Assessing the impact of variation in diagnostic coding among the three countries in the UK Biobank

1.BackgroundThe UK Biobank (UKB) study has linked hospital inpatient data collected from England, Scotland, and Wales, which use different clinical coding systems to record health outcomes. Scotland records up to 6 different diagnostic codes for one inpatient episode, compared with up to 20 in England, and 14 in Wales. We assessed the impact of the variations in diagnostic coding among countries on observed disease incidence rates. MethodsWe examined the number of diagnoses coded by each country, and then compared the incidence of three diseases between countries: Parkinsons disease (PD), type 2 diabetes (T2D), and dementia. We constructed Cox models for each disease, adjusting for "country". ResultsEngland appears to have the highest risk (hazard ratio, HR) for all three diseases, while Scotland appears to have the lowest risk: HR [95% CI] = 0.62 [0.54, 0.72] for PD, 0.49 [0.45, 0.54] for T2D, and 0.88 [0.78, 0.99] for dementia. ConclusionsThe observed incidence of these diseases and the estimated effect of "country" in Cox models are likely influenced by the clinical coding variations among countries. Researchers need to be aware of this and account for these variations in their analyses. 2. Key messages in a boxWhat is already known on this topic: O_LIThere are variations in the clinical coding systems in hospitals among the three countries in UKB. C_LIO_LIThese variations could impact research investigating geographic differences in disease incidence. What this study adds: C_LIO_LIWe assessed the impact of the variations in diagnostic coding among the three countries in UKB. C_LIO_LIWe showed that the observed disease incidence and the estimated effect of "country" are likely influenced by these variations. C_LI How this study might affect research, practice or policy: O_LIResearchers need to understand the provenance of data and account for the different diagnostic coding and other geographic variation specific to each of three countries, in order to reach robust conclusions that are not influenced by artefacts. C_LI