Journal of Biomechanics ()

Contents lists available at ScienceDirect

Journal of Biomechanics
journal homepage: www.elsevier.com/locate/jbiomech
www.JBiomech.com

Implicit modeling of screw threads for efcient nite element analysis

of complex bone-implant systems
Jason A. Inzana n, Peter Varga, Markus Windolf
AO Research Institute, Clavadelerstrasse 8, 7270 Davos Platz, Switzerland

art ic l e i nf o

a b s t r a c t

Article history:
Accepted 19 April 2016

Finite element analysis is commonly used to assist in the development and evaluation of orthopedic
devices. The physics of these models are simplied through approximations that enable more efcient
simulations, without compromising the accuracy of the relative comparisons between implant designs or
congurations.
This study developed and evaluated a technique to approximate the behavior of a nely threaded
screw using a smooth cylinder with the threads implicitly represented through interfacial contact conditions. This pseudo-threaded model was calibrated by comparing to simulations that explicitly modeled
the thread geometry with frictional contact. A parametric analysis was performed with a single screw-inbone system, ve loading directions, and three Young's moduli that span the range of cancellous bone
(200, 600, and 1,000 MPa). Considering that screw cut-out from cancellous bone is a critical clinical issue
in the osteoporotic proximal humerus, the pseudo-threaded model was compared with a bonded
interface to examine three different screw congurations in a 3-part proximal humerus fracture across 10
patients.
In the single screw-in-bone system, the pseudo-threaded model predicted the screw displacement of
the explicitly threaded model with 15% difference and estimated the strain distributions and magnitudes more accurately than a bonded interface. Yet, the relative comparisons of implant stability across
the three different screw congurations in the proximal humerus were not affected by the modeling
choice for the bonescrew interface. Therefore, the bonded interface could serve as a more efcient
methodology for making relative comparisons between implants that utilize the same thread prole.
& 2016 Published by Elsevier Ltd.

Keywords:
Bonescrew interface
Continuum nite element analysis
Proximal humerus
Fracture xation

1. Introduction
Orthopaedic implant development and evaluation is traditionally based on surgical experience in combination with engineering
expertize and experimental biomechanical testing. Computational
simulations, particularly through nite element (FE) analysis, can
be an important assistive tool in this process (Taylor and Prendergast, 2014). In contrast with laboratory experiments, simulations enable a more efcient screening process for new design
ideas or research questions and can provide dramatic cost savings
as well as a reduced need for valuable tissue samples.
To maximize the utility of FE simulations for large parametric
analyses, the physics of the model are often simplied through
approximations to help ensure numerical convergence and save
n

Corresponding author. Tel.: 41 81 414 23 29.

E-mail addresses: jason.inzana@aofoundation.org (J.A. Inzana),
peter.varga@aofoundation.org (P. Varga),
markus.windolf@aofoundation.org (M. Windolf).

computational time. Yet, it is critical that such approximations do

not compromise the accuracy of the conclusions that are drawn
from the analyses. In regard to the evaluation of orthopaedic
trauma devices, the physical behavior of the bonescrew interface
prior to osseointegration of the implant is not well understood
(Steiner et al., 2015). Many studies that utilize FE analysis to assess
the mechanical competence of bone-implant systems have modeled the bonescrew interface as perfectly bonded (Dubov et al.,
2011; Kennedy et al., 2013; Maldonado et al., 2003). While this
bonded approximation greatly simplies the numerical analysis, it
is more appropriate in the case of osseointegration rather than
assessment of primary implant stability, particularly in osteoporotic bone (Wirth et al., 2010). In contrast, some studies have
examined the effects of using frictional or frictionless sliding
contact at the interface between threaded screws and bone
(Huang et al., 2008; Karunratanakul et al., 2013; MacLeod et al.,
2012). Yet, the computational cost to explicitly model the geometry of nely threaded screws is orders of magnitude greater
than approximating the geometry as a smooth cylinder. High

http://dx.doi.org/10.1016/j.jbiomech.2016.04.021
0021-9290/& 2016 Published by Elsevier Ltd.

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i

J.A. Inzana et al. / Journal of Biomechanics ()

computational costs are inhibitory to parametric or statistical

population analyses that may require hundreds or thousands of
simulations of a complex bone-implant system, such as the
proximal humerus internal locking system (PHILOS) plate (DePuy
Synthes Inc.; Oberdorf, Switzerland), which may use up to 12
screws. Screws with ne threads are used for anchorage in both
cortical and cancellous bone at many anatomical sites, such as the
humerus, tibia, bula, and pelvis. Screw xation in the cancellous
bone of the osteoporotic proximal humerus is particularly challenging for surgeons (Kralinger et al., 2014; Roderer et al., 2011;
Sudkamp et al., 2009), making this site a vital target for research
and innovation.
This study examined the hypothesis that a nely threaded
screw could be approximated as a smooth cylinder, with the
threads implicitly represented through interfacial contact conditions. The interfacial modeling strategy was calibrated against the
threaded case using a single screw-in-bone system with bone
moduli in the range of cancellous bone. The calibrated modeling
strategy was then applied to a relevant full bone-implant system,
which compared the primary implant stability (i.e. xation stiffness) and peri-implant bone strains of three different screw

