Surgical and Non-Surgical Interventions in Complete Rotator Cuff Tears

Abstract

Background

This systematic review compares the efficacy and safety of surgical and non-surgical interventions for full-thickness rotator cuff tears.

Method

A systematic literature search was conducted in five databases. Randomized (RCTs) and non-randomized controlled trials of interventions (non-RCTs) for the surgical or non-surgical treatment of patients with traumatic or atraumatic full-thickness rotator cuff tears were included. The review protocol was published in the PROSPERO registry (CRD42018100343).

Results

Ten studies (three RCTs with 332 participants; seven non-RCTs with 650 participants) met the inclusion criteria. One year after treatment, shoulder function, measured with the 100-point Constant score, had improved by 6.7 points (95% confidence interval [2.3; 11.0]) and pain, measured with the 10-cm visual analog scale, by 1.1 cm (0.5; 1.7] in the full-thickness rotator cuff tears treated surgically compared with non-surgical treatment. In one study the difference in favor of surgery persisted after 10 years’ follow-up. For other outcomes, such as range of motion, muscle strength, quality of life, and adverse events, the data were sparse and the group differences were similar. The findings of the non-RCTs were comparable with those of the RCTs.

Conclusions

With regard to functional improvement and pain reduction, surgical treatment of full-thickness rotator cuff tears was superior to non-surgical treatment in the short and the long term. Whether the differences between the groups are relevant for individual cases is uncertain, as the measured results were distributed below and above the threshold of clinical relevance. The conclusions may not be applicable to rotator cuff tears over 3 cm in size or to young persons.

Tears of the rotator cuff of the shoulder joint are a commonly occurring musculoskeletal injury and may or may not be associated with symptoms. In the general population, the prevalence of rotator cuff tears exceeds 20% overall and is known to increase with age (1). Based on data from a systematic review, the prevalence of full-thickness rotator cuff tears in symptomatic persons between 44 and 50 years was found to be 35% when assessed with ultrasonography and 41% with magnetic resonance imaging (2). For those with asymptomatic tears, the prevalence of full-thickness lesions was 22% and 10%, respectively.

Rotator cuff tears can be caused by traumatic or atraumatic events, with the latter mostly being attributable to degenerative processes. Lesions are classified by size, site, number of tendons affected, degree of tendon retraction, and existing sequelae such as muscular atrophy.

Rotator cuff tears vary in the extent to which they impact shoulder function, activities of daily living, and/or quality of life (3, 4). Moreover, they often result in extended absence from work and relatively high healthcare costs (5).

The treatment options include both surgical and non-surgical (i.e., conservative) procedures. The primary aim of both approaches is to relieve pain and restore shoulder function. Surgery involves arthroscopy, minimally invasive (mini-open) techniques, or —less commonly—open interventions and employs a variety of methods such as suture, refixation, subacromial decompression, or reconstruction (with or without augmentation), as well as acromioplasty. Conservative treatment usually comprises physiotherapy, pain medication, and/or steroid injections (6).

Both conservative and surgical treatment options are frequently effective in ameliorating the symptoms. Despite the publication of several clinical studies and reviews comparing surgical and non-surgical approaches, there is still no consensus regarding the best strategy for treatment (7– 10). In practice, decisions regarding the choice of treatment are mostly guided by a variety of patient characteristics such as the patient’s age, the extent of comorbidities, the degree of functional impairment, the patient’s level of physical activity, and structural changes (11– 13). The German guideline on the management of rotator cuff tears advocates non-surgical treatment as the first-line therapy. An operation should be considered only if conservative treatment fails or in traumatic tears.

In view of the existing uncertainties regarding the best treatment, the aim of this systematic review was to compare the efficacy and safety of surgical and non-surgical interventions in persons with full-thickness rotator cuff tears.

Methods

The protocol of this systematic review was published in the PROSPERO registry (CRD42018100343) The reporting conformed with the Preferred Reporting of Systematic Reviews and Meta-Analyses Statement (PRISMA) (14).

We included patients with traumatic or atraumatic full-thickness rotator cuff tears confirmed by diagnostic imaging. Full-thickness tear was defined as complete separation of one or more tendons of a rotator cuff muscle. All surgical interventions were considered. Eligible comparison treatments included sham treatment, no treatment, watchful waiting, and any conservative measures such as physiotherapy and/or pharmacological treatment. Detailed information about the methods of the systematic review is provided in the eMethods and in eTable 1.

eTable 1. Search strategy in the Medline database.

