Variability of glenohumeral positioning and bone-to-tendon marker length measurements in repaired rotator cuffs from longitudinal computed tomographic imaging.

BACKGROUND: To address the need for more objective and quantitative measures of tendon healing in research studies, we intend to use computed tomography (CT) with implanted radiopaque markers on the repaired tendon to measure tendon retraction following rotator cuff repair. In our small prior study, retraction at 1-year follow-up averaged 16.1± 5.3 mm and exceeded 10.0 mm in 12 of 13 patients, and thus tendon retraction appears to be a common clinical phenomenon. This study's objectives were to assess, using 5 longitudinal CT scans obtained over 1 year following rotator cuff repair, the variability in glenohumeral positioning because of pragmatic variations in achieving perfect arm repositioning and to estimate the associated measurement variability in bone-to-tendon marker length measurements.
METHODS: Forty-eight patients underwent rotator cuff repair with intraoperative placement of radiopaque tendon markers at the repair site. All patients had a CT scan with their arms at the side on the day of surgery and at 3, 12, 26, and 52 weeks postoperatively. Glenohumeral position (defined by the orientation and distance of the humerus with respect to the scapula) and bone-to-tendon marker lengths were measured from each scan. Within-patient variation in glenohumeral position measurements was described by their pooled within-patient standard deviations (SDs), and variation in bone-to-tendon marker lengths by their standard errors of measurement (SEMs) and 95% confidence level minimally detectable distances (MDD95) and changes (MDC95).
RESULTS: The mean glenohumeral orientation from the 5 longitudinal CT scans averaged across the 48 patients was 12.6° abduction, 0.4° flexion, and -0.1° internal rotation. Within-patient SDs (95% confidence intervals) of glenohumeral orientation were 3.0° (2.7°-3.4°) in extension/flexion, 5.2° (4.6°-5.8°) in abduction/adduction, and 8.2° (7.3°-9.2°) in internal/external rotation. The SDs of glenohumeral distances were less than 1 mm in any direction. The estimated SEMs of bone-to-tendon lengths were consistent with a common value of 2.4 mm for any of the tendon markers placed across the repair, with MDD95 of 4.7 mm and MDC95 of 6.7 mm.
CONCLUSION: Apparent tendon retraction of 5 mm or more, when measured as the distance from a tendon marker's day of surgery location to its new location on a volumetrically registered longitudinal CT scan, may be considered above the usual range of measurement variation. Tendon retraction measured using implanted radiopaque tendon markers offers an objective and sufficiently reliable means for quantifying the commonly expected changes in structural healing following rotator cuff repair.

Healing following rotator cuff repair remains a significant clinical challenge. Recurrent tears have been reported for 10%-90% of the rotator cuff repairs,,,,, depending on the patient's age and tear size., Meta-analyses of the available literature has shown that the correlation between recurrent tears and clinical outcomes in the early to intermediate postoperative period (≤5 years) following rotator cuff repair is weak on average and therefore of uncertain clinical importance.,, The relationship is also inconsistent, as reports show that individually some patients with recurrent tears still have less pain and higher functional scores after surgery,,, whereas patients with a structurally intact repair may demonstrate persistent degenerative muscle changes and shoulder weakness.,,, The weak and at times inconsistent correlation between recurrent tears and clinical outcomes following rotator cuff repair suggests a need for a deeper and more nuanced understanding of structural tendon healing.

Magnetic resonance imaging (MRI),,, magnetic resonance arthrography,, and ultrasonography,, are routinely used for diagnosing rotator cuff tears and repair integrity after surgery. Excellent sensitivity and specificity have been reported for each technique in terms of its binary diagnostic reliability. A more granular classification of rotator cuff repair integrity using the 5-point Sugaya score can be used, with fair to moderate exact inter-rater agreement (simple κ= 0.31-0.49), and improved agreement when adjacent-category disagreements are penalized only minimally (weighted κ= 0.72). Although these modalities constitute the current standard-of-care for diagnostic imaging of the rotator cuff, none can discern the type or quality of tissue at the repair site, and thus each faces challenges of clinical interpretation. To address the need for more objective and precise measures of tendon healing, we and others have developed a means by which to measure tendon retraction following rotator cuff repair using implanted radiopaque markers on the repaired tendon and 3-dimensional (3D) radiographic imaging.,

