The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
1
A Patient-Specific Predictive Model Increases Preoperative Templating
Accuracy in Hip Arthroplasty.
Amir Pourmoghaddam, PhD; Marius Dettmer, PhD; Adam M. Freedhand, MD; Brian C.
Domingues, BSc.; Stefan W. Kreuzer, MD, MSc.
Affiliation:
Memorial Bone & Joint Research Foundation, Department of Orthopaedic Surgery, The
University of Texas Health Science Center at Houston β Medical School
1140 Business Center Drive, Suite 101
Houston, TX 77043
This is authorβs version. For publisher version please visit:
http://www.arthroplastyjournal.org/article/S0883-5403(14)00897-3/abstract
Abstract:
Application of digital radiography during preoperative templating has shown potential to reduce
complications in total hip arthroplasty. Digital radiography has significantly improved this process.
In this study, we aimed to further improve the digital templating by using a predictive model built
on patientsβ specific data. The model was significant in improving the accuracy of templating
within +/-1 size of acetabular component (2(1, π = 468) = 19.314, p<0.0001, =0.604, and
odds-ratio: 7.750 (95% CI 2.740-30.220)). We successfully achieved a 99% accuracy within +/- 2
of templated size. Additionally, patient demographics, such as height and weight, have shown
significant effects on the predictive model. The outcome of this study may help reducing the costs
of health care in the long term by minimizing implant inventory costs for each patient.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
2
Introduction:
Preoperative planning in total hip replacement is an essential part of the procedure [1,2] because
it prepares the surgical team and significantly reduces the surgical time by minimizing potential
complications [3]. The utilization of digital radiography in clinical settings has grown significantly
in the past few years. This technique is being used more commonly as a financially sound
alternative compared to analogue radiography, as it provides the opportunity to store an unlimited
number of images while reducing the costs of film storage and the need for recycling [4]. In
addition, advanced computer programs have been developed to analyze the obtained digital
information [5].
Historically, templating accuracy, particularly when utilizing analog radiography, has been
reported to be relatively low; recently, the literature shows that digital templating can be successful
in identifying the implant size within +/- 2 implant sizes [3β13]. In this study, we aimed to further
improve the accuracy of predicting the implant size in total hip replacement by considering some
anthropometric characteristics of the patient.
In todayβs digital world, due to the costs associated with expanding implant options and inventory
management, digital templating could be of great benefit for surgical preparation and for reducing
inventory in the field, thereby decreasing the overall cost of THA. Proper surgical preparation,
including digital templating, can reduce surgical time and potential complications, but digital
templating is rarely utilized to manage inventory flow in the field. Accurate templating requires
proper patient positioning and standardized X-ray techniques to generate images of sufficient
quality. A major concern with templating, particularly computer-assisted, image-processing
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
3
software, is the issue of βinaccurate magnification ratio.β Generally with computer-assisted
templating, an initial 20% magnification of the image is assumed. This ratio can be affected by
patient-related factors, such as obesity and body habitus, or technical factors, such as the distance
of the X-ray tube to the joint being X-rayed. Traditional preoperative planning utilizes a tube-to-
film a distance of 48 inches for X-ray imaging, resulting in an estimated magnification of 20%
compared to the actual size. However, the accuracy of such magnification has been questioned and
can vary widely [4]. Traditional templates provided by implant manufacturers come in different
magnifications and are overlaid on standard X-rays to provide the templated implant size. A similar
concept is utilized in digital radiography, in which the digital radiographs are adjusted for
magnification by utilizing a magnification marker or a constant magnification ratio and software
is used to create digital overlays of implant sizes to select the appropriate size and orientation of
the component. Error! Reference source not found. summarizes some of the recent studies
conducted in different institutions representing the relationship between the preoperative
templating and the actual implant sizes used for the patients. Although many of the previous studies
indicated high accuracy, other studies have raised concerns about relying on the outcome of
templating. Efe et al. reported that, in approximately 9% (15 out of 169) of their cases, the
intraoperative implant size could not be predicted, even within 2 sizes [14]. These concerns have
inspired our research to find other factors that might facilitate and improve the accuracy of
preoperative templating. Thus, the main goal of this study was to improve accuracy of preoperative
templating by using a predictive model that includes patient-specific demographics (i.e., height,
weight, Body Mass Index (BMI). Thus, the purpose of this study was to develop a prediction model
to enhance the accuracy of preoperative digital templating by considering a multitude of patient
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
4
variables, such as height, age, sex, BMI, etc. We hypothesized that anthropometric variables
impact the accuracy of the preoperative templating size of acetabular and femoral components.