Ovid MEDLINE(R) 1946 to May week 3 2018. Ovid MEDLINE(R) Epub Ahead of Print 25 May 2018. Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations 25 May 2018. Ovid MEDLINE(R) Daily Update 25 May 2018
	#	Searches	Results
A. Rotator cuff tear	1	rotator cuff injuries/	4586
	2	shoulder impingement syndrome/	1592
	3	((rotator cuff* or supraspinatus or infraspinatus or teres minor or subscapularis) adj3 (tear? or rupture* or injur* or disease or impingement)).ti.ab.	5864
	4	((shoulder or subacromial) adj impingement).ti.ab.	1105
	5	or/1–4	8711
B. Surgery	6	rotator cuff/su [surgery]	3134
	7	rotator cuff injuries/su	384
	8	Arthroscopy/	20 721
	9	(arthroscop* or acrom?oplast*).ti.ab.	26 810
	10	((rotator cuff* or supraspinatus or infraspinatus or teres minor or subscapularis or tendon?) adj3 (surg* or repair or fixation or refixation or suture? or reconstruct* or reinsertion or open)).ti.ab.	10 824
	11	or/6–10	40 277
A+B	12	5 and 11	4758
Humans only	13	exp animals/ not exp humans/	4 464 699
	14	12 not 13	4519
No editorials	15	editorial/	458 893
	16	14 not 15	4468

Results

Literature search

Figure 1 outlines the selection process for the studies identified by our literature search. Among the 7909 unique records screened, 155 were subjected to full-text screening. Of these, 10 studies (14 publications) were deemed eligible for inclusion (15– 28).

Figure 1.

PRISMA flow chart of study selection

PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Study characteristics

The characteristics of the three randomized (RCTs; six publications with 279 participants) (15– 20) and seven non-randomized controlled intervention studies (non-RCTs (eight publications with 654 participants) (21– 28) are shown in the Table. The studies were conducted in Europe, Asia, and North America and were published between 2000 and 2019. One RCT with an observation period of 10 years reported long-term outcomes (20).

Table. Study characteristics.

Reference	Study design	Country	Patient characteristics					Sample size		Longest follow-up (years)
			Definition of full-thickness tear	Mean age (SD). years		Mean symptom duration (SD). years		Surgical treatment	Non-surgical treatment
Kukkonen*1 2014/2015 (16, 17)	Multicenter RCT	Finland	Atraumatic (SSP. 100%)	65 (6)	64 (6)	2.3 (0.8)		Arthroscopy + acromioplasty 60	60	2
Lambers Heerspink 2015 (15)	Multicenter RCT	Netherlands	Atraumatic (SSP. 90%)	61 (7)	61 (7)	0.4–2.2	0.7–2.0	Mini-open technique + acromioplasty 25	31	1
Moosmayer*2 2010/2014/ 2019 (18– 20)	Single-center RCT	Norway	Traumatic/atraumatic (SSP. 74%) (≤ 3cm)	59 (8)	61 (8)	1.0 (1.6)	0.8 (0.8)	Mini-open technique +acromioplasty 52	51	10
Boorman 2014/18 (21, 22)	Prospective cohort study	Canada	Traumatic/atraumatic	58 (9)	61 (9)	2,7	2,2	Operation not precisely defined 23	70	5
De Carli 2017 (23)	Retrospective cohort study	Italy	Not specified	63 (4)	64 (3)	Not specified		Arthroscopy ± acromioplasty 20	18	1,5
Fabbri 2016 (24)	Retrospective cohort study	Italy	Not specified	50–68	55–71	Not specified		Arthroscopy ± acromioplasty 19	18	7
Lee 2016 (25)	Retrospective cohort study	Korea	Not specified					Arthroscopy ± acromioplasty 115*3	114	1
Vad 2002 (26)	Retrospective cohort study	USA	Not specified	59	63	0.1–1.4		Operation not precisely defined 36	40	3
Yamada 2000 (27)	Retrospective cohort study	Japan	Not specified	47–82	55–81	0.1–4.5	1.0–11	Operation not precisely defined 26	14	4
Yoo 2018 (28)	Retrospective cohort study	Korea	Not specified	64 (9)	65 (8)	3,4	2,5	Arthroscopy ± acromioplasty 104	33	1,2

*¹ For seven patients. both shoulders were included. with no indication of the distribution of the seven patients between the groups. Moreover. there were minor data discrepancies between the two publications regarding patient characteristics. The characteristics reported here were extracted from the 2015 publication.

*² There were minor discrepancies between the data reported in the three publications. The characteristics reported here were extracted from the 2014 publication.

