Background — The use of administrative health data and other secondary data sources to conduct research are increasing, and the quality of these data requires careful scrutiny to ensure that findings of studies based on them are accurate.
Methods — We conducted a multicenter, chart-abstraction study in Ontario, Canada to evaluate the ability of linked administrative databases to identify important baseline demographic and treatment information, changes in dialysis treatment modality over time, and the occurrence of important outcome events in incident dialysis patients. The medical record was considered the reference standard.
Results — Within administrative databases, demographic information was very well coded, as was the location where individuals started dialysis, the first treatment modality, the first outpatient modality, and the treatment that was in use 90 days after the start of therapy. The ability to accurately recreate an individual patient's entire dialysis treatment history using physician billing claims was somewhat limited. The treatment changes were often identified in the correct temporal sequence, but the dates that the events occurred did not agree well. Finally, important outcomes including the death and kidney transplantation were captured well, although the recovery of kidney function could not be evaluated because of poor inter-rater reliability.
Conclusions — This validation study provides important information concerning the ability to detect variables related to dialysis care using administrative datasets. Validation work should focus not only on the ability of secondary data to identify baseline comorbidities, but should also attempt to verify that other key variables required to conduct analyses are reliably captured.