Fractures Around Collared and Collarless Hip Stems Initiate in High Strain Regions: A Combined Cadaveric Experiment and Finite Element Analysis

ABSTRACT

Periprosthetic femoral fracture (PFF) is a common early complication after primary total hip arthroplasty. Collared stems reduce but do not eliminate the incidence of PFF, and exhibit variable biomechanical effectiveness, emphasizing our limited understanding of the local bone–implant interaction mechanics. Our goal was to elucidate the relationship between the local strains at the bone–implant interface and the experimental fracture patterns and loads for collared and collarless stems.

Six pairs of women cadaveric femurs implanted with a collarless or collared stem were loaded to failure under simulated stumbling to determine the location of PFF and the load to fracture, which we related to the collar‐calcar separation. Corresponding specimen‐specific FE models were developed to determine the strain at the bone‐implant interface and to predict the fracture onset load and location of fracture, which we related to the experimental location of PFF and the load‐to‐fracture.

Load to fracture was greater for collared stems and was inversely correlated with the collar‐calcar separation at the time of implantation (r = −0.80, p = 0.055). Fractures occurred in areas of high strain. The experimental fracture location coincided with the first yielding cortical element in five cases. The load at which the first cortical element yielded (i.e., onset of fracture was moderately correlated with the experimental fracture load (R ² = 0.43, RMSE = 1231 N).

Our results emphasize the importance of initial calcar contact to realize the benefits of the collar. Localized load transfer was a precursor of macroscopic fracture, which initiated in areas of high tensile strain.

1. Introduction

Periprosthetic femoral fracture (PFF) accounts for nearly 20% of revisions within 3 months of primary total hip arthroplasty (THA) in the United States [1]. Recent clinical data suggest that collared femoral stems are protective against PFF, showing a 5–7‐fold reduction in fracture incidence that has led to a dramatic increase in the use of collared stems [2]. However, since collared stems do not eliminate PFF [3], understanding when collars are protective remains important, particularly because up to 58% of patients who experience PFF are unable to regain their pre‐fracture mobility and have inferior outcomes, including 1‐year mortality rates as high as 22% [4]. Additionally, revision surgery requires expensive implants and longer operative time, both of which increase healthcare costs and strain [5, 6].

Biomechanical studies on paired cadaveric specimens showed that, compared to collarless stems, collars increase the torque to fracture between 18% and 160% [7] and the compressive load to fracture between 21% and 170% [8]. These studies hypothesized that collars help increase the stability of the femoral stem, thereby increasing the required load to fracture. However, the protective effect of collars is inconsistent in these studies [3, 7, 8, 9, 10], and some collared stems even resulted in lower torque to failure than their collarless counterparts [7]. While these prior studies [3, 7, 8, 9, 10] offer useful insights, they relied on simplified loading conditions that may not reflect physiologic loading characteristics of fractures, and they do not evaluate the local bone‐implant interaction. These limitations, along with the persisting incidence of PFF in clinical practice, emphasize our limited understanding of the biomechanical mechanisms driving PFF, which hampers the development of effective strategies to prevent this complication.

The goal of this study was to establish the biomechanical differences between collared and collarless implants. Specifically, we aimed to (1) quantify the differences in load to fracture between collared and collarless stems and (2) establish the difference in fracture location and pattern among these two groups. We hypothesized that collared stems would require higher loads to fracture, corresponding to lower strains at the bone‐implant interface. To achieve these goals, we tested cadaveric femurs under physiological loads to failure and developed corresponding specimen‐specific finite element (FE) models to study the bone‐implant interaction at the experimental fracture loads.

2. Methods

2.1. Study Design Overview

We conducted a combined experimental and computational Institutional Review Board‐approved study (Figure 1). Experimentally, paired cadaveric femurs were CT‐scanned, implanted with a collared or collarless stem, loaded to failure, and the fracture load and location were recorded. Computationally, specimen‐specific finite element (FE) models reproduced the experiment, with the implant position and specimen orientation derived from 3D scans of the experiment, and the load applied in the model matching the experimental fracture load. The strains at the bone‐implant interface obtained with the model were compared to the experimental fracture location.

Figure 1.

Overview of the study design. The experiment informs the computational model, and the outcomes from each branch are compared.

