Validation of computerized Swedish dog and cat insurance data against veterinary practice records

Large computerized medical databases offer great potential for epidemiological research. However, data-quality issues must be addressed. This study evaluated the agreement between veterinary practice records and computerized insurance data in a large Swedish claims database. For the year 1995, the company insured over 320,000 dogs and cats. A total of 470 hard-copy records were sampled from claims for health care (n = 236) and life insurance (n = 234). Computerized insurance data for these claims were accessed and records from the attending veterinary practices were collected. For health and life claims, respectively, 79.2 and 72.8% of practice records were retrieved. Variables compared between the computer and practice records were breed, sex, date of birth and diagnosis for the claim. The degree of agreement was categorized as agreement, minor disagreement, major disagreement or data missing. Multivariable logistic regression was used to examine factors associated with errors. The observed agreement for breed and sex was excellent. There was 28.9 and 33.8% minor disagreement for the date of birth for health and life claims, respectively. This was mainly because, prior to 1993, the date of birth was coded as the first of January when more complete information was unavailable. Major disagreement (different year of birth) was low for both health and life claims. For health and life claims, the observed agreement for diagnosis was 84.0 and 84.9%, minor disagreement was 6.4 and 4.3%, and major disagreement 5.9 and 9.4%, respectively. Although there was no difference based on size of the veterinary clinic involved, there was a tendency for better agreement between the practice record and the computerized insurance data for claims from clinics with computerized practice records (n = 40) than for clinics with manual practice records (n = 286). Rates of discrepancy were affected by the clerk who processed the claims. Records processed by two of the 21 clerks had significantly more disagreements. Given the nature of the data, it was only possible to calculate a measure of observed agreement. In general, the agreement between data in the insurance-company database and from the practice records was excellent for demographic data such as breed and sex and fair for diagnostic information and date of birth. In general, the data are adequate to support ongoing research with due considerations of certain limitations.