A method for estimating "persons" versus "cases" from hospital morbidity data in the absence of unique personal...

The authors present a method for estimating the number of persons (as opposed to the number of admissions) admitted to hospital for a defined time period for specific diagnoses when unique personal identifiers are not available. A simple computer algorithm (algorithm 1), employing a family-based scrambled health insurance number in conjunction with diagnosis, sex, and age was used to estimate the rates of readmission to hospital specific diagnoses. A proportion of the hospital morbidity records used in the study did not contain a health insurance number; therefore, estimated readmission ratios for specific diagnoses, calculated from admissions with health insurance numbers, were applied to all admissions for those diagnoses to obtain the estimates of total persons admitted to hospital in the given time period. The validity of the methodology was tested for selected diagnoses for a specific year by 1) comparing estimates of persons obtained by applying algorithm 1 with manual counts of persons obtained through examination of printed lists of all Ontario admissions to hospital for the specific diagnoses, and 2) comparing actual counts of persons admitted to a specific hospital with computer estimates of persons admitted to that hospital. The conclusions drawn were that the method, using algorithm 1, is a valid one for obtaining estimates of the number of persons admitted to hospital in a given year with a specific diagnosis and that the information derived is potentially useful for conducting epidemiologic studies. Readmission ratios were also estimated through the use of a second algorithm (algorithm 2) which was not dependent on a scrambled health insurance number. Sensitivity and specificity of these ratios were found to be low when compared with ratios estimated by algorithm 1.