Use of Network Meta-Analysis to Compare the Efficacy of Osteobiologics in Spinal Fusion

Commentary A bewildering array of bone graft preparations and synthetic materials intended to promote bone healing (“osteobiologics”) is available for lumbar spinal fusion, but it is unrealistic for a single randomized clinical trial to directly compare the safety and efficacy of all of them. In the recent publication by Ambrosio and coauthors, fusion rates, perioperative details, and complications were compiled from 43 published randomized controlled trials that included a total of >3,800 patients. Twenty-two different combinations of osteobiologics were identified, although some had been evaluated in only a single trial. The results showed that recombinant human bone morphogenetic protein (rhBMP)-2 use was associated with significantly improved fusion rates and fewer complications compared with iliac crest autograft. Several other preparations might be slightly better than iliac crest autograft alone for inducing fusion, but their quality of evidence was marginal at best. Most off-the-shelf preparations resulted in fewer complications than autologous iliac crest bone graft (AICBG), perhaps largely related to the avoidance of harvesting autograft. Few other conclusions could be made with confidence. Although many previous studies have arrived at similar conclusions, the unique aspects of this publication include (1) collection of data from more published trials than any other review to date, (2) detailed observations from each study, far more than could be thoroughly discussed in the manuscript itself, that are available in tables in the Appendix, (3) rigorous evaluation of the included studies with respect to quality, bias, and funding source, and (4) systematic review and network meta-analysis (NMA) methods that allowed both direct and indirect evidence to be gleaned from publications that had, for example, compared only a single osteobiologic agent with iliac crest bone graft. More specifically, a usual meta-analysis extracts direct comparative evidence from studies of similar design. NMA, however, is a statistical method that helps compare multiple different treatments simultaneously in a single review by combining direct and indirect evidence in a mapped network of existing trials1. Direct evidence is obtained from trials that compare, for example, Treatment A with Treatment B. Indirect evidence is obtained via common comparators. If there is no published trial that compares Treatment A with Treatment B, then the 2 treatments can still be compared indirectly if both have been separately compared with Treatment C. The use of direct and indirect evidence can be graphically displayed using a network graph, such as Figure 2 of the Ambrosio publication. That figure illustrates that iliac crest bone graft was the most commonly used osteobiologic, so iliac crest bone graft was selected as the reference treatment to which all other osteobiologics were compared. Other complex aspects of the NMA, such as sensitivity, intransitivity, and statistical heterogeneity, were also considered. Space for discussion was limited, but the Appendix tables provide a treasure trove of details available to the energetic reader to help identify issues for further study. For example, the tabulated data suggest a significantly lower fusion rate for locally harvested autologous bone combined with iliac crest bone graft than for iliac crest bone graft alone. There is no obvious explanation for that observation, although in correspondence the authors suggested that since local autograft may contain less cancellous bone and more soft tissue than iliac crest autograft, the combination might be dilutional. Review of the Appendix material also shows that at least 1 study demonstrated poor fusion results (9 of 30 patients) for the combination of autologous local bone and allograft, a surprising observation that also needs further study. This review also calls attention to information that is missing from many publications, prompting the suggestion that a working group in the spine community might develop a list of elements that should be included in future reports, such as the source and amount of local autograft; the source, amount, and preparation methods of allograft (including demineralized bone matrix); the amount of rhBMP-2 used; relevant perioperative, mid-term, and long-term complications (see eSupplement 2 of the Ambrosio paper); length of stay; operative time; patient-reported outcome measures; details regarding donors of biologic products; and a more uniform definition of fusion. In summary, although the general conclusions of the study represent only a modest advance over previous investigations, the details available in the Appendix tables could help identify more specific questions for future research, and the manuscript is an outstanding illustration of how treatment variables can be compared both directly and indirectly via an appropriately designed NMA.