Predictors of patient-reported outcomes after primary arthroscopic rotator cuff repair

Abstract

Background:

Patient-reported outcomes vary following rotator cuff repair surgery, yet the preoperative factors associated with patient outcomes are not fully understood. This study aimed to assess associations of preoperative patient, disease and surgical factors with 1-year Penn Shoulder Score (PSS) in patients undergoing primary arthroscopic rotator cuff repair (ARCR).

Methods:

Patients who underwent ARCR for superior-posterior rotator cuff tendon tears at Cleveland Clinic from February 2015-February 2022 with completed baseline PSS were included. We used multivariable identity-link beta regression and proportional odds models to fit 1-year PSS-Total and subscores to 23 prospectively identified patient, disease, and surgical factors and used R², Nagelkerke’s pseudo-R², and incremental changes in Akaike Information Criterion to respectively assess model overall predictive capacities and predictor relative importances, multiply-imputing missing data.

Results:

Of 3,483 cases of mean age of 58.8 ± 9.7 (standard deviation) years, 59% were males, 13% current smokers, 24% had chronic pain, 46% had used opioids within the prior year, 32% had at least one psychiatric diagnosis, and 5% had Worker’s Compensation insurance. Median (quartiles) preoperative Veterans Rand 12-Item Health Survey Mental Component Score (VR-12-MCS) and PSS were 53.2 (43.9, 60.5) and 40.0 (29.5, 52.0), respectively. 1-year PSS was provided by 2,491 patients (72%) with median 90.0 (76.0, 97.0) reflecting significant improvement. Lower preoperative PSS, VR-12-MCS, and nonprivate insurance (particularly Workmen’s Compensation) were the most important predictors of lower 1-year PSS and all subscores. Other (neither White nor Black) self-reported race, a chronic pain diagnosis, glenohumeral cartilage degeneration, and no acromioplasty were also statistically significantly associated with lower 1-year PSS. Model R² was 20%. Sensitivity analyses showed that inclusion of preoperative PSS and VR-12-MCS as predictors suppressed the statistical significance of several other patient factors.

Conclusion:

Patients generally reported excellent 1-year outcomes following primary ARCR. The most important predictors of 1-year PSS were baseline PSS, VR-12-MCS, and insurance type, with race and acromioplasty also among top predictors. However, the 23 factors examined in this study only account for a modest fraction of the variability in 1-year PSS, suggesting that other factors, for example, strength, structural healing, and patient biology, may contribute as well.

Level of evidence:

Level II; Prospective Cohort Design; Prognosis Study

Full-thickness rotator cuff (RC) tears are present in ≥25% of individuals by age 60 and ≥50% by age 80.²⁵ These tears often cause shoulder pain, weakness, and dysfunction, leading to more than 250,000 rotator cuff repairs (RCRs) annually in the United States, with costs exceeding $3 billion.^7,48 While RCR typically reduces pain and improves function with significant improvement in patient-reported outcome measures (PROMs), PROMs following RCR do show notable variablity, and some patients report suboptimal outcomes. Although 20%–30% of RCRs experience retears, early (1 year) postoperative PROMs do not correlate well with RCR healing status,^12,51 suggesting that other patient and surgical factors^19,27 influence the early clinical success of this surgery.

Numerous studies have explored patient and surgical factors associated with shoulder pain and function following RCR. Patient factors such as age, sex, comorbidities, mental health status, social determinants of health, and Workmen’s Compensation (WC) claim and disease/surgical factors such as preoperative shoulder PROMs and strength, tear size, multiple tendon involvement, muscle atrophy and fatty infiltration, and additional biceps or acromioclavicular (AC) joint procedures have been reported as predictors of clinical outcomes.^{16,24,29,35,38,46} However, such predictive associations can vary based on study inclusion criteria (cohort heterogeneity), sample size, the specific set of covariates considered in statistical models, and the methods used for outcome analysis. Many previous studies have been relatively small and homogeneous due to narrow inclusion criteria, often relying on retrospective data collection with limited variables, which reduces their power and generalizability. Few studies have used a strict cohort design that includes factors such as opioid use, chronic pain,⁴⁹ psychiatric diagnoses,^2,21 or residential Area Deprivation Index (ADI),²⁶ all of which have been linked to poor medical outcomes in other fields. There remains a need for large, rigorously inclusive cohort studies to better guide decision-making in RCR surgery.