congurations for xation of an unstable 3-part proximal humerus

fracture in ten subjects.
2. Materials and methods
2.1. Modeling of a single screw in homogenous bone-like material
2.1.1. Model geometries and material properties
An idealized screw was drawn in SolidWorks (Dassault Systmes SolidWorks Corp.;
Waltham, MA, USA) to represent the thread geometry of the DePuy Synthes 3.5 mm
locking screw. The screw was drawn with a major diameter of 3.5 mm and a triangular
threading, with a thread depth of 0.27 mm, pitch of 0.85 mm, and thread width of
0.35 mm at the triangle base (Fig. 1a). A simplied cylindrical screw model was created
with a 2.96 mm diameter to match the minor diameter of the threaded screw. The
screw models were virtually inserted to a depth of 12 mm in a 20 mm diameter, 15 mm
long bone cylinder. The bone cylinder and screw models were meshed using quadratic
tetrahedral elements and assigned homogenous, isotropic, linear elastic material properties using the Simpleware ScanIP software suite (Simpleware Ltd.; Exeter, UK). There
were 201,919 nodes in the threaded model and 52,217 nodes in the simplied model
after mesh convergence (Appendix A.1.). Poisson's ratio of the bone cylinder was 0.3 and
Young's modulus was assigned as 200, 600 or 1,000 MPa to simulate cancellous bone
(Morgan et al., 2003) or 17,000 MPa to simulate cortical bone (Bayraktar et al., 2004).
The screws were modeled as titanium alloy with a Young's modulus of 105 GPa and a
Poisson's ratio of 0.3.

Fig. 1. Principal strain distributions are more closely replicated by a pseudo-threaded interface than a tied (perfectly bonded) interface. A threaded screw (a) and a simplied
geometry screw (b) were inserted into a homogenous cylinder with Young's modulus in the range of cancellous bone (200, 600 or 1,000 MPa). Each screw was evaluated
under ve different loading directions, ranging from 0 to 180 (c). A frictional bonescrew interface was used for the threaded screw model. The bonescrew interface of the
simplied geometry screw (b) was modeled as uniformly tied (i.e. perfectly bonded) or as a cohesive-frictional interface, with an anisotropic cohesion model on the pseudothreads and isotropic friction in between. Contour plots of the maximum (absolute value) principal strain qualitatively demonstrate similarities and differences between the
models (dr). Bone E 600 MPa, friction coefcient 0.3 in the threaded and pseudo-threaded models, pseudo-thread thickness ratio 0.5, cohesive stiffness (Kss) 104 N/
mm3.

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i

J.A. Inzana et al. / Journal of Biomechanics ()

2.1.2. Interfacial contact properties, loading, and boundary conditions
The interaction between the threaded screw and bone cylinder was modeled as
a frictional interface using isotropic Coulomb friction. The friction coefcient was
varied (0.1, 0.3 or 0.5) to assess the sensitivity to this parameter (Huang et al., 2008;
MacLeod et al., 2012; Pessoa et al., 2011). The bonescrew interface for the simplied cylindrical screw model was simulated with either a tied interface or with
the pseudo-threaded interfacial modeling strategy. The tied models employed tie
constraints between the bone and simplied screw surfaces. For the pseudothreaded models, a helical set of nodes was selected such that the helix pitch and
thickness approximated the geometry of the threaded screw (Fig. 1b). The node
density at the bonescrew interface was 29 nodes/mm2. The mesh density was
varied to achieve an interfacial density of 9 nodes/mm2 and 49 nodes/mm2 to
assess the sensitivity of the pseudo-threaded model to this interfacial node density.
Isotropic Coulomb friction was applied in between the pseudo-threads, while the
pseudo-thread interaction was modeled as anisotropic cohesion. The friction
coefcient, cohesive stiffness coefcient, and thickness of the pseudo-thread region
were based on the calibration study described below. All frictional contact was
enforced with the penalty method using a small sliding formulation with the
allowable elastic slip set to 0.005. Normal contact overclosure was modeled with
the default linear penalty method in Abaqus v6.13 (Simulia; Dassault Systmes;
Waltham, MA, USA).
A 50 N concentrated force was applied to a control node that was positioned on
the screw axis, 1.5 mm above the screw head. The nodes on the top surface of the
screw head were constrained to this control node through a kinematic coupling.
The force angle was varied from 0 (axial pulling) to 180 (axial pushing) in 45
increments and pinned boundary conditions were applied to the bottom and sides
of the bone cylinder (Fig. 1c). Each FE model was solved using Abaqus software
v6.13 with 4 cores on a server with Intel Xeon X5690 (3.5 GHz) processors and
sufcient RAM to avoid using swap space.

2.2.2. Loading, boundary conditions, and interface conditions

A published in vitro experimental loading protocol for testing proximal
humerus xation (Unger et al., 2012) was simulated (Fig. 5b; details in Appendix
A.3.). The locked screw-plate interface was simulated as bonded (Synek et al.,
2015). The bonescrew interface was modeled either as tied or pseudo-threaded
using the optimal parameters from the single screw study: Kss equal to 104 N/mm3,
a friction coefcient of 0.3, and a pseudo-thread thickness of 0.18 mm. Potential
contact between the bone and plate was modeled as isotropic Coulomb friction
with a friction coefcient of 0.1 (Nedoma and Stehlik, 2011).
2.2.3. Data analysis
Cylindrical ROIs that were 15 mm long and 8 mm in diameter were used to
calculate the maximum principal compressive strain around each screw tip. This
strain value was then averaged over the humeral head screws. Two virtual points
were positioned 10 cm lateral to the humeral head center. One point followed the
rigid motion of the proximal section of the PHILOS plate and the other point followed the lesser tubercle. Similar to the study of Unger et al. (2012) the angle
formed by the divergence of these two points was used to quantify the stability of
the humeral head fragment relative to the plate. The three screw congurations
were compared within each bonescrew interface using a repeated measures oneway ANOVA with Tukey's correction for multiple comparisons. Data were conrmed to satisfy assumptions of a Gaussian distribution based on the D'AgostinoPearson normality test or a non-parametric equivalent (Friedman test) was used.