We broadly defined tendon retraction as the translation of the repaired tendon away from its position of initial fixation on the bone with or without formation of a recurrent defect. In a study involving 13 patients with rotator cuff repair, rotator cuff tendon retraction at 1-year follow-up after repair averaged 16.1 ± 5.3 mm (range 6-23 mm) and exceeded 10.0 mm in 12 of 13 patients. Thus, tendon retraction appears to be a common clinical phenomenon, yet only 30% (4 of 13) of repairs were classified as structurally failed by MRI. This contrast between the continuous and categorical results suggests that “failure with continuity”—tendon retraction without the formation of a recurrent defect—might partially explain how patients with apparently structurally intact repairs on MRI might experience undesirable clinical outcomes.,,,

In another study of accuracy and repeatability of tendon retraction measurements in 6 patients with rotator cuff repair, we observed that variation in repositioning the arm for repeat scanning could introduce variation in tendon retraction length measurements. The root mean square difference between bone-to-tendon length measurements on computed tomographic (CT) scans repeated twice on the same day, in arm positions such as the hand on umbilicus or hand at the side, was 1.3 mm. Because tendon retraction in the clinical setting will derive from lengths measured on CT scans repeated longitudinally, we expect that pragmatic variation in glenohumeral position across longitudinal scans would also introduce variability into tendon retraction measures. Therefore, the objectives of this study were to (1) quantify the variability in glenohumeral positioning among 5 longitudinal CT scans obtained with the arm at the side in the first year following rotator cuff repair and (2) estimate the associated measurement variability in bone-to-tendon marker length measurements due to pragmatic variations in achieving perfect arm repositioning across the scans.

Materials and methods

Patient longitudinal CT scanning

In this study, 48 patients participating in an ongoing, IRB-approved study (IRB 16-089, https://clinicaltrials.gov/ct2/show/NCT02716441) at our institution were analyzed. These patients had undergone arthroscopic repair of a rotator cuff tear involving the supraspinatus tendon with or without extension into the infraspinatus tendon. Following rotator cuff repair, radiopaque markers—either 1-mm tantalum beads (Halifax Biomedical, Mabou, Nova Scotia, Canada) in 18 patients or FibermarX tissue markers (Viscus Biologics LLC, Cleveland, OH, USA) in 30 patients—were tied to the superficial surface of the tendon. Tantalum beads were predrilled, threaded onto 5-0 Prolene and tied to a 2-0 Ti-Cron suture that was subsequently tied to the tendon surface. Markers were placed medial to the repair sutures at approximately 1-cm intervals in the anterior-posterior direction using a standard suture lasso and arthroscopic knot-tying technique. Given the variability in tear size, shape, and location across the cohort, between 2 and 5 radiopaque tendon markers were used per patient. Each tendon marker was named based on its anatomic location relative to the repair sutures and myotendinous junction. Specifically, T1, T3, T5, and T7 were used to name tendon markers just medial to the repair sutures, based on the involvement of the supraspinatus (T1, T3) and/or infraspinatus tendons (T5, T7). Similarly, T2, T4, and T6 were used to name markers placed on the repaired tendon more medially, near the myotendinous junction. The locations and appropriate labeling of the tendon markers were defined by visual inspection intraoperatively and subsequently confirmed against the bony anatomy on the day of surgery CT (CTDOS) for each patient.

All the patients underwent 5 longitudinal CT scans: the CTDOS and at 3 (CTWK003), 12 (CTWK012), 26 (CTWK026), and 52 weeks (CTWK052) postoperatively. A low-dose scan protocol was performed (100-120 kVp, 45-60 mAs) on Siemens Definition Edge, Siemens Somatom Sensation 64 (0.6-mm slices; Siemens Medical Solutions, Malvern, PA, USA), GE Revolution EVO 64 (0.625-mm slices; GE Healthcare, General Electric Company, Boston, MA, USA), or Philips Brilliance 16 (0.8-mm slices; Philips Electronics N.A., Andover, MA, USA) scanners.,, At the time of each CT scan, patients were instructed to lie supine and position the operated arm at the side, with hand on the thigh and thumb facing upward for consistency in arm positions across the longitudinal CT scans. The field of view was adjusted to enclose the shoulder joint, including the clavicle, scapula, and entire humerus. The entire image volume was reconstructed using a soft tissue kernel for image processing and analysis. Image processing was performed using custom software (RotatorCuffEvaluation, version 2.1.1; Cleveland Clinic, OH, USA) to load, segment, visualize, and register longitudinal CT images, identify the bony landmarks to define anatomic coordinate systems, and compute the 3D coordinates of the radiopaque markers with respect to the anatomic coordinate systems. The specific operations are described below.