*³ The study included both patients with partial-thickness tears patients with full-thickness tears. Only the patients with full-thickness tears were considered for this systematic review.

RCT. Randomized controlled trial; SD. standard deviation; SSP. supraspinatus

Patient characteristics

All included studies were restricted to adults. The mean age of the participants ranged between 58 and 70 years, and 12 to 78% were female.

Characteristics of the tears

Two RCTs explicitly excluded tears of traumatic origin

(15– 17). In the long-term RCT by Moosmayer et al., 57% of the participants presenting with a traumatic tear had experienced preceding episodes of symptoms (18– 20). Moreover, this study was limited to patients with small to moderate tears (≤ 3 cm). In contrast to the RCTs, only 1 non-RCT specified tear characteristics (21, 22).

Surgical intervention

The procedures performed in the studies were either shoulder arthroscopy (16, 17, 23– 25, 28) or a minimally invasive (mini-open) technique (15, 18– 20) with or without acromioplasty. The surgical intervention was followed by a period of sling or splint wearing with or without passive motion exercises. Three studies did not define the precise nature of the intervention.

Non-surgical intervention

Patients in the non-surgical group received up to 24 sessions of physiotherapy (16– 20). In their RCT, Lambers Heerspink et al. (15) additionally administered one to three corticosteroid injections and oral analgesics. Moosmayer et al. (18– 20), in contrast, allowed no analgesia. No other studies reported whether any co-interventions were used. The RCTs allowed non-surgically treated patients to cross over to the surgery group if the treatment outcome was unsatisfactory (secondary surgery).

Evidence from RCTs

Efficacy

Shoulder function

Shoulder function was measured with the 100-point Constant score at 1, 2, 5 and 10 years (Figure 2a and eTable 2a). All effect estimates showed (statistically) significant differences in favor of surgery compared with non-surgical treatment. At 12 months, the mean Constant score for shoulder function after surgery exceeded that after conservative treatment by 6.7 points (95% confidence interval [CI] [2.3; 11.0]; 257 patients; 85% atraumatic; three RCTs; moderate certainty of evidence). The largest effect, with a mean difference (MD) of 11.8 points, was observed at 10 years (95% CI [5.2; 18.4]; 91 patients; 43% atraumatic; 1 RCT; moderate certainty of evidence).

Figure 2a.

Shoulder function (Constant score. 0–100 points. 100 = best outcome). Forest plots (based on data from RCTs)

(A) Random sequence generation (selection bias)

(B) Allocation concealment (selection bias)

(C) Blinding of participants and personnel (performance bias)

(D) Blinding of outcome assessment (detection bias)

(E) Incomplete outcome data (attrition bias)

(F) Selective reporting (reporting bias)

(G) Other forms of bias not covered by A–F

* Studies (15– 20)

Chi². Chi-squared test; CI. confidence interval; fixed. fixed-effects model analysis; IV. inverse variance; I². Higgins and Thompson heterogeneity test; mean difference. difference between baseline and follow-up measurements; SD. standard deviation; +. low risk of bias; -. high risk of bias; ?. unclear risk of bias

Shoulder pain

The mean intensity of shoulder pain at 1 year, measured with the 10-cm visual analog scale (VAS), was 1.1 cm lower in surgically treated patients than in those who received non-surgical interventions (95% CI [0.5; 1.7]; 257 patients; 85% atraumatic; three RCTs; low certainty of evidence) (Figure 2b and eTable 2a). As for the endpoint shoulder function, the greatest effect in favor of surgically treated patients was observed at 10 years (MD 2.0 cm [1.0; 3.0]; 91 patients; 43% atraumatic; one RCT; low certainty of evidence).

Figure 2b.

Shoulder pain (visual analog scale. 0–10 cm. 0 cm = best outcome). NB: The algebraic sign of the mean differences were inverted so as to match the other scores. for which higher numbers show better outcomes (e.g.. the Constant score).

(A) Random sequence generation (selection bias)

(B) Allocation concealment (selection bias)

(C) Blinding of participants and personnel (performance bias)

(D) Blinding of outcome assessment (detection bias)

(E) Incomplete outcome data (attrition bias)

(F) Selective reporting (reporting bias)

(G) Other forms of bias not covered by A–F

* Studies (15– 20)

Chi². Chi-squared test; CI. confidence interval; fixed. fixed-effects model analysis; IV. inverse variance; I². Higgins and Thompson heterogeneity test; mean difference. difference between baseline and follow-up measurements; SD. standard deviation; +. low risk of bias; -. high risk of bias; ?. unclear risk of bias

Range of motion and muscle strength

For the endpoints range of motion and muscle strength (both measured with the Constant subscore), the group differences were small at every time of measurement (moderate certainty of evidence, Figure 2c and eTable 2a). Goniometer measurements (pain-free flexion) at 10 years showed that surgical treatment was more beneficial (MD 15.6° [0.8; 32]; 91 participants; 43% atraumatic; one RCT; moderate certainty of evidence).

