Accuracy of implant placement after pre-operative planning using Blueprint Software in inlay and onlay reverse total shoulder arthroplasty systems: A cadaver study

Abstract

Background:

Accurate implant placement is desirable in reverse total shoulder arthroplasty (rTSA). Prior studies have focused on the accuracy of patient-specific glenoid implant placement, with little data on humeral preparation. This study quantified both glenoid and humerus implant placement, with specific focus on differences between inlay and onlay humeral implant systems.

Methods:

Twenty cadaver shoulders were preoperatively planned in Blueprint Software. After the rTSA was placed, computed tomography (CT) scans were obtained and 3-dimensional models of the bones and implants were created. Errors were assessed for factors like glenoid guide pin location, reaming depth, inclination, and version, and humeral implant height, inclination, and version, by comparing the models from planned and performed surgeries.

Results:

Glenoid guide pin placement was accurate, with error of mean±SD 1.6±0.7 mm. While glenoid reaming depth appeared accurate (0.9±1.4 mm), variability between specimens ranged from −2.5 mm of over-reaming to 4.8 mm of under-reaming. Consequent glenoid inclination and version showed small mean error in inlay and onlay systems (1.4±3.9° and −1.4±4.2°, respectively) but large ranges among specimens (15.8° and 20.0°, respectively). Likewise, humeral implant height relative to the bone had small mean error (0.5±3.4 and −0.8±1.1 mm, respectively). The inlay height variation was nearly 3x that of the onlay (11.8 and 3.6 mm, respectively). Error in effective humeral inclination was again small (0.9±3.3° and −1.8±3.7°, respectively), with large overall variation (11.4° and 12.4°, respectively). Humeral implant version was the most error-prone, where specimen-specific variation ranged nearly 40° between the plan and procedure. The glenoid or humeral preparation error did not differ between the inlay and onlay systems for any metric (p≥0.059).

Conclusion:

This study quantified the error in placing glenoid and humeral implants relative to a preoperative plan for both humeral inlay and onlay systems. With the exception of humeral implant version, the mean placement error was generally small. In contrast, specimen-specific implant placement had large errors, even when the average appeared acceptable. Device manufacturers should evaluate the ability of their preoperative planning softwares (like Blueprint) and instrumentation to control these variables in practice, especially humeral implant version, and specifically account for anatomic humeral torsion as it likely has a substantial impact on humeral implant placement. These procedure-to-procedure variations have biomechanical implications that could affect shoulder function and complications and negatively impact outcomes of patient-specific rTSA if left unchecked.

Reverse total shoulder arthroplasty (rTSA) was originally designed to restore shoulder function in the presence of rotator cuff tear arthropathy, which is now one among an expanding list of indications.^7,18,36,44 The rTSA is now the most utilized shoulder replacement procedure.^5,36,45 As a non-anatomic joint replacement, the relative positions of the scapula and humerus implants are essential to success.^{2,9,15,19,30,48} Variations of implant position have been shown to affect shoulder biomechanics via alterations in impingement-free range of motion and forces applied to the glenohumeral joint.^{1,16,17,21,28,32,43,62}

Due to the importance of accurate implant placement, patient-specific surgical instrumentation was developed for the glenoid.^31,33 Three-dimensional (3D) preoperative planning with patient specific guides decreased variability and prevented large errors in setting glenoid baseplate inclination as measured on two-dimensional (2D) radiographs.²³ Placement of the guide pin, and baseplate inclination and version, was also more accurate than non-specific guides in 3D assessments (e.g., mean placement errors <1–2 mm, <5° for inclination and version).^6,20,56,57 However, there is limited data showing that patient specific humeral implant guides are more accurate than conventional procedures.⁶⁰ This study assessed only the 2D stem offset in the humeral canal, and data on the 3D accuracy of humeral component is generally lacking. This may affect the ongoing debate over humeral inlay and onlay systems since factors like implant design, instrumentation, and surgical preparation can each influence accuracy.^24,38,55,62

Our goal was therefore to quantify the 3D accuracy in placement of the augmented glenoid and humeral inlay and onlay rTSA systems compared to preoperative plans using human cadaver shoulders. Specimen-specific glenoid component guides were the gold standard to validate methods against prior studies, whereas humeral preparation followed standard operative techniques without specimen-specific guides. We hypothesized that glenoid preparations relative to the plan would be comparable to prior studies: placement <2 mm mean error, inclination and version <5° mean error.^6,20,56,57 We also hypothesized that accuracy of humeral implant placement would be influenced by inlay and onlay implant design. Specifically, implant height, inclination, and version relative to the humeral shaft would differ between systems due to seating within or upon cancellous bone, respectively.