2.2. Cadaveric Testing

2.2.1. Specimen Preparation

We procured six pairs of women cadaveric femurs representative of THA patients (age: 70–78 years, BMI: 36 ± 11 kg/m²), with no history of cancer or trauma to the lower extremities. This sample size would suffice to detect a difference in fracture load equal to the average difference observed by Demey et al. [8] (mean ± standard deviation: 5663.6 ± 2757.6 N) with 80% power. We chose women for being at a higher risk of fracture due to the decreased bone density. We excluded specimens with any history of lower extremity trauma or a history of cancer. Prior to any manipulation, a radiopaque 50 mm by 20 mm plastic L‐bracket and three 2.4 mm CT‐opaque localizers (Model 211, Beekley Medical, Bristol, CT) were attached to the femoral shaft (Figure 2). The L‐brackets were used to design specimen‐specific, 3D‐printed guides for accurate specimen positioning during testing, and the CT‐opaque beads allowed for specimen registration across scans. Specimens were then CT‐scanned (Discovery CT750 HD, GE Healthcare, Chicago, IL. Acquisition: 200 mA, 120 kV with bone reconstruction kernel, slice spacing of 0.625 mm, and field of view of 462 mm on a 512 × 512 matrix) to obtain the three‐dimensional (3D) computational representation of their native anatomy. A density calibration phantom (Mindways Software, Austin, TX) was included in each CT scan to enable conversion from Hounsfield Units (HU) to dipotassium phosphate (K2HPO4) equivalent bone mineral density (BMD).

Figure 2.

Specimens were potted by engaging the L‐bracket on the femoral shaft into the specimen‐specific 3D‐printed guide. This enabled the vertical MTS loading axis to apply the stumbling load orientation relative to the implanted femur.

One specimen of each pair was randomly assigned a collared Avenir Complete Standard stem (Zimmer Biomet, Warsaw, IN), and the contralateral side received the collarless variant of the same design. In total, collared stems were used for three left and three right specimens. Implantation was performed by a fellowship‐trained orthopedic surgeon (A.H.) using fluoroscopy to confirm his choice of implant size. All matched pairs were within one stem size of one another (Table 1). The implanted specimens were 3D‐scanned with a resolution of 0.2 mm (HandySCAN, Creaform, Levis, Quebec, Canada) to determine the implant position in the femur and enable virtual implantation in the computational models (see detailed description in Section 2.3‐1).

Table 1.

Avenir complete standard stem size for each paired specimen.

	Spec 1	Spec 2	Spec 3	Spec 4	Spec 5	Spec 6
Collarless	2	5	6.5	3	4	5
Collared	3	6	6	4	4	5

2.2.2. Loading the Specimens

We considered the peak loads during a stumbling incident (file jb4541a from OrthoLoad [11]) that were measured in vivo with an instrumented implant. This scenario was consistent with PFF events reported by patients at our institution. To apply this load, we created the same reference frame as reported in OrthoLoad [11], where the origin is placed at the center of the replaced femoral head, the z axis corresponded to the line connecting the intersection point of the femoral neck and shaft with the deepest point of the fossa intercondylaris, and the x axis was parallel to the posterior condylar axis and pointed to the right. In this reference frame, we determined the direction of the loading vector from the experimentally measured peak force from OrthoLoad [11] [−0.3605, −0.1516, −0.9204]. Then, we oriented the implanted femur such that this loading vector corresponded with the loading axis of a uniaxial load frame (Model 825, MTS, Eden Prairie, MN). Finally, we secured specimen‐specific, 3D‐printed guides to the L‐brackets attached to the femoral shaft to achieve the target femur orientation when potting the specimens in epoxy‐based filler (Bondo, 3 M, Maplewood, MN) (Figure 2).

Specimens were secured in an aluminum fixture and fixed to a uniaxial servo‐hydraulic testing machine with step clamps and C‐clamps (Figure 1). The displacement of specimens under loading was measured with a digital image correlation (DIC) system (ARAMIS Adjustable, Zeiss, Oberkochen, Germany). The DIC cameras were placed roughly 45 cm from the specimen during testing to directly visualize the expected areas of fracture at the medial and posterior regions of the proximal femur. A mirror was mounted behind the specimen to visualize the anterior side. To replicate the experimental setup in the computational models, we 3D‐scanned the specimens within the testing setup prior to loading.

An initial compression was applied at 0.05 mm/s up to 2 mm to ensure seating of the femoral head in the polyethylene liner, followed by a 5‐second hold. All specimens were then loaded to failure under displacement control at 2 mm/s, except for three specimens (Specimen 1/Collarless and Specimen 5/Collarless and Collared) that were tested in load control at 1000 N/s, which resulted in a displacement rate of 1.93 mm/s. A post‐fracture 3D scan was taken of each specimen to document the fracture location.

