The most common cause of shoulder pain is rotator cuff pathology [1]. Population studies estimate rotator cuff tears to be present in 20%–22% of individuals in the adult population [2]. There are several factors that influence the outcome of rotator cuff repair (RCR), including age, tear size, muscle quality, sex, and fatty degeneration [3-6]. Tendon integrity is correlated with better strength and function, highlighting the importance of factors that may affect healing [4,7].

Despite the widespread use of non-steroidal anti-inflammatory drugs (NSAIDs) following RCR, there are few clinical studies assessing their safety, and their use remains highly controversial [8-12]. Expert opinion is split; when polled, only 51% of surgeons in the American Shoulder and Elbow Surgeons and the Association of Clinical Elbow and Shoulder Surgeons allow patients to take NSAIDs in the first four weeks after surgery [13].

NSAIDs are an effective analgesia following RCR, reduce opioid consumption [14], and are used as part of multimodal analgesia [15]. However, the effect of NSAIDs on tendon-to-bone healing has been investigated in a number of animal models, and reduced strength at the healing site has been demonstrated repeatedly [16-20]. The results of human studies investigating the topic vary, and consequently recommendations have been mixed varying from no cyclooxygenase-2 (COX-2) use [8,21], safe for short durations [22], use without specific restrictions [10,23], or inconclusive evidence [11,12,24]. Large trials analysing NSAID use with patient reported outcome measures (PROMs) are lacking, as is analysis of outcomes based on duration of NSAID use [11,21,24,25].

We compared those taking NSAIDs postoperatively with those who did not, as well as investigating the relationship between duration of NSAID use and outcome, with follow-up to a minimum of 5 years postoperatively. Primary outcomes were functional PROMs and pain scores, and secondary outcomes were self-reported complications such as retear rate, infection and stiffness. We hypothesise that longer NSAID use will be associated with poorer PROMs and higher retear rates.

The study complied with the requirements of the National Ethics Committee, and approval was granted (No. NTX/07/04/034/AM04), and written informed consent was obtained from all participants.

Prospective data collection began at recruitment, with consent at the time of surgical booking. Ninety-two surgeons from a mixture of small and large centres participated from March 1, 2009, to December 31, 2010. Arthroscopic, mini-open and open approaches were used per surgeon preference, as was the postoperative rehabilitation protocol, either surgeon-directed or through formal physiotherapy. Inclusion criteria were completion of an operative day questionnaire survey, written informed consent and postoperative follow-up. NSAID use, smoking status, ethnicity, sex, work and recreational activities, age and whether the tear was related to trauma were among patient factors recorded. Categories of postoperative NSAID use were 0–2 weeks, 3–6 weeks, 7 weeks to 3 months or over 3 months. This was not further categorised into COX nonselective vs. COX-2 inhibition, nor daily dose and frequency of NSAIDs use during that time. A validated functional assessment score that is shoulder specific, the Flex Shoulder Function (Flex SF) was used, and is highly rated compared to other shoulder scores [26,27]. Better function is reflected by a better score, with a maximum of 50. Pain was assessed with a visual analog scale (VAS) from 1 to 10 over the previous month, comprising four questions.

The primary operating surgeon completed a standardised questionnaire on the day of surgery. Technical data were recorded including presence of a labral tear, and surgical approach. Using comparison to an instrument of known size, such as a probe, tear size was recorded in both the anteroposterior dimension and retraction. Tear area was the tear size in the anteroposterior dimension multiplied by the retraction. Tendon quality was graded as very good (normal thickness), good (some deterioration), thin or poor.

Flex SF and VAS pain scores were included in follow-up questionnaires completed at 6 months, and thereafter at 1, 2 and a minimum of 5 years postoperatively. The primary reason patients were lost to follow-up was change to contact details over the study duration.

