Preoperative Planning Software Does Not Accurately Predict Range of Motion in Reverse Total Shoulder Arthroplasty

Abstract

Background:

The purpose of this study was to determine whether preoperative planning software (PPS) accurately predicts clinical range of motion (ROM) in patients with reverse total shoulder arthroplasty 1 year postoperatively with preoperative and postoperative computed tomography (CT) scans.

Methods:

This was a retrospective study of 16 reverse total shoulder arthroplasty patients with preoperative and postoperative (CT) scans obtained at least 1 year postoperatively. Clinical ROM was measured in abduction, external rotation at resting abduction, extension, and flexion at a minimum of 1 year postoperatively. All clinical measurements were obtained before generation of PPS ROM values. Using postoperative CT scans, the achieved implant component positions were quantified and then replicated in PPS on the preoperative CT scans. The preoperative predicted ROM was then recorded, both with and without osteophyte removal. Bland-Altman plots were generated within each motion comparing the differences between clinically measured motion and software-predicted motion.

Results:

The variation in clinically measured ROM in abduction, external rotation at resting abduction, extension, and flexion were 118± 27 (65° to 180°), 33 ± 16 (10° to 75°), 56 ± 8 (50° to 65°), and 137 ± 25 (80° to 160°), respectively. Clinically measured motion differed greatly from PPS-predicted ROM, with mean differences of 33 ± 29 (−32 to 93) for abduction, 44± 25 (−38 to 57) for external rotation, 44± 25 (−35 to 65) for extension, and 54 ± 50 (−51 to 147) for flexion with no significant correlations between clinically measured and PPS-predicted ROM (P > 0.05). With humeral or humeral and glenoid osteophyte resection, correlations for only flexion became significant (P = 0.002 for both).

Conclusion:

The passive glenohumeral impingement-free ROM generated from PPS incompletely predicts clinically measured active humerothoracic ROM, possibly because of the unmeasured factors of soft-tissue tension, muscular strength, humeral torsion, resting scapular posture, and, most importantly, scapulothoracic motion.

Level of Evidence:

IV.

Preoperative planning software (PPS) has become common in templating for both anatomic and reverse total shoulder arthroplasty (rTSA).¹ Using preoperative computed tomography (CT) scans and osseous segmentation to automatically create 3D models, PPS allows the surgeon to estimate the passive, impingement-free range of motion (ROM) based on the selection and placement of various implant components. This tool has the potential to improve outcomes after rTSA by allowing the surgeon to predict areas of impingement and alter component position and size to avoid impingement to optimize ROM.² The PPS generates a predicted ROM by measuring the impingement-free arc in multiple planes and directions.³ Specifically, it generates motions between two 3D bone-implant shapes relative to one another within specified planes, with the limit of the arc defined when shapes overlap or a geometric conflict between 3D structures occurs. Many surgeons use this feature preoperatively to alter their desired implant size and position to maximize impingement-free arc of motion. However, it remains unknown whether the software-generated preoperative ROM predicts the in vivo clinically achieved postoperative ROM.

Thus, the purpose of this study was to determine whether PPS predicted clinically measured ROM. The hypothesis was that the ROM values generated by PPS would be predictive of clinically measured ROM after rTSA.

Methods

Patient Selection

This was a retrospective study. The surgical logs of the senior surgeon (R.Z.T.) were searched for all patients who underwent an rTSA (CPT 23472) with a postoperative CT of the shoulder (CPT 73200). The senior author routinely obtained postoperative CT scans of all patients who underwent concomitant glenoid bone grafting to evaluate graft and baseplate positions. The decision to perform an rTSA with a bone graft was based on the ability to correct baseplate inclination to at least neutral tilt on a standing true anteroposterior radiograph of the shoulder and to within 10° of retroversion on an axillary radiograph, without reaming beyond 5 to 10 mm of glenoid bone stock to gain correction. The goal of reaming was to correct to 100% baseplate seating. If these goals could not be achieved with reaming alone, then rTSA with bone grafting was selected. Bone grafting was not performed for pure lateralization of the baseplate. This group was then restricted to patients who had a documented CT scan before and after the operation and were more than 1-year postoperative. Patients without a preoperative or postoperative CT scan of sufficient quality were excluded from this study, along with any patient who had undergone an rTSA revision.

Measurement Methodology

For all the patients who met the inclusion criteria for this study, an orthopaedic surgeon (R.Z.T.) measured active shoulder ROM in clinic with a goniometer in abduction, external rotation, extension, and flexion and was blinded to the PPS ROM measurements. The PPS autogenerates impingement-free ROM, measuring external rotation in adduction and flexion at zero degrees of rotation, with neutral being defined relative to a reference position of the bones as defined from their 3D morphology.

