Background: The infectious risk of human tissue allografts is unknown. Healthcare claims data from large populations may be useful for quantifying and monitoring rates of infection after tissue allograft transplantation, but require accurate algorithms to identify allograft implantations and infections. We explored the feasibility of identifying allograft implantations and infections using claims data.
Methods: We identified all patients with procedure codes for anterior cruciate ligament (ACL) surgery within four HMO Research Network sites during calendar years 2000-2008. Charts were randomly selected for review to determine the frequency of allograft versus autograft use. We then flagged patients with claims-based indicators of possible infection including suggestive ICD9 diagnosis codes, procedures, antibiotic prescriptions, specialty consultations, emergency department visits, and readmissions within the 90 days following surgery. We reviewed charts of patients with and without infection indicators and calculated the sensitivity and positive predictive value of individual indicators as well as combinations of multiple indicators.
Results: We identified 6,615 patients with codes for ACL surgeries in four HMO Research Network sites. On review of randomly selected charts, allografts were utilized in 128/536 (24%) surgeries and autografts in 392/536 (73%) surgeries. Implant type was inadequately documented in the remaining 16 charts (3%). Possible claims-based infection indicators were present in 1,827/6,615 (28%) surgeries. We reviewed 765 charts with indicators of possible infection and 475 charts without infectious indicators. There were 30 confirmed deep tissue infections, all amongst patients with possible infection indicators. Correcting for sampling weights, we estimated 1.6 versus 1.0 infections per 100 surgeries in allograft versus autograft implants (crude odds ratio 1.7, 95% CI 1.1-2.8). The sensitivity of individual indicators ranged from 0 to 70% and positive predictive value from 0 to 24%. Combining indicators increased sensitivity to >90% but positive predictive values remained poor.
Conclusions: Procedure codes for ACL surgeries are not specific for allograft versus autograft implants. Claims-based indicators of possible infections reliably capture all patients with serious infections but also flag many false positives. We are now validating the sensitivity and positive predictive value of claims-based infection indicators using data from two additional HMO Research Network sites.