Validation of death and suicide recording on the THIN UK primary care database

PURPOSE: Data from a cohort study were used to investigate death and cause of death (COD) recording on the THIN UK primary care database. METHODS: Subjects from 118 practices had a body mass index recorded, 2002-2003 (index date), no recent pregnancy or emotional event, or cancer. Coded and free text entries were searched for a COD for deaths dated < or = 485 days after index date. External documents were requested when date or COD were unclear, suicide was possible, and for a random sample of 40 patients with a COD identified. RESULTS: Five of 1399 deaths dated within 1 year of the index date had not died (positive predictive value 99.6%) and 4 of 222 dated within 366-485 days had died during the year (sensitivity 99.7%). A database COD was identified for 887 (63.4%) with external documents received for 597 (92.8% requested). Of 40 compared with external sources, the underlying COD was on the database in 33 (82.5%), and could be identified as such in 26 (65%). The date was within 1 day of that on external documents in 504 (94.9%) with a date of death but 3 (5.7%) with only a transfer-out of practice date. One of seven suicides was recorded as such. CONCLUSION: Any database record of death or its date is reliable, but transfer-out date is often later. Studies of COD or acute, potentially fatal conditions will miss cases unless a COD is obtained for all fatalities. Most suicides are not noted as such on the electronic record.