3. Results
3.1. Analysis of a single screw in a homogenous bone-like material

2.1.3. Calibration of the pseudo-thread modeling parameters

The pseudo-thread modeling parameters were calibrated through a 33 full
factorial analysis. The friction coefcient between pseudo-threads was equal to 0.3,
1.0, or 1.7 and the contact thickness of the pseudo-threads was equal to 0.5, 1.0, or
1.5 times the original thickness (0.35 mm). The cohesive stiffness coefcients on
the pseudo-thread nodes equaled 102, 103, or 104 N/mm3 along the screw axis (Kss)
and zero in the normal (radial) direction (Knn) and the circumferential direction
(Ktt). Each parameter combination was simulated under ve loading directions (0,
45, 90, 135, and 180) and three bone modulus values in the range of cancellous
bone (200, 600, and 1,000 MPa).
An objective function was utilized to score the difference in stiffness and
principal strains between the pseudo-threaded models and the threaded models
(Appendix A.2.). A generalized linear model with second order terms was t to the
overall difference score as a function of the three parameters using R software (R
Foundation for Statistical Computing; Vienna, Austria). The model was interpolated
to determine the parameter values that achieved the minimum difference score.

2.2. Comparison of the pseudo-threaded and tied interfaces in a whole bone-implant

system
2.2.1. Generation of the subject-specic nite element models
Bone models were developed from high-resolution peripheral quantitative
computed tomography scans (HR-pQCT; XtremeCT, Scanco Medical; Bassersdorf,
Switzerland) of 10 left-sided proximal humeri. The PHILOS plate (DePuy Synthes
Inc.; Oberdorf, Switzerland) was virtually instrumented onto each bone according
to the DePuy Synthes technique guide. Three clinically relevant combinations of
humeral head locking screws were simulated in each subject (Fig. 5a). The screws
were modeled as idealized cylinders with a 3.0 mm diameter. Virtual osteotomies
were performed that simulate an unstable three fragment fracture with medial
comminution (AO 11-B3.2). Further details on model geometry are available in
Appendix A.3.
The models were meshed using quadratic tetrahedral elements with a mean
edge length of 0.97 mm and 720,000 770,000 nodes (Appendix A.1.). The node
density at the bonescrew interface was 12 nodes/mm2. The bone Young's modulus
was assigned element-wise according to the BMD-modulus relationship employed
by Dragomir-Daescu et al. (2011):
E 146641:49 MPa
where E is Young's modulus and is the element's apparent BMD obtained from
the down-sampled HR-pQCT scan. Poisson's ratio was set as 0.3. Considering that
this BMD-modulus relationship was derived from proximal femur data, a dramatically different relationship (E 175463 MPa, based on (Carter and Hayes, 1977))
was also simulated to assess the sensitivity to this modeling choice. The titanium
plate and screws were assigned a linear elastic Young's modulus of 105 GPa with
0.3 as Poisson's ratio. Implant placement, virtual osteotomies, meshing and material property assignments were performed using ScanIP software (Simpleware Ltd.;
Exeter, UK).

3.1.1. Mesh and friction coefcient sensitivity analysis

Varying the friction coefcient from 0.1 to 0.3 to 0.5 on the
threaded bonescrew interface resulted in 5.1% 71.7% and 6.0%
73.4% variability in the screw head displacement and maximum
principal compressive strain, respectively, averaged over the three
cancellous bone moduli and ve loading directions. Therefore, the
friction coefcient was kept as 0.3 during subsequent analyses.
Varying the mesh density of the pseudo-threaded models, such
that the interfacial surface node density ranged from 9 nodes/mm2
to 49 nodes/mm2, changed the displacement and strain outcomes
by 3.7% 70.8% and 4.8% 75.0%, respectively over the three bone
moduli and ve loading directions, suggesting negligible mesh
dependence.
3.1.2. Calibration of the pseudo-thread modeling parameters
The minimum difference score was achieved when Kss equaled
104 N/mm3, the friction coefcient equaled 0.3, and the pseudothread thickness was 0.18 mm. Although this combination was on
the edge of the parameter space, the score variations were converging in this region and did not exceed 10%. Thus, further optimization would not substantially improve the prediction.
3.1.3. Comparison of tied, pseudo-threaded, and threaded interface
models
The average analysis time was 255 740 min for the threaded
models, 5.4 7 0.9 min for the pseudo-threaded models, and
2.8 70.4 min for the tied models. Qualitatively, the principal strain
distributions of the threaded models were more accurately
represented by the pseudo-threaded interface compared to the
tied interface on the simplied screw geometry (Fig. 1dr), especially during off-axis loading (45, 90, and 135).
For the cancellous bone moduli (i.e. 200, 600, and 1,000 MPa),
the difference in maximum principal strains, relative to the
threaded models, was 85% lower for the pseudo-threaded interface compared to the tied interface in the off-axis loading cases;
however, the differences were more similar for axial load cases
(Fig. 2). The differences in displacement of the screw head ranged
from 1% to 5% for the pseudo-threaded interface and from 4% to
41% for the tied interface, relative to the threaded models (Fig. 4a).