2.2.3. Load‐Displacement Post‐Processing

The DIC‐acquired data were recorded at 10 Hz to ensure capturing the fracture event. The MTS‐acquired loading data were acquired at 100 Hz and down‐sampled from 100–10 Hz (Matlab, Mathworks, Natick, MA) to match the frequency of the displacement data. Then, DIC‐acquired loading and displacement data were filtered with a third‐order, low‐pass Butterworth filter with a normalized cutoff frequency of 0.1, which was empirically determined to avoid high‐frequency noise while maintaining the curve information. The loading and displacement data were synchronized using the 5‐second hold. Then, the instance of fracture was visually determined by analyzing the DIC videos, and the corresponding load was considered the fracture load.

2.3. Computational Modeling

2.3.1. Geometrical Modeling

We created FE models reproducing the specimen‐specific implantation and testing. First, the femur geometry was segmented from the pre‐implantation CT scans (Mimics, Materialise, Leuven, Belgium) to obtain the 3D representation of the periosteal surface. We also segmented the 3D geometry of the CT‐opaque localizer beads. The computer aided design (CAD) models of the stems provided by the manufacturer were defeatured to have a smooth surface and reduce computational cost. This modification had no noticeable effect on the model outcomes. We reproduced the implantation by performing a best‐fit alignment of the neck and trunnion between the implant CAD and the pre‐testing 3D scan and by aligning the centers of the CT‐opaque localizer beads from the pre‐implantation CT‐scan to the pre‐testing 3D scan. After alignment, we performed a Boolean subtraction of the implant from the bone to create a line‐to‐line fit of the implant within the bone, and created the neck cut from the experimental neck cut visible in the 3D scan.

2.3.2. Meshing

Both the femur and implant were meshed with linear tetrahedral elements (Abaqus 2018, Dassault Systemes, Velizy‐Vilacoublay, France). Based on a mesh convergence analysis (Supporting Information S1: Figure S1, Appendix) we chose an element size of 1 mm for the implant, while for the femur we increased the element size from 0.85 mm at the interface with the implant to 1.5 mm at 175 mm distal to the head center, and 2 mm at the distal end of the bone (200 mm distal to the head center).

2.3.3. Materials

The implant was assigned linear elastic, isotropic, and homogeneous titanium alloy properties (E = 110 GPa, ν = 0.33) [12]. The bone was modeled as linear elastic, isotropic material with ν = 0.3 and non‐homogenous elastic modulus, obtained by (1) interpolating the CT HUs to the FE mesh on an element‐by‐element basis, (2) converting the HU to apparent density using the calibration phantom and empirical relationships [13], and (3) converting the density to elastic modulus using empirical relationships across distinct density ranges [14, 15]. To limit the number of material properties created in the input file, elements were grouped into 50 MPa bins.

2.3.4. Interactions

Surface‐to‐surface contact was defined between the implant (contact) and bone (target), with a friction coefficient of 0.64 [16]. The implant and bone were modeled with a line‐to‐line fit (i.e., no press‐fit or gaps).

2.3.5. Loading and Boundary Conditions

The specimen‐specific experimental fracture load was applied to a point at the head center, which was coupled to the trunnion. Load was applied along the vertical (i.e., loading) axis, constraining the motion of this point in all other directions to mimic the experimental condition. The distal end of the bone was rigidly fixed in all degrees of freedom. All simulations were static implicit analyses accounting for non‐linear geometrical effects and were run in Abaqus/Standard.

2.4. Outcome Metrics

The metrics of the in vitro experiments were the load at the first visible fracture in the DIC video, the anatomical quadrant relative to the trunnion axis where the fracture occurred (Figure 3a), and the collar‐calcar gap distance, measured as the minimum distance between the cut and the most anterior, middle, and most posterior points of the collar (Figure 3b).

Figure 3.

(a) Quadrant classification for fracture location. Medial quadrant is divided into anteromedial and posteromedial halves. In this representative example, fracture occurs in the posteromedial quadrant. (b) Red dots indicate the location where gap distance was measured between collar and calcar. The minimum distance was recorded.

To determine the relationship between fracture load and the local mechanics of a collar, the percentage change in load to fracture from collarless to collared specimen pairs was compared to the collar‐calcar gap distance.

The primary metric of the computational analysis was the distribution of absolute maximum principal strain at the bone–implant interface, which we related to the yield strain of femoral trabecular bone (7000 µε in compression, 6100 µε in tension) to determine the area of the interface with the highest risk of bone mechanical failure [14].