The Cleveland Clinic Health System’s Outcomes Management and Evaluation (OME)⁶ database is an established, valid tool for prospective collection of standardized data including preoperative demographic, disease-specific and surgical data, and shoulder-specific validated preoperative and 1-year PROMs in patients undergoing RCR.^9,39,41,42 A 1-year endpoint was adopted by our institution based on literature showing minimal differences between 1-year and 2-year outcomes for most elective orthopedic procedures,^30,31,43 combined with the added cost and significant loss to follow-up incurred with 2-year follow-up. Using 1,442 patients undergoing RCR included in OME, we previously found that large/massive tear size, lower mental health status (Veterans Rand 12-Item Health Survey Mental Component Score [VR-12-MCS]), and several additional patient and disease-specific factors (eg, female sex, opioid use in the 3 months prior to surgery) were associated with lower preoperative patient-reported Penn Shoulder Score (PSS).⁴² In another study of 518 OME arthroscopic rotator cuff repair (ARCR) patients investigating the effect of surgeon on 1-year outcomes, we observed that preoperative PSS and VR-12-MCS were the dominant and only statistically significant predictors of 1-year PSS improvement following ARCR, after controlling for surgeon and 14 patient and disease-specific factors.³⁹

The purpose of the present study is to investigate the associations of preoperative and intraoperative factors with 1-year postoperative PROMs in a much larger cohort of patients undergoing primary ARCR. In addition to the larger cohort, this present study includes a more comprehensive set of patient, disease-specific, and surgical factors, including opioid use over the prior year, additional assessment of mental health status indicated by presence of a psychiatric diagnosis, and socioeconomic status assessed by their residential ADI. Based on our prior work showing significant correlations of preoperative patient and disease-specific factors with preoperative PSS,⁴² we also performed sensitivity analyses to assess the extent to which preoperative PSS and VR-12-MCS account for the statistical effects of other patient and disease-specific variables on 1-year PSS.

Materials and methods

Rotator cuff repair surgical cohort

Patients enrolled in the OME database⁶ (Institutional Review Board 06–196) for surgery for an RC tendon tear at the Cleveland Clinic Health System between February 1, 2015 and February 28, 2022 were considered for the study and included if undergoing primary ARCR of superior-posterior RC tears, defined as a tear of the supraspinatus, infraspinatus, and/or teres minor tendons. Patients with large/massive subscapularis tears, adhesive capsulitis, a stretched or torn capsule, undergoing concomitant labral repair, or with an os acromiale were excluded. Patients with no preoperative PSS were also excluded.

Outcomes

PSS was chosen as the primary outcome because it assesses pain, function, and satisfaction more thoroughly than other PROMs; is validated for assessment of RC pathology²²; and is considered an excellent research instrument.¹⁴ Its subdomains include pain (3 items, 0–30 points), function (20 items, 0–60 points), and satisfaction (1 item, 0–10 points), with higher scores representing less pain, better function, and higher satisfaction.²² PSS was the primary and its subscores (pain, function, and satisfaction) are the secondary outcomes for modeling.²² Several additional outcomes (modified American Shoulder and Elbow Surgeons [ASES] score, Single Assessment Numeric Evaluation [SANE] score, 1-year Patient Acceptable Symptom State [PASS], return to work by 1 year, and additional surgery within 1 year) were also collected.

Predictor selection

No outcome-driven variable selection was performed. We pre-specified, for multivariable modeling of 1-year PSS and its subscores, 23 preoperative patient and disease-specific/surgical factors as possible predictors: these included 2 patient-reported baseline measures (PSS and mental health as assessed by VR-12-MCS),⁴² 12 general patient factors (age, body mass index [BMI], comorbidities as assessed by the Charlson comorbidity index [CCI],³³ years of education, ADI,^1,20 sex, race, smoking status, opioid use in the prior year, presence of chronic pain and psychiatric diagnoses, and insurance status), and 9 disease-specific/surgical factors (prior ipsilateral shoulder surgery, RC tear type, RC tear size, repair technique, subscapularis status, biceps status, AC joint status, glenohumeral cartilage status, and whether acromioplasty was performed) (Table I). ADI, which ranks neighborhoods at census block levels by socioeconomic disadvantage, was obtained from the University of Wisconsin neighborhood atlas (2021 ADI).^1,20 Opioid use in the prior year was based on an opioid prescription in the patient’s electronic medical record (EMR) between 12 months and 24 hours before surgery. Psychiatric diagnosis (yes/no) was determined by the presence of at least one of 37 International Classification of Diseases, Ninth Revision diagnosis codes of depression, anxiety, post-traumatic stress disorder, psychosis, or bipolar disorder, and chronic pain diagnosis (yes/no) was determined by an International Classification of Diseases, Ninth Revision diagnosis code of 338.2 (chronic pain) and/or 304.0x (opioid dependence) in the EMR.⁴⁰

Table I.

Preoperative PROMs, general patient characteristics, and surgical characteristics of the ARCR cohort (n = 3,483, observed data).

