Development of an algorithm to classify primary care electronic health records of alcohol consumption: experience using data linkage from UK Biobank and primary care electronic health data sources

Abstract

ObjectivesDevelop a novel algorithm to categorise alcohol consumption using primary care electronic health records (EHRs) and asses its reliability by comparing this classification with self-reported alcohol consumption data obtained from the UK Biobank (UKB) cohort.DesignCross-sectional study.SettingThe UKB, a population-based cohort with participants aged between 40 and 69 years recruited across the UK between 2006 and 2010.ParticipantsUKB participants from Scotland with linked primary care data.Primary and secondary outcome measuresCreate a rule-based multiclass algorithm to classify alcohol consumption reported by Scottish UKB participants and compare it with their classification using data present in primary care EHRs based on Read Codes. We evaluated agreement metrics (simple agreement and kappa statistic).ResultsAmong the Scottish UKB participants, 18 838 (69%) had at least one Read Code related to alcohol consumption and were used in the classification. The agreement of alcohol consumption categories between UKB and primary care data, including assessments within 5 years was 59.6%, and kappa was 0.23 (95% CI 0.21 to 0.24). Differences in classification between the two sources were statistically significant (p<0.001); More individuals were classified as ‘sensible drinkers’ and in lower alcohol consumption levels in primary care records compared with the UKB. Agreement improved slightly when using only numerical values (k=0.29; 95% CI 0.27 to 0.31) and decreased when using qualitative descriptors only (k=0.18;95% CI 0.16 to 0.20).ConclusionOur algorithm classifies alcohol consumption recorded in Primary Care EHRs into discrete meaningful categories. These results suggest that alcohol consumption may be underestimated in primary care EHRs. Using numerical values (alcohol units) may improve classification when compared with qualitative descriptors.