Table 1 β Examples of previous studies investigating the accuracy of preoperative templating of femoral and
acetabular components.
Study Exact Size (%) Β± 1 size (%) Β± 2 size (%)
Total Femoral Acetabular Femoral Acetabular Femoral Acetabular
Whiddon et. al (2011) β
Digital
51 61 39 90 78 96 96
Whiddon et. al (2011) β
Manual measurement
51 33 31 82 67 100 88
Shaarani et. al(2013) 100 38 36 80 75 98 98
Maratt (2012) 20 NA NA 75 73 93 96
Maratt (2012) β Acetate 20 NA NA 93 63 98 86
Methods:
A retrospective review of preoperative radiographs for 468 individuals (224 females/ 244 males)
who received total hip arthroplasty was conducted. The preoperative templated sizes were
compared to actual implant sizes used at the time of surgery. The data was collected from August
2012 to December 2013 at a single institution.
The average age was 59.96 Β±12.50 years, 436 diagnosed with osteoarthritis, 53 with avascular
necrosis, 13 with failed THA, 2 with infection, 4 post trauma, and 13 with failed hemi arthroplasty.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
5
All patients underwent direct anterior total hip arthroplasty. The level of the femoral osteotomy
was performed based on preoperative templating. Acetabular reaming and cup impaction was
performed in standard fashion. Supplemental screw fixation was utilized when needed. The final
acetabular component size was selected based on the final reamer size and a quality press fit in the
bleeding bone. An attempt was made to restore the native hip center of rotation in all cases. During
the study period, two different implant manufacturers for the acetabular component (Stryker
Orthopedics and Corin Inc.) were utilized. Acetabular reaming was performed according to the
manufacturersβ recommendation to achieve a solid interference fit.
All preoperative radiographs were taken with a standardized X-ray source-to-image distance of 1
m when the patients were at the standing position. All radiographs were taken by a single team of
radiology technicians. However, no magnification marker was used in these radiographs for the
purpose of referencing.
For THA templating in the anteroposterior view the pelvis image was centered over the pubic
symphysis, while the hip was internally rotated between 10Β° to 15Β°. Figure 1 depicts a sample of
the templated joints that was used in predicting the implant size. The digital radiographs were all
analyzed utilizing the TraumaCadTM software system (TraumaCad, BRAINLAB, Westchester, IL,
USA) [15]. The initial implant model was adjusted and magnified by a factor of 120% to provide
an acceptable overlay on the hip joint. All intraoperative data was collected prospectively using
online, web-based, data-entry software from an IRB-approved joint registry (IRB # HSC-GEN-
09-0143), including the final implant sizes selected by the surgeon.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
6
Figure 1 β A sample of templated hip joints by using TraumaCadTM
Statistical model
A multiple regression model was used to develop the predictive model and to investigate the
contribution of each factor to the model to predict the actual size of the implant from the
preoperative measurements. These variables included, preoperative acetabular size, preoperative
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
7
femoral size, body mass index (BMI), age, gender, height, and weight. In each model, a backward
stepwise algorithm was used to identify the variables, resulting in significant model demonstration.
This algorithm initially enters all variables into the model but systematically removes the
predictors that do not have significant contribution in defining the changes in the model. In
addition, we used a multiple regression model that include all the variables of interest in this study
to explore their effects as well and provide a more standardized equation for future studies. Finally,
to assess the improvement in the accuracy of the templating we used a nonparametric McNemarβs
test to compare the binomial accuracy outcome (i.e., Yes vs. No) between the templating alone
method versus utilizing the model. The effect size and odds ratio of the McNemarβs test was
calculated for the significant differences. In the cases in which the number of cells in the
McNemarβs test was less than 5 we used an exact McNemarβs test. A significance level of .05 was
assumed in this study. The analysis was conducted by using SPSS 21.0.0 (SPSS Inc., Chicago,
Illinois, USA).
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
8
Results:
Table 2 includes the demographics of the individuals who participated in this study.
Table 2- patientsβ preoperative and intraoperative measurements.
Average SD
Age (years) 61.01 12.50
Height (cm) 172 10.67
Weight (kg) 84.83 20.13
BMI 28.5 5.26
Tempalated Acetabulum size (mm) 54.63 4.07
Tempalated Femoral size (mm) 4.73 1.81
Actual Acetabulum size (mm) 55.61 3.77
Actual Femoral size (mm) 4.82 1.77
The Acetabular component
The backward stepwise model has demonstrated that, for the acetabular component size, four
significant predictors were achieved, which were templated acetabulum size, height, BMI, and
templated femoral size. This model had the π 2 = .0.797 with Adjusted π 2 = .795 with standard
error of (ππΈ = 1.726). Gender and weight were not significant factors in this model. The outcome
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
9
of a full regression model is demonstrated in the acetabular model, in which gender and weight
are also included in the prediction.
