OBJECTIVE: To compare two methods for identifying adverse events using routinely recorded hospital abstract data in all public and private hospitals in Victoria, Australia. METHODS: Secondary analysis of data on all admissions in the period 1 July 2000-30 June 2001 (n = 1,645,992) to estimate the rates of adverse events using International Classification of Diseases 10th Revision Australian Modification codes alone and in combination with an "incidence" data flag indicating complicating diagnoses which arise after hospitalization; rates of incidence and pre-existing adverse events, and rates for same-day and multi-day admissions. RESULTS: In total, 8% of all admissions were recorded with an adverse event. Use of ICD codes alone identified only 59% of the events identified using the combined method, giving a prevalence rate of only 5%. Incident cases, that is, those occurring in the index admission, represented 68% of identified adverse events. The adverse events incidence rate for multi-day admissions was significantly higher at 12%, compared with the same day rate of 0.4%. CONCLUSION: An "incidence flag" is essential to identify those adverse events for which a hospital has unambiguous responsibility. Using such a flag, secondary analysis of administrative data can provide hospital quality assurance programmes with a comprehensive view of all adverse events (not just "sentinel" events) at a reasonable cost and with more timely results than more intensive methods can achieve. Although the method is likely to underestimate the true rate of adverse events (in particular, by not capturing adverse events which only manifest after discharge), in this study of Australian hospitals, rates of adverse events were found to be similar to those derived from studies using manual review of patient records.
OBJECTIVE: To compare two methods for identifying adverse events using routinely recorded hospital abstract data in all public and private hospitals in Victoria, Australia. METHODS: Secondary analysis of data on all admissions in the period 1 July 2000-30 June 2001 (n = 1,645,992) to estimate the rates of adverse events using International Classification of Diseases 10th Revision Australian Modification codes alone and in combination with an "incidence" data flag indicating complicating diagnoses which arise after hospitalization; rates of incidence and pre-existing adverse events, and rates for same-day and multi-day admissions. RESULTS: In total, 8% of all admissions were recorded with an adverse event. Use of ICD codes alone identified only 59% of the events identified using the combined method, giving a prevalence rate of only 5%. Incident cases, that is, those occurring in the index admission, represented 68% of identified adverse events. The adverse events incidence rate for multi-day admissions was significantly higher at 12%, compared with the same day rate of 0.4%. CONCLUSION: An "incidence flag" is essential to identify those adverse events for which a hospital has unambiguous responsibility. Using such a flag, secondary analysis of administrative data can provide hospital quality assurance programmes with a comprehensive view of all adverse events (not just "sentinel" events) at a reasonable cost and with more timely results than more intensive methods can achieve. Although the method is likely to underestimate the true rate of adverse events (in particular, by not capturing adverse events which only manifest after discharge), in this study of Australian hospitals, rates of adverse events were found to be similar to those derived from studies using manual review of patient records.