The effect of NSAIDs on postfracture bone healing: a meta-analysis of randomized controlled trials

Objectives: To determine whether nonsteroidal anti-inflammatory drugs (NSAIDs) have an adverse effect on bone healing by evaluating all available human randomized controlled trials (RCTs) on this subject. Data Sources: A comprehensive search of electronic databases (PubMed, MEDLINE, and Cross-References) until October 2018 comparing the occurrence of nonunion in patients who received NSAIDs to the control group through RCTs. Study Selection: Inclusion criteria were English-only studies, and the type of studies was restricted to RCTs. Data Extraction: Two authors independently extracted data from the selected studies, and the data collected were compared to verify agreement. Data Synthesis: Nonunion was the main outcome evaluated in each study. Regression analysis was used to estimate the relative risk comparing the duration and the type of NSAIDs by calculating the odds ratio (OR) for dichotomous variables. Studies were weighed by the inverse of the variance of the outcome, and a fixed-effects model was used for all analyses. Conclusions: Six RCTs (609 patients) were included. The risk of nonunion was higher in the patients who were given NSAIDs after the fracture with an OR of 3.47. However, once the studies were categorized into the duration of treatment with NSAIDs, those who received NSAIDs for a short period (<2 weeks) did not show any significant risk of nonunion compared to those who received NSAIDs for a long period (>4 weeks). Indomethacin was associated with a significant higher nonunion rate and OR ranging from 1.66 to 9.03 compared with other NSAIDs that did not show a significant nonunion risk.


Introduction
Tissue damage triggers the processes of coagulation, inflammation, and healing. The healing process for soft tissue around bone is defined by regeneration and repair that matures into scar tissue, whereas the bone tissue itself has a unique ability to regenerate shape, strength, and preinjury function. Fracture healing is a combination of a complex and sequential set of events that depends on the stability of the fracture. [1,2] There are many factors that influence the healing process, including environmental factors, drugs, and physiological changes. One of these factors is the use of NSAIDs, which is commonly used in patients as an anti-inflammatory as well as an analgesic. NSAIDs effect on bone healing has been studied for many years, with results showing debatable effects on bone healing.
There have been multiple studies published evaluating the benefits and consequences of NSAIDs in bone healing of both animal and human tissue in vitro and in vivo. The studies that showed that NSAIDs have a significant effect on bone healing are categorized by mechanism of action of NSAIDs in Table 1.  There have been multiple studies published evaluating the benefits and consequences of NSAIDs in bone healing of both animal and human tissue to clinical RCTs. Subsequently, there are animal studies that showed NSAIDs do not have a significant effect on bone healing, which are listed in Table 2. [35][36][37][38][39][40][41][42][43][44][45][46][47][48][49] In conjunction to several RCTs, there has been meta-analysis of strong animal studies, yet the effect of NSAIDs on bone healing is controversial and many surgeons avoid these medications because of the concern of the possible delay in healing. [13,[50][51][52] The purpose of a meta-analysis of only randomized control human trials is to provide a greater understanding of the true effect of NSAIDs on human bone healing, and the risk of nonunion. Secondary outcome measures, duration of NSAIDs use, and type of NSAIDs will also be included to determine its effect on healing complications.

Source of Funding: nil.
We disclose that the authors of this study have no commercial or financial association with regards to this project and, therefore, have no affiliation that may pose a conflict of interest with the manuscript enclosed. We had full access to all the data in the study and take responsibility for the integrity and accuracy of the data, as well as the decision to submit for publication.
There were no external funding sources for this study.

Material and methods
This meta-analysis was conducted following the Preferred Reporting Items for Systemic Reviews and Meta-Analyses Statement. The study was deemed exempt from Institutional Review Board and Animal Use Committee Review. Informed consent was not applicable in our study. A comprehensive metaanalysis of the available clinical evidence was performed, with regards to NSAIDs exposure and nonunion risk in humans. Studies were excluded if long bone fractures were not studied, such as dental and spine.
A literature search was conducted through PubMed and Medline databases and restricted to English literature. The search terms used were "anti-inflammatory agents," "fracture," "healing," with expanded search terms with Boolean operators being used. The period of search included any studies until October 2018. Manual searching of studies was also done through review articles and other relevant material, through the snowballing technique. Gender and age group were not limited. Neither the type of NSAIDs nor the duration postoperatively was used as a restriction to the study. The type of study was restricted to human RCTs only. The summary of the study flow is provided in Figure 1.
Two reviewers reviewed each title independently and only relevant abstracts were included after the screen, which included 10 studies. From these studies, 3 were excluded. One was a study on regenerative periodontal treatment for which the healing process is different from bone healing and was, therefore, excluded.
The data extraction from each study includes year of publication, randomization method, patient, and treatment characteristics. The studies were evaluated specifically for publication bias using a funnel plot. RevMan software (version 5.3, The Cochrane Collaboration) was used for the analysis. Treatment effects were estimated by calculating the OR with Table 1 Animal studies which showed effect on bone healing with the use of NSAIDs.  Animal studies which showed no effect on bone healing with the use of NSAIDs.
Farii et al OTA International (2021) e092 www.otainternational.org 95% confidence interval (CI) for dichotomous variables. Studies were weighed by the inverse of the variance of the outcome, and a fixed-effects model was used for all analyses.