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i

J.A. Inzana et al. / Journal of Biomechanics ()

Fig. 2. Maximum principal strains are more closely replicated by the pseudo-threaded interface than the tied interface in the Young's modulus range for cancellous bone
(200, 600, and 1,000 MPa). The maximum principal compressive (a) or tensile (b) strain within each concentric circle region of interest (ROI) surrounding the screw was
normalized to the strain observed in the threaded model at the greatest radial distance (dotted line 1). The normalized strains in (a and b) are the mean and standard
deviation across the ve loading directions and three bone modulus values. The differences in strain from the pseudo-threaded models and tied models, relative to the
threaded models, were summed over the ROIs and averaged across the three different cancellous bone moduli (c and d). Values are means and error bars are standard
deviations.

Fig. 3. Maximum principal strains are more closely replicated by the pseudo-threaded interface than the tied interface in the Young's modulus range for cortical bone (
17,000 MPa). The maximum principal compressive (a) or tensile (b) strain within each concentric circle region of interest (ROI) surrounding the screw was normalized to the
strain observed in the Threaded model at the greatest radial distance (dotted line 1). The normalized strains in (a and b) are the mean and standard deviation across the ve
loading directions. The differences in strain from the pseudo-threaded models and tied models, relative to the threaded models, were summed over the ROIs (c and d).

Similar results were observed when the bone cylinder was

simulated as cortical bone (E17,000 MPa). The pseudo-threaded
model closely estimated the threaded model strains, except directly

adjacent to the screw where strains were overestimated (Fig. 3). The
differences in screw head displacements, relative to the threaded
models, ranged from 1% to 10% for the pseudo-threaded models and

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i

J.A. Inzana et al. / Journal of Biomechanics ()

Fig. 4. Screw displacements in an explicitly threaded model are more closely replicated by the pseudo-threaded interface than the tied interface. The displacement
magnitude of the screw head under each loading direction was normalized to the displacement of the threaded model. The normalized displacement was averaged over the
three cancellous bone modulus values (200, 600, and 1,000 MPa) that were used for model calibration (a). The calibrated model was further tested with a bone modulus of
17,000 MPa (b). Values are mean and error bars are standard deviation.

from 4% to 28% for the tied models over the ve loading directions
(Fig. 4b).
3.2. Comparison of the pseudo-threaded and tied interfaces in a
whole bone-implant system
The average analysis time was 296 747 min for the pseudothreaded models and 7.571.5 min for the tied models. Similar to
the single screw simulations, the pseudo-threaded models had
consistently higher strains (49%720%) and humeral head rotations (30%7 10%) compared to the tied models.
When the proximal humerus fracture was managed with the
addition of calcar screws (Screws 8, 9; With Calcar), the maximum
principal compressive strains (Fig. 5c) and the humeral head
rotation relative to the plate (Fig. 5d) were signicantly reduced
compared to the other two screw combinations (Without Calcar,
Without Calcar 5,6). Addition of the fth and sixth screws
(Without Calcar 5,6) signicantly reduced the humeral head
rotation, but not the bone strain, compared to the four most
proximal screws alone (Without Calcar). The conclusions regarding the effects of different screw patterns on humeral head stability, based on p o0.05, were not affected by the use of a tied
versus pseudo-threaded bonescrew interface.
Principal strain contour plots from the instrumented humerus
models qualitatively demonstrate that the groups with the fth
and sixth screws (Without Calcar 5,6) or the eighth and ninth
screws (With Calcar) had reduced strain concentrations in the
proximal region and an overall reduction in peri-implant bone
strains compared to the group using only the four most proximal
screws (Without Calcar; Fig. 6).
When the simulations were performed with the alternative
BMD-modulus relationship (E 175463), the statistical comparisons across the three screw combinations were identical between
a tied and pseudo-threaded interface (Supplemental Figure S1).
The With Calcar group had signicantly lower principal compressive strains than the Without Calcar 5,6 group (p 0.029)
and lower humeral head rotations compared to the Without Calcar
group (p 0.0001).

4. Discussion
This study developed a computationally efcient modeling
technique for the bonescrew interface that avoids modeling a
nely threaded geometry. The pseudo-threaded models accurately
reproduced the apparent stiffness of the threaded model geometry

