One of the most difficult diagnostic challenges in clinical orthopaedics is
the determination of whether a fracture has healed. Ideally, symptoms of pain
and a lack of evidence of healing on radiographs would allow surgeons to
detect all nonunions. However, plain radiographs can often be difficult to
interpret as overlying hardware obscures the bone, and callus is not always
visible. In addition, pain is a poor discriminator—a large number of
patients, even those with a healed fracture, have pain and dysfunction after
trauma1-3.
The identification of a tibial nonunion is extremely important because most
patients with a nonunion require surgery to achieve healing, while a delayed
union heals without intervention.
Computed tomography scanning offers a potentially more accurate method to
discriminate fracture union from nonunion. Despite its widespread use, to our
knowledge, the diagnostic accuracy of computed tomography in the assessment of
bone-healing has never been evaluated. Without information regarding the
specificity and sensitivity of a diagnostic test, surgeons do not have
adequate data to allow them to incorporate the test into their clinical
practice. We sought to determine the diagnostic accuracy of computed
tomography scanning in the detection of tibial nonunion.
Study Population
The study was approved by our institutional review board. The protocol at
our institution is to obtain a computed tomography scan of the tibia only if
the plain radiographs and clinical findings of fracture-healing are equivocal.
All patients undergoing computed tomography scanning of the lower extremity
over a three-year period were entered into a prospective database. We
identified forty-eight patients who had a computed tomography scan of the
tibia performed specifically for the assessment of fracture-healing
(Fig. 1). The remaining
computed tomography scans of the tibia were done exclusively on fresh
intra-articular fractures for preoperative planning. Thus, we were able to
identify all patients who underwent computed tomography scanning of the tibia,
regardless of whether they underwent surgery.
Three patients were excluded because they had been followed for less than
six months; four patients, because the scan was performed to assess healing of
an intra-articular ankle fracture; and four patients, because no medical
information was available. Two patients were excluded because the scan was
obtained before three months had elapsed from the date of the fracture (which
was too early to rigorously classify the fracture as a nonunion).
Thus, we identified thirty-five patients who had undergone computed
tomography scanning for the assessment of a possible tibial nonunion and had
been followed clinically for a minimum of six months. A review of the medical
records confirmed that, for all thirty-five patients, the computed tomography
scan was obtained because the orthopaedic surgeon could not determine union
from nonunion on the basis of plain radiographs and clinical examination.
Assessment with the Gold Standard
The medical records were reviewed to determine the final clinical diagnosis
for all thirty-five patients according to the so-called gold standard (union
at the time of surgery or after six months of clinical observation).
Twenty-five patients had assessment of the degree of tibial fracture-healing
(union or nonunion) by direct inspection at the time of surgery. The remaining
ten patients attended the orthopaedic clinic for a minimum of six months
without evidence of hardware complication or breakage and clinically were
considered to have had fracture union.
Plain Radiography
Orthogonal radiographs of the tibia were made in a standard fashion. They
were reviewed by an orthopaedic traumatologist who was blinded to the clinical
data. The number of bridging cortices was recorded, and an overall assessment
of union was made as well.
Computed Tomography Scanning Technique
The studies were performed on eight multidetector-row computed tomography
scanners. Seven scanners were made by General Electric Medical Systems
(Waukesha, Wisconsin), and they ranged from four to sixteen detector-rows. One
was a sixteen detector-row scanner built by Siemens Medical Solutions
(Forchheim, Germany). Detector configurations varied according to the machines
used and at the discretion of the supervising radiologist. Coronal and
sagittal reformats were performed for all thirty-five patients. The scans were
obtained with use of 2.5-mm slice thickness. Reconstruction intervals were
performed to a thickness of 1.25 mm.