CT image analysis

Scapular and humeral coordinate systems were defined from the reconstructed CTDOS images using bony landmarks (Fig. 1). Similar to the International Society of Biomechanics guidelines for describing glenohumeral joint motion, the scapular coordinate system was defined by the 3 bony landmarks of the glenoid center, trigonum scapulae, and angulus inferior. The humeral coordinate system was defined by the 3 bony landmarks of the humeral head center and the medial and lateral epicondyle. Any tendon markers (T1-T7) were detected in the scapular coordinate system using a threshold level of ∼1500 Hounsfield units, and 4 virtual humeral bone markers (V1, V2, V3, and V4) were defined at the surface of the lateral edge of the greater tuberosity in the humeral coordinate system (Fig. 1).

Figure 1

Representative shoulder computed tomography (CT) scan depicting the scapular and humeral bone models, respective bony landmarks and anatomic coordinate systems, virtual humeral markers (V1-V4), and tendon markers (T1, T3, T5, T7). DOS, day of surgery; HC, humeral head center; LE, lateral epicondyle; ME, medial epicondyle; GC, glenoid center; TS, trigonum scapulae; AI, angulus inferior; Ant, anterior; Lat, lateral; Sup, superior; HCS, humeral coordinate system; SCS, scapular coordinate system.

Spatial registration of the CT volume to a particular bone model was performed using a modified maximization of mutual information algorithm., Each postoperative CT volume was first spatially registered to the humerus bone model of CTDOS, and the coordinates of the humeral bony landmarks and virtual humeral markers were transferred to each postoperative CT volume using the transformation matrix at each time point. Second, each postoperative CT volume was spatially reregistered to the scapula bone model of CTDOS. The previously defined coordinates of the humeral bony landmarks and virtual humeral markers for each scan were transferred into the scapular coordinate system, and the humeral coordinate system was recalculated. These steps resulted in the humeral coordinate system and virtual humeral markers at each postoperative time point being expressed in the scapular coordinate system defined on the CTDOS.

Variation in glenohumeral position

Glenohumeral position was measured in terms of glenohumeral orientation and glenohumeral distance. Glenohumeral orientation was expressed in terms of abduction/adduction, flexion/extension, and internal/external rotation of the humerus relative to the scapula. To accomplish this, we projected the 3D orientation of the humerus onto each plane of the scapular coordinate system and calculated the projected 2D angles between each pair of corresponding axes. Glenohumeral distance was defined as the distance from the glenoid center to the humeral head center (the origins of the 2 coordinate systems) in medial/lateral, anterior/posterior, and superior/inferior directions in the scapular coordinate system. An example of variation in glenohumeral position from the longitudinal scans for a given patient is shown in Figure 2.

Figure 2

(Top) Representative longitudinal computed tomographic (CT) images of a given patient throughout the year after rotator cuff repair demonstrate variations in glenohumeral position during scanning. (Bottom) The SEMs of bone-to-tendon marker lengths due to the variation in glenohumeral position were estimated by the within-patient SDs of repeated measurements of bone-to-tendon marker lengths between the coordinates of the virtual humeral markers (V1, V2, V3, V4) on each of the 5 longitudinal CT scans and the coordinates of the tendon markers set at their day of surgery (DOS) locations (T1DOS, T3DOS, T5DOS, T7DOS). The colors of the virtual humeral markers represent the time of the longitudinal CT images (yellow: DOS; green: WK003; turquoise: WK012; purple: WK026; and pink: WK052). WK, week; Ant, anterior; Lat, lateral; Sup, superior; SCS, scapular coordinate system.

Measurement variability in bone-to-tendon marker length measurement

Bone-to-tendon marker length was measured between the coordinates of the virtual humeral markers (V1-V4) on each of the 5 serial CT scans and the coordinates of the tendon markers set at their locations on CTDOS (ie, V1T1DOS, V1T2DOS, V2T3DOS, V2T4DOS, V3T5DOS, V3T6DOS, V4T7DOS). An example of variation in bone-to-tendon marker lengths resulting from variation in glenohumeral position across the longitudinal scans for a given patient is shown in Figure 2.

