Keywords
canine hip dysplasia - distraction index - Norberg angle - radiography
Introduction
Canine hip dysplasia is a common orthopaedic disease in dogs.[1] The prevalence varies in different breeds between 2 and 80%.[2] Canine hip dysplasia is a polygenetic and multifactorial condition[3]
[4]
[5]
[6] and heritabilities of 0.14 to 0.43 are reported.[7]
[8] Phenotypic breeding stock selection is aimed to reduce the incidence based on the
genetic component. Increased hip joint laxity is one of the most important factors
in the assessment of canine hip dysplasia. There are numerous radiographic methods
for the detection of canine hip dysplasia in the world.[9] The most widely used method in Europe is the five grade (A-E) Fédération Cynologique
Internationale (FCI) scheme,[10] which is based on evaluation of various radiographic findings, including signs for
osteoarthritis, and the Norberg angle (NA) as an objective indicator for hip laxity.
A line between both femoral head centres and the corresponding craniolateral acetabular
margins on each side form the NA.[11] In contrast to the FCI method, the PennHIP method relies on the identification of
osteoarthritis and, for those without signs for osteoarthritis, assessment of the
passive hip joint laxity expressed by the distraction index (DI).[12] Laxity is measured on radiographs with a distraction device causing the femoral
head to displace laterally. The DI is calculated using the distance between the acetabular
and the femoral head centre divided by the radius of the femoral head.
The FCI grading system has relatively poor interobserver agreement[13]
[14] although the reproducibility of the NA seems to be sufficient.[15] For the PennHIP method, a study was published and showed high within- and between-examiner
repeatability.[16] One study showed a high repeatability of DI measurements when comparing the official
results to results of trained researchers.[17] A recent study revealed substantial variability for the NA but not for the DI.[18]
Measurements should be both reliable and valid to evaluate the radiographic phenotype.
Accuracy, also referred to as validity, demonstrates how close a measurement is to
the true value based on the gold standard. Reliability, also referred to as precision
or consistency, determines how close the measurements are to each other and is therefore
negatively correlated to variability. Reliability can be evaluated by repeated measurements.[19]
To evaluate the reliability of radiographic measurements, different factors have to
be taken into account. An error may derive from differences in the radiograph due
to positioning, projection or different forces applied during acquisition. This effect
can be assessed by acquiring two identical sets of radiographs and is also referred
to as repeatability, also termed intraoperator reliability or -agreement, if the radiographs
is taken by the same person or reproducibility (also termed interoperator reliability
or agreement) if the radiographs are taken by different persons. Furthermore, an error
can be derived from the measurement itself. This can be evaluated measuring twice
using the same radiograph and is also termed repeatability (intraobserver or intrarater
reliability or agreement) or reproducibility (interobserver or interrater reliability
or agreement) depending if the measurements are made by the same or different persons.[19]
In the available literature, to date there is no study that directly compares the
reliability between measurements of NA and DI in a structured and comparable form
that takes repeatability and reproducibility into consideration. The aim of the study
was to evaluate intraoperator-reliability as well as intra- and interobserver reliability
of the NA and DI measurements.
Materials and Methods
A total of 59 dogs that were presented for official hip screening were included after
the owner's consent was given. The dogs had to fit the minimum weight requirement
of 8 kg for evaluation with the PennHIP distractor. To comply with the FCI criteria
for official screening, the minimum age was 12 months. All animals underwent injection
anaesthesia using dexmedetomidine (0.01–0.02 mg/kg dexdomitor 0.5 mg/mL; Orion Pharma
GmbH, Hamburg, Germany), medetomidine (0.01–0.04 mg/kg Dorbene Vet 1 mg/mL, Zoetis
Deutschland GmbH, Berlin, Germany) or diazepam (0.1–0.5 mg/kg Ziapam 5 mg/mL, Ecuphar
GmbH, Greifswald, Germany) intravenously followed by the administration of propofol
(1–8 mg/kg Narcofol 10 mg/mL, CP-Pharma GmbH, Burgdorf, Germany) until the dogs were
fully anaesthetized with adequate muscle relaxation.[20]
For each dog five radiographs were taken in the same order on a direct digital radiography
system (Siemens Axiom Luminos dRF; Siemens Healthcare AG, Erlangen, Germany) without
the use of positioning devices. All radiographs were obtained by the same PennHIP-certified
veterinarian. A standard ventrodorsal projection of the pelvis with extended hips
also known as the FCI position 1 and the ventrodorsal projection of the pelvis with
limbs in neutral position with distraction of the femoral joint using a PennHIP distractor
(PennHIP distraction view) were repeated, while the PennHIP compression view was performed
once. Images were anonymized by a person not involved in scoring of the radiographs
and evaluations were performed at the earliest 1 month after acquisition of the images.