Results
A total of 384 studies were initially screened, where 10 studies were considered for inclusion. From those, 2 were excluded because they discussed pain management post spinal fusions, and a third study on healing of enamel matrix, was excluded because the healing process differs from long bones. Additionally, the study that included the results from Study to Prospectively Evaluate Reamed Intramedullary Nails in Patients with Tibial Fractures trial described qualitative results; however, it did not provide quantitative information on their conclusions on NSAIDs use for bone healing. A total of 6 randomized control studies were considered for final analysis (Table 3). [53][54][55][56][57][58] The data from these 6 studies were extracted for analysis.

Study characteristics and quality
From all of the studies that were included, there were 609 patients included where 290 were exposed to an NSAID compared to 319 control patients. Included in these studies are Adolphson et al and Drendel et al, which had no events of nonunion in either arms of the studies. [53,57] Based on the Cochrane handbook for metaanalysis, the standard practice for the calculations of OR and risk ratio is to exclude the studies where there are no events in the treatment group and in the control. [59] In the calculation table, it can be seen that the studies are included; however, they do not have weight attributed to the total analysis on the 6 studies, as described in Figure 2. From these studies, 3 considered long bone  [55] 112 Prophylaxis post acetabular fracture ORIF + long bone fracture  [58] 98 Acute acetabular fracture treated operatively  fractures, Colles fracture, [53,56] or simple arm fractures (radius, ulna, or humerus), 2 studies considered acetabular fractures, [55,58] and 1 studied elective hallux valgus surgery. [54] Each study was an RCT comparing NSAIDs with a control. Adolphson et al studied piroxicam 20 mg per day, Brattwall et al studied COX-2 inhibitors etoricoxib (120 mg for 4 days then 90 mg for 3 days) or tramadol (200 mg for 7 days), [53,54] Davis and Ackroyd [56] studied flurbiprofen 150 to 300 mg daily for 2 weeks, Drendel et al [57] studied ibuprofen, whereas Burd et al and Sagi et al studied indomethacin from 3 days to 6 weeks. [55,58] The diagnosis of nonunion was made through either X-ray of the fracture and CT scan or telephone follow-up and file review. From the studies included, the ORs were well distributed in terms of publication bias (Fig. 3).
A power analysis on the selected studies and the subgroups was conducted to understand the limitation from having a limited number of cases per study and its significance. The power refers to the probability that the test will find a statistically significant difference when such a difference exists. To calculate this, a type I error level (alpha) of 0.05 was used. A type II error (beta) of 0.2 was used that corresponds to a power of 0.8. Based on a high heterogeneity group, with a minimum power of 0.8, ideally it would need 5 RCTs of 50 patients in each group, 10 RCTs of 30 patients in each group, or 15 RCTs of 20 patients in each group.
In the case of this meta-analysis, there are 4 RCTs with an average of 43 patients per group. Using these values, with an effect size of 0.5, this is considered to have a power of 100%, which means for large differences, this is a significant metaanalysis. However, with the effect size of 0.2, the power is 74.6%, which is less than 80%, which means for small differences, this number of studies is not sufficient to find significance.

General effect of NSAIDs on bone healing and formation of nonunion
From those studies, there were 48 nonunion (36 for NSAIDs vs 12 for non) , defined as a bone that fails to heal, of which 36 were exposed to NSAIDs and 12 were not. The 4 studies resulted in an OR of 3.47, with a 95% CI of 1.68 to 7.13. The significance level is P = .001, as described in Figure 2. This indicates that there is an increased risk of nonunion with NSAIDs exposure which is significant.

Effect of NSAIDs for short duration compared to long duration on bone healing
A subgroup analysis was performed classifying the duration of NSAIDs administration, with a short duration defined as less than 2 weeks and a long duration defined at more than 2 weeks. Short duration studies included 2 studies, [54,57,58] with the OR of 1.56, 95% CI between 0.48 and 5.10, and P value of 0.484, as described in Figure 4, whereas the 4 studies included in the long duration, including 1 study, Sagi et al, which quantified both short and long duration, [53,55,56,58] , had an OR of 5.27, 95% CI between 2.34 and 11.88 and the P value of less than .0001, as described in Figure 5. This indicates that although short use of NSAIDs has no significant effect on bone healing, long duration of more than 2 weeks has a significant higher rate of nonunions.