Variable	Value
Preoperative PROMs
*PSS-Total (1.4% incomplete)	40.0 (29.5, 52.0)
*VR-12-MCS	53.2 (43.9, 60.5)
General patient characteristics
*Age, yr	58.8 ± 9.7
*BMI	30.4 ± 6.5
*CCI (2.2% missing)	0 (0, 1)
*Education (yr)	14.1 ± 2.8
*ADI (3.3% missing)	51.5 ± 25.4
Sex
Female	1,435 (41.2%)
Male	2,048 (58.8%)
Race (self-identified, 3.6% missing)
White	2,797 (83.3%)
Black	436 (13.0%)
Other	124 (3.7%)
Smoking status
Current	463 (13.3%)
Quit	1,126 (32.3%)
Never	1,893 (54.4%)
Prior-year opioid use
Yes, 3 mo-12 mo	1,028 (29.5%)
Yes, <3 mo	564 (16.2%)
No	1,891 (54.3%)
Chronic pain diagnosis
Yes	845 (24.3%)
No	2,638 (75.7%)
Insurance status (0.6% missing)
Private	1,977 (57.1%)
Medicare	994 (28.7%)
Medicaid	300 (8.7%)
Worker compensation	191 (5.5%)
Psychiatric diagnosis
Yes	1,099 (31.6%)
No	2,384 (68.4%)
Disease-specific/surgical characteristics
Prior ipsilateral shoulder surgery
Yes	171 (4.91%)
No	3,312 (95.1%)
RC tear type
Full-thickness	2,803 (80.5%)
Partial-thickness	680 (19.5%)
RC tear size
Large/massive	1,221 (35.1%)
Medium	1,759 (50.5%)
Small	503 (14.4%)
Repair technique
Double-row	2,368 (68.0%)
Single-row	1,115 (32.0%)
Subscapularis status
Torn and repaired	417 (12.0%)
Normal/not repaired	3,066 (88.0%)
Biceps status
Tenodesis/tenotomy	1,504 (43.2%)
Normal/no treatment/débridement	1,979 (56.8%)
AC joint status
Treated	224 (6.4%)
Normai/no treatment	3,259 (93.6%)
Glenohumeral cartilage status
Either G3/G4	283 (8.1%)
Both normal/G1/G2	3,200 (91.9%)
Acromioplasty
Yes	1,438 (41.3%)
No	2,045 (58.7%)

PROMs, patient-reported outcome measures; ARCR, arthroscopic rotator cuff repair; PSS, Penn Shoulder Score; VR-12-MCS, Veterans Rand 12-Item Health Survey Mental Component Score; BMI, body mass index; CCI, Charlson Comorbidity Index; ADI, area deprivation index; RC, rotator cuff; AC, acromioclavicular.

Missingness proportions for variables with more than one missing data point are provided in parentheses.

^*Results are presented as counts (%) for categorical variables and as mean ± standard deviation (SD) or median [quartiles], depending on degree of skewing, for numeric variables.

Statistical analysis

Continuous variables were summarized by means and standard deviations when they were reasonably symmetric and medians (quartiles) when they were otherwise, and categorical variables were summarized by frequency counts (%). Associations among predictors were assessed before considering outcomes, grouping related categories where possible (Table I). For example, glenoid and humeral cartilage status were combined into a single variable, simplifying the Outerbridge classification into 2 categories: little or no cartilage degeneration (normal/Grade 1/Grade 2) vs. substantial degeneration (Grade 3/Grade 4) on either joint surface.²⁸

Preoperative differences in patients with and without 1-year PROMs were tested by Student’s t-test, Wilcoxon rank sum test, and chi-square test as appropriate, and preoperative to postoperative changes in PROMs by Wilcoxon signed-rank tests. Missing data were multiply-imputed assuming they were missing at random.³⁷ Multivariable models were fit to 1-year PSS-Total, PSS-Pain, PSS-Function (identity-link beta regression models), and 1-year PSS-Satisfaction (proportional odds model). Differences in means (beta regression models) or odds ratios (proportional odds model) were presented with 95% confidence intervals and P values. Overall model predictive capacity was measured using R² for beta regression models and Nagelkerke’s pseudo-R² for the proportional odds model. To reduce false positives in multiple comparisons, we only considered predictors that showed a significant association with the PSS-Total, before examining its subscores. Additionally, we applied the Benjamini-Yekutieli method to control the false discovery rate below 20%.³ Further details on the methods of multiple imputation, statistical modeling, and multiple comparison control are included in the accompanying Supplement to this article.

Relative importances of each variable in explaining variation in 1-year PSS were assessed by calculating and ranking the increases in Akaike Information Criterion (AIC)¹⁷ upon removal of that variable from the full model. Pairwise Pearson correlations of 1-year PROMs (PSS, modified ASES, and SANE scores) were also calculated to empirically describe their overlap. As sensitivity analyses, we fit 2 additional models to PSS-Total, one without preoperative PSS and another without both preoperative PSS and VR-12-MCS, to assess the extent to which their inclusion might be suppressing, through mediation, the effects of other patient and disease-specific variables.