Before the study was conducted, every observer trained measuring the DI and the NA
in 10 cases with known official results. The FCI and Swiss scheme scoring of the hips
as well as measurements of the NA and the DI was performed twice, after a 2-month
interval, by a first year imaging resident with 5 years of experience in diagnostic
imaging, once by an European specialist in veterinary diagnostic imaging and member
of the German association of scrutineers and one intern without experience in veterinary
diagnostic imaging. The measurements were made in the same digital environment in
the same order by all observers, using specific tools for measurement of the NA and
DI provided by the commercial software (Dicom PACS, Oehm & Rehbein GmbH, Rostock,
Germany) used in the institution. The ‘distraction index tool’ consists of two circles
that can be manually adjusted to fit the femoral head and the acetabulum and automatically
calculates the DI value. The ‘Norberg angle tool’ consists of two circles that need
to be drawn over each femoral head and a line that needs to be adjusted to the cranial
acetabular edge on each side. The NA for each side is displayed subsequently.
Results were stored for each hip joint separately in an excel spreadsheet (Office
2010 Excel; Microsoft, Redmond, Washington, United States). Statistical analysis was
conducted using commercial statistical software (MedCalc; MedCalc Software, Ostend,
Belgium). Intraclass correlation coefficient (ICC) was calculated to evaluate reliability
of intraoperator, intraobserver as well as interobserver measurements. This test allows
comparison between samples of different scales, such as the NA (degree) and the DI
(unitless) values.[10]
[12] An ICC of 1 indicates perfect agreement, whereas an ICC of 0 indicated not more
than random agreement. Intraclass correlation coefficient values less than 0.5, between
0.5 and 0.75, between 0.75 and 0.9 and greater than 0.90 can be interpreted as poor,
moderate, good and excellent reliability, respectively.[21] Cohens weighted kappa was calculated to compare the observer agreement between the
categorical FCI classification and classification made using the Swiss scoring scheme.[22] A kappa of 1 indicates perfect agreement, whereas a value of 0 indicates not more
than random agreement and negative values represent a negative correlation. Values
of 0.21 to 0.40, 0.41 to 0.60, 0.61 to 0.80 and greater than 0.81 can be interpreted
as fair, moderate, substantial and as almost perfect agreement, respectively.[23]
Results
The 59 dogs included 20 different breeds (10 German Shepherd Dogs, 7 Labrador Retriever,
6 Golden Retriever, 4 Doberman Pinscher, 4 Flat Coated Retriever, 3 Small Münsterländer,
3 Belgian Shepherd Dogs, 3 Entlebucher Mountain Dogs, 2 Akita, 2 Australian Shepherd
Dogs, 2 Border Collies, 2 Nova Scotia Duck Tolling Retriever, 2 Schnauzer, 2 Vizsla,
2 White Shepherd Dogs, 1 Pyrenean Shepherd Dog, 1 Bernese Mountain Dog, 1 German Wirehaired
Pointer, 1 Eurasian Dog, 1 Keeshond). Of all dogs 32.2% (n = 19) were scored FCI grade ‘A’ (no evidence of hip dysplasia), 42.4% (n = 25) were scored FCI grade ‘B’ (borderline), 18.6% (n = 11) were scored FCI grade ‘C’ (mild hip dysplasia) and 6.8% (n = 4) FCI grade ‘D’ (moderate hip dysplasia.
Results of the statistical analysis for intraoperator reliability, intraobserver reliability
and interobserver reliability are provided in [Table 1].
Table 1
Comparison of intraclass correlation coefficient for the reliability of Norberg angle
and distraction index
|
Norberg angle
|
Distraction index
|
|
Intraoperator
|
0.962
|
0.892
|
|
Intraobserver
|
0.975
|
0.986
|
|
Interobserver
|
0.969
|
0.972
|
Intraoperator Reliability
Intraclass correlation coefficient for the NA was 0.962 with a 95% confidence interval
from 0.941 to 0.975 and for the DI 0.892 with a 95% confidence interval from 0.833
to 0.931.
Intraobserver Reliability
Intraclass correlation coefficient for the NA was 0.975 with a 95% confidence interval
from 0.964 to 0.983 and for the DI 0.986 with a 95% confidence interval from 0.979
to 0.990.
The weighted kappa for the agreement between both measurements for the classification
according to the FCI scheme was 0.699 with a 95% confidence interval from 0.609 to
0.789 and for the classification according to the Swiss scheme 0.661 with a 95% confidence
interval from 0.556 to 0.767.
Interobserver Reliability
Intraclass correlation coefficient between all three observers for the NA was 0.969
with a 95% confidence interval from 0.957 to 0.978 and for the DI 0.972 with a 95%
confidence interval from 0.950 to 0.983.
Intraclass correlation coefficient between both experienced observers (AB and JK)
for the NA was 0.983 with a 95% confidence interval from 0.969 to 0.990 and for the
DI 0.980 with a 95% confidence interval from 0.972 to 0.986.
Intraclass correlation coefficient between one experienced and one unexperienced observer
(AB, SW) for the NA was 0.936 with a 95% confidence interval from 0.895 to 0.959 and
for the DI 0.947 with a 95% confidence interval from 0.865 to 0.973.