Effect of indomethacin NSAIDs compared to other NSAIDs on bone healing
A subgroup analysis was performed classifying the NSAIDs with the exception of indomethacin. From the studies, 4 studies administered NSAIDs that were not indomethacin. [53,54,56,57] The OR was found to be 2.58 with 95% CI 0.65 to 10.30, the P value of .18, therefore, it is not significant, as described in Figure 6. This means that NSAIDs that did not include indomethacin did not have a significant effect on nonunion, compared to studies that included only indomethacin as NSAIDs, [55,58] had an OR of 3.87 with 95% CI 1.66 to 9.03, and a P value of .002, as described in Figure 7. This means that indomethacin used as an NSAID has a significant effect on healing leading to nonunions.   NSAIDs are generally classified by chemical structure or mechanism of action. Some of the commonly used classes are salicylates, propionic acid derivatives, acetic acid derivates, enolic acid derivatives, and selective COX-2 inhibitors. It can be understood that there is a difference in effect on bone healing with the NSAIDs due to the variation of the mechanism of action of NSAIDs.
There have been about 50 animal studies that were published on the effect of NSAIDs showing a variety of results. These studies are described in Tables 1 and 2. The animal studies included NSAIDs from different classes of NSAIDs. Animal studies investigated various NSAIDs during short-term and longterm use while looking at end point factors including mechanical properties of bone, histological grade, and radiological appearance. From these studies, most had greater support that NSAIDs delay or prevent bone healing, whereas others denied that effect. However, the question remains if this is true in humans. A thorough meta-analysis of human studies done by Wheatly et al, [61] which included pediatric and adult bone healing and was not limited to RCT studies, showed that NSAIDs exposure increased delayed union or nonunion. However, there is no significance in the pediatric population, or with low-dose or short-duration use.
The purpose of this meta-analysis was to analyze RCTs that studied the use of NSAIDs in bone healing. This meta-analysis of 6 RCTs evaluated 609 patients, of whom 290 were exposed to an NSAID compared to 310 control patients. From these studies, subcategories were further studied. When comparing all the included studies, there was a significant increased risk of nonunions when exposed to NSAIDs with an OR of 3.47. Adolphson et al studied piroxicam for 8 weeks and found that none of the patients had nonunions, similar to control. Similarly, Drendel et al studied ibuprofen compared to acetaminophen, where both groups did not have any nonunions; however, the main purpose of the study was not to investigate bone healing, and, therefore, there was no strong evidence. Since these studies have zero events, the OR cannot be statistically calculated. Brattwall et al [54] conducted an RCT that studied COX-2 inhibitors for a short time and found no significant difference of nonunions compared to control. Burd et al [55] conducted an RCT that studied indomethacin for long duration and found a significant formation of nonunion. In contrast, Davis et al studied flurbiprofen for long duration and found no significant difference in bone healing. [56] Lastly, Sagi et al [58] also studied indomethacin for both short and long durations, finding that short duration did not show a significant difference, whereas long duration did.
These studies were subcategorized to have a better understanding of the type and duration of NSAIDs that had an influence on bone healing. After classifying the duration of the NSAIDs, it was found that short duration (less than 2 weeks) had an OR of 1.56 and was not significantly different, whereas long duration NSAIDs use had an OR of 5.27 and was significantly different compared to control. This means that short use of NSAIDs as a means of pain management might not have a significant effect on risk of nonunion formation. In contrast, long duration does have a significant increased risk of nonunion. This is an important factor that can influence the clinical management of patients.
A second subgroup that was studied was the type of NSAIDs used in the RCTs. Indomethacin, an acetic acid derivative, is a nonselective COX inhibitor. The studies that did not include indomethacin had an OR for nonunion of 2.58, which was not significant, whereas the studies that included only indomethacin had an OR of 3.87, which was significant. This may be a good indicator as to which NSAIDs to prescribe in order to decrease the risk of nonunion.

Limitations
The present meta-analysis does have its limitations. The studies that were included were limited to only 6 RCTs, which limit the analysis that is possible to have a better understanding of NSAIDs on bone healing. Among these studies, the primary outcome of nonunion was not standardized in terms of outcome measurement, some used X-ray or CT scan, whereas others used clinical evaluation for healing. For this reason, extraction of the data was limited to what was made available. Additionally, this metaanalysis did not consider the heterogeneity of the population and the targeted fracture sites that were chosen for the study. This would require a greater availability of studies. The studies that were included in the meta-analysis included various fracture types. For example, Adolphson, Davis, and Drendel included Colles and pediatric fractures of the upper extremities. Such fractures are uncommon to have nonunion; therefore, to equate the risk of nonunion of these fractures to, for example, long bones in adults would underestimate the number of possible nonunions. Additionally, the elective surgeries of hallus valgus described in Brattwal have a different healing potential compared to the high www.otainternational.org risk of nonunion in long bones. For these reasons, the limitation of the number of available RCTs on this topic has made the heterogeneity of the studies large.

Conclusions
Based on the available literature of NSAIDs exposure on bone healing, it can be concluded that there is a need for further studies to have a better understanding of the mechanism of action of different types of NSAIDs. From this meta-analysis, it can be concluded that NSAIDs that do not include indomethacin can be used for pain management without having a significant effect on bone healing and, additionally, the use of NSAIDs for short duration, less than 2 weeks, does not show a statistical increase in nonunions.