Methods

Specimen preparation

Imaging and rTSA procedures from a prior study comparing humeral inlay and onlay biomechanics were analyzed.⁶² Details on the specimen preparation can be found therein. Right cadaver shoulders (N=10 inlay, N=10 onlay) were preoperatively planned by two orthopedic surgeons (RZT, GW) (Blueprint; Stryker, Kalamazoo, MI, USA) and 3D models with specimen-specific position and orientation of the bones and implants were provided by the manufacturer (planned group). The software allowed 3D planning of glenoid baseplate placement with respect to 3D measures of inclination, version, and seating contact area. The software also allowed 3D planning of humeral component height, inclination, and version, but did not include the distal humerus. Therefore, version could not be quantified relative to the elbow/forearm. As a result, planned version was also guided by the 2D axial images, attempting to approximate the osteotomy plane with the anatomic margin plane (AMP). A single board certified, fellowship trained shoulder and elbow surgeon (RZT) that annually performs over 200 shoulder arthroplasties implanted each using specimen-specific glenoid guides fabricated from the plan. Standard operative technique for the humerus was an osteotomy at the anatomic neck as visualized intraoperatively, placing the implant version relative to the cut and verifying it relative to the forearm with a rod set at 20° retroversion. Postoperative computed tomography (CT) scans were obtained with metal artifact reduction (performed group).

The scapula, humerus, and implants were segmented from CT to generate 3D models for the performed group (Mimics; Materialise, Plymouth, MI, USA). In 3-Matic (Materialise), the performed scapula model was aligned to the planned model using N-point and Global registration. This process was repeated for humerus models. Rigid body associations carried segmented implants with their respective bones. The manufacturer-provided 3D implant model was then aligned to the performed implant model to remove segmentation artifact that could influence calculations. Comparisons were made between planned and performed implants relative to local coordinate systems of each bone (Figure 1).^29,50 A proximal humerus coordinate system referenced the lesser tuberosity since medical imaging and preoperative planning routinely omits the distal humerus and reliable humeral head centers are not present in pathologic or implanted bones.⁵⁰

Figure 1:

Bony coordinate systems. (A) The scapular plane was defined by the trigonum spinae, glenoid center, and inferior angle (dotted line). The transverse plane was perpendicular to the scapular plane, passing through the trigonum and glenoid center. The anteroposterior (AP) plane was perpendicular to both the scapular and transverse planes, passing through the glenoid center. (B) The shaft axis (dashed yellow) was created from a cylinder fit to the proximal humeral shaft. The humeral AP plane passed through the shaft axis and lesser tuberosity. The mediolateral (ML) plane was perpendicular to the humeral AP plane and passed through the shaft axis. The humeral transverse plane was perpendicular to both the humeral AP and ML planes and passed through the lesser tuberosity.

The accuracy of bone-to-bone registration was determined using Root-Mean-Square (RMS) error across surfaces calculated with the Analyze tool in 3-Matic. The scapulae error was 0.8±0.7 mm and humeri was 1.2±0.6 mm, which is below the level of measurements reported above and did not likely influence the differences between planned and performed metrics.

Scapular measurements

The glenosphere baseplate guide pin location on the glenoid was analyzed (Figure 2A, 2B). Pin placement was found by projecting the centroid of the central hole onto the scapular anteroposterior (AP) plane. The Euclidean distance between the planned and performed guide pin was calculated, excluding the effects of glenoid reaming and preparation.