Figure 2c.

Muscle strength (Constant score. 0–25 points. 25 = best outcome)

(A) Random sequence generation (selection bias)

(B) Allocation concealment (selection bias)

(C) Blinding of participants and personnel (performance bias)

(D) Blinding of outcome assessment (detection bias)

(E) Incomplete outcome data (attrition bias)

(F) Selective reporting (reporting bias)

(G) Other forms of bias not covered by A–F

* Studies (16– 20)

Chi². Chi-squared test; CI. confidence interval; fixed. fixed-effects model analysis; IV. inverse variance; I². Higgins and Thompson heterogeneity test; mean difference. difference between baseline and follow-up measurements; SD. standard deviation; +. low risk of bias; -. high risk of bias; ?. unclear risk of bias

Quality of life

Quality of life was reported in one RCT. Group differences, measured with the SF-36 score, were small (18– 20).

Adverse events

The risk of adverse events showed no significant difference between the groups (relative risk 1.4; 95% CI [0.5; 4.0]; 255 participants; 85% atraumatic; three RCTs; low certainty of evidence, eFigure and eTable 2a). According to the study authors, none of the reported adverse events were related to treatment.

eFigure.

Adverse events (longest follow-up)

(A) Random sequence generation (selection bias)

(B) Allocation concealment (selection bias)

(C) Blinding of participants and personnel (performance bias)

(D) Blinding of outcome assessment (detection bias)

(E) Incomplete outcome data (attrition bias)

(F) Selective reporting (reporting bias)

(G) Other forms of bias not covered by A–F

* Studies (15– 20)

Chi². Chi-squared test; CI. confidence interval; fixed. fixed-effects model analysis; I². Higgins and Thompson heterogeneity test; M-H. Mantel–Haenszel procedure; SD. standard deviation; ?+. low risk of bias; -. high risk of bias; ?. unclear risk of bias

Retear

In one RCT in which only atraumatic tears were considered, the retear rate at 1 year was 74% (15). Moosmayer et al. (19, 20) reported increasing numbers of full- or partial-thickness retears in the course of time: 21% at 2 years, 26% at 5 years, and 33% at 10 years.

Secondary surgery

Only the RCT by Moosmayer et al. (20) stated the number of patients who initially received non-surgical treatment but switched to the surgical group owing to lack of treatment success. Overall, 27% crossed over (12 patients within 2 years and two patients between 5 and 10 years after randomization) (18– 20).

Evidence from non-RCTs

Efficacy

Shoulder function was better after surgical than non-surgical interventions for up to 5 years following treatment, as measured with the 100–point Constant score (e.g., at 4.5 years: MD 7.6 points [1.2; 14]; 35 patients; etiology unknown; one non-RCT [24]; very low certainty of evidence). No meaningful group differences were observed for the other endpoints (eTable 2b).

eTable 2. GRADE evidence profile.