The preoperative and postoperative CT images were downloaded in Digital Imaging and Communications in Medicine (DICOM) format and imported into Mimics (Materialise). The CT images were segmented to generate 3D models for the preoperative bone, postoperative bone, and postoperative implant components (Figure 1, A and B). The positional relationship between postoperative implant components and postoperative bone was maintained for subsequent analysis for both the scapula and the humerus.

Figure 1.

Preoperative (A) and postoperative (B) computed tomography (CT) images uploaded into Mimics and 3D models generated for each humerus, scapula, and postoperative components.

Within 3-Matic (Materialise), the preoperative bone model remained stationary, thus maintaining its position relative to the original CT coordinate system. The postoperative model (bone + implant) was then registered to the preoperative model through point alignment of anatomic landmarks. An iterative closest-point algorithm then provided a best fit of the postoperative bone model onto the position of the preoperative bone model. Postoperative implants transformed coordinate systems with the postoperative bone (Figure 2, A). The postoperative bone model was then removed, leaving just the glenosphere and humeral component in their appropriate position relative to the preoperative bone model (Figure 2, B). The implant components were then exported in stereolithography (STL) format back to Mimics with the preoperative bone.

Figure 2.

Postoperative 3D models aligned to the position of the preoperative models (A). The postoperative bone would then be removed, leaving the postoperative components in position relative to the preoperative bone (B).

A mask of the implant was created such that the position of the component models then appeared in a new DICOM file as a gray outline, indicating the position of the given implant component within the preoperative bone (Figure 3). The DICOM file was loaded into the PPS, where two board-certified and fellowship-trained shoulder surgeons (R.Z.T. and P.N.C.) created a ‘postoperative plan’ using the gray outline to guide actual component placement. Three iterations of preoperative plans were made, which included no osteophyte resections, resections of only the humeral side, and osteophyte resections of the humeral and glenoid sides. Resections were done by board-certified and fellowship-trained shoulder surgeons (R.Z.T. and P.N.C.) to simulate the conditions that the PPS would be used in practice. Osteophyte removal was performed with both surgeons present, and the extent of osteophyte removal was agreed upon by the two surgeons for each scan. When the plan was finalized, the PPS provided estimated glenohumeral ROM values for abduction, external rotation, extension, and flexion.

Figure 3.

Radiographs showing postoperative components now in position relative to the preoperative bone, outlined on preoperative DICOM files in black, thus providing an outline for component placement in preoperative planning software. Additional image in yellow for enhanced visibility.

Statistical Methods

Descriptive statistics were calculated. To test the hypothesis that PPS-predicted ROM was associated with a clinically measured ROM, Spearman rho correlations were calculated between the clinically measured ROM and the software-predicted ROM, both before and after osteophyte removal. Bland-Altman plots were also generated. All analyses were conducted in Excel 16 (Microsoft, Redmond, WA) and SPSS 28 (IBM, Armonk, NY).

Results

Inclusion criteria were met by 22 shoulders in 21 patients. Clinical ROM was obtained on 16 shoulders in 15 patients. Seven female and nine male shoulders as well as six left shoulders and 10 right shoulders were evaluated (Table 1). The mean ± standard deviation age of patients was 76.7 ± 9.4 years. The included patients were 36 ± 21 (12 to 85) months postoperative. Postoperatively, no patient in this cohort had complications (infection, fracture, and instability event) and no patients required a revision surgery.

Table 1.

Demographic Information of Volunteers

Patient	Age	Sex	Side	Time Postoperatively (mo)
1	84	F	L	22
2	77	M	L	31
3	76	M	R	60
3	76	M	L	53
4	85	F	R	54
5	67	F	L	16
6	70	M	R	39
7	55	F	R	16
8	86	M	R	12
9	84	F	R	54
10	78	M	R	16
11	73	M	R	34
12	74	F	R	34
13	96	F	L	31
14	68	M	R	12
15	78	M	L	85

Descriptive statistics showed that the iterations of PPS-generated ROM mean values were lower than those of clinically measured ROM values in all motions and PPS minimums of zero degrees were measured in every motion with the exception of abduction (Table 2). Clinically measured abduction ROM values had a larger standard deviation, maximum value and minimum value compared PPS values. External rotation at 0° abduction in clinic produced a mean ROM more than double that of PPS and a standard deviation that was greater than PPS iterations without osteophyte resection and humeral osteophyte resection, but less than the standard deviation of PPS values when glenoid and humeral osteophytes were resected. Extension clinic measurements showed a decreased standard deviation and maximum, but a decreased standard deviation when compared with PPS iterations. Flexion showed a similar trend with increased mean and decreased standard deviation in clinical measurements compared with PPS values. There was a much greater tendency for PPS to generate a ROM value of zero degrees as opposed to a clinic-measured value. No clinic measurements of zero degrees were obtained; however, external rotation, extension, and flexion all had a minimum of zero degrees recorded in all iterations, with the exception of flexion with humeral and glenoid osteophyte resection.