and represented the peri-implant bone strains of the threaded

model more accurately than a tied interface. The improvements in
modeling accuracy were especially true for the off-axis loading
cases (45, 90, and 135), which are more likely to represent
physiological screw loads than pure axial loading. The greater
difference of the tied interface relative to the threaded model is
largely attributable to the articial tension that develops (positive
strain, Figs. 1o, p and q and 6f). The pseudo-thread cohesive
stiffness normal to the screw surface is zero, but some articial
tension can develop due to shear displacement along the screw
axis that is associated with screw bending and sliding (Fig. 1j and
k).
Although the pseudo-thread modeling parameters were calibrated in a modulus range corresponding to cancellous bone (200
1,000 MPa), the pseudo-threaded model accurately reproduced
the apparent stiffness (screw head displacement) and peri-implant
strains of the threaded model in simulated cortical bone
(E 17,000 MPa) as well. Therefore, the pseudo-threaded modelling strategy is not sensitive to the bone Young's modulus and
could be applicable to both cancellous and cortical bone.
The effects of the bonescrew interface model were examined
in a preliminary study of a clinically relevant research question
related to xation of a proximal humerus fracture. The same
conclusions regarding the relative xation stability could be drawn
from using both the tied interface and the pseudo-threaded
interface. Although some variation existed in regard to the magnitude of differences between groups, the population-based trends
and the statistical conclusions were consistent. Similarly, MacLeod
and colleagues modeled tibial shaft xation using a tied or frictional interface and observed no signicant differences in the local
peak strains or global stiffness ( o5% deviation) between these
different interface modeling strategies, but substantial differences
in the strain distributions were observed (MacLeod et al., 2012). In
contrast, Karunratanakul et al. (2010) observed that a perfectly
bonded interface was 50% stiffer than when the threaded geometry was modeled with frictional or frictionless sliding contact.
However, it is important to recognize that the model of Karunratanakul et al. (2010) was more highly constrained than that of
MacLeod et al. (2012), which may partly explain the discrepancy in
results. This observation of an overestimated stiffness with a
bonded interface model is consistent with the ndings of the
current study, which also simulated rigidly constrained boundary
conditions. Yet, the relative comparisons between the three different screw congurations (With Calcar vs. Without Calcar vs.
Without Calcar 5,6) in the proximal humerus were not sensitive
to the bonescrew interface model (tied vs. pseudo-thread).

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i

J.A. Inzana et al. / Journal of Biomechanics ()

Fig. 5. In a whole bone-implant system, relative comparisons between xation strategies that use the same screw type may be independent of the bonescrew interface
modeling strategy. An unstable 3-part proximal humerus fracture, stabilized with a PHILOS plate and one of three locking screw combinations (a), was simulated according
to a published experimental loading setup (b). The maximum principal compressive strain in the cancellous bone elements was measured from an 8 mm diameter  15 mm
long cylindrical region of interest around the tip of each screw, and this strain value was averaged over the 46 screws in the humeral head (c). Additionally, virtual points
were created to follow the average motion of the proximal section of the plate and the lesser tubercle (Humeral Head) to track the overall rotation of the humeral head
relative to the plate (d). The three screw congurations were compared within each bonescrew interface model using a repeated measures one-way ANOVA with Tukey's
correction for multiple comparisons. The multiplicity-adjusted p values for the comparisons within each model type are included in (c and d). Note that the strain and
deformation values are higher in the Pseudo-threaded models compared to the Tied, but the statistical comparisons within each model type are similar.

Calibration of the pseudo-thread parameters was based on a

fully threaded screw in a continuum model rather than directly
comparing with experimental data. Previous studies have
demonstrated that pull-out strength and stiffness can be predicted
by local bone properties such as the trabecular bone volume
fraction, bone mineral density, or local Young's modulus of the
bone near the screw (Tingart et al., 2006; Wirth et al., 2011). Wirth
et al. (2012) observed a good correlation (r2 0.78) between the
pull-out stiffness estimated by experimentally validated micro-FE
methods and continuum models. Therefore, it is likely that a
threaded continuum model with modulus values based on the
local bone mineral density could provide reasonable estimations of
experimental stiffness. In contrast to stiffness, the strain distributions observed in the study by Wirth et al. (2012) were substantially different between the continuum and micro-FE models,
suggesting that bone-implant continuum models should focus on
stiffness-based outcomes.
In the proximal humerus part of this study, the calcar screws
(Screws 8 and 9) signicantly improved primary implant stability
and reduced the peak bone strains around the screws. Similarly,
previous experimental studies have demonstrated the biomechanical benets of the calcar screws for primary xation stability (Bai
et al., 2014; Ponce et al., 2013; Zhang et al., 2014). Further, the use

of such medial support screws has signicantly reduced the rates

of xation failure and better maintained the anatomical reduction
in clinical studies (Gardner et al., 2007; Zhang et al., 2011).This
good correspondence with experimental and clinical ndings
provides initial corroboration of the simulation strategies utilized
herein.
There are a few limitations of this study. First, the calibration of
the pseudo-thread model parameters was performed for a single
thread prole and may not be directly translatable to other thread
proles without re-calibration. Further, the thread depth was
small (0.27 mm) and the simplication to a at cylindrical surface
may not be appropriate for large thread depths. Second, the
simulations were limited to elastic analyses and failure of the
bonescrew interface was not evaluated. Third, a threaded screw
model could not be evaluated for comparison in the proximal
humerus study due to the unmanageable computational costs.
Since the relative comparisons between the three screw patterns
were not sensitive to the tied vs. pseudo-thread interface, it is
likely that similar outcomes would have been observed with a
threaded, frictional interface. Finally, the BMD-modulus relationship used for the proximal humerus is based on data from the
proximal femur, but varying this relationship did not alter the

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i

J.A. Inzana et al. / Journal of Biomechanics ()

Fig. 6. Contour plots of the maximum (absolute value) principal strains demonstrate the redistribution and overall reduction in strain with the addition of screws 5 and 6 (b,
e, h, and k) or the calcar screws (screws 8 and 9; a, d, g and j) compared to the four most proximal screws only (c, f, i and l). Subject 1 was the most osteoporotic
(BMD69 mg HA/cm3) and is shown with a tied bonescrew interface (ac) and a pseudo-threaded interface (df). Note the large, articial tensile strain regions (positive
strain) in the tied case (arrows, c). The high compressive strain region (negative strain, black) around the calcar screw (a anf d) in osteoporotic Subject 1 is attributable to the
lack of bone (i.e. only marrow) in this region.

comparisons between the interface models or the apparent biomechanical benet of the calcar screws.

