PURPOSE: To define periods of acceptable mortality reporting in primary care and to demonstrate through examples the implication for research using automated medical data. METHODS: Annual death counts were obtained for each primary care practice participating in The Health Improvement Network "THIN" (UK). Expected counts were calculated from national death rates, accounting for the practice's age/sex structure. The standardized mortality ratio (SMR) was calculated with 95% confidence intervals (CI). A visual review process was undertaken to assign the year from which the practice had acceptable mortality reporting (AMR). The process involved reviewer pairs who were blinded to each other's decisions. Patterns of death reporting were checked. The AMR year was applied as a filter to THIN data to assess its impact on the SMR. RESULTS: For most practices the SMR was relatively stable and the AMR year was easily identified with 86% agreement between the blinded reviewer pairs. Applying the AMR to THIN removed under-reporting of death. However, the total computerized follow-up reduced from 37 to 32 million patient-years. Problematic death recording patterns included some practices keeping only live patient records when converting their software systems thereby creating 'immortal periods' prior to this moment, and peaks occurring when practices updated the vital status of their patients' records. CONCLUSIONS: This is the first time that an external standard has been used to assess completeness of mortality in automated primary care data. The resulting AMR year provides a natural filter for research and avoids biases associated with 'immortal periods', record updating and under-reporting.
PURPOSE: To define periods of acceptable mortality reporting in primary care and to demonstrate through examples the implication for research using automated medical data. METHODS: Annual death counts were obtained for each primary care practice participating in The Health Improvement Network "THIN" (UK). Expected counts were calculated from national death rates, accounting for the practice's age/sex structure. The standardized mortality ratio (SMR) was calculated with 95% confidence intervals (CI). A visual review process was undertaken to assign the year from which the practice had acceptable mortality reporting (AMR). The process involved reviewer pairs who were blinded to each other's decisions. Patterns of death reporting were checked. The AMR year was applied as a filter to THIN data to assess its impact on the SMR. RESULTS: For most practices the SMR was relatively stable and the AMR year was easily identified with 86% agreement between the blinded reviewer pairs. Applying the AMR to THIN removed under-reporting of death. However, the total computerized follow-up reduced from 37 to 32 million patient-years. Problematic death recording patterns included some practices keeping only live patient records when converting their software systems thereby creating 'immortal periods' prior to this moment, and peaks occurring when practices updated the vital status of their patients' records. CONCLUSIONS: This is the first time that an external standard has been used to assess completeness of mortality in automated primary care data. The resulting AMR year provides a natural filter for research and avoids biases associated with 'immortal periods', record updating and under-reporting.
Authors: Kevin Haynes; Warren B Bilker; Tom R Tenhave; Brian L Strom; James D Lewis Journal: Pharmacoepidemiol Drug Saf Date: 2011-07-13 Impact factor: 2.890
Authors: Koen B Pouwels; F Christiaan K Dolk; David R M Smith; Julie V Robotham; Timo Smieszek Journal: J Antimicrob Chemother Date: 2018-02-01 Impact factor: 5.790
Authors: Koen B Pouwels; F Christiaan K Dolk; David R M Smith; Timo Smieszek; Julie V Robotham Journal: J Antimicrob Chemother Date: 2018-02-01 Impact factor: 5.790