Abstract
Background/Aims Infection following implantation of an orthopedic allograft can be devastating. Infections caused by contaminated allograft tissues are believed to be rare, but clusters — groups of infections with a common cause — have been discovered and reported repeatedly. Discovering clusters is challenging, however, because most orthopedic infections are attributed to surgical infection rather than allograft contamination. Statistical cluster detection tools may facilitate recognition of allograft contamination patterns. It is not known, however, whether the combination of number of surgeries, suppliers, hospitals, and infection rates in the US would offer sufficient power to detect allograft contamination signals, even if the FDA were to establish a national surveillance system. We therefore sought to model the feasibility of detecting clusters through active surveillance of allograft infection.
Methods We simulate data that might be collected through such a surveillance system. We consider the ability of scan statistics to detect synthetic clusters of infection that reflect either manufacturing contamination or hospital error. We evaluate the power of scan statistics to detect infection clusters under a range of scenarios with varying factors including number of allograft implantations, baseline manufacturing contamination rate, baseline hospital error rate, elevated manufacturing contamination rate, elevated hospital error rate, and cluster size. The simulation settings are specified according to published statistics and with inputs from experts with subject matter knowledge in an attempt to mimic real-life settings.
Results Power increases with the number of allograft implantations, the cluster size, the relative contamination rate between the cluster and the baseline, and the relative surgical error rate between the cluster and the baseline. Power ranges from 0.1 under very conservative assumptions to nearly 1.0 with the greatest prevalence and contamination rates, with realistic scenarios resulting in power > 0.8. For example, there is power of 0.80 to detect a medium supplier (2,000 tissues per year) with a contamination relative risk of 8 compared to other suppliers, presuming 60K procedures per year, 20 suppliers, 1025 hospitals, 0.1% baseline contamination rate, and 0.1% baseline hospital error rate.
Discussion An active allograft surveillance system covering most US hospitals would have good power to detect clusters of allograft infection.