Statistical analysis

Variation across the 5 longitudinal CT scans was summarized, for each glenohumeral position measure and each bone-to-tendon marker length, by the within-patient standard deviation (SD) of the repeated measurements, estimated from 1-way analyses of variance with scans grouped by patient, and by 95% confidence intervals (CIs) obtained from the associated underlying chi-square distributions. The within-patient SD for each glenohumeral position measure describes its net variability from the combination of pure measurement variability, such as would be seen from replicate scans with glenohumeral position held fixed, and true scan-to-scan changes in glenohumeral position. In contrast, the bone-to-tendon lengths on each scan were measured from the virtual humeral markers to the tendon marker location identified on the CTDOS, after very precise registration of the scapulae and a deterministic translation of the tendon marker location on the CTDOS to the current scan's coordinate system. Hence, variation in these length measurements reflects essentially pure measurement variability, and the within-patient SD in this case is an estimate of the standard error of measurement (SEM) of length. The 95% confidence level minimal detectable distance (MDD) from a fixed point, and minimal detectable change (MDC) between 2 longitudinal length measurements with independent measurement errors, were estimated as and , respectively. Parallel boxplots were used to depict measurement spreads for individual patients, and the Brown-Forsythe test to screen for heteroscedasticity (heterogeneous SDs among patients). Statistical computations were performed using the FREQ, UNIVARIATE, and GLM procedures of SAS, version 9.4 (SAS Institute, Inc., Cary, NC, USA).

Results

Patient demographics, tear characteristics, and tendon marker characteristics are summarized in Table I. The average anterior-posterior tear size in this cohort was 2.1 ± 1.0 cm and tears were predominantly isolated to the supraspinatus tendon (32 of 48 patients = 67%). Forty-four of 48 patients (92%) had 2, 3, or 4 implanted tendon markers (range 1-5). The additional operative time to apply the markers at the end of the case was approximately 10-15 minutes. The 48 patients were scanned on 8 different CT scanners (CTDIvol: median 1.78 mGy, IQR 1.75-4.71 mGy; scan length: median 37.7 cm; IQR 35.3-40.0 cm; estimated effective radiation dose per scan: median 0.99 mSv, IQR 0.90-2.28 mSv) by 16 clinical imaging personnel at 6 locations in our health care system.

Table I

Distributions of patient demographics, tear characteristics, and the number of implanted radiopaque tendon markers in 48 patients undergoing rotator cuff repair surgery

Patient demographics
Age, yr, mean ± SD	59 ± 8
Sex, n (%)
Male	25 (52)
Female	23 (48)
BMI, mean ± SD	29.9 ± 6.1
Laterality
Right	32 (67)
Left	16 (33)
Tear characteristics
Size, cm, mean ± SD
AP	2.1 ± 1.0
ML	1.2 ± 0.5
Location, n (%)
SS	32 (67)
SS+IS	16 (33)
Tendon markers, n (%)
No. of markers placed
1	2 (4.2)
2	15 (31.2)
3	19 (39.6)
4	10 (20.8)
5	2 (4.2)

BMI, body mass index; AP, antero-posterior; ML, medial-lateral; SS, supraspinatus only, SS+IS, both supraspinatus and infraspinatus.

Mean ± standard deviation for continuous variables, counts and percentages for categorical variables.

The mean glenohumeral orientation from the 5 longitudinal CT scans averaged across the 48 patients was 12.6° abduction, 0.4° flexion, and –0.1° internal rotation (Figs. 3 and 4, Table II). Within a patient, the range of glenohumeral orientation from the 5 CT scans was as small as 5.2° (patient 5) and as large as 25.1° (patient 30) in abduction/adduction, as small as 1.1° (patient 12) and as large as 13.7° (patient 9) in flexion/extension, and as small as 5.2° (patient 39) and as large as 48.2° (patient 16) in internal/external rotation (Fig. 3, Supplementary Appendix S1-S3). For 40 of 48 patients (83%), the highest absolute abduction angle was measured on the CTDOS (Fig. 3, Supplementary Appendix S1). The within-patient SD (95% confidence interval) was least for extension/flexion at 3.0° (2.7°-3.4°) and highest for internal/external rotation at 8.2° (7.3°-9.2°) (Table II). Heteroscedasticity was not statistically significant for any orientation measurement.