Certainty of evidence							Summary of findings
Number of patients (studies). follow-up	Risk of bias	Inconsistency	Indirectness	Imprecision	Other aspects	Certainty of evidence	Number of patients		RR (95% CI)	Baseline (control group)	MD (surgically treated)
							Control	Surgery
a: Randomized controlled trials. Study setting: Adults with full-thickness rotator cuff tears from Finland. Norway. Netherlands
Shoulder function (Constant score. 0–100 points. 100 points = best result)
257 (3). 1 year	Not serious	Not serious	Not serious	Serious*1	None	MODERATE	131	126	-	38.4–57.1	6.7 higher*2 (2.3 to 11.0 higher)
210 (2). 2 years	Not serious	Not serious	Not serious	Serious*1	None	MODERATE	105	105	-	38.4–57.1	4.4 higher*2 (0.1 to 8.8 higher)
101 (1). 5 years	Not serious	Not serious	Not serious	Serious*1	None	MODERATE	50	51	-	38,4	8.7 higher*2 (1.3 to 16.1 higher)
91 (1). 10 years	Not serious	Not serious	Not serious	Serious*1	None	MODERATE	48	43	-	38,4	11.8 higher*2 (5.2 to 18.4 higher)
Shoulder pain (VAS. 0–10 cm. 0 is best result)
257 (3). 1 year	Serious*3	Not serious	Not serious	Serious*1	None	LOW	131	126	-	2.7–6.3	1.1 lower*2 (0.5 to 1.7 lower)
210 (2). 2 years	Serious*3	Not serious	Not serious	Serious*1	None	LOW	105	105	-	2.7–5.3	0.9 lower*2 (0.3 to 1.5 lower)
101 (1). 5 years	Serious*3	Not serious	Not serious	Serious*1	None	LOW	50	51	-	5.3	1.3 lower*2 (0.5 to 2.1 lower)
91 (1). 10 years	Serious*3	Not serious	Not serious	Serious*1	None	LOW	48	43	-	5,3	2.0 lower*2 (1.0 to 3.0 lower)
Range of motion (Constant subscore. 0–40 points. 40 is best result)
110 (1). 1 year	Not serious	Not serious	Not serious	Serious*4	None	MODERATE	55	55	-	29,3	0.1 lower (0.5 lower to 0.3 higher)
109 (1). 2 years	Not serious	Not serious	Not serious	Serious*4	None	MODERATE	55	54	-	29.3	0.2 lower (0.5 lower to 0.2 higher)
Range of motion (Goniometer. pain-free flexion in degrees)
91 (1). 10 years	Not serious	Not serious	Not serious	Serious*5	None	MODERATE	48	43	-	81,9	15.6 higher*2 (0.8 to 32 higher)
Muscle strength (Constant subscore. 0–25 points. 25 is best result)
212 (2). 1 year	Not serious	Not serious	Not serious	Serious*4	None	MODERATE	106	106	-	8.1–8.4	0.4 lower (2.0 lower to 1.3 higher)
210 (2). 2 years	Not serious	Not serious	Not serious	Serious*4	None	MODERATE	105	105	-	8.1–8.4	0.1 higher (1.6 lower to 1.8 higher)
101 (1). 5 years	Not serious	Not serious	Not serious	Serious*4	None	MODERATE	50	51	-	8,1	1.3 higher (1.2 lower to 3.8 higher)
Adverse events: longest follow-up
255 (3)	Serious*3	Not serious	Not serious	Serious*4	None	LOW	5/130 (3.8%)	7/125 (5.6%)	1.4 [0.5; 4.0]	38 per 1000	17 more per 1000 (20 fewer to 127 more)
b: Non-randomized controlled intervention studies. Study setting: Adults with full-thickness rotator cuff tears from Canada. Italy. Korea. Japan. USA
Shoulder function (Constant score. 0–100 points. 100 is best result)
38 (1). 1.5 years	Very serious*6	Not serious	Not serious	Serious*7	None	VERY LOW	18	20	-	47	11.6 higher*2 (8.8 to 14.3 higher)
35 (1). 4.5 years	Very serious*6	Not serious	Not serious	Very serious*7	None	VERY LOW	18	17	-	Not reported	7.6 higher*2 (1.2 to 14.0 higher)
Shoulder pain (VAS. 0–10 cm. 0 is best result)
229 (1). 1 year	Very serious*6	Not serious	Not serious	Serious*4	None	VERY LOW	114	115	-	7,1	0.1 lower (0.5 lower to 0.3 higher)
137 (1). 0.8 years	Very serious*6	Not serious	Not serious	Very serious*8	None	VERY LOW	33	104	-	Not reported	0.5 lower (1.5 lower to 0.5 higher)
Range of motion (JOA subscore. 0–30 points. 30 is best result)
40 (1). 4 years	Very serious*6	Not serious	Not serious	Very serious*8	None	VERY LOW	14	26	-	18	2.7 higher (1.5 lower to 6.9 higher)

CI. confidence interval; JOA. Japanese Orthopedic Association; MD. mean difference considering the change between the baseline and follow-up measurement of each treatment arm; RR. risk ratio; VAS. visual analog scale

*¹ Insufficient precision (wide confidence interval). downgraded by one level: concerns regarding clinical relevance of results. the smallest clinically relevant difference for the endpoint shoulder pain (measured with the Constant score) is 10.4 points and for pain (measured with the VAS). ≥ 1.4 cm

*² Effect in favor of surgical treatment

*³ Risk of bias. downgraded by one level: concerns regarding lack of blinding. particularly for subjective endpoints (pain and adverse events)

*⁴ Insufficient precision (wide confidence interval). downgraded by one level: wide 95% CI consistent with both the possibility of benefit (improving symptoms) and the possibility of harm (worsening symptoms)

*⁵ Insufficient precision (very wide confidence interval). downgraded by one level

*⁶ Risk of bias. downgraded by two levels: serious confounding. selection bias. and lack of blinding

*⁷ Insufficient precision (wide confidence interval). downgraded by two levels: concerns regarding clinical relevance of results and low number of patients

*⁸ Insufficient precision (wide confidence interval). downgraded by two levels: wide 95% CI consistent with both the possibility of benefit (improving symptoms) and the possibility of harm (worsening symptoms). plus low number of patients (endpoint range of motion) or variable group size (endpoint pain)

Adverse events

The retear rate within 1.5 years of surgery was 10% (23). The proportion of non-surgically treated patients with tear progression ranged between 11% and 67% (23– 25).