Figure 2:

Glenoid component measurements. (A) A circle was fit to the center pin hole of the glenoid implant (red), and the center was projected onto the scapular AP plane. (B) For the planned implant (blue dot) and the performed implant (red square) the Euclidean distance between the two was calculated. (C) The centers of rotation (COR) were the centroid of each glenosphere. (D) The Euclidean distance between the planned COR (blue dot) and performed COR (red square) on the scapular AP plane was calculated. (E) Reaming distance was the difference between the planned and performed COR along the scapular axis. (F) The baseplate line (yellow) was the intersection of the glenoid baseplate plane with the scapular plane (red). Beta (β) was the angle between the best fit line of the supraspinatus fossa and the baseplate line. (G) Superior view of the scapula showing the glenoid baseplate version: the angle between the scapular AP plane (blue) and the glenoid baseplate line (black) in the transverse plane.

To consider glenoid preparation the center of rotation (COR) of the glenospheres was analyzed (Figure 2C). Like the guide pin, CORs were projected onto the scapular AP plane and the Euclidean distance between them was calculated (Figure 2D). To account for reaming depth, the distance between CORs along the scapular axis was found (defined trigonum spinae to glenoid center, a positive value for performed-planned was lateral offset from the plan) (Figure 2E).

Glenoid inclination was measured. Analysis of the suprascapular fossa highlighted the area of maximum curvature to which an inertial axis was fit.⁸ A plane was fit to the baseplate face and its intersection with the fossa line projected onto the scapular plane was the β angle (Figure 2F). Neutral inclination was defined as β=90° where a positive value for performed-planned was inferior inclination relative to the plan). Glenoid version was measured as the angle between the baseplate plane and the scapular AP plane as projected onto the scapular transverse plane (a positive value for performed-planned was anteversion offset from the plan) (Figure 2G).

Humeral measurements

The height of the humeral implant was measured. Circles were fit to the planned and performed implant trays, then the difference between centroids was calculated along the humeral shaft axis (a positive value for performed-planned was proximal offset from the plan) (Figure 3A).

Figure 3:

Humeral component measurements. (A) A circle was fit to the planned (blue) and performed (red) implant tray, and the height difference between the circle centroids (blue circle, red square) was calculated along the shaft axis. (B) Effective humeral implant inclination of the planned and performed implant was defined as the angle between the vector normal to the implant tray plane and the shaft axis. (C) Effective humeral implant version for both planned and performed implants was determined as the angle between the vector normal to the implant tray plane and the humeral anteroposterior (AP) plane, as viewed along the shaft axis.

Effective humeral implant inclination was determined as the angle between the vector normal to the implant tray plane and the shaft axis (a positive value for performed-planned was superior inclination (valgus) offset from the plan) (Figure 3B). Finally, effective humeral implant version was calculated. The angle between the vector normal to the implant tray plane and the humeral AP plane was determined as viewed along the humeral shaft axis (a positive value for performed-planned was anteversion offset from the plan) (Figure 3C).

Analyses revealed large variability between planned and performed humeral implant version. To investigate the origins, version of the humeral osteotomy plane was measured directly from the plan, and from a plane fit to the cortical rim in the performed models (Figure 4A). For comparison to the accuracy of creating osteotomy planes, the AMP was also determined (Figure 4B). Definition of the AMP was subjective and may affect its ability to clock humeral version, but as a self-consistent reference the effects were minimized by using the average of multiple observers defining the plane. Since the models were devoid of articular cartilage, two independent observers with expertise in shoulder anatomy (RZT, HBH) created AMPs in 3-Matic by selecting 3 landmarks around the AMP periphery.⁵¹ Unit vectors normal to these planes were averaged to calculate version of the AMP. The epicondylar axis was calculated to serve as a reference to the axis of the forearm in setting humeral implant version, and was readily calculated from the anatomy as imaged (Figure 4C). While this is not available in most clinical scans, it may be an important source of error in setting humeral implant version that has not been studied in detail. Anatomic humeral torsion, the degree of twist in the humeral shaft that develops during adolescence, was defined as the angle between the vector normal to the AMP and the epicondylar axis as viewed along the humeral shaft axis.¹¹ Comparisons of version between planned and performed humeral osteotomies and AMP, and their correlations to anatomic humeral torsion, were conducted to isolate sources of error that accumulated in setting humeral implant version. All resultant measurements for the glenoid and humeral implants can be found in the Supplement.