A total of 2,533 patients participated in the study; 72% of these patients took anti-inflammatories postoperatively and 28% did not. With regard to duration, 18% took NSAIDs for up to 2 weeks, 21% for 3–6 weeks, 12% for 7 weeks to 3 months and 21% for over 3 months. Follow-up at a minimum of 5 years was 81% for the cohort. The average age of study participants was 57 years and 71% were male; 12% were smokers and 91% were European. Surgical approach was arthroscopic in 17.9%, mini-open in 38.5%, and open in 43.6%.

At baseline there was no functional difference between the two groups. From 6 months postoperative through 5 years postoperative, function remained significantly higher in the group who did not take NSAIDs (P<0.001) (Fig. 1); however, there was no difference in functional improvement between the NSAID and non-NSAID groups between baseline and 5 years. When analysed by duration of NSAID use, at baseline those that would go on to use NSAIDs for 7 weeks or longer had lower function than those that would not. Throughout the remainder of follow-up, those using NSAIDs for longer than 7 weeks had poorer Flex SF scores (Fig. 2). Those using NSAIDs for a duration of 3–6 weeks had poorer function at the 6- and 12-month follow-up visits. Those using NSAIDs for longer than 3 months had less improvement in Flex SF between baseline and 5 years.

Pain was significantly higher preoperatively in the group taking NSAIDs postoperatively compared to those who did not (4.8 vs. 4.5, P=0.02). Pain was significantly higher in this group 6–24 months postoperative (P<0.001), although no difference was seen at 5 years postoperative (Fig. 3). Preoperatively, those that would go on to use NSAIDs >7 weeks had worse pain (P<0.05). This group experienced more pain for 12 months postoperatively; however, by 24 months, and also at 5 years, only those using NSAIDs for >3 months reported higher pain than the non-NSAID group. Also, at 5 years, those using NSAIDs <2 weeks had better pain scores than the non-NSAID group (Fig. 4). There was no difference in improvement in pain between baseline and 5 years based on duration of NSAID use.

The rate of stiffness or frozen shoulder symptoms was lower in the non-NSAID group compared to those taking NSAIDs (20.4% vs. 30.1%, P<0.001). Infection rate was nearly doubled in the NSAID group compared to the non-NSAID group (2.6% vs. 1.4%, P=0.13); however, this was not statistically significant. Retear rate was not different between the two groups. The NSAID group was younger than the group that did not take NSAIDs (56.7 vs. 58.3, P<0.001).

Intraoperative tear size was recorded as well as number of tendons involved, and there was no difference in tear area, anteroposterior tear size, or tear retraction between patients that would go on to use NSAIDs and those that did not (Table 1). To assess the impact of potential confounding, NSAID use was correlated with other factors recorded (Table 2). None were significantly associated with NSAID use (P<0.05). Improvement in Flex SF was not associated with age (P=0.30, Pearson R statistic 0.032)

Our results demonstrated that from a similar baseline, those taking NSAIDs had poorer functional outcomes out to 5 years postoperatively, though no difference in improvement from baseline to 5 years. The use of NSAIDs for longer durations, notably over 6 weeks, was associated with decreased shoulder function throughout follow-up, and reduced improvement in function if taken for more than 3 months.

Although there is a paucity of human data on the subject, there are similar findings across animal experiments. These include decreased load to failure and impaired collagen production in a NSAID-treated (COX 1 and 2, except for ibuprofen) rat patella tendon-to-bone healing model [20]. Poorer enthesis repair in several animal models has been found [28], as well as inconsistent fibrocartilaginous regrowth in a rat supraspinatus model [16]. Mechanisms of these effects are thought to be through COX-2 decreasing prostaglandin synthesis critical for angiogenesis [29], with the blood supply for healing coming chiefly from the bone [30], as well as impaired bone response [28].