Acetabular component model
π΄ππ‘πΈπ π‘ππππ‘ππ = 4.532 + 0.661 Γ π΄ππ‘ππππ + 0.202 Γ πΉππππππ + 0.067 Γ π»πππβπ‘ β 0.024 Γ ππππβπ‘
+ 0.138 Γ π΅ππΌ +
In which π΄ππ‘πΈπ π‘ππππ‘ππ is the estimated size of acetabulum cup, π΄ππ‘ππππ is the preoperative
templated acetabulum size from digital radiography, πΉππππππ is the preoperative templated
femoral size, and is the residual error term. Gender is defined as a binomial variable (Female =
0 and Male = 1).
This model resulted in an estimated acetabular size of 54.96 Β±3.41 mm. This model was significant
overall to predict the acetabular size (Mean Square = 1331.40, F(6,461) = 302.338, p <.0001).
The Femoral component
The backward stepwise model indicated that the femoral component size could only be
significantly determined by the preoperative femoral component measurements. This model had a
π 2 = .727 with Adjusted π 2 = .723 and standard error of (SE=0.935). Femoral component model
was developed based on a fully factored model, including the preoperative measurements of the
patients to develop a prediction model for estimating the femoral size.
Femoral component model
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
10
πΉπππΈπ π‘ππππ‘ππ = 3.387 + 0.016 Γ π΄ππ‘ππππ + 0.814 Γ πΉππππππ β 0.018 Γ π»πππβπ‘ + 0.018 Γ ππππβπ‘
β 0.061 Γ π΅ππΌ +
In which πΉπππΈπ π‘ππππ‘ππ is the estimated size of the femoral stem.
This model resulted in an estimated femoral size of 4.82 Β±1.51 mm. This model was significant
overall to predict the femoral size (Mean Square = 178.32, F(6,460) = 204.152, p <.0001).
Figure 2 β The accuracy of the model to retrospectively predict the implant sizes in the current study. The femoral
component size was predicted accurately in 97.2% of cases of within Β±2 size and more than 99.1% for acetabular
component size within Β±2 size.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
11
For the acetabular component, two cases were predicted within +/- 3 size and two cases were
predicted within +/-4 size. However, for the femoral component, the model predicted with less
accuracy, and in 11 cases the implant size was correctly predicted in +/- 3 size and in two cases
within +/- 4 size as depicted in Figure 3.
Figure 3 β The distribution of the prediction error for both Acetabulum cup size and femoral stem size.
Accuracy improvements
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
12
The results of the McNemarβs test and McNemarβs exact test are summarized in Appendix 1. The
outcomes suggest that using the model overall improved accuracy of the templating as summarized
in
Table 3. The improvement in accuracy was significant in the acetabular component within +/- 1 size
(2(1, π = 468) = 19.314, p<0.0001, =0.604, and odds-ratio: 7.750 (95% CI 2.740-30.220)).
The details of the test and cross-tab tables are presented in the Appendix.
Discussion:
The use of digital radiography and preoperative templating is increasingly common and can
improve the success of joint replacement surgery [3]. Pre-surgical planning is critical in the
performance of joint replacement surgery and should be used to optimize patient outcomes. Pre-
surgical planning can also be used to enhance OR efficiency and implement cost savings through
inventory and supply chain management. Operating room efficiency depends on a coordinated
team effort, the use of specialized equipment, and the shared knowledge of pertinent patient
information. We routinely use preoperative templating to streamline our surgical efforts, minimize
the use of unnecessary equipment, optimize patient outcomes, and minimize cost.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
13
In this study we used a multiple regression model to identify factors that would allow us to predict
component size during hip replacement surgery to within +/- 1 size. For the acetabulum, the
significant factors were the templated acetabular size, templated femoral size, weight, and height.
For the femoral component, the preoperative femoral templating was the only significant factor.
In our retrospective review of 468 patients, the exact acetabular implant size was identified by the
model with an accuracy of 48.93%, which was slightly higher than results achieved by using only
preoperative digital templating size to predict the exact acetabular cup size (i.e., 48.08%).
Application of the model was further justified when the gap between the accuracy of identifying
proper acetabular cup within +/- 1 size of the actual implant was improved significantly by utilizing
the model. This statistically significant increase was a more than 5% improvement (27 cases in
this study). The high effect size of the findings (indicated by >0.4) and the significance of the
improvement suggest that utilizing the model will enhance the accuracy of preoperative
templating.