Table 2.

Descriptive Data of Range-of-Motion Values Obtained in Clinic and the Three Iterations of Preoperative Planning Software Values Reported as Mean ± Standard Deviation (Range).

Motion	Abduction	External Rotation	Extension	Flexion
Clinically measured	118 ± 28 (65 to 180)	34 ± 16 (10 to 75)	58 ± 6 (50 to 65)	139 ± 25 (80 to 160)
Without osteophyte resection	84 ± 17 (55 to 128)	9 ± 14 (0 to 43)	12 ± 26 (0 to 100)	60 ± 41 (0 to 131)
With humeral osteophyte resection	86 ± 18 (64 to 142)	9 ± 14 (0 to 43)	12 ± 26 (0 to 100)	72 ± 44 (0 to 131)
With humeral and glenoid osteophyte resection	85 ± 20 (55 to 142)	16 ± 21 (0 to 78)	34 ± 45 (0 to 120)	84 ± 29 (13 to 131)

When evaluating the differences between clinical ROM and PPS-predicted ROM, there was a mean ± standard deviation (range) difference in abduction of 33° ± 29° (−32 to 93), in external rotation of 18° ± 23° (−38 to 57), in extension of 45° ± 25° (−35 to 65), and in flexion of 54° ± 51° (−51 to 147). For instance, abduction ranged from −32° (underpredicted) to +93° (overpredicted), where all other planes demonstrated similar variability between measurement techniques. The Bland-Altman plot for abduction demonstrated a positive relationship between mean clinical ROM and the difference (ie, as the clinical ROM increased, the difference between clinically achieved and PPS-predicted ROM also increased) (Figure 4). A similar relationship was observed for external rotation, with the exception of one outlier (Figure 5). For extension, the Bland-Altman plot demonstrated one clustered group but four other patients in whom increased ROM values negatively associated with increased mean differences (Figure 6). For flexion, the Bland-Altman plots did not show any specific relationship between mean ROM and mean difference between ROM techniques (Figure 7). Thus, there was no statistically significant correlation between PPS and clinically measured ROM values as indicated by P values of Spearman correlation tests (rho), as PPS and clinical values without osteophyte resection were P = 0.180 (0.365) for abduction, P = 0.643 (0.130) for external rotation, P = 0.151 (0.404) for extension, and P = 0.112 (−0.428) for flexion. With humeral osteophyte resection, values were P = 0.176 (0.369) for abduction, P = 0.643 (0.130) for external rotation, P = 0.151 (0.404) for extension, and P = 0.002 (−0.724) for flexion. Humeral and glenoid osteophyte resection values were P = 0.180 (0.365) for abduction, P = 0.458 (0.207) for external rotation, P = 0.794 (−0.077) for extension, and P = 0.002 (−0.737) for flexion as indicated by P values of Spearman correlation tests (Table 3).

Figure 4.

Graph showing a Bland-Altman plot of abduction measurements

Figure 5.

Graph showing a Bland-Altman plot of external rotation measurements

Figure 6.

Graph showing a Bland-Altman plot of extension measurements

Figure 7.

Graph showing a Bland-Altman plot of flexion measurements

Table 3.

Correlation Between Preoperative Planning Software and Clinically Generated Range-of- Motion Values [p-values (Rho)]

P of Spearman Correlation Tests (Rho)
Motion	Abduction	External Rotation	Extension	Flexion
Without osteophyte resection	0.180 (0.365)	0.643 (0.130)	0.151 (0.404)	0.112 (−0.428)
With humeral osteophyte resection	0.176 (0.369)	0.643 (0.130)	0.151 (0.404)	0.002 (−0.724)
With humeral and glenoid osteophyte resection	0.180 (0.365)	0.458 (0.207)	0.794 (−0.077)	0.002 (−0.737)

Discussion

The PPS mostly underpredicted in vivo clinical postoperative ROM in patients with rTSA. In abduction, most of the data support that increased clinical ROM correlates with greater difference between clinically measured and PPS ROM values. This could be because PPS only considers glenohumeral motion, whereas clinical motion includes both glenohumeral and scapulothoracic motion (ie, humerothoracic motion). However, in select cases, clinical motion was less than that predicted by PPS, and the range of differences across this cohort of 15 patients was large. The overprediction of clinical ROM may be because of soft-tissue tension or scarring that limits glenohumeral motion to less than the impingement-free arc of motion. These results suggest that while PPS may be useful for preoperative planning to avoid impingement-related complications, it incompletely captures clinically achieved ROM.

