% (95% confidence interval,
32 to 86) and a specificity of 67% (95% confidence interval, 30 to 93). The
positive predictive value was 73% (95% confidence interval, 39 to 94), and the
negative predictive value was 55% (95% confidence interval, 23 to 83). There
were eight true-positive diagnoses, six true-negative diagnoses, and three
false-positive diagnoses.
Nuclear arthrography had a sensitivity of 8% (95% confidence interval, 0 to
36) but a specificity of 100% (95% confidence interval, 66 to 100). The
positive predictive value was 100% (95% confidence interval, 0 to 100), and
the negative predictive value was 43% (95% confidence interval, 22 to 66).
There was one true-positive diagnosis, twelve false-negative diagnoses, and
nine true-negative diagnoses, but there were no false-positive diagnoses.
Bone scintigraphy had a sensitivity of 77% (95% confidence interval, 46 to
95) and a specificity of 78% (95% confidence interval, 40 to 97). The positive
predictive value was 83% (95% confidence interval, 52 to 98), and the negative
predictive value was 70% (95% confidence interval, 35 to 93). There were ten
true-positive diagnoses, seven true-negative diagnoses, and two false-positive
diagnoses.
Receiver operating characteristic curves showed that plain radiography and
bone scintigraphy had a relatively better diagnostic performance than
subtraction arthrography and nuclear arthrography
(Fig. 2-B).
Interobserver Variability
With plain radiography, the intraclass correlation coefficient was 0.29 for
cemented acetabular components (fair interobserver agreement) and 0.53 for
uncemented components (moderate interobserver agreement). With subtraction
arthrography, the overall intraclass correlation coefficient was 0.71. With
nuclear arthrography, the overall intraclass correlation coefficient was 0.24
(fair agreement), which was mainly due to a poor agreement in the assessment
of the cemented acetabular components (intraclass correlation coefficient,
0.07). With bone scintigraphy, the overall intraclass correlation coefficient
was 0.43 (moderate agreement), which was mainly due to a fair intraclass
correlation coefficient for the cemented components (intraclass correlation
coefficient, 0.39) (Table
III).
Accuracy of Combined Imaging Examinations
Multivariate regression analysis was used to determine whether the
diagnostic accuracy was improved when plain radiography was used in
combination with other imaging techniques. Subtraction arthrography and bone
scintigraphy both had a significant predictive value for acetabular cup
loosening when used together with radiography (p < 0.05) (see Appendix).
The combination of a positive subtraction arthrogram and a positive radiograph
was associated with a loose cup in 87% of the hips, whereas the combination of
a negative subtraction arthrogram and a negative radiograph was associated
with a solid cup in 95% of the hips. The combination of a positive nuclear
arthrogram and a positive radiograph was predictive, but not significantly, of
loosening of the acetabular component (p = 0.26).
In clinical practice, plain radiography is the first imaging modality used
when evaluating the status of a prosthesis, and it has been consistently
reported to be an accurate technique to assess cemented and uncemented
acetabular
components6,9,13,15,21,25.
The reported accuracy of plain radiography has ranged from 50% to
98%25,26.
Lieberman et al. reported an accuracy of 97% for the evaluation of cemented
acetabular components, and they stated that plain radiography is the most
effective method to detect component
loosening6. In a
study of sixty-four patients, Miniaci et al. found that plain radiography was
a more accurate method than both subtraction and nuclear arthrography for the
detection of loose cemented prosthetic
components9. Our
data, which showed that plain radiography had an accuracy of 82% for
uncemented cups and 86% for cemented cups, are in agreement with published
reports. Importantly, we found only moderate interobserver agreement for plain
radiography (intraclass correlation coefficient, 0.37).
Earlier studies have also described considerable interobserver variability
when acetabular radiolucencies were
evaluated18.
Kramhoft et al. found a kappa value of 0.48 for radiologists and 0.37 for
orthopaedic surgeons with regard to the radiographic evaluation of loose or
fixed acetabular
cups19, and an even
lower interobserver agreement between orthopaedic surgeons was found for the
radiographic evaluation of fixation of cemented total hip
prostheses27. These
data indicate that while the overall performance of plain radiography in the
evaluation of acetabular components is superior to the other methods,
diagnostic accuracy is not optimal and is compromised by considerable
interobserver variability. Improved clinical training and new radiographic
scoring systems may help to reduce this interobserver
variability27.
Bone scintigraphy was the second-best diagnostic test, with an accuracy of
76% for cemented cups and 77% for uncemented cups, and with moderate
interobserver agreement. These findings are consistent with previous reports,
in which bone scintigraphy was found to have an accuracy ranging from 67% to
90% in cemented
cups6,8,17.
For uncemented cups, an accuracy of 67% has been
reported8. We found
that bone scintigraphy improved the accuracy of the diagnosis of a loose
acetabular component when it was combined with plain radiography. In this
regard, our findings differ from the findings of Ginai et al., who found no
additional value of bone scintigraphy and plain radiography in combination
with subtraction
arthrography16.
Nuclear arthrography was a poor imaging modality to evaluate the acetabular
component, with an accuracy of 69% for cemented components and 45% for
uncemented components, and low interobserver agreement.
These disappointing results could be due to the masking effect of the
intra-articular injected tracer, which, in the case of the acetabular
component, may obscure signs of component
loosening28,29.
The interobserver agreement for subtraction arthrography had an intraclass
correlation coefficient of 0.73 for cemented components and an intraclass
correlation coefficient of 0.65 for uncemented prostheses and was a better
diagnostic tool for cemented acetabular cups than for uncemented cups. These
findings agree with the results of a recent study by Cheung et al., who
reported a sensitivity of only 50% and an accuracy of 68% for subtraction
arthrography of uncemented acetabular
components15. They
concluded that this technique is not useful in the assessment of uncemented
prostheses. The accuracy of subtraction arthrography has been reported to
range from 76% to 90% for cemented prostheses and from 63% to 93% for
uncemented prostheses. Ginai et al. found subtraction arthrography to be a
better diagnostic tool than other imaging modalities, but this has not been
confirmed by other
investigators16. We
found subtraction arthrography to be of additional value when combined with
plain radiography. It may be a useful technique in patients in whom septic
loosening of a hip prosthesis is
suspected30.
In the present study and in earlier published reports, we found
considerable interobserver variability for these diagnostic imaging
techniques. Since interobserver variability may depend on the experience and
expertise of the observers, clinical training, and scoring systems, future
studies are needed to address these topics and to assess their effect on the
diagnostic performance of these imaging modalities.
In conclusion, no single imaging technique is ideal for the assessment of
acetabular component loosening. In this study, plain radiography had the best
diagnostic accuracy but a low interobserver agreement. We recommend a
combination of plain radiography and bone scintigraphy or subtraction
arthrography for the evaluation of acetabular cup loosening. Additional
studies are needed to evaluate the effect of interobserver variability on
diagnostic accuracy.
A table showing the results of the multivariate regression analysis is
available with the electronic versions of this article, on our web site at
(go to
the article citation and click on "Supplementary Material") and on
our quarterly CD-ROM (call our subscription department, at 781-449-9780, to
order the CD-ROM). ?