The weighted Kappa for the agreement between both experienced observers (AB and JK)
for the classification according to the FCI scheme was 0.687 with a 95% confidence
interval from 0.596 to 0.778 and for the classification according to the Swiss scheme
0.681 with a 95% confidence interval from 0.588 to 0.774. The weighted Kappa for the
agreement between one experienced and one unexperienced observer (AB and SW) for the
classification according to the FCI scheme was 0.465 with a 95% confidence interval
from 0.344 to 0.585 and for the classification according to the Swiss scheme 0.514
with a 95% confidence interval from 0.392 to 0.635.
Discussion
Repeated radiographs and measurements were performed to evaluate reliability of DI
and NA.[19] The intraoperator reliability of the DI was slightly lower (ICC 0.892), but still
a good, almost excellent result. The NA seems to generate slightly more precise results
in between two repeated radiographs. Although our operators are PennHIP-certified,
they are much more trained in the more frequently used standard ventrodorsal radiograph
compared with the distraction radiographs. This experience may influence the repeatability.
Subjectively distraction radiographs are more difficult because besides patient positioning,
additional attention has to be paid to the handling of the distraction device. The
slight differences in repeated radiographs may derive from a combination of various
factors such as the forces applied, pelvic tilting, muscle relaxation, central beam
position or other unknown random effects.[20]
[24]
[25]
The ICC showed minimally better results for the intra- and interobserver-reliability
of the DI compared with the NA. This complies with a recent study where variability
of the NA was higher than of the DI.[26] In contrast to the other study, we found no substantial difference and excellent
reliability (ICC > 0.90) for both methods and the differences seem negligible. Based
on the small sample size of only 10 dogs in the other study, their higher variability
for the NA might be caused by outliers. Another main influence on intra- and interobserver-reliability
is probably caused by the precise definition of measurement points with special focus
on common anatomic variations. The availability or the lack of a detailed and in-depth
description of measurement points and procedures, also with special regard to anatomical
variants, may contribute to the variations in the results of various studies of inter-observer
agreement. Norberg angle and DI are based on the measurement of perfect circles. Based
on our agreement for the NA, the femoral head circle was defined by two points on
the cranial and craniolateral projected surface and one point on the centre of the
caudomedial projected surface of the femoral head on the radiograph, neglecting and
bridging the depression or flattening of the acetabular fossa and the junction to
the femoral neck. Neither the femoral head nor the craniolateral acetabular rim of
the facies semilunata were always projected as perfect circle segments on radiographs, this can be due
to distortion caused by divergence of the X-ray or just normal anatomical variation.[27] But we were able to fit freely adjustable circles to these structures by approximation.
In our experience, it was frequently hard to precisely define the measurement point
of the caudolateral acetabular edge for the DI as well as the craniolateral acetabular
edge for the NA. This can be explained variability in the visibility of the measurement
points in different radiographs, probably mainly due to anatomic variation and positioning.
Another feature that might influence the precision of the measurements is the severity
of osteoarthritis in the population. It is probably easier to generate reliable results
in hips without evidence of osteoarthritis.
For the measurement process, digital environment may play an important role, like
thin or thick, dotted or continuous tool-line, screen-size and level of magnification.
Use of a three-point circle as alternative to freely adjustable circles might also
have an influence.[27] We used standard commercially available 24-inch high definition flat panel screen
computer monitors with high, but undefined zoom levels of the radiographic image and
thin continuous coloured tool-lines (1px) in our setting.
Comparing the interobserver reliability of NA and DI, there was no substantial difference
related to the level of experience and both methods show excellent reliability (ICC > 0.90).
The interobserver agreement of the FCI scheme and the Swiss scheme is similar. There
was almost no difference in the comparison between experienced observers with a good
agreement (Kappa 0.687 and 0.681, respectively). In the comparison between one experienced
and one unexperienced observer, the agreement was still moderate. Kappa for the FCI
scheme was considerably lower than for the Swiss scheme (0.465 and 0.514, respectively).
This implies the Swiss scoring scheme enables better results in unexperienced observers
than the FCI system. It has to be considered that in our study only three different
observers scored the images. To make a recommendation, follow-up studies should be
performed with a higher number of observers. And even if it is unlikely to have unexperienced
observers in an official hip screening scenario, it is obviously easier for the beginner
to adopt and successfully implement the structured approach of the Swiss scoring scheme
than the categorical FCI grading system. It is probably easier and more consistent
to work through a table of predefined anatomical structures, with a description and
predefined scoring of individual findings that sum up to a final result than to match
a complex joint into a single category based on a global description.
Conclusion
The intraoperator reliability was slightly better for the NA than for the DI. Intra-
and interobserver reliability showed excellent results for both, the NA and the DI.
Therefore, both methods can be considered highly and equally reliable. The influence
of the positioning seems to have slightly more impact on the result than the measurement
itself. The FCI and the Swiss scheme seem to be equally reliable in experienced observers,
but based on the better results for the unexperienced observer, we suggest novices
at hip scoring to favour the Swiss scoring system.
