Identifying in-hospital venous thromboembolism (VTE): a comparison of claims-based approaches with the Rochester Epidemiology Project VTE cohort.

摘要

BACKGROUND: Efforts to identify hospital-acquired complications from claims data by applying exclusion rules to discharge diagnosis codes exhibit low positive predictive value (PPV). The PPV improves when a variable is added to each secondary diagnosis to indicate whether the condition was "present-on-admission" (POA) or "hospital-acquired". Such indicator variables will soon be required for Medicare reimbursement. No estimates are available, however, of the proportion of hospital-acquired complications that are missed (sensitivity) using either exclusion rules or indicator variables. We estimated sensitivity, specificity, PPV, and negative predictive value (NPV) of claims-based approaches using the Rochester Epidemiology Project (REP) venous thromboembolism (VTE) cohort as a "gold standard." METHODS: All inpatient encounters by Olmsted County, Minnesota, residents at Mayo Clinic-affiliated hospitals 1995-1998 constituted the at-risk-population. REP-identified hospital-acquired VTE consisted of all objectively-diagnosed VTE among County residents 1995-1998, whose onset of symptoms occurred during inpatient stays at these hospitals, as confirmed by detailed review of County residents' provider-linked medical records. Claims-based approaches used billing data from these hospitals. RESULTS: Of 37,845 inpatient encounters, 98 had REP-identified hospital-acquired VTE; 47 (48%) were medical encounters. NPV and specificity were >99% for both claims-based approaches. Although indicator variables provided higher PPV (74%) compared with exclusion rules (35%), the sensitivity for exclusion rules was 74% compared with only 38% for indicator variables. Misclassification was greater for medical than surgical encounters. CONCLUSIONS: Utility and accuracy of claims data for identifying hospital-acquired conditions, including POA indicator variables, requires close attention be paid by clinicians and coders to what is being recorded.