Figure 4:

Version of the humeral osteotomy and articular margin planes (AMPs). (A) Version of the planned osteotomy was the angle between the vector normal to the planned cut plane and the humeral AP plane, as viewed along the shaft axis. The performed osteotomy plane was fit to the segmented 3D model and the normal vector was used to calculate version. (B) Two observers manually selected landmarks to define the AMP (e.g., yellow, blue), and their normal vectors were averaged to define an unbiased native AMP (green). (C) The native anatomic humeral torsion (red) was then calculated as the angle between the vector normal to the AMP (green), and the epicondylar axis (black), as viewed along the shaft axis.

Statistical analysis

Descriptive statistics were calculated. When assessing differences between planned and performed metrics within specimens, two tailed one sample t-tests compared differences to the nominal value 0 (perfect agreement). When comparing metrics between inlay and onlay systems, two-tailed independent t-tests were used. Pearson correlation was used to determine if anatomic humeral torsion was related to the errors in setting humeral implant version.⁵⁴ Sample size and specimen demographics were fixed based on data from the prior study.⁶² Since glenoid preparation was the same in all specimens, N=20 were capable of achieving Power=0.8 with mean differences of 3±4 in each metric (e.g., mm, degrees) for the one sample t-test (Cohen’s d = 0.75). When comparing metrics between inlay and onlay, N=10 specimens per group were capable of achieving Power=0.8 with mean differences of 3±2 in each metric (e.g., mm, degrees) for the independent t-test (Cohen’s d = 1.5). A p≤0.05 was considered significant for all comparisons. Analyses were conducted in SPSS (version 21; IBM, Armonk, NY, USA). Power estimates were calculated using G*Power 3.1.¹²

Results

Each group contained 5 female and 5 male specimens. Only donor height differed by sex, with no differences in age, body mass index, or weight.⁶² All glenoids were prepared the same way, therefore their data were pooled. The 2D distance between planned and performed guide pins was 1.6 mm, ranging 2.5 mm (Table 1). The average 2D COR distance was 1.8 mm with a range of 3.7 mm. The reaming depth difference was 0.9 mm with a range of 7.3 mm, tending to be more lateral than planned. The performed implant had a β angle difference of 1.4° greater than the plan, ranging 15.8°. The difference in glenoid implant version was −1.4° more retroverted than the plan, ranging 20.0°. Neither inclination nor version reached statistical significance. For completeness, comparisons between inlay and onlay did not differ, supporting the ability to pool results.

Table 1:

Error between planned and performed implant placement (N=20 glenoid implants).

Variable	Mean±SD error	Min error	Max error	Range	p
Guide pin distance	1.6±0.7 mm	0.5 mm	3.0 mm	2.5 mm	<0.001
COR distance	1.8±0.9 mm	0.0 mm	3.7 mm	3.7 mm	<0.001
Reaming depth	0.9±1.4 mm	−2.5 mm	4.8 mm	7.3 mm	0.007
Inclination (β angle)	1.4±3.9°	−6.3°	9.5°	15.8°	0.130
Version	−1.4±4.2°	−14.0°	6.0°	20.0°	0.143

Error: performed – planned implant placement

p: vs. 0, which denoted perfect agreement between the plan and surgery

The humeral preparation compared inlay and onlay systems (Table 2). For inlay, the height difference was −0.5 mm (more distal than the plan), ranging 11.8 mm. For onlay, the height difference was −0.8 mm (more distal than the plan), ranging 3.6 mm. Height did not differ within or between implant systems. For differences in effective humeral head inclination, the inlay implant was 0.9° superiorly inclined (valgus) relative to the plan, ranging 11.4°. For onlay, the implant was −1.8° inferiorly inclined (varus) relative to the plan, ranging 12.4°. Effective inclination did not differ within or between implant systems. The effective implant inclination was also compared to the respective neck shaft angle. The inlay system was 2.2° more superiorly inclined (valgus) relative to 132.5°, ranging 10.5°. The onlay system was −0.6° inferiorly inclined (varus) relative to 135°, ranging 10.3°. Inlay reached statistical significance, while onlay did not. The systems did not differ.

Table 2:

Error between planned and performed implant placement (N=10 inlay, N=10 onlay implants).