Assessment of Computed Tomography Scans
Three reviewers (two musculoskeletal radiologists [A.K. and H.S.] and one
fellowship-trained orthopaedic trauma surgeon [T.B.]) independently reviewed
the computed tomography scans. The reviewers were blinded to the clinical
outcome and to additional radiographic data, including previous radiographs or
computed tomography scans. The reviewers assessed the location of the
fracture, type of fixation, and number of bridging cortices. To assess the
number of bridging cortices, the axial cut of the computed tomography scan was
divided into quadrants. The reviewers graded each scan as either fracture
union or nonunion. In the case of disagreement, the majority opinion (two of
three reviewers) was taken as the proper assessment. Interrater reliability
was calculated with use of the intraclass correlation coefficient two-way
mixed model for consistency.
Statistical Methods
The data were analyzed with use of SPSS software (version 10.0; SPSS,
Chicago, Illinois). Sensitivity, the ability of computed tomography to detect
nonunion of a fracture, was calculated by dividing the total number of
patients with a nonunion on computed tomography scans by the total number of
patients with a true-positive nonunion (nonunion at the time of surgery).
Specificity, the ability of computed tomography to exclude nonunion, was
calculated by dividing the number of patients with union that was correctly
diagnosed on computed tomography scans by the number of patients with union
according to the so-called gold standard. Diagnostic accuracy was measured by
the kappa statistic, which measures the strength of agreement of the findings
on computed tomography with the gold-standard assessment of union. A kappa
value of >0.8 indicates almost perfect agreement, while a kappa value of
<0.4 indicates fair-to-poor agreement.
We also calculated the area under the receiver operating characteristic
curve. The area under the receiver operating characteristic curve represents
the probability that computed tomography scans correctly discriminate between
patients with fracture union and those with nonunion, where 0.5 is chance
discrimination and 1.0 is perfect discrimination.
The final study cohort consisted of twenty-two patients with nonunion and
thirteen patients with union. The characteristics of the study cohort are
shown in Table I. The mean time
from the injury to the computed tomography scan was 9.7 months (range, three
to twenty-nine months). The mean number of surgeries on the tibia prior to
computed tomography examination was 2.6 procedures (range, zero to eight
procedures). Twenty-seven patients (77%) had hardware in place at the time of
computed tomography scanning.
Table II summarizes the
accuracy of plain radiography in the detection of nonunion. As expected in
this difficult group of fractures, plain radiography led to the incorrect
classification of fifteen fractures (43%). The kappa statistic for plain
radiography compared with the so-called gold standard was 0.14, indicating
poor diagnostic accuracy.
In contrast, computed tomography scanning had very good accuracy in the
detection of nonunion in this complex group of fractures
(Table III). Computed
tomography scanning was 100% sensitive for nonunion. The kappa statistic for
computed tomography scanning compared with the gold standard was 0.67,
indicating good diagnostic accuracy. The intraclass correlation coefficient
was 0.89, indicating almost complete agreement among the three observers about
the diagnosis.
Figure 2 depicts the
receiver operating characteristic curve for computed tomography scanning in
the detection of nonunion. The area under the receiver operating
characteristic curve was 89.9%. This indicates that computed tomography
accurately discriminated between subjects with fracture union and those with
nonunion. A clinical example of the sensitivity of computed tomography
scanning is shown in Figure
3.
Computed tomography scanning, however, displayed a limited specificity of
only 62%. For five patients, the diagnosis of nonunion was made on the basis
of computed tomography scans but the patients were found to have union. For
two patients, the surgeons ignored the findings on the computed tomography
scans and the fractures were later determined to be healed clinically. Three
of the five patients actually underwent surgery with the intent to treat the
nonunion, and the surgeon found that the fracture had united
(Fig. 4). None of these three
patients required further surgery.
A summary showing the agreement among the findings on plain radiography,
computed tomography scanning, and the clinical gold standard is shown in
Table IV.
In this study of patients with a possible tibial nonunion and equivocal
findings on radiographs, computed tomography scanning demonstrated a very good
diagnostic accuracy. However, computed tomography scanning had limited
specificity and tended to lead to a diagnosis of nonunion when, indeed, the
fracture was healed in some patients.
There are very limited data in the literature on diagnostic tests to
evaluate tibial fracture-healing or fracture-healing in general.