Inability to isolate glenohumeral motion from scapulothoracic motion in clinic indicates a potentially notable discrepancy in the measurement methodology when compared with the isolated glenohumeral calculations done by PPS. Scapulothoracic contributions to humeral thoracic motion may increase after rTSA.⁴ Within our data set, the range of differences between PPS-predicted ROM and clinically achieved ROM was as large as 93° in abduction, 57° in external rotation, 65° in extension, and 147° in flexion. These values varied differently between motions. This suggests that no easy ‘correction factor’ could be applied to account for scapulothoracic motion. To truly predict ROM, future software may need to determine methods whereby scapulothoracic motion can be accounted for in the predicted ROM.

In addition, the predicted free arc of motion that is measured within PPS has the potential to be affected by the soft tissues. Two potential inaccuracies may arise from the soft tissues. First, the full impingement-free arc of motion may not be reached because of soft-tissue tension, which is influenced by component selection and positioning.^5,6 Indeed, these factors may work against one another. For instance, with more lateralization and distalization, there may be a larger impingement-free arc but also more soft-tissue tension, potentially resulting in less clinical motion despite greater PPS-predicted motion. This effect has been shown biomechanically where increasing sphere size has been correlated with a reduced ROM in a cadaveric shoulder simulator.⁷ Second, the PPS does not account for muscular strength while cuff muscle atrophy, tendon retraction, and fatty infiltration are common in rTSA patients. As before, selection and placement of implant components also influence muscle moment arms and, therefore, their force-producing capabilities.^8,9 Thus, while there may be an available arc of motion, the patient may not have muscle available to actively power the arm into the arc.

Component wear and scapular notching may also create inaccuracies in PPS prediction of ROM. After component placement, notching can occur, which may increase the impingement-free arc of motion beyond initial predictions.¹⁰ Similarly, wear may occur on the polyethylene rim, which may also increase the impingement-free arc of motion, essentially by reducing the constraint within the system. Thus, the achieved arc of motion may be greater than predicted, especially at a mean of 3 years postoperatively measured within our study. This factor could lead PPS to underpredict motion in select cases.

The final potential discrepancy between PPS and the clinical measurement methodology is the effect of humeral rotation and resting scapular posture. For instance, clinically measured shoulder rotational measurements are of the forearm relative to the thorax, with the forearm pointing anteriorly defined as zero degrees. PPS rotational measurements are of the humerus relative to the center of the glenosphere, with the resting posture configuration set in that the scapula is at 30° internal rotation and the humerus is at 10° abduction with the arm along the body, with zero defined as the resting posture within the CT scan. There are multiple sources of variation between these methods of measurement, including rotational deformities of the thorax such as kyphosis, differences between patients in resting scapular posture, and humeral torsion. These differences affect how the neutral position is defined, and therefore, the relative deviations from neutral may be biased. Users of PPS using the ROM feature should be aware of the presence of confounding variables and adjust use accordingly.

This study has several limitations. The patient pool was established from existing records of patients who had both preoperative and postoperative CT scans, which limited the sample size of this study. The postoperative CT scans were only obtained for patients with notable glenoid deformity where glenoid bone grafting was performed, introducing selection bias. Segmentation and registration were done by a single author and may be subject to error, but followed reliable protocols for preparation of 3D models from CT image stacks. Clinical ROM values were measured with a goniometer and were subject to variability and human error. A previous study documented intraobserver (11° to 16°) and interobserver (14° to 24°) goniometer repeatability for multiple shoulder activities, which is well below the effect size measured in this study.¹¹ Clinically, the movement of glenohumeral motion was not isolated from scapulothoracic motion as it is in PPS. The 3D models were produced by two surgeons and also subject to human error. Osteophyte removal in PPS was subjective and performed by two surgeons by consensus to best approximate bone removal typical of a surgical case. PPS measures passive ROM, and active ROM was measured and recorded in clinic.

Conclusion

The passive glenohumeral impingement-free ROM generated from PPS incompletely predicts clinically measured active humerothoracic ROM, possibly because of the unmeasured factors of soft-tissue tension, muscular strength, humeral torsion, resting scapular posture, and, most importantly, scapulothoracic motion. Optimization of PPS likely requires inclusion of soft tissues, distal humerus morphology, scapulothoracic and glenohumeral ROM, and kinematic considerations to accurately predict clinical ROM. Owing to these extensive confounding variables, PPS-generated ROM values do not correspond with clinically measured ROM.