Variable	Mean±SD error	Min error	Max error	Range	p
Implant height (inlay vs. onlay)					0.824
Inlay	−0.5±3.4 mm	−5.6 mm	6.2 mm	11.8 mm	0.629
Onlay	−0.8±1.1 mm	−2.4 mm	1.2 mm	3.6 mm	0.051
Effective inclination (inlay vs. onlay)					0.103
Inlay (vs. humeral shaft)	0.9±3.3°	−7.0°	4.4°	11.4°	0.425
Onlay (vs. humeral shaft)	−1.8±3.7°	−5.8°	6.6°	12.4°	0.154
Relative inclination (inlay vs. onlay)					0.059
Inlay (vs. 132.5°)	2.2±3.0°	−4.1°	6.4°	10.5°	0.045
Onlay (vs. 135°)	−0.6±3.2°	−3.7°	6.6°	10.3°	0.564
Implant version (inlay vs. onlay)					0.535
Inlay	−7.2±7.5°	−19.7°	0.6°	20.2°	0.014
Onlay	−10.1±12.6°	−29.4°	10.3°	39.8°	0.031
AMP version (inlay vs. onlay)					≥0.460
Inlay	9.1±9.7°	−10.4°	24.7°	35.1°	0.016
Planned vs. AMP	2.7±9.4°	−11.2°	14.6°	25.8°	0.394
Performed vs. AMP	−6.4±8.4°	−19.1°	3.0°	22.1°	0.040
Performed vs. Planned
Onlay
Planned vs. AMP	8.8±7.7°	−0.2°	24.4°	24.6°	0.006
Performed vs. AMP	−0.9±9.2°	−15.2°	11.7°	26.9°	0.733
Performed vs. Planned	−9.7±11.6°	−28.0°	8.5°	36.5°	0.027
Osteotomy version (inlay vs. onlay)					0.692
Inlay	−0.8±2.6°	−4.5°	4.4°	8.9°	0.359
Onlay	−0.3±3.0°	−7.9°	3.0°	10.9°	0.692

Error: performed – planned implant placement

p (inlay vs. onlay): implant vs. implant

p (implant): vs. 0, which denoted perfect agreement between the plan and surgery

For differences in humeral implant version, the inlay was −7.2° more retroverted than the plan, ranging 20.2°. For onlay, the implant was −10.1° more retroverted than the plan, ranging 39.8°. There were no differences between the implant systems.

To understand the origins of variability in humeral implant version, additional analyses evaluated the osteotomy planes and the AMP. For inlay, the planned osteotomy was 9.1° more anteverted than the AMP, ranging 35.1°. The performed osteotomy was 2.7° more anteverted than the AMP, ranging 25.8°. The performed osteotomy was −6.4° more retroverted than the plan, ranging 22.1°. For onlay, the planned osteotomy was 8.8° more anteverted than the AMP, ranging 24.6°. The performed osteotomy was −0.9° more retroverted than the AMP, ranging 26.9°. The performed osteotomy was −9.7° more retroverted than the plan, ranging 36.5°. There were no differences between systems for any of these comparisons.

Version of the planned osteotomy was identical to the planned implant for both systems since planning software placed the implant flush with the cut (i.e., 0±0°). For inlay, the performed implant version was −0.8° more retroverted than the performed osteotomy, ranging 8.9°. For onlay, the performed implant version was −0.3° more retroverted than the performed osteotomy, ranging 10.9°. The inlay and onlay systems did not differ.

Anatomic humeral torsion had very strong correlation to the difference between the planned osteotomy plane and AMP across specimens (R=−0.96, p<0.001). There was a moderate correlation to the difference between the performed osteotomy and AMP (R=−0.45, p=0.023). There was moderate to weak correlation of torsion to the difference between the planned and performed osteotomies (R=0.40, p=0.087) and difference between the performed osteotomy and final implant version (R=−0.20, p=0.411), but these did not reach statistical significance.

