Validation of computerized Swedish horse insurance data against veterinary clinical records

The aim was to evaluate the agreement between computerized insurance data in a large Swedish horse insurance database and the information in the corresponding clinical records (CR). A random sample of 400 veterinary care and 140 life claims was included. Information on name of the horse, breed and gender, year of birth, specific diagnosis and system diagnosis (e.g. joints, digestive and skeletal) was compared between sources. The concordance for demographic variables was categorized as agreement, disagreement or data missing. For diagnostic information, the categories were agreement, minor disagreement and major disagreement and for system information agreement or disagreement. There were missing values for demographic information in the CR, varying from 2% for name to 16% for breed. The overall agreement for demographic information was >94% (disregarding missing data), 92% for system and 84% for specific diagnosis. For veterinary care and life claims, the observed agreement for diagnosis was 85 and 83%, minor disagreement 6 and 5%, and major disagreement 9 and 12%, respectively. Using the CR data as gold standard, for the systems evaluated (joints, digestive, skeletal, skin and hooves), sensitivity varied between 62% (skin) and 89% (digestive) whereas the specificity was >96% for all systems. The positive predictive values ranged from 86% (skin) to 97% (digestive). Logistic regression analysis was used to examine factors associated with agreement for diagnosis. Analyses were performed separately for veterinary care and life claims. Factors examined were type of visit (clinic/field), treating veterinarian/clinic (categorized as district veterinarians, private practitioners, small clinics, medium clinics and, for the clinics with > or =20 claims, the specific clinics), computerized or manual CR, processing clerk, whether the CR was included in the paper file, if the claim was rejected or reimbursed, system diagnosis and if a immediate settlement (in analysis for veterinary care claims) or death certificate (in analysis for life claims) was included in the paper file. For veterinary care claims, in the logistic regression model type of visit was significantly associated with agreement, with clinic visits generating better agreement than field visits.