We also predicted location and load magnitude at fracture initiation by identifying the first cortical element (density > =1 g/cm³) near the metaphyseal wedge of the implant to exceed yield strain [17, 18, 19, 20, 21, 22]. We focused on tensile yield as fracture involves surface separation driven by tensile forces. For some models, this required increasing the load magnitude beyond the experimentally determined values until the first element exceeded yield or a pre‐established limit of 10 kN was reached. The location of this element (i.e., fracture initiation) was visually compared to the quadrant where the experimental fracture was observed (Figure 3a). We also quantified the agreement between experimental and computationally predicted fracture loads through Pearson correlation analysis and by calculating the root mean square error (RMSE).

2.5. Statistical Analysis

The difference in load to fracture between collared and collarless stems was assessed with pairwise t‐test at a significance of 0.05.

3. Results

3.1. Experimental Load to Fracture

All specimens exhibited Vancouver Type A fractures initiating at the calcar and propagating distally. One specimen (Specimen 1/Collared) developed a Type B1 clamshell‐like fracture in the diaphysis, secondary to a Type A calcar crack (Figure 4). Specimen 4/Collarless and Specimen 6/Collared showed two fractures occurring in the same frame of the video.

Figure 4.

Experimental fracture patterns for paired specimens with collarless and collared stems. Yellow arrow indicates fracture location(s). For specimens with two arrows, both fractures occurred in the same frame of the video.

The load to fracture for collared (range: 2167–5964 N) and collarless stems (range: 2189–5800 N) was similar (p = 0.56, Figure 5). While most specimens experienced an increase in load to fracture with a collar, Specimen 4 showed no change, and Specimen 2 had a lower load with a collar as compared to collarless.

Figure 5.

Comparison of fracture loads between collarless and collared stems. Fracture load increases for most paired specimens when using a collared stem, but this effect is not universal.

3.2. Collar–Calcar Contact

All specimens showed a collar–calcar gap (0.1–1.9 mm). While specimens with a larger gap tended to have a greater increase in load to fracture relative to their non‐collared counterparts, no relationship was found between the gap and the load to fracture (r = −0.80, p= 0.06) (Figure 6).

Figure 6.

(Left) Visualization of the collar‐calcar alignment for each specimen. (Right) Percentage change in fracture load from collarless to collared stem vs. collar–calcar gap distance.

3.3. Comparison of Computational and Experimental Outcome Metrics

For all specimens, the largest strains were concentrated along the medial aspect of the bone–implant interface near the neck cut (Figure 7). The strains in this location were predominantly compressive and exceeded the yield strain of cancellous bone (black areas in Figure 7). These high‐strain regions generally matched the experimental fracture locations.

Figure 7.

Comparison of experimental fracture patterns to computational data at the fracture load. For each row (collarless and collared), the top side shows a 3D scan of the experimental fracture (indicated by the black arrow), and the bottom side shows the computational strains at the bone implant interface, with the red arrow indicating the location of the first cortical element (density > = 1 g/cm³) to exceed tensile yield strain. Labels beneath each image indicate the fracture location (Abbreviation: AM = anteromedial, DM = direct medial, PM = posteromedial). For specimens with two fracture locations, the more prominent fracture is shown. Specimen 4/Collared did not have a failing cortical element at the fracture load. Some specimens are mirrored for visualization purposes.

The experimental fractures occurred in the anteromedial quadrant for five specimens, in the posteromedial quadrant for five specimens, directly medial for three specimens, and in the posterior quadrant for one specimen (Figure 7). Computationally, the first cortical element to exceed tensile yield strain was located in the anteromedial quadrant for five specimens and in the posteromedial quadrant for six specimens (Figure 7). Specimen 4/Collared did not have a yielding cortical element before reaching 10 kN. The location of the first cortical element to exceed tensile yield strain directly matched the experimental location of fracture in five cases and was adjacent to the experimental fracture quadrant in another five specimens.

Excluding the only specimen that did not show a yielding element below the pre‐established limit of 10 kN (Specimen 4/Collared), the FE‐predicted load showed a moderate correlation with the experimentally determined load (R ²= 0.43, RMSE = 1231 N, Figure 8).

Figure 8.

FE‐predicted vs. experimental fracture loads. Data points with thick outlines indicate collared specimens. The dashed line indicates perfect 1:1 agreement.

4. Discussion

Despite advances in surgical technique and implant design, PFF remains a common and poorly understood complication after THA. This study aimed to investigate the load transfer between implant and bone to elucidate the local mechanics of periprosthetic femoral fracture, and to determine the influence of a collar. To this end, we considered loads representative of a stumbling event and related the location of the fracture determined experimentally with the bone strains determined through specimen‐specific computational modeling.