Discussion

Average error in rTSA glenoid component placement with preoperative planning and specimen (i.e., patient) specific guides was similar to previous studies, demonstrating that our methods were robust in quantifying the glenoid component. Application to the humerus was considered equally reliable given that the same protocols were utilized. We then analyzed humeral preparation for error in 3D metrics including height, inclination, and version in implant placement relative to the plan. The average error in effective humeral implant neck shaft angle and humeral implant version was found to be statistically significant for the inlay, but the onlay only reached statistical significance for version. Regardless of metric, there was wide specimen-specific variation in implant placement. The range of error was 5 to 20 times greater than the average error. Discrepancies in planning, osteotomy execution, and implant instrumentation and placement compounded to produce significant specimen-specific errors, which were obscured when averaged across the entire population. The variability in humeral version, specifically, was likely accentuated by anatomic humeral torsion that was not considered in preoperative planning or the procedure, since osteotomy version and therefore final implant version were constrained by proximal humeral anatomy (i.e., AMP). Ultimately, our data shows a stack up of errors exists in the glenoid, and for both inlay and onlay humeral systems. This procedure-to-procedure variability could have biomechanical implications on functional outcomes of patient-specific rTSA, and must not be overlooked.

Studies typically focus on averages. While mean±SD describes population trends, central tendency masks discrepancies that may have profound effects on patient-specific surgery. For example, the difference in glenoid baseplate guide pin placement was 1.6±0.7 mm, similar to prior studies.^6,56,57 Standard deviations were small, and our range was only 2.5 mm (<10% the glenoid width¹⁰). Thus, one can confidently locate the baseplate on the glenoid face when using specimen/patient-specific instrumentation within realistic limits of manual drills and reamers. Glenoid reaming depth tells a different story. Here the specimens were placed 0.9±1.4 mm relative to the plan, which appears accurate, but ranged from −2.5 mm of medialization (over-reaming) to 4.8 mm of lateralization (under-reaming). While it is no surprise that reaming affects glenosphere lateralization, the data highlights how it can easily exceed the first increment available in commercial rTSA systems (often ~4 mm via options in glenosphere diameter or pure lateral offset). Glenosphere lateralization, and similarly humeral lateralization, is arguably the most studied variable in rTSA, with many biomechanical implications and resultant effects on clinical and functional outcomes.^{14,16,22,32,35,37,38,41,42,46,47,52} More accurate control of relative reaming depth is needed to improve glenoid baseplate medial/lateralization when using a patient-specific plan and guide.

Glenoid inclination and version also demonstrate this effect, but more dramatically. Herein the difference between planned and performed procedures was 1.4±3.9° for β and 1.4±4.2° for version, both accurate and in agreement with prior studies.^6,56,57 Yet these differences in inclination ranged from 9.5° (superior) to −6.3° (inferior), and version from 6.0° (ante) to −14.0° (retro). The negative effects of superior glenoid inclination are known,^13,28,53 where up to 16° of patient-specific error could lead to complications. Similarly, more retroverted glenoids are less likely to sublux, with increased external rotation and total range of motion,⁴ where 20° of error could impact function. This variability arises from the angulation of cannulated reamers relative to the glenoid during preparation and rotational alignment of an augmented baseplate that influences the orientation of the baseplate itself. The latter led some commercial systems to include a secondary guide pin to control rotational alignment of the augmented baseplate. This was not present in the studied system.

Regarding implant height relative to the proximal humerus, mean differences did not differ between inlay and onlay (0.5±3.4 mm versus −0.8±1.1 mm, respectively). But again, specimen-specific variation tells a different story. Here the inlay ranged nearly 12 mm (−5.6 mm distal to 6.2 mm proximal) whereas onlay was 3x more consistent with only 3.6 mm of variation (−2.4 mm distal to 1.2 mm proximal). This discrepancy could arise through placement of the humeral cutting guide (which was not patient-specific, but a universal intramedullary guide), the reaming depth of the metaphysis, and/or the impaction of the humeral stem. The primary goal of inlay is large metaphyseal fill with less importance placed upon exact restoration of humeral height as this can be adjusted with polyethylene spacers. The primary goal of onlay is to seek metadiaphyseal stability with the smallest fill ratio that allows stability while exactly replicating the osteotomy. Consequently, inlay stems can sit inset or proud to achieve the largest metaphyseal fill independent of final height. In contrast, onlay stems are more likely to have accurate replication of height because the smallest stem providing stability allows fine adjustment while the tray still registers the osteotomy itself. Regardless, even 5 mm of discrete humeral lengthening increases forces on the glenohumeral joint up to 37%,⁶¹ where lengthening is a secondary consequence of glenosphere distalization, humeral insert thickness, and neck shaft angle.^14,41,42 Without a priori control of humeral distalization, patients may be at increased risk for functional deficits or complications like acromial fracture due to deltoid over-tensioning.^46,61