Figure 3

Box and whiskers plot shows minimum, first quartile, median (black line within the box), third quartile, and maximum of repeated measurements of glenohumeral orientation in abduction/adduction (top panel), flexion/extension (middle panel), and internal/external rotation (bottom panel) from 5 longitudinal computed tomographic (CT) scans in each patient. Individual observations within a patient are represented by the time point expressed in the legend. The mean of the repeated measures within a patient is represented with a light-blue line. The mean of the mean glenohumeral orientation from the 5 longitudinal CT scans across the 48 patients was 12.6° abduction, 0.4° flexion, and –0.1° internal rotation. We observed that the highest absolute abduction angle was measured on the day of surgery (DOS) in 40 of 48 patients (83%). WK, week.

Figure 4

Box and whiskers plot shows minimum, first quartile, median (black line within the box), third quartile, and maximum of repeated measurements of glenohumeral distance in medial/lateral (top panel), anterior/posterior (middle panel), and superior/inferior (bottom panel) from 5 longitudinal computed tomographic (CT) scans in each patient. Individual observations within a patient are represented by the time point expressed in the legend. The mean of the repeated measures within a patient is represented with a light-blue line. The mean of the mean glenohumeral position from the 5 longitudinal CT scans across the 48 patients was 25.5 mm in lateral, –1.5 mm in anterior, and 6.4 mm in superior direction. WK, week.

Table II

The mean (range) of glenohumeral orientation and position averaged across the 5 longitudinal CT scans from 48 patients

Glenohumeral position	Mean (range) of patient means from repeated measures	Within-patient SD	95% CI	Heteroscedasticity∗P value
Glenohumeral orientation, degrees
Abduction (+) / adduction (–)	12.6 (–5.3 to 29.9)	5.2	4.6-5.8	.99
Flexion (+) / extension (–)	0.4 (–12.7 to 15.4)	3.0	2.7-3.4	.75
External (+) / internal (–) rotation	–0.1 (–18.1 to 22.9)	8.2	7.3-9.2	.75
Glenohumeral distance, mm
Lateral (+) / medial (–)	25.5 (19.1 to 31.0)	0.4	0.3-0.4	.06
Anterior (+) / posterior (–)	–1.5 (–7.0 to 2.2)	0.9	0.8-1.1	.61
Superior (+) / inferior (–)	6.4 (–0.6 to 15.8)	0.9	0.8-1.0	.0003†

CT, computed tomographic; SD, standard deviation; CI, confidence interval.

Variation in glenohumeral position was assessed by the within-patient SD and 95% CI of orientation and distance measurements.

From Brown-Forsythe test.

Statistically significant heteroscedasticity.

Similarly, the mean glenohumeral distance from the 5 longitudinal CT scans averaged across the 48 patients was 25.5 mm lateral, –1.5 mm posterior, and 6.4 mm superior from the glenoid center to humeral head center (Table II). Within a patient, the range of glenohumeral distance from the 5 CT scans was as small as 0.2 mm (patient 40) and as large as 2.5 mm (patient 30) medial/lateral, as small as 0.4 mm (patient 31) and as large as 4.5 mm (patient 44) anterior/posterior, and as small as 0.05 mm (patient 13) and as large as 6.9 mm (patient 30) superior/inferior to the origin of the scapula (Fig. 4). The SDs (95% confidence intervals) of glenohumeral distances were similar in the anterior/posterior (0.9 mm [0.8-1.1 mm]) and superior/inferior (0.9 mm [0.8-1.0 mm]) directions and lower in the medial/lateral direction (0.4 mm [0.3-0.4 mm]) (Table II). Heteroscedasticity was nonsignificant in the anterior/posterior and medial/lateral directions, but there was evidence that measurement variability and hence reliability in the superior/inferior direction itself varies across patients (P = .0003).