Finally, we implemented online calculators based on our primary models for 1-year PSS-Total and each subscore. These calculators report the predicted mean scores, along with a range where the central 80% of patient responses are predicted to fall.

Data management and analysis used R version 4.0,³⁴ particularly the mice⁴ betareg,⁸ and rms packages.¹³

Results

Primary arthroscopic rotator ruff repair surgical cohort

3,967 cases undergoing primary ARCR at Cleveland Clinic facilities between February 2015 and February 2022 were captured in the OME database, of whom 188 were excluded because they had a large/massive subscapularis tear (n = 84), adhesive capsulitis (n = 47), an os acromiale (n = 35), a stretched/torn capsule (n = 16), or underwent concomitant labral repair (n = 6). Another 296 cases were excluded because they had no preoperative PROMs. Ultimately, 3,483 cases of superior-posterior ARCR, performed by 36 surgeons, met the inclusion criteria and were included in the analysis cohort (Fig. 1).

Figure 1.

STROBE diagram of the inclusion and exclusion of rotator cuff repair patients. STROBE, STrengthening the Reporting of OBservational studies in Epidemiology; RC, rotator cuff; RCRs, rotator cuff repairs; PROMs, patient-reported outcome measures.

General preoperative patient and disease-specific characteristics

Table I presents the preoperative characteristics of the cohort. Patients averaged 58.8 ± 9.7 years of age, with a BMI of 30.4 ± 6.5, 14.1 ± 2.8 years of education, an ADI of 51. 5 ± 25.4, and median CCI of 0 (0, 1), VR-12-MCS of 53.2 (43.9, 60.5), and PSS of 40 (29.5, 52.0). Patients were most commonly White (83.3%), male (58.8%), without history of prior ipsilateral shoulder surgery (95.1%), and with full-thickness (80.5%), medium-sized (50.5%) tears repaired using a double-row repair technique (68%). Additionally, 45.6% were current or former smokers, 31.6% had a psychiatric diagnosis, 24.3% had a chronic pain diagnosis, and 45.7% had an opioid prescription within 12 months preoperatively. Although 43.2% underwent tenodesis or tenotomy of the long head of the biceps tendon and 41.3% underwent concomitant acromioplasty, most cases had a subscapularis tendon that was normal or not repaired (88%), an AC joint that was normal or not treated (93.6%), and little or no glenohumeral arthritis (91.9%).

Data missingness and multiple imputation

Rates of missing data for predictors ranged from 0.6% for insurance status to 3.6% for race (Table I). For 1-year PSS, 28.5% were nonrespondents, with an additional 0.6% having incomplete components of the score. Non-respondents were younger, less educated, from more disadvantaged areas, and had lower preoperative VR-12-MCS and PSS than respondents (Supplementary Table S1). They were also more likely to be non-White, smokers, Medicaid/WC insured, opioid users, and/or have a psychiatric diagnosis. Median imputed 1-year PSS for nonrespondents was 3.1 points lower than the observed scores for respondents (Supplementary Table S2).

1-year outcomes and multivariable analysis

All shoulder-specific PROMs showed clinically significant improvements from preoperative to 1 year postoperatively, with 1-year PSS-Total of 90.0 (76.0, 97.0), ASES-Total of 90.4 (76.7, 98.2), and SANE of 90.0 (80.0, 97.0) (Table II). VR-12-PCS improved by a median of 10.2 (3.6, 16.7) points; however, median VR-12-MCS change was only 0.2 (5.2, 7.0) points. In addition, 83.9% of respondents achieved an acceptable symptom state (PASS), 87.8% of patients returned to work, and only 3.5% underwent additional surgery within a year (Table II). 1-year PSS-Total and its pain and function subscores were highly correlated with their ASES counterparts (r = 0.95, 0.89, and 0.98, respectively), as was 1-year PSS-Total with 1-year SANE score (r = 0.84).

Table II.

Preoperative and 1-year postoperative measures of shoulder and general health status and recovery in patients undergoing primary arthroscopic rotator cuff repair (observed data).

