Capture recapture as a method of determining the completeness of tuberculosis notifications

Notification of tuberculosis cases is often incomplete but combining data from several sources can provide a more accurate estimate of the number of cases. Data for the city of Liverpool were collected over an eight-year period from three sources: notifications, microbiological records and in-patient discharge coding data. Capture-recapture (CR) techniques were used to estimate the total number (including unreported cases) of tuberculosis cases in the city. By creating a log-linear model from the pattern of case replication between data sets, a model of best fit was created from which the number of cases present in the population, but not identified in any of the data sets, was estimated. False positive diagnoses were found in 67/516 (13%) of notifications and in 65/241 (27%) of in-patient codings. After excluding these, the total combined number of cases from all data sources was 473. CR methods identified only twelve extra cases (2.5%) making the estimated number of true cases total 485. Of these, in-patient codings identified 36.3%, microbiological records 56.3% and notifications 92.6%. It was concluded that notification of tuberculosis is very complete in Liverpool. Capture-recapture methods can be used to assess completeness of notification data in other settings.