Variability in bone-to-tendon marker length measurement

The mean bone-to-tendon marker lengths from the 5 longitudinal CT scans averaged across scans and patients are shown in Table III by marker location across the repair. The respective estimated SEMs ranged from 2.2-2.6 mm for all markers placed, were not statistically distinguishable by marker location, and there was no statistically significant heteroscedasticity. A common estimate for the SEM, pooled across marker locations, was 2.4 mm. Corresponding MDD95 values range from 4.3-5.1 mm with a common estimate of 4.7 mm, whereas MDC95 values range from 6.1-7.2 mm with a common estimate of 6.7 mm (Table III).

Table III

The mean (range) of bone-to-tendon lengths averaged across the 5 longitudinal CT scans from 48 patients

Bone-to-tendon length		Mean (range) of means from repeat measures, mm	Estimated SEM, mm	95% CI, mm	Estimated MDD95	Estimated MDC95	Heteroscedasticity∗P value
Endpoint	n
V1T1DOS	30	20.3 (6.9-29.2)	2.6	2.3-2.9	5.1	7.2	.76
V1T2DOS	9	31.8 (25.5-42.5)	2.2	1.9-2.4	4.3	6.1	.51
V2T3DOS	44	25.6 (14.8-34.6)	2.5	2.2-2.8	4.9	6.9	.98
V2T4DOS	6	33.2 (25.9-38.9)	2.5	2.2-2.8	4.9	6.9	.06
V3T5DOS	35	27.6 (15.4-47.0)	2.2	2.0-2.5	4.3	6.1	.96
V3T6DOS	2	29.3 (28.2-30.5)	2.3	2.0-2.6	4.5	6.4	.80
V4T7DOS	13	27.2 (10.6-47.2)	2.5	2.2-2.8	4.9	6.9	.50

CT, computed tomographic; SEM, standard error of measurement; CI, confidence interval; MDD, 95% confidence level minimal detectable distance; MDC, 95% confidence level minimal detectable change.

The variability in bone-to-tendon length measurements was assessed by the SEM, its 95% CI, and the MDD95 and MDC95 between 2 distance measurements. Bone-to-tendon length was measured between the coordinates of the virtual humeral markers (V1-V4) on each of the 5 longitudinal CT scans and the coordinates of the tendon markers set at their day of surgery locations on CTDOS (T1DOS-T7DOS).

From Brown-Forsythe test.

Discussion

To address the need for more objective and quantitative tendon healing measures, we have developed a quantitative method to measure tendon retraction following rotator cuff repair using implanted radiopaque markers on the repaired tendon and longitudinal 3D CT imaging., We have previously demonstrated the biocompatibility of barium sulfate–infused polypropylene radiopaque tissue markers following 8-12 weeks' implantation in a porcine model, where no radiographic leaching, calcification, or local adverse events were observed. Now, applying and interpreting marker-derived tendon retraction appropriately in the clinical setting requires appreciation of this method's measurement variability. To that end the current study's objectives—recognizing that variation in glenohumeral position during longitudinal scanning likely introduces variability in length measurements—were to assess variation in glenohumeral position among 5 longitudinal CT scans in the first year following rotator cuff repair, and estimate the variabilities in bone-to-tendon marker length measurements due to pragmatic variations in achieving perfect arm repositioning for these scans.

We found that glenohumeral orientation was on average reproducible within approximately 8° of internal/external rotation, 5° of abduction/adduction, and 3° of flexion/extension in a given patient, whereas the center of the humeral head within a patient varied less than 1 mm in any direction in its distance from the center of the glenoid. These findings demonstrate that variation in glenohumeral position on longitudinal CT scanning arises almost exclusively from rotations of the humerus. Several factors may affect reproducibility of glenohumeral position for repeated CT scanning, one being simply the imaging technician's attention to positioning of the arm for scanning in the clinical setting. Sixteen clinical imaging personnel and 8 CT scanners at 6 locations were involved in this study. One day prior to each scan, imaging staff were reminded to “position study patients with their arm at the side,” but some variation in glenohumeral position found here is likely explainable simply by inconsistent execution of this instruction. Such variability would likely have been higher without this instruction.

The reproducibility of glenohumeral positioning for longitudinal CT scanning might also be affected by shoulder pain, swelling, body habitus, and/or comfort during lying supine, particularly when such factors change over time. For example, we observed that the highest absolute abduction angle was measured on the CTDOS in 40 of 48 patients (83%) despite removing the sling for scanning (Figs. 3 and 4), possibly explained by immediate postoperative fluid retention in the shoulder. Furthermore, acquiring good-quality images of a shoulder requires positioning the patient off-center on the scanner bed, and the ease of achieving an optimal position varies by the scanner's bore size, patient's girth, and pain at the time. Such patient factors likely contribute to the variation in glenohumeral position reported here.