Appendix
A.1. Mesh convergence and sensitivity analyses

Conict of interest disclosure

The authors are not compensated and there are no other
institutional subsidies, corporate afliations, or funding sources
supporting this work unless clearly documented and disclosed.

Acknowledgments
This investigation was performed with the assistance of the AO
Foundation via the AOTRAUMA Network (Grant number
AR2008_01). Jason Inzana was supported in part by a Whitaker
International Program Post-doctoral Research Scholarship.

For the calibration study in the simplied bone cylinder, there

were 52,217 nodes in the simplied model and 201,919 nodes in
the threaded model. Mesh convergence analysis was performed
with the number of nodes ranging from 20,024 to 232,256 nodes
in the simple model and from 135,986 to 383,545 nodes in the
threaded model. The displacement of the screw head control node
and the average principal compressive strain in the cylindrical
regions of interest around the screw were compared and averaged
over all ve loading directions (0180) using a Young's modulus
of 600 MPa for the bone cylinder. The threaded model was simulated with isotropic Coulomb friction (m 0.3) and the simple
model was simulated with tied contact conditions between the
screw and bone. The screw head displacement and average principal compressive strain of the tied simple model with 52,217

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i

J.A. Inzana et al. / Journal of Biomechanics ()

nodes varied by only 0.26% 70.05% and 4.24%73.1%, respectively,

compared to the model with 232,256 nodes, demonstrating sufcient mesh convergence. The screw head displacement and
average principal compressive strain of the threaded model with
201,919 nodes varied by only 0.35%70.13% and 3.5% 72.0%,
respectively, compared to the model with 383,545 nodes,
demonstrating sufcient mesh convergence.
For the proximal humerus study, mesh convergence analysis
was performed in three subjects (low BMD Subject 1, medium
BMD Subject 2, and high BMD Subject 5) over a range of
approximately 570,0002.1 million nodes by varying the maximum target edge length between 4.0, 3.0, 2.0, 1.0, and 0.75 mm.
The minimum target edge length for the quadratic tetrahedral
elements was held constant at 0.4 mm. The models were simulated with the four most proximal screws only (Without Calcar)
and a tied bonescrew interface. In the models with a target
maximum edge length of 3.0 mm, the humeral head rotation and
mean principal compressive strain around the humeral head
screws varied by 3.7% 73.0% and 8.5% 7 7.3% (averaged over the
3 subjects), respectively, compared to the nest density models.
Although the strain criteria ranged up to a 15% deviation from the
nest models, this was determined acceptable for convergence
considering the primary outcome metric is focused on stiffness
and any relative comparisons that are based on strain are not
sensitive to this deviation in accuracy. Therefore the study models
were meshed with a target minimum and maximum edge length
of 0.4 mm and 3.0 mm, respectively, which resulted in models
with an average of 720,0007 70,000 nodes (range: 620,000
840,000).
A.2. Optimization criteria for the pseudo-thread calibration
The difference in the displacement of the screw head control
node was calculated for each pseudo-threaded model relative to
the threaded model (dM; Eq. (1)). The maximum principal strains
in the bone elements were measured in ten concentric circle
regions of interest (ROI), ranging from the screw minor diameter
up to 2.5 mm away. The maximum was dened as the mean of the
top fth percentile of elements. The difference in the maximal
strain, relative to the threaded model, was calculated by summing
the difference over the ROIs (EP; Eq. (2)).





dM
 dM Thread


dM

Pseudo  thread
1


dM Thread

EP

10


X

EP

Pseudo  thread;ROI  EP Thread;ROI

EP Thread;ROI

ROI 1

where dM is the displacement magnitude and EP is the maximum

principal compressive or tensile strain within each ROI. These
differences were summed over the ve load cases and three bone
moduli to give summed difference metrics for the displacement
(MdM; Eq. (3)) and strain (MEP; Eq. (4)).

MdM

5
X

3

X

dM
YM;Load

Load 1 YM 1

MEP

5
X

3

X

EP3
EP1
YM;Load YM;Load

Load 1 YM 1

where YM,Load is the difference in the principal compressive strain

(EP1), principal tensile strain (EP3), or displacement magnitude
(dM) at each load and Young's modulus (YM). The overall difference function (Eq. (5)) was designed to achieve a balanced optimization of the strain and displacement differences.