We did not find an increased rate of self-reported retear in the NSAID group compared with the non-NSAID group (5.6% vs. 6.5% respectively, P=0.86). This conflicts with the findings of Oh et al. [8], who performed a randomised controlled trial of 180 patients and found increased risk of re-rupture following RCR in those taking the COX-2 inhibitor celecoxib. In that study, 82 patients had magnetic resonance imaging (MRI) or ultrasound postoperatively, and re-rupture rate was 37% in the celecoxib group compared to 7% in the ibuprofen group and 4% in the tramadol group (P=0.009) [8]. Burns et al. [21] randomised 40 patients to celecoxib or placebo following RCR and found only a 50% healing rate in the celecoxib group compared to a 70% healing in the placebo group on 1 year postoperative MRI (P=0.35) [21]. We hypothesised that the lack of routine reimaging performed in our study led to retears being underdiagnosed. It was impractical to perform routine reimaging given the large number of patients and observational nature of this study. Regarding age, it is not surprising that there was an age difference (58.5 vs. 56.7 in the non-NSAID and NSAID groups respectively), given the potential gastrointestinal and renal side effects that increase with age. However, age did not influence improvement in shoulder function, hence the rationale to not perform age-adjustment in our analysis, nor to adjust for other factors not significantly associated with NSAID use.

The largest study to investigate the effect of NSAIDs on the outcome of RCR to date was a retrospective review by Kraus et al. [23] with 463 patients undergoing arthroscopic RCR by three surgeons. They were categorised into two groups, those who had taken ibuprofen/NSAIDs after surgery (182 patients) and those who had not (281 patients). They found no difference between these groups in terms of improvement or final VAS and PROM scores. Possible explanations include the difference in sample size, the use of multiple approaches in our study compared to arthroscopic only, the number of surgeons involved, and the lack of quantitative analysis by NSAID.

Regarding the investigation of duration of NSAID use and outcome, this question has not been well addressed in the literature, despite widespread NSAID use. In a recent meta-analysis of RCTs investigating the effects of NSAID use on clinical outcomes in RCR [25], a total of 298 patients were randomised to NSAID use and 209 to control. In none of the seven studies within this meta-analysis were NSAIDs used for more than 3 weeks, and given the size of the individual studies they may not have had adequate power to detect differences in retear rates or clinical results; in addition, the average follow-up period was one year postoperatively.

This is the largest prospective cohort trial of RCR to date, and the only nationwide study. There was a high follow-up rate of 81%. We enrolled a large cohort of shoulder surgeons representing population-wide skills, versus just shoulder specialists, suggesting good applicability to a wider orthopaedic community.

The type of NSAID used, including whether non-selective or COX-2 specific, was not recorded, nor was drug dosage or compliance. These were observational data, and while the associations do not prove cause and effect, the findings are consistent with other studies addressing the effects of NSAIDs on the outcomes of RCR. Routine follow-up imaging would have demonstrated whether healing rates differed based on NSAID use, but this was impractical with such a large cohort. As our complication data were self-reported, results should be interpreted with caution. Patient self-reporting is likely to capture most clinically significant retears that have been imaged but will miss silent re-tears in patients who have not been reimaged. There was no separation into superficial or deep infection. Stiffness or frozen shoulder symptoms were also self-reported and could have occurred at any timepoint postoperatively; self-reporting is generally less accurate than if clinician diagnosed.

This study has found an association between extended NSAID use and poorer shoulder function following RCR. Based on these data, we would suggest exercising caution in extended NSAID use postoperatively, in particular for longer than 6 weeks. However, there is potential for indication bias, with those having greater postoperative pain using NSAIDs longer. Collected data did not establish dosage or type of NSAIDs used. Pain outcomes were not consistently worse long-term with NSAID use, and improvement from baseline to 5 years postoperatively was similar for pain and function. However, NSAID use longer than 3 months was associated with reduced functional improvement. Further randomised studies are required to determine the safe duration of NSAID use postoperatively following RCR. While this is observational data, it is supported by experimental findings and increasing clinical data.