The variability of bone-to-tendon marker length measurements due to variation in glenohumeral position during longitudinal CT imaging cannot be directly measured because we cannot isolate measurement uncertainty from real structural changes of the tendon over time. Hence, we estimated the variability of bone-to-tendon marker length measurements using the virtual humeral markers on each of the 5 serial CT scans and the coordinates of the tendon markers set at their locations on CTDOS. The resulting SEMs of bone-to-tendon marker lengths associated with variations in glenohumeral position were approximately 2.4 mm (range, 2.2-2.6 mm) and statistically indistinguishable between markers, regardless of the location of the measurement from anterior to posterior across the greater tuberosity. The 2.4-mm value exceeds the analogous SEM of , derived from the 1.3-mm SD of differences we previously reported from repeated CT imaging in arm positions such as the hand on umbilicus or hand at the side. This is unsurprising, because the prior study's repeated scans were taken in a well-controlled research setting on the same day whereas the current scans were obtained by multiple personnel in different clinical settings and separated by weeks and months, conducive to more variable positioning.

Of the 13 measurement variables studied, only glenohumeral distance in the superior/inferior direction exhibited statistically significant heteroscedasticity. The within-patient SD for this variable target is not a common value for all patients, but rather the quadratic mean (ie, the root mean square) of the distribution of individual within-patient repeated measurement SDs. The highly variable measurements of patients 19 and 30 were the most conspicuous drivers of this result (Fig. 4, bottom panel). Interestingly, patient 30 also had the most variable distribution of medial/lateral and abduction/adduction measurements. Several possible reasons could explain the higher variability in glenohumeral repositioning in these patients as discussed above.

Our findings provide valuable guidance for future research studies using implanted markers to quantify rotator cuff tendon retraction, that is, movement of the repaired tendon away from the humeral head footprint following surgery, which is fundamentally derived from longitudinal measurements of length using 1 of 3 approaches. The first approach is to measure and compute the difference of bone-to-tendon marker lengths on the CTDOS and on any postoperative scan in each CT scan's coordinate system without bony registration., The advantage of this approach is that no software or expertise is needed to register the longitudinal CT scans. The disadvantages are the added measurement error associated with needing to localize the positions of the reference markers from the CTDOS on all subsequent scans, and the inability to describe the direction of tendon retraction in an anatomic coordinate system. The second approach measures and computes the difference of bone-to-tendon marker lengths between CTDOS and any later postoperative time, following registration of longitudinal CT volumes to either the scapular or humeral bone models based on the user's interest for a specific anatomic coordinate system. The advantages of volumetric registration, which itself is accurate to 0.1 mm for position and 0.1° for orientation, 13 are the ability to transfer the locations of reference markers from the CTDOS onto longitudinal scans within 0.2mm accuracy (unpublished data), describe the direction of tendon retraction in an anatomic coordinate system, and measure variation in glenohumeral position as well. With either of these first 2 approaches, however, measurement variation must be accounted for in both distance measurements. Based on our data, the minimally detectable retraction measured in this manner would be 6.7 mm (MDC95).

Third, rotator cuff tendon retraction can also be estimated by measuring a single distance between a tendon marker's location on the CTDOS and its later location at any subsequent time point following scapular registration of longitudinal CT volumes. This approach of measuring tendon marker-to-tendon marker length limits variation to only a single distance measurement from a tendon marker's reference position on the CTDOS to its position at any later time. We believe the minimally detectable distance (MDD95) for measuring a single bone-to-tendon marker length estimated at 4.7 mm in this study is a reasonable estimate for tendon marker-to-tendon marker MDD, and may actually be a conservative estimate because tendon markers affixed to the compliant rotator cuff tendon medial to its bony insertion are probably less likely to be influenced by variations in humeral position than bone markers rigidly attached to the greater tuberosity. As a point of reference, we note that others have arbitrarily assigned 5 mm as the level of uncertainty of rotator cuff tendon tear size measurements made on MRI and ultrasonography.