Table 3 depicts the improvement in accuracy of identifying the appropriate acetabular size while
using the model. Similar outcome was found in predicating the actual femoral stem size, and the
application of the model improved the accuracy of identifying the proper size (
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
14
Table 3).
Our results resemble the findings of Della Valle et al., who reported 99% prediction within one
size for the acetabular component and 99% within two sizes for the femoral component [16]. Our
results were similar to previous studies, including those conducted with TraumaCad [10].
Table 3 depicts the differences between the predicted model versus the actual implant size
(acetabular and femoral). The error is larger in the acetabular cup size. This may be due to the
limited number of commercially available acetabular cup sizes.
Table 3 β The accuracy of the identifying the proper implant component size while utilizing the model versus the
preoperative templating size. Total number of cases was 468.
Acetabular component size
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
15
Method Exact Β±1 size Β±2 size
N % N % N %
Model 229 48.93 437 93.38 464 99.13
Preop Size 225 48.08 410 87.60 462 98.70
Change 4 0.85 27 5.77 2 0.43
Femoral Component Size
N % N % N %
Model 216 46.15 415 88.74 462 98.7
Preop Size 209 44.65 414 88.52 460 98.48
Change 5 1.07 1 0.22 2 0.44
In our study we were limited by the total number of personnel who performed the preoperative
templating; hence, we were not able to evaluate the interrater reliability factor and measure
potential errors that may be added to the model if different evaluators are used such as those
reported by Maratt et al. [8]. However, our assumption in the current study was that the
employment of digital radiography and application of standard templating techniques could
minimize this potential error. Future studies may focus on retrospectively estimating
preoperative implant sizes by using different trained technicians. Another limitation is in the
difficulty in templating patients with severe bone loss or proximal femoral or acetabular
deformity. We analyzed the patients whose predicted model was more than +/- 2 sizes different
from the actual implant size used at the time of surgery and found that, in each case, there was
severe distortion of the hip anatomy that made templating difficult and subsequently inaccurate.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
16
An example of such case is depicted in
Figure 4. These cases typically required preoperative CT scan to assess the anatomy, assess bone
stock for implant fixation, and identify specific component needs (i.e., augments, cages, revision
stems, and the need for bone grafting). Thus, employing preoperative templating should be used
with caution in more complex cases in which the hip anatomy is severely distorted.
All cases were templated the week before the actual procedure. Single instrument trays were
prepared, consisting of a basic set of surgical tools in addition to the specifically sized implants
and size-matched tools. Only specifically required instruments were standard components of the
tray. Additionally, there was a general backup set in case of significantly inaccurate templating
results (size according to templating off by more than two sizes). However, there was a 99%
success rate using the templating technique, which meant the backup was rarely required.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
17
Figure 4 β Anterior-posterior view X-ray of an example case that was not templated within 2 implant sizes. The
margins of the acetabulum and femoral head are not clearly shown in the X-ray due to severe deformity of the limb.
In conclusion, this study presented a prediction model to better estimate the actual size of implants.
We routinely use this information to create patient-specific trays to simplify patient care, improve
OR efficiency, and optimize patient outcomes. This practice can also produce significant cost
savings in the delivery of joint replacement care.
Acknowledgements
The authors would like to thank Mr. David Balderree for his assistance with preparing the collected
data. We would also like to thank Mr. Brian Caballero and Ms. Lauren Hennecke for their
assistance with pre and post-operative data collection. Finally, we are grateful to Dr. Ashish Arya
for assistance with editing this manuscript and providing insightful comments.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
18
Appendix
Femoral
Exact Size
Model
Accurate
No Yes
Templated
Acc
ura
cy
No 235 24
Yes 17 192
(2(1, π = 468) = 0.878, p=0.3487)
+/- 1 Femoral Size
Model
Accurate
No Yes
Templated
Acc
ura
cy
No 48 6
Yes 5 409
Exact McNemarβs test results indicated that the two-sided probability was p=.5 thus in +/- 1 size no
significant improvement was provided by the model.
π = 2 Γ (116
) Γ 0.52 Γ 0.50 = 0.45
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
19
+/- 2 Femoral Size
Model
Accurate
No Yes
Templated A
ccura
cy
No 6 2
Yes 0 460
Exact McNemarβs test results indicated that the two-sided probability was p=.5 thus in +/- 2 size no
significant improvement was provided by the model.
