Achieving solid osseous fusion remains a challenge in spine surgery. Autologous iliac crest bone graft remains the gold-standard graft material. In most cases, the fusion rates achievable with modern surgical techniques and autologous iliac crest bone graft are >90%. In specific scenarios, including long-segment fusion surgery and in patients with comorbidities, the fusion rates are substantially lower than this benchmark. Commercially available recombinant human bone morphogenetic protein (rhBMP) has rapidly been adopted as an alternative that can avoid the morbidity of iliac crest harvest and yet achieve similar, or possibly better, fusion rates than autograft bone.

A number of recent publications have questioned the potential risks associated with the use of rhBMP. A series of papers showing conflicting results has followed: some describing serious complications associated with rhBMP, and some reporting minimal or no effect on complication rates. In fact, two independent groups analyzed individual patient-level data supplied by the manufacturer of rhBMP-2 (Medtronic Sofamor Danek) and reached similar, but not identical, conclusions.

This controversy is the context for the paper by Jain et al., which describes the use of rhBMP in cervical spine surgery and the in-hospital complications associated with it as identified in the Nationwide Inpatient Sample (NIS) database. Several broad questions arise from this paper.

The number of studies using large national databases to identify rare complications is growing. This trend makes sense. These databases provide access to large amounts of data on large numbers of patients relatively inexpensively. They provide great statistical power to identify even small trends in incidence or prevalence of complications, and the additional demographic data that they contain allow for multivariate analysis to try to identify predictive factors. However, serious limitations remain. First and foremost, the NIS data set contains no patient-reported outcomes. If our primary objective is to improve the quality of our patients’ lives, then an analysis that provides no ability to measure the patients’ sense of treatment success by definition cannot truly measure the primary outcome of our care. Second, the data set is based on billing and coding-level data and therefore many important details are not available. For instance, as Jain et al. point out, the surgeon’s indications for rhBMP use and the dose of rhBMP used are not available. If these data were available, they might dramatically change the conclusions of the paper. Third, if the billing and coding data are inaccurate or limited (for instance, in identifying certain complications), then the analysis will be inaccurate or limited too. The fact that the database used in this study includes only in-hospital complications limits the analysis since most anterior cervical surgery involves only a short hospitalization. As the authors point out, an analysis of the prevalence of swallowing difficulty or dysphagia after surgery is limited when the data set includes only patients whose swallowing difficulty was so severe as to appear in the hospital-level billing or coding data. Fourth, and perhaps most concerning, the data input for these databases are site-specific and not consistent across sites. The reliability and validity of much of these data have not been established in a consistent manner. These questions apply to this study specifically and to the growing number of database-level studies generally.