There are a couple of noteworthy limitations to the work reported here. First, although this study estimated uncertainty introduced in length measures from variable positioning of the glenohumeral joint, it does not address measurement uncertainty from imprecision in tendon marker localization over time. In particular, tendon markers that derive from radiopaque suture knots may orient variably on longitudinal scans, leading to variation in the location of their centroids. Because the knot stacks comprising the markers are approximately 4 mm tall, we estimate any added variability from localization of their centroid to be within ±2 mm. Furthermore, based on our observations using these markers in animal models, fibrous tissue likely forms around the suture knots such that their orientation is not continuously changing beyond the early postoperative period, so it is very possible that this source of variation becomes negligible after 3-4 weeks. It should also be noted that this study did not assess the validity of tendon retraction measurements. Errors could arise from tendon markers moving, slipping or breaking away from the location where placed, and validity will be assessed in future work.

In summary, the aim of this article was to measure the variability in glenohumeral positioning among longitudinal CT scans and estimate associated measurement variability introduced in marker-based measurements of tendon retraction. Repeating a standard shoulder CT imaging protocol 5 times over 1 year postoperatively in the clinical setting resulted in within-patient SDs of glenohumeral orientation of 8° of internal/external rotation, 5° of abduction/adduction, and 3° of flexion/extension. The associated estimated SEM of bone-to-tendon length was 2.4 mm with an MDD95 of 4.7 mm, which we believe are also reasonable estimates for the SEM and MDD95 of tendon marker-to-tendon marker lengths that can be used as estimates of tendon retraction. Therefore, apparent changes in tendon retraction of 5 mm or more, when measured as the distance from a tendon marker's location at CTDOS to its new location on a volumetrically registered longitudinal CT scan, may be considered above the usual range of measurement variation.

Conclusion

Tendon retraction measured from implanted radiopaque tendon markers and longitudinal low-dose CT scanning offers a sufficiently reliable means for objectively measuring the commonly expected changes in structural healing following rotator cuff repair, though its potential clinical significance must be demonstrated and is the aim of our ongoing work (https://clinicaltrials.gov/ct2/show/NCT02716441). We anticipate that incorporating the magnitude, timing, location, and direction of tendon retraction as well as the continuity of the repaired tissue from traditional imaging will yield an improved understanding of rotator cuff tendon healing, allowing for advances in treatment strategies that improve surgical healing and clinical outcomes and result in more durable rotator cuff repairs over time. At the current time, we anticipate that tendon retraction measured from implanted radiopaque tendon markers and longitudinal low-dose CT scanning will find its primary utility as a research tool. However, if its significance as a clinically meaningful measure of structural healing following rotator cuff repair is demonstrated by our ongoing research, future work to understand the mechanisms of tendon retraction and to develop broadly applicable clinical methods for measuring tendon retraction following rotator cuff repair would be justified.

Acknowledgments

The following orthopedic surgeons contributed the cases analyzed in this study: Lutul Farrow, MD, Joseph Iannotti, MD, PhD, Eric Ricchetti, MD, and Mark Schickendantz, MD.

Disclaimer

Sambit Sahoo, or his immediate family, received payments or pecuniary, in kind, or other professional or personal benefits including stock, honoraria, or royalties (collectively, “Benefits”) or any commitment or agreement to provide such benefits from the following commercial entities related to the subject of this article: Viscus Biologics, LLC.

Andrew Baker, or his immediate family, received payments or pecuniary, in kind, or other professional or personal benefits including stock, honoraria, or royalties (collectively, “Benefits”) or any commitment or agreement to provide such benefits from the following commercial entities related to the subject of this article: Viscus Biologics, LLC.

Joseph Iannotti, or his immediate family, received payments or pecuniary, in kind, or other professional or personal benefits including stock, honoraria, or royalties (collectively, “Benefits”) or any commitment or agreement to provide such benefits from the following commercial entities related to the subject of this article: Viscus Biologics, LLC.

Kathleen Derwin, or her immediate family, received payments or pecuniary, in kind, or other professional or personal benefits including stock, honoraria, or royalties (collectively, “Benefits”) or any commitment or agreement to provide such benefits from the following commercial entities related to the subject of this article: Viscus Biologics, LLC.

All the other authors, their immediate families, and any research foundations with which they are affiliated have not received any financial payments or other benefits from any commercial entity related to the subject of this article.