Risk of bias

The nature of studies comparing surgical with non-surgical interventions generally precludes the blinding of participants. Consequently, it cannot be excluded that the observed group differences are, to an indeterminable extent, attributable to the influence of patients’ awareness of their treatment rather than to the effects of the interventions. Details of the risk of bias assessment are displayed in Figure 2a–c and the eFigure (RCTs) and in eTable 3 (non-RCTs).

Discussion

This is the first systematic review considering not only the best available evidence from RCTs—including recently published 10-year follow-up results (19, 20)—but also results from non-RCTs. Although our findings suggest that surgery may have a more favorable outcome than non-surgical treatment with regard to various endpoints, the measured differences are not always clinically relevant. For instance, the upper and lower boundaries (95% CI) of the MD for the endpoint shoulder function were 0.04 and 18.4 points. Applying the published value of 10.4 points (29) as minimal clinically important difference, the lower boundaries of the 95% CI were below this threshold at all follow-up times. As another example, the boundaries of the 95% CI for the mean reduction in pain were between 0.3 and 3.0 cm in favor of surgical intervention. Applying a minimal clinically important difference estimate of 1.4 cm (30), it again has to be assumed that this threshold was not attained in all patients.

Any surgical intervention carries a potential risk for complications such as infections, deep venous thromboses, pneumonia or peripheral nerve damage (31, 32). Given that the figures in our review are based exclusively on data from the primary studies, the rates may have been underestimated due to poor reporting or the unsystematic assessment of adverse events in the clinical studies. Therefore, the potential risks must be weighed against the available evidence of benefit. Moreover, data for other endpoints, such as range of motion, muscle strength, and quality of life, were very limited and the group differences slight.

Surgery is often carried out to prevent tear progression (33). A recently published review reported a progression rate of 40% in conservatively treated full-thickness tears after 4 years’ observation (34). In our systematic review, the proportions of patients with tear progression (categorized as present or absent), varied from 11% to almost 70%. The progression of tears in non-surgically treated patients needs to be balanced against the retear rate in patients treated surgically. This also differed widely, between 10% and 74%, in accordance with descriptions elsewhere of retear rates ranging from 0% to 94% (35). In the long term, the crossover rates to secondary surgery was below 30%, suggesting that the majority of people in the non-surgical group accepted their treatment results.

Although various outcomes were assessed in the pool of available studies, the potential for amalgamating data for statistical analysis was limited. The reported effects are often based on the results of single studies, e.g., long-term data for shoulder function or pain, or data on quality of life, and are therefore associated with imprecision, i.e., wide confidence intervals. Particularly in non-RCTs, the quality of reporting was poor and often inconsistent. For example, the individual components of the non-surgical treatment were usually not specified, and the additional use of oral pain medications or steroid injections was insufficiently described (21, 22, 24, 26, 28). Acromioplasty was applied widely, but the surgical interventions did not adhere to any uniformly discernible standards. Moreover, the comparability of the results across the studies is limited by differences in tear characteristics or inadequate description. For instance, Kukkonen et al. (16, 17) included only isolated supraspinatus tears, whereas the other RCTs also included infraspinatus and subscapularis tears, albeit in small numbers. Massive rotator cuff tears, another prognostic factor (36), were also not (adequately) considered in the available study pool.

Conclusions

With regard to the endpoints shoulder function and shoulder pain, surgery (performed arthroscopically or by means of a minimally invasive technique, including acromioplasty) was superior to non-surgical treatment (also in the long term) for the management of full-thickness rotator cuff tears. Whether the differences between the groups are clinically relevant for the individual patient, however, is uncertain, as the measured results were distributed below and above the clinical relevance threshold. For other outcomes such as range of motion, muscle strength, quality of life, and adverse events, only very limited data were available and the group differences were minimal.

In many studies the rotator cuff defect was inadequately characterized, and subgroup analyses for different surgical techniques or conservative treatments were impossible because the results often came from individual studies and the data were thus limited.