Scoreijk

M EP
ijk


min M EP
IJK

MdM
ijk



min M dM
IJK

where Scoreijk is the difference score for parameter combination [i,

j, k]; and min(MIJK) is the minimum summed difference among
the set of evaluated parameter combinations.
A.3. Bone-implant simulations in the proximal humerus
High-resolution peripheral quantitative computed tomography
(HR-pQCT; XtremeCT, Scanco Medical; Bassersdorf, Switzerland)
scans were previously collected for 10 left-sided cadaveric proximal humeri (4 male, 6 female; age range 7292 years, 84 77
years) at an isotropic voxel size of 82 mm and a power of 60 kVp.
The scans were converted from Hounseld units (HU) to bone
mineral density (BMD) using a standard Scanco calibration phantom. The BMD of the humeral head was measured according to
Krappinger et al. (Krappinger et al., 2012) and ranged from 69
152 mg HA/cm3 with a mean and standard deviation of
102 729 mg HA/cm3. Each HR-pQCT scan was down-sampled to
an isotropic voxel size of 0.328 mm and segmented such that the
cancellous/marrow and cortical compartments were separated
using a specialized ll algorithm (Pahr and Zysset, 2009) implemented in Medtool v3.8 software (Dr. Pahr Ingenieurs e.U.; Pfaffsttten, Austria).
The PHILOS plate (DePuy Synthes Inc.; Oberdorf, Switzerland)
was virtually instrumented onto each bone according to the DePuy
Synthes technique guide, placing the plate 58 mm distal to the
superior aspect of the greater tubercle and 25 mm lateral to the
bicipital groove. Three clinically relevant combinations of humeral
head locking screws were simulated in each subject (Fig. 5a). One
group was instrumented with only the four most proximal screws
in the plate (Without Calcar). In addition to the four most proximal
screws, the fth and sixth screws were added in one group
(Without Calcar 5,6) and the calcar screws (eighth and ninth)
were added in the third group (With Calcar). The screws represented the DePuy Synthes 3.5 mm locking screw, modeled as
idealized cylinders with a 3.0 mm diameter to represent the inner
diameter of the screw. Each screw length was selected such that
the tip was approximately 6 mm below the subchondral bone
surface, according to the PHILOS surgical technique guide. Virtual
osteotomies were performed that simulated an unstable three
fragment fracture with medial comminution (AO 11-B3.2). A
wedge of bone, with an inclination of 7.5, was removed from the
surgical neck region. The greater tubercle was then separated from
the humeral head with a 1 mm virtual cut running through the
bicipital groove and medial to the rotator cuff insertion points.
The virtual bone-implant models were simulated according to
an in vitro loading protocol for testing proximal humerus xation
(Unger et al., 2012) (Fig. 5b). Specically, the distal 5 mm of the
humeral shaft were constrained to a control node through a
kinematic coupling. A 300 N compressive force was applied to the
control node, which was constrained against rotation and allowed
to translate only in the proximal-distal direction (Y axis). The
articular surface nodes were constrained to a second control node,
which was located at the center of mass of the articular surface.
This control node was connected to a stiff beam element (200 GPa,
Poisson's ratio0.3, circular cross-section radius 15 mm)
through a joint element that was open to rotation only in the
coronal plane (allowing varus bending). The opposite end of the
beam element simulated a ball bearing device, allowing rotation
about the Y axis and translation in the XZ plane.

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i

J.A. Inzana et al. / Journal of Biomechanics ()

Appendix A. Supplementary material

Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.jbiomech.2016.04.021.

References
Bai, L., Fu, Z., An, S., Zhang, P., Zhang, D., Jiang, B., 2014. Effect of calcar screw use in
surgical neck fractures of the proximal humerus with unstable medial support:
a biomechanical study. J. Orthop. Trauma 28, 452457.
Bayraktar, H.H., Morgan, E.F., Niebur, G.L., Morris, G.E., Wong, E.K., Keaveny, T.M.,
2004. Comparison of the elastic and yield properties of human femoral trabecular and cortical bone tissue. J. Biomech. 37, 2735.
Carter, D.R., Hayes, W.C., 1977. The compressive behavior of bone as a two-phase
porous structure. J. Bone Jt. Surg. Am. 59, 954962.
Dragomir-Daescu, D., Op Den Buijs, J., McEligot, S., Dai, Y., Entwistle, R.C., Salas, C.,
Melton 3rd, L.J., Bennet, K.E., Khosla, S., Amin, S., 2011. Robust QCT/FEA models
of proximal femur stiffness and fracture load during a sideways fall on the hip.
Ann. Biomed. Eng. 39, 742755.
Dubov, A., Kim, S.Y.R., Shah, S., Schemitsch, E.H., Zdero, R., Bougherara, H., 2011. The
biomechanics of plate repair of periprosthetic femur fractures near the tip of a
total hip implant: the effect of cablescrew position. Proc. Inst. Mech. Eng. 225,
857865.
Gardner, M.J., Weil, Y., Barker, J.U., Kelly, B.T., Helfet, D.L., Lorich, D.G., 2007. The
importance of medial support in locked plating of proximal humerus fractures.
J. Orthop. Trauma 21, 185191.
Huang, H.L., Hsu, J.T., Fuh, L.J., Tu, M.G., Ko, C.C., Shen, Y.W., 2008. Bone stress and
interfacial sliding analysis of implant designs on an immediately loaded maxillary implant: a non-linear nite element study. J. Dent. 36, 409417.
Karunratanakul, K., Kerckhofs, G., Lammens, J., Vanlauwe, J., Schrooten, J., Van
Oosterwyck, H., 2013. Validation of a nite element model of a unilateral
external xator in a rabbit tibia defect model. Med. Eng. Phys. 35, 10371043.
Karunratanakul, K., Schrooten, J., Van Oosterwyck, H., 2010. Finite element modelling of a unilateral xator for bone reconstruction: importance of contact
settings. Med. Eng. Phys. 32, 461467.
Kennedy, J., Feerick, E., McGarry, P., FitzPatrick, D., Mullett, H., 2013. Effect of calcium triphosphate cement on proximal humeral fracture osteosynthesis: a
nite element analysis. J. Orthop. Surg. 21, 167172.
Kralinger, F., Blauth, M., Goldhahn, J., Kach, K., Voigt, C., Platz, A., Hanson, B., 2014.
The inuence of local bone density on the outcome of one hundred and fty
proximal humeral fractures treated with a locking plate. J. Bone Jt. Surg. Am. 96,
10261032.
Krappinger, D., Roth, T., Gschwentner, M., Suckert, A., Blauth, M., Hengg, C., Kralinger, F., 2012. Preoperative assessment of the cancellous bone mineral density
of the proximal humerus using CT data. Skelet. Radiol. 41, 299304.
MacLeod, A.R., Pankaj, P., Simpson, A.H., 2012. Does screwbone interface modelling matter in nite element analyses? J. Biomech. 45, 17121716.
Maldonado, Z.M., Seebeck, J., Heller, M.O.W., Brandt, D., Hepp, P., Lill, H., Duda, G.N.,
2003. Straining of the intact and fractured proximal humerus under
physiological-like loading. J. Biomech. 36, 18651873.
Morgan, E.F., Bayraktar, H.H., Keaveny, T.M., 2003. Trabecular bone modulusdensity relationships depend on anatomic site. J. Biomech. 36, 897904.