We observed a variable effect of the collar on the load to fracture under a stumbling load scenario. While most specimens showed increased load to fracture with collared stems, one was unaffected, and another showed a decrease in the load to fracture. Prior biomechanical studies have also found variable effects of a collar under compressive and torsional loading [7, 8]. Further, our results agree with clinical reports that indicate that fracture can still occur with collars [3]. We observed that the increase in the load to fracture was associated with the contact between collar and calcar. For example, Specimen 2 had the largest collar–calcar gap and experienced a 38% reduction in fracture load compared to its collarless counterpart. The DIC video of this specimen revealed that the fracture occurred simultaneously with collar–calcar contact, suggesting limited load transfer through the collar prior to fracture. This finding emphasizes the importance of achieving initial contact to realize the benefits of the collar. Lamb et al. similarly reported a reduced risk of PFF, under torsional loads, when such contact was achieved, noting that the chance of achieving contact, and thereby increasing fracture load, reduced nearly four‐fold with each millimeter of separation at initial implantation [23].

We observed that experimental fractures propagated distally from the calcar in patterns commonly observed by our surgeon authors (A.H., P.S., E.G.) and in agreement with prior in vitro tests [9, 24]. Fractures occurred in concentrated areas of high strain, suggesting that localized load transfer is a precursor to macroscopic fracture. While interfacial strain distributions provided a general sense of fracture location, the first cortical element to exceed tensile yield strain identified a precise point of fracture initiation. This element was consistently located near the metaphyseal wedge of the implant, at one of its small‐radius corners, further demonstrating that locations of localized load transfer are at increased risk of failure. Our model results support continued investigation of tensile yield strain as a promising indicator of fracture initiation. Prior biomechanical studies have used a similar strategy for predicting fracture load and location with good agreement to experimental data, but often restrict their analyses to whole‐bone behavior without implants [17, 18, 19, 20]; only Miles et al. and Hennicke et al. modeled fracture propagation around femoral stems [21, 22]. While our model focuses only on fracture initiation, it presents important advantages over these prior studies, including computational cost‐efficiency, use of non‐homogenous bone properties, and a larger sample size.

This study has limitations. We used a static load case at one time point of a stumbling load cycle, and therefore, did not capture accumulated strains across the activity. However, we used the worst‐case time point in that load cycle. Also, we considered perfect bone‐implant contact, ignoring press‐fit and over‐broaching. This could be resolved by modeling the bone geometry based on a post‐broach, pre‐implantation CT scan. Additionally, we considered simplified bone material properties (linearly isotropic) and can incorporate transversely isotropic and viscous or plastic behaviors in future studies [18, 19, 20, 22, 25, 26]. Lastly, during testing, we only produced one Vancouver Type B‐1 fracture, a common pattern associated with PFF. This specimen was the only one where testing continued beyond fracture initiation, similar to the in‐vivo situation where a patient cannot abruptly halt their motion, whereas in all other specimens, testing was terminated immediately upon fracture detection. Our computational models focused only on fracture initiation and did not account for damage accumulation and crack propagation. In this way, our approach cannot predict the actual fracture pattern, but only the initiation of the fracture. In this way, while our simplified criterion for determining fracture initiation, based on previous studies [17, 18, 19, 20, 21, 22], correctly captured the inter‐specimen trends in load to fracture, it underpredicted the experimental fracture load by more than 10% on two specimens and overpredicted the fracture load by more than 10% on five specimens.

In conclusion, the variability in load‐to‐fracture between collarless and collared stems highlights the critical role of implant positioning and collar–calcar contact in the failure mechanics of periprosthetic fracture. Our computational models revealed that fracture initiation is likely driven by localized load transfer, with interfacial strains and tensile forces emerging as promising indicators of PFF risk. In future research, we aim to leverage these models to investigate the biomechanical implications that various patient, surgical, and implant design variables have on the risk of periprosthetic fracture in an in silico clinical trial [27], where we will systematically evaluate the effectiveness of fracture protection for patient‐specific cases who suffered PFF.

Author Contributions

Ryan T. Helbock and Clarisse Zigan contributed to project design, data collection and analysis, and writing of the manuscript. Andrew Hughes, Joseph D. Lipman, Timothy M. Wright, Peter K. Sculco, Elizabeth B. Gausden, Sony Manandhar, Ghislain Maquer, Jeff Bischoff, and Fernando J. Quevedo Gonzalez contributed to project design, interpretation of data, and revision of the manuscript. All authors read and approved the final submitted manuscript.

Supporting information

Supporting File

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.