Variable	Preoperative value	1-yr change	1-yr value
PSS-Total	40.0 [29.5, 52.0]	43.0 [28.2, 56.0]	90.0 [76.0, 97.0]
PSS-Pain	13.0 [8.0, 17.0]	13.0 [8.0, 17.0]	28.0 [24.0, 30.0]
PSS-Function	25.9 [18.0, 33.3]	24.1 [15.0, 32.5]	54.0 [45.7, 58.9]
PSS-Satisfaction	2.0 [0.0, 3.0]	6.0 [3.0, 8.0]	9.0 [7.0, 10.0]
ASES-Total	45.0 [31.7, 58.3]	38.8 [24.2, 51.9]	90.4 [76.7, 98.2]
ASES-Pain	25.0 [15.0, 35.0]	20.0 [10.0, 30.0]	50.0 [40.0, 50.0]
ASES-Function	20.0 [13.3, 26.7]	20.3 [12.9, 26.7]	43.3 [36.7, 48.3]
SANE	40.0 [25.0, 50.0]	45.0 [30.0, 60.0]	90.0 [80.0, 97.0]
VR-12-MCS	53.2 [43.9, 60.5]	0.23 [5.2, 7.0]	56.3 [48.1, 60.3]
VR-12-PCS	35.1 [29.0, 41.2]	10.2 [3.6, 16.7]	47.9 [40.2, 54.2]
PASS	–	–	83.9%
Return to work	–	–	87.8%
Additional surgery	–	–	3.5%

PSS, Penn Shoulder Score; ASES, American Shoulder and Elbow Surgeons; SANE, single assessment numeric evaluation; VR-12-MCS, Veterans Rand 12-Item Health Survey Mental Component Score; PASS, patient acceptable symptom state.

Results are presented as median [quartiles] for numeric variables and percentages with affirmative responses for dichotomous (yes/no) variables.

Tables III, S3, and Figure 2 detail the effects of predictors on 1-year PSS and its subscores for ARCR patients. Model R²s were 20% for PSS-Total, Pain, and Function, and 9% (pseudo-R²) for PSS-Satisfaction, indicating relatively limited fit, that is, explanatory power. Seven of 23 predictors were statistically significant for 1-year PSS-Total after adjusting for multiple comparison testing (Table III). Among patient-reported measures, preoperative PSS-Total and VR-12-MCS were significantly directly associated with 1-year PSS-Total scores. Specifically, the predicted mean 1-year PSS-Total score of a patient with preoperative PSS-Total of 29.5 (lower quartile) was 4.3 (95% confidence interval: 3.6, 4.9) points below that of a patient with pre-operative PSS-Total of 52 (upper quartile). Similarly, the predicted mean 1-year PSS-Total score of a patient with preoperative VR-12-MCS of 43.9 was 3.1 (2.3, 4.0) points lower than that of a patient with preoperative VR-12-MCS of 60.5 (upper quartile). Among general patient characteristics, patients with non-White or non-Black race, Worker’s Compensation Insurance, or with preoperative chronic pain diagnosis had predicted mean 1-year PSS-Total scores that were 5.7 (2.1, 9.2), 9.0 (5.7, 12.4), or 1.5 (0.4, 2.6) points, respectively, below those of patients from their respective reference categories. Finally, among disease/surgical factors, a patient with grade 3 or 4 cartilage degeneration of the glenoid and/or humerus had predicted mean 1-year PSS-Total score of 2.3 (0.5, 4.0) points lower, and a patient undergoing acromioplasty had predicted mean 1.5 (0.6, 2.4) points higher, than patients who did not, after controlling for all other variables.

Table III.

Estimated effects (differences in means) of preoperative PROMs, general patient characteristics, and surgical factors on 1-year PSS-Total, with 95% confidence intervals for each predictor, calculated using the full model in patients undergoing primary arthroscopic rotator cuff repair.

Variable (reference category)	Change or level	Estimated difference	P value
Preoperative PROMs
PSS-Total	22.5 point increase	4.2 (3.6, 4.9)	<.001
VR-12-MCS	16.7 point increase	3.1 (2.3, 3.9)	<.001
General patient characteristics
Age	13-yr increase	0.1 (0.7, 0.9)	.733
BMI	7.7 kg/m2 increase	0.2 (0.4, 0.8)	.452
CCI	1 unit increase	−0.4 (−0.8, 0.0)	.031*
Education	4y increase	0.5 (0.2, 1.2)	.197
ADI	39 unit increase	−0.9 (−1.7, 0.1)	.024*
Sex (vs. male)	Female	−0.7 (−1.7, 0.2)	.136
Race (vs. White)	Black	−1.4 (−3.1, 0.3)	.003
	Other	−5.7 (−9.2, 2.1)
Smoking status (vs. never)	Quit	−0.1 (−1.1, 0.8)	.074
	Current	−1.7 (−3.5, 0.2)
Prior-year opioid use (vs. no)	Yes, 3 mo-12 mo	−0.4 (−1.6, 0.9)	.052
	Yes, <3 mo	−1.5 (−2.7, 0.3)
Chronic pain diagnosis (vs. no)	Yes	−1.5 (−2.6, 0.3)	.011
Psychiatric diagnosis (vs. no)	Yes	0 (−1.1,1.1)	.995
Insurance status (vs. private)	Medicare	−0.9 (−2.1, 0.4)	<.001
	Medicaid	−3.9 (−6.7, 1.0)
	Worker compensation	−9.0 (−12.3, 5.7)
Disease/surgical characteristics
Prior ipsilateral shoulder surgery (vs. no)	Yes	−2.1 (−5.1, 1.0)	.185
RC tear type (vs. partial-thickness)	Full-thickness	1.3 (0.0, 2.5)	.046*
RC tear size (vs. small)	Medium	−0.4 (−1.8, 1.0)	.732
	Large/massive	−0.3 (−1.8, 1.3)
Repair technique (vs. single-row)	Double-row	0.5 (0.6, 1.5)	.404
Subscapularis status (vs. normal/not repaired)	Torn and repaired	−0.7 (−2.2, 0.7)	.308
Biceps status (vs. normal/no treatment/débridement)	Tenodesis/tenotomy	0.1 (0.8, 1.0)	.800
AC joint status (vs. normal/no treatment)	Treated	0.7 (1.0, 2.5)	.412
Glenohumeral cartilage status (vs. both normal/G1/G2)	Either G3/G4	−2.3 (−4.0, 0.5)	.012
Acromioplasty (vs. no)	Yes	1.5 (0.6, 2.3)	.001

