Effect of NSAIDs on Bone Healing Rates: A Meta-analysis

Benjamin M. Wheatley, MD; Kyle E. Nappo, MD; Daniel L. Christensen, MD; Ann M. Holman, MLS; Daniel I. Brooks, PhD; Benjamin K. Potter, MD


J Am Acad Orthop Surg. 2019;27(7):e330-e336. 

In This Article


A computerized search was performed using MEDLINE (PubMed and Ovid platforms), Embase, Cochrane Database of Systematic Reviews, Joanna Briggs and EBM Reviews (Ovid platform), and Web of Science. The searches were performed on October 25 to 26, 2016. The lead author identified keywords, and the medical librarian identified corresponding Medical Subject Heading terms and Emtree thesaurus terms. In addition, ClinicalTrials.gov, the clinical trials database, was searched for relevant studies (Search strategies: see Appendix 1, Supplemental Digital Content 1, http://links.lww.com/JAAOS/A150, for MEDLINE (PubMed platform), Appendix 2, Supplemental Digital Content 1, http://links.lww.com/JAAOS/A150, for Embase, and Appendix 3, Supplemental Digital Content 1, http://links.lww.com/JAAOS/A150, for MEDLINE and Joanna Briggs and EBM Reviews [Ovid platform]). Results were limited by publication date: January 1, 1990, to October 25, 2016, but not by language. We included studies of pediatric and adult patients with healing bone including fracture, osteotomy, and fusion surgeries and postoperative or post-injury NSAID exposure versus control patient cohorts without NSAID exposure. The outcomes of interest were nonunion, delayed union, and pseudarthrosis with a minimum length of follow-up of 6 months. Because of variability in outcome reporting, nonunion, delayed union, and pseudarthrosis were considered collectively as adverse bone healing events and were not differentiated during the final analysis. All relevant randomized controlled trials and cohort and case-control studies were included.

Each title from the performed search was assessed for inclusion based on the criteria mentioned earlier (Figure 1). The relevant abstracts were then independently reviewed by two authors for inclusion, and if there was any doubt regarding the relevance and applicability, the full-text article was obtained. The full-text articles of the selected abstracts were then appraised by two authors for inclusion in the meta-analysis. Three authors then independently extracted relevant information including age, sex, smoking status, bone involved, NSAID class, dosage, duration of exposure, length of follow-up, and definition of delayed union or nonunion. In studies that included patients with multiple fractures or osteotomies of long bones, each bone was considered a separate instance, with the exception of the spine, which was considered as one osteosynthesis site regardless of the number of levels fused. Study quality was assessed independently by three authors using the Newcastle-Ottawa Scale (NOS) for nonrandomized studies and the Jadad scale for randomized studies.[14,15] Nonrandomized studies with an NOS score of five or greater on the nine point scale and randomized studies with a Jadad score of two or more on the five point scale were included. Any conflicts in quality assessment were resolved by a consensus between the three reviewing authors. Studies with zero events in both the exposed and nonexposed groups were excluded from analysis.[16] The senior author then adjudicated final assessment for inclusion or exclusion of selected articles and data.

Figure 1.

PRISMA flow diagram showing included studies.

This meta-analysis was performed in accordance with the QUORUM[17] and PRISMA guidelines.[18] All data were analyzed with R,[19] using the metafor package.[16] Because of the varied study designs and significant statistical heterogeneity (Q test statistic = 62.15; P < 0.0001 and I 2 = 77.25%), a maximum likelihood random-effects model was used to minimize variance. A pooled effect estimate, expressed as an odds ratio (OR) with 95% confidence intervals (CIs), was obtained using a maximum likelihood random-effects model. The number of delayed union or nonunion events and the nonevents from each study was used to calculate the OR of each study. The study quality, patient age, sex, bone involved, duration of exposure, and length of follow-up were identified as potential sources of heterogeneity. Separate subgroup analyses were performed in studies of pediatric and adult patients that were determined to involve a low dose or short duration of NSAID treatment. A low dose was defined as <125 mg/d of diclofenac, 150 mg/d of indomethacin, or 120 mg/d of ketorolac, whereas a short duration was defined as <2 weeks of treatment.

Meta-regression[20] was used to evaluate the risk of delayed union or nonunion in pediatric versus adult patient populations and by the bone involved in adult studies. Statistical significance of meta-regression analyses was evaluated with the Wald-type test implemented in the metafor package. Alpha was set at 0.05 for all analyses.