Nedoma, J., Stehlik, J., 2011. Mathematical and Computational Methods and Algorithms in Biomechanics: Human Skeletal Systems. Wiley, Hoboken, New Jersey,
p. 112.
Pahr, D.H., Zysset, P.K., 2009. From high-resolution CT data to nite element
models: development of an integrated modular framework. Comput. Methods
Biomech. Biomed. Eng. 12, 4557.
Pessoa, R.S., Coelho, P.G., Muraru, L., Marcantonio Jr., E., Vaz, L.G., Vander Sloten, J.,
Jaecques, S.V., 2011. Inuence of implant design on the biomechanical environment of immediately placed implants: computed tomography-based nonlinear three-dimensional nite element analysis. Int. J. Oral. Maxillofac.
Implant. 26, 12791287.
Ponce, B.A., Thompson, K.J., Raghava, P., Eberhardt, A.W., Tate, J.P., Volgas, D.A.,
Stannard, J.P., 2013. The role of medial comminution and calcar restoration in
varus collapse of proximal humeral fractures treated with locking plates. J. Bone
Jt. Surg. Am. 95, 111117.
Roderer, G., Erhardt, J., Kuster, M., Vegt, P., Bahrs, C., Kinzl, L., Gebhard, F., 2011.
Second generation locked plating of proximal humerus fractures a prospective multicentre observational study. Int. Orthop. 35, 425432.
Steiner, J.A., Ferguson, S.J., van Lenthe, G.H., 2015. Computational analysis of primary implant stability in trabecular bone. J. Biomech. 48, 807815.
Sudkamp, N., Bayer, J., Hepp, P., Voigt, C., Oestern, H., Kaab, M., Luo, C., Plecko, M.,
Wendt, K., Kostler, W., Konrad, G., 2009. Open reduction and internal xation of
proximal humeral fractures with use of the locking proximal humerus plate.
Results of a prospective, multicenter, observational study. J. Bone Jt. Surg. Am.
91, 13201328.
Synek, A., Chevalier, Y., Baumbach, S.F., Pahr, D.H., 2015. The inuence of bone
density and anisotropy in nite element models of distal radius fracture
osteosynthesis: evaluations and comparison to experiments. J. Biomech. 48,
41164123.
Taylor, M., Prendergast, P.J., 2014. Four decades of nite element analysis of
orthopaedic devices: where are we now and what are the opportunities? J.
Biomech. 48, 767778.
Tingart, M.J., Lehtinen, J., Zurakowski, D., Warner, J.J., Apreleva, M., 2006. Proximal
humeral fractures: regional differences in bone mineral density of the humeral
head affect the xation strength of cancellous screws. J. Shoulder Elb. Surg. 15,
620624.
Unger, S., Erhart, S., Kralinger, F., Blauth, M., Schmoelz, W., 2012. The effect of in situ
augmentation on implant anchorage in proximal humeral head fractures. Injury
43, 17591763.
Wirth, A.J., Goldhahn, J., Flaig, C., Arbenz, P., Muller, R., van Lenthe, G.H., 2011.
Implant stability is affected by local bone microstructural quality. Bone 49,
473478.
Wirth, A.J., Muller, R., van Lenthe, G.H., 2010. Computational analyses of small
endosseous implants in osteoporotic bone. Eur. Cells Mater. 20, 5871.
Wirth, A.J., Muller, R., van Lenthe, G.H., 2012. The discrete nature of trabecular bone
microarchitecture affects implant stability. J. Biomech. 45, 10601067.
Zhang, L., Zheng, J., Wang, W., Lin, G., Huang, Y., Zheng, J., Edem Prince, G.A., Yang,
G., 2011. The clinical benet of medial support screws in locking plating of
proximal humerus fractures: a prospective randomized study. Int. Orthop. 35,
16551661.
Zhang, W., Zeng, L., Liu, Y., Pan, Y., Zhang, W., Zhang, C., Zeng, B., Chen, Y., 2014. The
mechanical benet of medial support screws in locking plating of proximal
humerus fractures. Plos. One 9, e103297.

Please cite this article as: Inzana, J.A., et al., Implicit modeling of screw threads for efcient nite element analysis of complex boneimplant systems. Journal of Biomechanics (2016), http://dx.doi.org/10.1016/j.jbiomech.2016.04.021i