PROMs, patient-reported outcome measures; PSS, Penn Shoulder Score;VR-12-MCS, Veterans Rand 12-Item Health Survey Mental Component Score; BMI, body mass index; CCI, Charlson comorbidity index; ADI, area deprivation index; RC, rotator cuff; AC, acromioclavicular.

The effects for continuous variables (preoperative PSS-Total, preoperative VR-12-MCS, age, BMI, CCI, education, and ADI) are comparing predicted means at their respective quartiles (75th vs. 25th percentiles).

Examples of interpretation of PSS-Total scores (identity-link beta regression models).

• A patient with preoperative PSS-Total of 52 (75th percentile) has 1-year PSS-Total scores that are 4.2 points higher on average than a patient with preoperative PSS-Total of 29.5 (25th percentile), after controlling for all other variables.

• A patient with grade 3 or 4 arthritis of the glenoid and/or humerus has 1-year PSS-Total scores that are 2.3 points lower on average than a patient who does not, after controlling for all other variables.

^*Nonsignificant after multiple comparison control.

Figure 2.

Forest plots of adjusted effect estimates for each predictor: differences in means for 1-year PSS-Total and Pain and Function subscores and odds ratios (ORs) for Satisfaction subscore, with change interval or category comparison in the row label. PSS, Penn Shoulder Score; VR12-MCS, Veterans Rand 12-Item Health Survey Mental Component Score; BMI, body mass index; CCI, Charlson comorbidity index; ADI, Area Deprivation Index; RC, rotator cuff.

The same 7 predictors were also significantly associated with 1-year PSS Pain, Function, and Satisfaction subscores, except for chronic pain diagnosis which was not associated with 1-year PSS-Function, and glenohumeral cartilage degeneration which was not associated with 1-year PSS-Satisfaction (Supplementary Table S3). The directions of associations with 1-year subscores were consistent with those observed for 1-year PSS-Total score (Fig. 2).

Relative importance of predictors in influencing 1-year PSS and subscores

AIC analysis revealed that preoperative PSS, VR-12-MCS, insurance status, and race were consistently among the top 5 predictors for 1-year PSS-Total and each of its subscores. The fifth most predictive factor was acromioplasty for PSS Total, Pain, Function, whereas a chronic pain diagnosis ranked among the top 5 predictors for PSS-Satisfaction (Fig. 3). The top 5 variables explained 90% of the model’s ability to predict 1-year PSS-Total.

Figure 3.

Relative variable importances of patient demographic and disease-specific characteristics on 1-year PSS (A) Total and (B) Pain, (C) Function, and (D) Satisfaction subscores, based on increases in Akaike information criterion (AIC) upon removal from the full model. PSS, Penn Shoulder Score; MCS, Mental Component Score; ADI, Area Deprivation Index; CCI, Charlson comorbidity index; AC, acromioclavicular; BMI, body mass index; Psy, psychiatric.

Sensitivity analysis without preoperative PROMs

The predictive capacity of the 1-year PSS-Total score model decreased from 20% to 14% when preoperative PSS was removed and dropped further to 11% when VR-12-MCS was also removed. Without preoperative PSS in the model, ADI, sex, and prior-year opioid use became significant prognostic factors for 1-year PSS. When both preoperative PSS and VR-12-MCS were removed, CCI, education, smoking, psychiatric diagnosis, and RC tear type also emerged as significant prognostic factors for 1-year PSS (Supplementary Table S4). AIC plots for these additional models are presented in Supplementary Figure S1.

Risk calculators

The risk calculator based on our primary 1-year PSS-Total and subscore models is implemented and publicly available online in the Cleveland Clinic Risk Calculator Library at https://riskcalc.org/Predicting1YearPROMSAfterRotatorCuffRepair.