Humeral inclination showed sizable effects in both implant systems when measured at the tray relative to the humeral shaft. Again, we saw inlay and onlay systems appeared to be accurately placed (0.9±3.3° and −1.8±3.7°, respectively), which was similar to a prior study reporting that patient-specific guides provided 2.6±1.1° error relative to the humeral canal.⁶⁰ But relative to the plan, the inlay was up to −7.0° inferiorly and 4.4° superiorly inclined (valgus), and when compared to the intended 132.5° neck shaft angle, ranged from −4.1° to 6.4°. Onlay showed similar effects, and was up to −5.8° inferiorly or 6.6° superiorly inclined (valgus) relative to the plan, and ranged from −3.7° to 6.6° relative to the intended 135°. With variability of 10–12°, this rivals the scale of increments in neck shaft angle in commercial systems (e.g., 135–155°). Metaphyseal constraint does not appear to factor into this variability given that errors and ranges were similar between systems. Instead, these short-stem implants are less constrained by the humeral canal, therefore varus/valgus alterations in inclination up to 5° are common as there is less distal control.

Most variable was humeral implant version. The inlay was −7.2±7.5° more retroverted than planned, ranging from −19.7° (retro) to 0.6° (ante). Similarly, the onlay was −10.1±12.6° more retroverted than planned, ranging from −29.4° (retro) to 10.3° (ante). Both were statistically significant, and ranges of 20.2° and 39.8° are clinically significant.⁴ The variation in humeral implant version has a likely origin in clinical imaging (i.e., CT) that typically ends at the humerus midshaft. Similarly, the surgical procedure was performed with an osteotomy to the anatomic version of the AMP, using the forearm only as a secondary check to confirm it was within reason. Using the forearm as a guide is neither accurate nor precise in setting implant version, with errors on the order of 5–10° and differences by sex.²⁷ The discrepancy due to limited imaging field of view means that preoperative planning of humeral version is only relative to the proximal humerus and does not take into consideration humeral torsion. Even perfect agreement between the AMP, osteotomy, and implant in the proximal humerus would be subject to inter-specimen/patient variability when aligning to the forearm given the uncontrolled influence of anatomic humeral torsion. The strong to moderate correlations between torsion and many variations between planning and placement support this as a primary source of clinical error.

Further analysis of humeral version showed that in both systems the planned osteotomy was generally more anteverted than the AMP whereas the final implant version was more retroverted than the plan. Osteotomy at the AMP has 10–15° of error, and is influenced by which aspect is used to guide the cut, which alone is a significant source of error.⁵¹ The validity of defining the AMP on CT is subject to error as the transition from cartilage to bone can only accurately be visualized during surgery. The only way to create a gold standard AMP is to map this during surgical dissection, which was unable to be performed in this study. Defining the AMP on CT models motivated use of two observers to create an average AMP, and resulted in errors in agreement with the prior study. Further, the planning software excluded the distal humerus (even though it was available in the CT scan) so the implant version was set relative to the same model AMP with the same expected amount of error. Interestingly, once the humeral osteotomy was created the implant version relative to the performed osteotomy was less than 1±3° and ranged <10° in both systems, demonstrating the final placement was accurate to the osteotomy.

Stack-up of specimen-specific variation in planning implant version, performing the AMP osteotomy, and placing the implant, each with up to 10–15° of variance, created the differences in the final implant version versus the plan. In some cases, these errors cancelled each other out, and in others they compounded. Without more accurate and consistent control of input CT data, and referencing within planning software and to intra-operative humeral cutting guides, clinicians cannot currently be expected to place a humeral implant accurately and consistently to a plan.