Further research into full-thickness rotator cuff tears should consider the following:

• Presentation of the results with regard to thresholds of clinical relevance

• Evaluation of various treatment options for subgroups, considering prognostic factors such as tear size and etiology, number of tendons involved, and symptom duration

• Establishment of the optimal treatment procedures for young persons, particularly as we identified no studies examining this population

• Stringent adherence to guidelines that serve to improve the reporting of clinical studies, such as the CONSORT (Consolidated Standards of Reporting Trials) statement (37) and/or the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement (38)

Supplementary Material

eMethods

Patient population

Included were children, adolescents, and/or adults with traumatic or atraumatic full-thickness rotator cuff tears treated in the framework of randomized (RCTs) and non-randomized controlled intervention trials (non-RCTs). We defined a rotator cuff tear as the complete rupture of one or more tendons attached to one of the rotator cuff muscles, irrespective of tear etiology, time of diagnosis, age, and any existing comorbidities. The diagnosis had to have been confirmed by diagnostic imaging.

Intervention

Any surgical intervention was considered. Studies in which the surgical procedure was supplemented by a biological intervention (e.g., platelet-rich plasma, growth factors, stem-cell therapy, or other cell-based therapies) were excluded. Shoulder joint replacement was also not considered as an intervention.

Comparison treatment

Eligible comparison treatments included sham treatment, no treatment, watchful waiting, or any conservative form of treatment such as physiotherapy or pharmacological treatment, including steroid injections. Studies comparing different surgical interventions were excluded.

Endpoints

We considered the following endpoints: shoulder function, pain, range of motion, muscle strength, quality of life, any adverse events (e1– e3), retear of the repaired tendon, and secondary surgery, i.e., the option for non-surgically treated patients with an unsatisfactory outcome to switch to the surgery group. To capture time-dependent variability, all available numerical data were extracted for endpoints were measured at multiple time points.

Literature search

The literature searches were conducted according to the PRESS (Peer Review of Electronic Search Strategies) criteria (e4) by a search specialist (IK) (time of original search: May 2018). Additionally, a search alerts for the Medline database was set up to receive notifications of any relevant new studies (last update: August 2019).

The following electronic databases were searched:

The search strategies for the Medline database can be seen in eTable 1. Alongside the electronic database searches, we conducted a hand search of the reference lists of relevant studies and systematic reviews to identify additional publications. Moreover, further potentially relevant studies were sought with the aid of the PubMed similar articles function (e5) and cited reference searching using the Web of Science Core Collection. No restrictions were imposed on time or language of publication.

Study selection process

The titles and abstracts of the citations identified by the searches were initially screened to determine which records could be classified as definitely relevant in conformity with the predefined inclusion and exclusion criteria. The full texts of all potentially relevant articles were scrutinized before a final decision on inclusion or exclusion (full-text screening). The full texts were also independently checked for eligibility by 2 reviewers (CB, VT, AD or BNS.) The complete selection process was carried out by two independent reviewers (CB, VT, or AD).

Data extraction, risk of bias assessment, GRADE assessment

Data extraction from published studies was undertaken by one author (CB or CS) and checked by a second author (VT or BNS). The following information was extracted: details of patient characteristics including the tear, details of the intervention and the comparison treatment, and data on the reported endpoints including method of measurement.

The risk of bias in RCTs was judged separately for each endpoint according to the methods defined in the Cochrane Handbook for Systematic Reviews of Interventions for RCTs (e6). For non-RCTs, assessment of the risk of bias ensued with the ‘Risk of Bias in Non-randomised Studies of Interventions’ (ROBINS-I) instrument (e7). We took account of prognostic factors (confounders) that may influence intervention assignment, such as age, tear size, number of tendons involved, fatty degeneration, tear etiology, duration of symptoms, diabetes, extent of disability, shoulder range of motion, shoulder strength, level of activity, tendon retraction, bone mineral density, patient preference, obesity, and psychological/psychosocial factors (e8, e9). Alongside evaluation of the risk of bias, the certainty of evidence of selected outcomes was assessed using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) method and presented in the form of an evidence profile (e10– e12).

Statistical analyses

The treatment effect for each continuous endpoint, measured on a continuous scale, was expressed as the mean difference (with 95% confidence interval [CI]) of the average change between the baseline and follow-up measurements. The effect estimate for dichotomous endpoints, i.e., events that do or do not occur for a patient, was expressed as the relative risk (RR) with 95% CI. The results of the individual studies were summarized quantitatively for each endpoint in the form of a fixed-effect model meta-analysis—provided results were available from at least two studies. The fixed-effect model was used owing to the small number of studies. The unit of analysis was the individual patient. Data synthesis and graphic presentation of the result (forest plots) were achieved with the aid of the software Review Manager (RevMan; version 5.3) (e13).