Discussion

This study aimed to investigate preoperative patient, disease-specific, and surgical factors associated with 1-year PSS in a large prospective cohort of patients undergoing primary ARCR. Consistent with prior literature,⁴³ patients generally reported excellent clinical outcomes following primary ARCR. Patients achieved a median 1-year PSS of 90 points, increasing by a median of 43 points, both a clinically and statistically significant improvement from preoperative values (minimal clinically important difference for PSS is 11 points).²² In addition, 84% of respondents reached a PASS, 88% returned to work, and only 3.5% had additional surgery within 1 year. These results demonstrate that ARCR is a successful surgery for most patients, at least in the short term.

The study’s models showed that higher preoperative PSS was the primary predictor of higher 1-year PSS following ARCR. This is consistent with other studies that have found strong associations between preoperative and 1-year postoperative PROMs after ARCR, including SANE, ASES score, and the Western Ontario Rotator Cuff index.^18,32 Preoperative PROMs have also been identified as a significant predictor of the change in PROMs from baseline to 1 year.^15,46 Notably, patients with higher preoperative PROMs generally exhibit lower change scores. This is due to 2 statistical phenomena: the “ceiling effect,” where patients with higher preoperative PROMs have less room for improvement, and “mathematical coupling”, a statistical artifact in which change scores are negatively correlated with baseline scores because the baseline value is embedded in the calculation (ie, 1-year PSS–baseline PSS). Although change scores are generally more symmetrically distributed than absolute scores and thus often more suitable than absolute scores for modeling with standard statistical methods, they ignore both ceiling effects and distinctions between very different scenarios such as 40 point-increases in PSS-Total from 20 to 60 vs. from 60 to 100 points. Hence, we used absolute PSS scores and subscores as outcomes and applied beta regression modeling (or proportional odds modeling for PSS-Satisfaction) to better handle skewness and accommodate ceiling effects.

Preoperative VR-12-MCS was the second most important predictor of 1-year PSS, consistent with reports that lower mental health status at the time of surgery is associated with a higher risk of poor outcomes for various orthopedic procedures.^10,40,47 We previously showed that preoperative VR-12-MCS was significantly associated with preoperative PSS in patients undergoing RCR.⁴² However, since VR-12-MCS captures just a 4-week snapshot of recent mental status, the extent to which it serves as a true indicator of mental health or, instead, a measure of distress related to preoperative shoulder pain and disability is uncertain. In this study, preoperative VR-12-MCS remained an important predictor of 1-year PSS, even when controlling for an established psychiatric diagnosis. Furthermore, in sensitivity analysis that excluded preoperative VR-12-MCS, presence of a psychiatric diagnosis, along with BMI, CCI, smoking, and RC tear type, emerged as independent predictors of 1-year PSS. Together, these findings suggest that preoperative VR-12-MCS captures mental health influences by both mental illness and shoulder disease, along with other general health conditions. Interestingly, while shoulder-specific (PSS, ASES, and SANE) and overall physical health (VR-12-PCS) PROMs showed substantial improvements over 1 year, VR-12-MCS remained unchanged, indicating it may reflect an underlying mental health condition rather than temporary distress linked to shoulder problems or general physical conditions.

Race (particularly, self-identifying as neither White nor Black), insurance (particularly, receiving Worker’s Compensation and, to a lesser degree, Medicaid), and having a chronic pain diagnosis were also independently associated with lower 1-year PSS-Total. Our results are consistent with studies showing that social determinants of health influence clinical outcomes after RCR through various mechanisms such as insurance coverage and access to healthcare services.^24,36,52 When preoperative PSS was removed from the model, factors such as higher ADI, female sex, and prior-year opioid use became significant predictors for lower 1-year PSS. Similarly, when both preoperative PSS and VR-12-MCS were excluded, factors such as higher CCI, less education, smoking, psychiatric diagnosis, and full-thickness tear type also emerged as significant predictors of lower 1-year PSS. This suggests that preoperative PSS and VR-12-MCS partially account for the effects of several patient factors on 1-year PSS. It also highlights how the selection of variables can affect the estimated effects of predictors in multivariable models. This may explain some of the differences in findings across studies, especially when results are compared without fully addressing or recognizing confounding factors. While adjusting for confounding typically increases the validity of associations, it can sometimes obscure the effects of closely related predictors.

Notably, we did not find that prior ipsilateral shoulder surgery, RC tear size, subscapularis status, biceps status, or RCR technique were associated with 1-year PSS. Although systematic reviews and meta-analyses have identified RC tear size as a predictor of PROMs after RCR,^15,38,46 individual studies using multivariable models report conflicting results.^11,45 This demonstrates again how associations found in multivariable models can vary depending on the predictors included and the degree of adjustment for confounding factors. While preoperative RC pathology has been consistently linked to postoperative RCR healing status,^35,38 we did not evaluate 1-year RCR healing in this study.

