Methodology for the determination of the reliability of database derived data

Methodology to access the reliability of databases was presented. As an illustration of the proposed methodology, a simple random sample of 150 cases was extracted from a medical database at Children's Hospital of Michigan, Department of Cardiovascular Surgery in Detroit, Michigan. The database was reviewed for errors to determine overall reliability. Precluded by logistical limitations, the investigator was able to assess reliability as related to data processing errors. Data processing errors were defined as errors made when (1) recording data onto paper database forms, (2) recording errors into the database (referred to as keyboard errors) e.g. typographical and spelling errors and (3) recording errors, which are errors made when recording into the database, e.g., transpositions, omissions, or redundancies. The methodology employed was the Pearson Product moment coefficient to determine the coefficient of equivalence between collection forms and electronic forms. The reliability coefficient generated was .99 (p = .00). The investigator was also able to identify who was making errors, the location of those errors, and the type of errors made. The methodology offers inquirers a snapshot method to report on database reliability status in a manner in which a concise evaluative statement could be made regarding the database reliability status. Researchers are encouraged to ascertain reliability information on databases prior to publication of research or making decisions based on database derived data.