Traditionally, the combination of pain and lack of healing on successive
radiographs is coupled with clinical judgment to lead to the diagnosis of a
nonunion4.
Unfortunately, a number of patients with a healed tibial shaft fracture
continue to have pain three years after the injury, making pain a poor
discriminator of union from
nonunion1. Hammer et
al., in a study of 208 patients, showed that plain radiography was no more
accurate than a coin flip in determining tibial diaphyseal fracture
union5. Bhandari et
al. surveyed 444 practicing orthopaedic surgeons and found a wide variation in
the criteria used to diagnose
nonunion6. Thus,
better tests and criteria are needed. To the best of our knowledge, the
present study is the first to rigorously analyze a clinically available
diagnostic test for nonunion.
A principal question in any study of diagnostic accuracy is the
generalizability of the results. By starting with a database of all patients
who underwent computed tomography scanning in our institution, we were able to
identify patients with a possible nonunion, whether they were treated
operatively or nonoperatively. We used computed tomography to selectively
evaluate the fractures that had equivocal findings on plain radiographs with
regard to union. A review of the medical records confirmed that equivocal
radiographic findings and concern for nonunion were the indications for
computed tomography. This is precisely the group of patients for whom most
orthopaedic surgeons would use computed tomography scanning.
In an attempt to clarify the clinical role of computed tomography scanning
in the assessment of tibial fracture-healing, we performed the first study of
which we are aware that compares the diagnostic test with a so-called gold
standard. Previous studies have noted that computed tomography can detect
nonunion, but they did not fully integrate the test into the clinical
picture7,8.
We found computed tomography scanning to be a reliable test with high
interobserver agreement. The intraclass correlation coefficient was 0.89,
indicating that different observers would be highly likely to interpret the
scans in a similar fashion. The computed tomography scan was also highly
sensitive as it detected all of the patients who had a true nonunion, as
confirmed by intraoperative assessment. Computed tomography scanning also
accurately discriminated union from nonunion, as evidenced by the large area
under the receiver operating characteristic curve. Finally, we found that
computed tomography scanning accurately delineated the osseous architecture
even though the majority of patients had hardware in place.
The main drawback to the computed tomography scan in the assessment of
tibial fracture-healing is its somewhat low specificity. Because the treatment
of nonunion is often surgical, surgeons need a test with high specificity as
they would like to avoid taking a patient to the operating room only to find a
healed tibial fracture. (This approach is in contrast to cancer screening,
where high sensitivity is required to detect all tumors.) Evaluation with
computed tomography led to the incorrect classification of nonunion in three
patients, who proved to have a healed fracture on operative inspection. Two
additional patients were classified with computed tomography scanning as
having nonunion, but the surgeons chose to follow the patients and the
fractures healed uneventfully. The computed tomography scan delineates the
osseous architecture to a very fine degree, and it can occasionally depict
clefts in bone that are of uncertain clinical importance. Understanding this
limitation to computed tomography scanning may prevent surgeons from embarking
on unnecessary surgery.
Our study has a few limitations. The computed tomography scans were
reviewed in a relative information vacuum for the purposes of study rigor;
most radiologists and surgeons would review the computed tomography scan in
conjunction with the previous plain radiographs and other clinical data. The
study is retrospective in nature and is subject to recall bias. We did not
perform an intrarater reliability study because the reliability of computed
tomography scanning was established by Grigoryan et
al.9. The study
cohort is small. However, we believe that most nonunions can be diagnosed on
the basis of plain radiographs, history, and clinical examination alone; thus,
the population of patients with equivocal nonunion is small.
In summary, we present information regarding the evaluation of tibial
fracture-healing with use of computed tomography scans that orthopaedic
surgeons can incorporate into their clinical practice. The computed tomography
scan is extremely sensitive for detecting tibial nonunion. However, it has
limited specificity, and surgeons must couple computed tomography and clinical
findings to minimize the risk of making a false-positive diagnosis of
nonunion. ?