While most factors related to the RC were not significant predictors of 1-year PSS in our model after adjusting for other covariates, concomitant acromioplasty during ARCR emerged as a significant predictor of 1-year PSS-Total and each subscore (Fig. 3). This finding, coupled with our previous study indicating a link between acromial pathology and worse preoperative PROMs,⁴² suggests that acromioplasty is associated with improved PROMs following ARCR. The literature on acromioplasty’s impact in RCR is mixed,^23,44 with some studies finding no clear benefit. However, a recent systematic review of randomized controlled trials indicates that acromioplasty during ARCR can enhance postoperative PROMs and decrease reoperation rates.⁵⁰ While our study supports the idea of acromioplasty providing a modest benefit on 1-year PROMs, it is important to note that this relationship may be confounded by variations in surgical practice among participating surgeons, including differences in both the extent of acromial pathology addressed in patients and the frequency of performing acromioplasty during ARCR.

Finally, we found that evidence of glenohumeral cartilage degeneration at surgery was associated with a lower 1-year PSS. While glenohumeral cartilage degeneration was not found to be associated with preoperative symptoms,⁴² its significant association with 1-year PROMs may be a consequence of the persistence of this pathology in the shoulder that is not addressed by RCR. These results generally align with the clinical expectation of poorer outcomes with underlying degenerative cartilage changes over time.⁵

The present study has several notable strengths. It uses data from a large prospective cohort that captures a wide range of patient, disease, and surgical factors relevant to ARCR. The nearly 100% enrollment rate, with a high rate (72%) of follow-up for 1-year PROMs and multiple imputation of any missing data, minimizes patient selection bias and ensures a comprehensive evaluation of factors affecting ARCR outcomes. Preoperative and intraoperative patient factors, identified through literature review or expert judgment, were prospectively selected and incorporated into multivariable models to identify statistically significant associations. To address skewness and ceiling effects in the 1-year PSS data, we employed beta regression and proportional odds modeling. Finally, we have implemented the primary multivariable models as a publicly available online calculator. Although the effects of several significant variables were small and perhaps not clinically significant in isolation, these variables occur in patients in combinations with additive effects in our multivariable models. For example, a non-White/non-Black patient with WC insurance, at the lower quartiles of both baseline mental health status (VR-12-MCS = 43.9 and PSS-Total = 29.5), and not having acromioplasty would have, on average, a 23.5-point (5.7 + 9.0 + 3.1 + 4.2 + 1.5) lower 1-year PSS-Total than a White patient with private insurance, at the upper quartiles of baseline mental health status (VR-12-MCS = 60.5 and PSS-Total = 52.0), and having acromioplasty, assuming common values of all other modeled factors.

The study also had several limitations. First, all patients were from a single tertiary hospital network, which may limit the generalizability of the findings to other populations. Second, nearly 30% of the cohort had missing 1-year PROMs, with significant differences in preoperative characteristics between respondents and nonrespondents. To address this, we used multiple imputation to estimate 1-year PSS in nonrespondents. Third, the study did not include other potential disease-related predictors of 1-year PROMs, such as tear etiology/chronicity, RC muscle pathology (like atrophy or fatty infiltration), or shoulder bony anatomy. Additionally, we only modeled patient-reported PSS, without considering objective functional outcomes such as strength and range of motion, or structural outcomes such as RCR healing status. Indeed, the relatively low predictive capacities (20% for PSS Total, Pain, and Function and 9% for PSS-Satisfaction) and the wide 80% prediction intervals suggest that there is considerable variability in patient outcomes, even when the 23 predictors we used are fixed, indicating that other unmeasured factors also play an important role in outcomes after RCR. Moreover, some preoperative characteristics, like opioid use and psychiatric diagnosis, were collected retrospectively from the EMR, and limited information was available on the actual drug usage or the extent, duration, or severity of the diagnoses. Finally, the study assessed associations rather than causality. Future research using causal analysis methods is needed to better understand causal relationships between predictors and outcomes.

Conclusion

In summary, 23 preoperative patients’ disease-specific and surgical factors were analyzed as predictors of 1-year PSS in a large cohort of primary ARCR patients. The most important predictors were preoperative PSS, VR-12-MCS, and insurance status, with race and acromioplasty also ranking among the top predictors for 1-year PSS-Total and all subscores. Sensitivity analyses showed that preoperative PSS and VR-12-MCS partially account for the effects of other patient and disease factors, highlighting how the selection of covariates can impact multivariable model results and the need for careful interpretation when comparing studies using different sets of covariates.

Supplementary Material

Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jse.2025.05.002.