C-C5-04: Can Claims Data Accurately Identify Graft Type and Infections Following Anterior Cruciate Ligament Repairs?

Abstract

> 85%, while the completion rate for anticonvulsants ranged from 30-75%. The lowest test completion rate was for phenobarbital levels to monitor phenobarbital (30%). For all cardiovascular and anticonvulsant drugs, the proportion of recommended tests not ordered by the clinician ranged from 5% to 60%. The lowest test order prevalence was for phenobarbital level for phenobarbital use (35%), followed by valproic acid level for valproate use(48%), and carbamazepine level for carbamazepine use (60%). Rates of patient noncompletion of ordered tests for all drugs was generally <10%. Conclusion: Completion of laboratory monitoring tests for high-risk cardiovascular medications was higher than for high-risk anticonvulsants according to HEDIS guidelines. Clinician ordering behavior exhibited more variation than patient adherence to test orders. Underestimation of HEDIS quality of care monitoring due to patient non-adherence is minimal for cardiovascular medications, but higher for anticonvulsants. Background: The frequency of human tissue allograft contamination leading to infection is unknown. Attributing infections to allograft contamination versus surgical factors is elusive. National claims data may facilitate study of allograft-associated infections but first require accurate algorithms to identify implantations and infections. We explored the feasibility of using claims data to identify anterior cruciate ligament (ACL) allograft implants and infections. Methods: We selected candidate diagnosis, procedure, visit type, and antibiotic prescription criteria to identify ACL allograft repairs and infections. Candidate criteria were categorized by likelihood of infection and putative pathogen type. Criteria were then applied to HMO Research Network Virtual Data Warehouse files for 2000-2008 from six sites. We defined the infection risk period as 90 days following ACL repair. We reviewed charts flagged by each criterion to determine implant type and to assess for infection. Sensitivity and positive predictive values (PPV) were calculated. Results: Preliminary results are available from five of six sites. Procedure codes 29888 and 81.45 (ACL repair, unspecified) flagged 11,202 episodes of care. Diagnosis codes compatible with infection flagged 139 episodes (1.2%) of which 71 were categorized as high probability (0.63%). Microbiology and infection management procedure codes flagged 1,029 episodes (9.2%) and 68 episodes (0.6%) respectively. Antibiotics were prescribed for 1,523 episodes (14%) of which 580 were deemed high likelihood prescriptions (5.2%). There were 1011 hospitalizations within 90 days of ACL implantation (9.0%), 787 emergency department visits (7.0%), and 43 infectious disease specialist visits (0.4%). Over 1500 chart reviews are planned of which 107 have been completed thus far. The PPV for allograft implantation is 33% (95% CI 25-43%). Sufficient data are currently available to confirm 14 infections. Antibiotic prescribing was the most sensitive criterion while infectious disease visit had the highest PPV (sensitivities 86% and 7%, PPVs 35% and 100% respectively). Conclusions: Claims data do not discriminate between allografts and autografts in ACL repair but do identify a subset of patients with possible infections. Once chart