The lack of precise control over planning humeral implant version (i.e., not considering humeral torsion, inability to accurately define the AMP on CT models) and performing humeral implant version (i.e., non-specific guides without rotational control relative to anatomy) could explain why clinical studies also differ. Some show patients with more retroversion have better outcome scores and range of motion compared to patients with less implant retroversion,^26,58 others have failed to show clinical differences associated with humeral version,⁵⁹ while still others suggest matching the humeral implant version to proximal humeral anatomy to improve clinical function and outcomes.^34,40,49 One group even concluded that setting humeral version relative to the scapula maximizes range of motion.³⁹ Biomechanically, anteversion maximizes teres minor moment arms but increases inferior impingement, while 40° retroversion increases the muscle length but decreases the moment arm and increases anterior impingement.³ Humeral retroversion also impacts resting abduction angles, with trends toward influencing passive external rotation and anterior dislocation forces.²¹ The impact of humeral version, and precise means to plan and set version during the procedure, are open questions that must be resolved.

This study has limitations. Registration of the 3D bone and implant models it is not perfect. However, the RMS error was small relative to detected effect sizes, so the data can be trusted as representative of planned and performed surgeries. Even with sub-millimeter CT slice thickness and metal artifact reduction protocols, errors in implant segmentation could still affect the accuracy of measurements. To minimize this, we used 3D implant models fitted to the segmentations, which helped reduce error across the entire implant volume. Ultimately, any misalignment—typically within a few degrees—was still much smaller than the specimen-specific variability observed in most metrics. Additionally, these surgeries were performed on cadavers and not real patients, so pathologic anatomy was not present and the repercussions of inaccurate implant placement could not be elucidated. However, the errors identified in this study may underestimate the true variability observed in clinical populations where pathologic anatomy further obscures the intended landmarks for planning and surgery. Our methods were validated by results consistent with prior work in the glenoid using similar sample sizes, and also addressed limitations of prior studies. Although the sample size was likely too small to statistically detect the subtle differences in many metrics, the main finding is that specimen- or patient-specific surgery involves a wide range of large errors that must be addressed to improve the accuracy of personalized procedures. Relying on measures of central tendency can obscure these individual variations in implant placement.

Our study analyzed 3D models for the major parameters relevant to performance of the procedure, and included the first detailed 3D analysis of the humeral implant placement to consider both inlay and onlay systems. There was no metal artifact as seen in previous work, in part due to metal artifact reduction scans, and because manufacturer-provided 3D implant models were compared. Despite utilization of patient specific guides, and thus improved overall accuracy of glenoid implant placement, there is generally a lack of clinical response to the use of patient-specific instrumentation.^6,23,25 Until accurate planning is reflected in accurate placement as highlighted herein, clinical studies may not be able to detect the true impact on patients. These findings apply only to the specific preoperative planning software and implant systems tested and may not be applicable to all rTSA systems. However, these technologies represent the current state of the art in rTSA surgery and likely reflect variability in current clinical practice. Therefore, it is not surprising that the literature has difficulty identifying differences in clinical outcomes between inlay and onlay systems in living patients, particularly when planning and surgical errors are confounding factors contributing to variability among individuals. The author who performed the preoperative planning and surgeries (RZT) is a paid consultant for the implant manufacturer, which provided the 3D models generated by the planning software for analysis. The potential for bias was minimized by having other members of the study team conduct the analyses.

Ultimately, the clinical choice of inlay or onlay humeral components is to strike a balance to maximize postoperative range of motion, avoid excessive overall lateralization, and avoid early arc impingement. While not yet definitively proven in the literature, this may also lower rates of postoperative conjoint tendinitis and acromion fracture. In light of the present data, it would not be surprising if much of the controversy surrounding the clinically detectable effects of a given implant parameter arises from averaging functional and patient-reported outcomes across a cohort with large patient-specific errors in implant placement.

Conclusions

This study quantified the error in placing glenoid and humeral implants relative to a preoperative plan for both humeral inlay and onlay systems. With the exception of humeral implant version, the mean placement error was generally small. In contrast, specimen-specific implant placement had large errors, even when the average appeared acceptable. Device manufacturers should evaluate the ability of their preoperative planning softwares (like Blueprint) and instrumentation to control these variables in practice, especially humeral implant version, and specifically account for anatomic humeral torsion as it likely has a substantial impact on humeral implant placement. These procedure-to-procedure variations have biomechanical implications that could affect shoulder function and complications and negatively impact outcomes of patient-specific rTSA if left unchecked.