All data were analyzed on an intention-to-treat basis and/or according to recently published recommendations for addressing missing data (e14, e15). In cases where missing data could not be reliably replaced (e.g., for non-attenders), we referred to the number of patients for whom data from the corresponding time point were reported.

Heterogeneity of the study results was evaluated and quantified on the basis of I² and the statistical chi-squared test. A chi-squared p value < 0.10 or an I² ≥ 75% was considered as relevant heterogeneity (e16).

Subgroup analyses were planned to examine whether estimates could have been affected by potential clinical factors (effect modifiers). The aim would have been to uncover differences between patient groups and treatment characteristics. However, the limited available study data precluded such analyses.

• Medline

• Medline Daily Update

• Medline In-Process & Other Non-Indexed Citations

• Medline Epub Ahead of Print (via Ovid)

• Science Citation Index Expanded

• Conference Proceedings Citation Index-Science

• BIOSIS Citation Index (via Web of Science)

• Cochrane Library (via Wiley)

• Embase (via Embase.com/Elsevier)

• SPORTDiscus (via Ebsco)

eTable 3. Risk of bias in non-randomized studies.

Study	Bias due to confounding*1	Selection bias in group allocation	Bias due to classification of intervention*2	Bias due to deviation in intervention phase	Bias due to missing data	Bias in outcome assessment	Bias due to selective reporting of outcomes	Overall assessment
Boorman 2014/18	Critical Treatment assignment after 3 months' conservative treatment: responders (non-surgical treatment) and non-responders (surgical treatment) – no adjustment	Low 93 patients included prospectively	Low	No information	Serious Proportion of missing data: 32%	Moderate to serious Outcome assessor and surgeon blinded. patients not blinded	Unclear	CRITICAL
De Carli 2017	Critical Treatment assignment according to contraindications (to surgery) or patients' preferences – no adjustment	Low 40 consecutive patients included retrospectively*3	Low	No information	Low to moderate Proportion of missing data: 2%	Serious No blinding described	Unclear	CRITICAL
Fabbri 2016	Critical Treatment assignment depended on clinical symptoms – no adjustment	Serious 63 patients included retrospectively*3	Low	No information	Serious Proportion of missing data: 11–18% (reasons not transparently described)	Moderate to serious Outcome assessor blinded. patients not blinded	Unclear	CRITICAL
Lee 2016	Serious Approaches to controlling for predefined prognostic factors were described. but few of these factors were actually addressed adequately – no adjustment	Serious Patients from three centers included retrospectively (inclusion based on sex. age. side of tear. duration of symptoms. and etiology)*3	Low	No information	Serious Missing data: '[...] sample size was too small due to missing data [...]'	Critical No blinding described; telephone interview to assess pain and range of motion (including patients who were lost to follow-up)	Unclear	SERIOUS
Vad 2002	Critical Treatment assignment after ≥ 6 months' conservative treatment; assignment according to patients' preferences – no adjustment	Serious 78 patients included retrospectively*3	Low	No information	Low All patients included were analyzed	Serious No blinding described	Unclear	CRITICAL
Yamada 2000	Critical Treatment assignment according to patients' preferences – no adjustment	Serious Patients from two hospitals included retrospectively: treated non-surgically (1979–99); treated surgically (1982–97)*3	Low	No information	Low All patients included were analyzed	Serious No blinding described	Unclear	CRITICAL
Yoo 2018	Critical Treatment assignment after ≥ 3 months' conservative treatment; assignment according to patients' preferences – no adjustment	Low 137 consecutive patients included retrospectively*3	Low	No information	Low All patients included were analyzed	Critical No blinding described; endpoints evaluated by telephone interview	Unclear	CRITICAL

*¹ Confounders. prognostic factors: age. tear size. number of tendons involved. fatty degeneration. etiology of tear. duration of symptoms. diabetes. extent of disability. shoulder range of motion. shoulder strength. level of activity. tendon retraction. bone mineral density. patient preference. obesity. and psychological/psychosocial factors. For more details see eMethods

*² Due to the nature of the comparison groups (surgical and non-surgical treatment). misclassification of the intervention can be excluded.

*³ Retrospective studies yielded insufficient information on the intake of additional medications. such as pain relievers. or on cointerventions and whether these were balanced across the groups.