π = 2 Γ (22
) Γ 0.52 Γ 0.50 = 0.5
Acetabular
Exact Size
Model
Accurate
No Yes
Templated
Acc
ura
cy
No 198 45
Yes 41 184
(2(1, π = 468) = 0.105, p=7463)
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
20
+/- 1 Size
Model
Accurate
No Yes
Templated A
ccura
cy
No 27 31
Yes 4 406
(2(1, π = 468) = 19.314, p<0.0001, =0.604, and odds-ratio: 7.750 (95% CI 2.740-30.220))
+/- 2 Acetabular Size
Model
Accurate
No Yes
Templated
Acc
ura
cy
No 2 4
Yes 2 460
Exact McNemarβs test results indicated that the two-sided probability was p=.5 thus in +/- 2 size no
significant improvement was provided by the model.
π = 2 Γ (64
) Γ 0.52 Γ 0.50 = 0.47
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
21
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
22
Reference
[1] Valle A Della. Preoperative planning for primary total hip arthroplasty. J Am β¦
2005;13:455β62.
[2] Eggli S, Pisan M, MΓΌller ME. The value of preoperative planning for total hip
arthroplasty. J Bone Joint Surg Br 1998;80:382β90.
[3] Davila J a, Kransdorf MJ, Duffy GP. Surgical planning of total hip arthroplasty: accuracy
of computer-assisted EndoMap software in predicting component size. Skeletal Radiol
2006;35:390β3. doi:10.1007/s00256-006-0106-4.
[4] Pullen WM, Whiddon DR. Accuracy and reliability of digital templating in primary total
hip arthroplasty. J Surg Orthop Adv 2013;22:148β51.
[5] Shaarani SR, McHugh G, Collins D a. Accuracy of digital preoperative templating in 100
consecutive uncemented total hip arthroplasties: a single surgeon series. J Arthroplasty
2013;28:331β7. doi:10.1016/j.arth.2012.06.009.
[6] Iorio R, Siegel J, Specht LM, Tilzey JF, Hartman A, Healy WL. A comparison of acetate
vs digital templating for preoperative planning of total hip arthroplasty: is digital
templating accurate and safe? J Arthroplasty 2009;24:175β9.
doi:10.1016/j.arth.2007.11.019.
[7] Whiddon DR, Bono J V, Lang JE, Smith EL, Salyapongse AK. Accuracy of digital
templating in total hip arthroplasty. Am J Orthop (Belle Mead NJ) 2011;40:395β8.
[8] Maratt JD, Srinivasan RC, Dahl WJ, Schilling PL, Urquhart AG. Cloud-based
preoperative planning for total hip arthroplasty: a study of accuracy, efficiency, and
compliance. Orthopedics 2012;35:682β6. doi:10.3928/01477447-20120725-05.
[9] Steinberg EL, Eitan S. Pre-operative Planning Using the Traumacad TM Software System.
US Radiol 2010;2:87β90.
[10] Steinberg EL, Shasha N, Menahem A, Dekel S. Preoperative planning of total hip
replacement using the TraumaCadTM system. Arch Orthop Trauma Surg 2010;130:1429β
32. doi:10.1007/s00402-010-1046-y.
[11] Wedemeyer C, Quitmann H, Xu J, Heep H, von Knoch M, Saxler G. Digital templating in
total hip arthroplasty with the Mayo stem. Arch Orthop Trauma Surg 2008;128:1023β9.
doi:10.1007/s00402-007-0494-5.
The Journal of Arthroplasty
Available online 26 November 2014 doi:10.1016/j.arth.2014.11.021
23
[12] Hossain M, Lewis J, Sinha a. Digital pre-operative templating is more accurate in total hip
replacement compared to analogue templating. Eur J Orthop Surg Traumatol
2008;18:577β80. doi:10.1007/s00590-008-0356-z.
[13] Gamble P, de Beer J, Petruccelli D, Winemaker M. The accuracy of digital templating in
uncemented total hip arthroplasty. J Arthroplasty 2010;25:529β32.
doi:10.1016/j.arth.2009.04.011.
[14] Turgay E, ZAyAT B El. Precision of preoperative digital templating in total hip
arthroplasty. Acta OrthopΓ¦dica β¦ 2011;77:616β21.
[15] King RJ, Makrides P, Gill J a, Karthikeyan S, Krikler SJ, Griffin DR. A novel method of
accurately calculating the radiological magnification of the hip. J Bone Joint Surg Br
2009;91:1217β22. doi:10.1302/0301-620X.91B9.22615.
[16] GonzΓ‘lez Della Valle A, Slullitel G, Piccaluga F, Salvati E a. The precision and usefulness
of preoperative planning for cemented and hybrid primary total hip arthroplasty. J
Arthroplasty 2005;20:51β8. doi:10.1016/j.arth.2004.04.016.