Research Fellow London School of Hygiene & Tropical Medicine London, United Kingdom
Background: Validity of disease definitions used in electronic health record (EHR) studies is not always assessed, nor consideration given to how exposure/outcome definition validity may affect study findings. This is particularly important in common conditions with varied course and severity, e.g., eczema. Different eczema phenotypes may need to be considered as different diseases, as they are likely to come with different prognoses, treatment options, and adverse outcomes. Equally important is considering the context in which outcomes are recorded for people with eczema, where more frequent consultation for eczema makes mental illness diagnosis more likely. Similar findings in both routinely- and prospectively-collected data increases confidence in disease definitions.
Objectives: Study 1: Replicate granular eczema phenotypes identified in cohort data in primary care data Study 2: Compare mental illness disease definitions in cohort and primary care data
Methods: We conducted two studies, both using linkage between primary care EHR data, and questionnaire and survey responses: 1) Avon Longitudinal Study of Parents and Children (ALSPAC), a cohort of people in southern England born in the 90s; and 2) UK Biobank, a cohort of half a million people, aged 40-69 years at recruitment.
In both studies, we assessed agreement in disease presence between routinely- and prospectively-collected data sources.
Study 1: We aim to replicate previously established eczema phenotypes (ALSPAC) using eczema-related variables in linked EHR data to predict phenotypes (lasso regression) and evaluate whether EHR phenotypes can be used for research.
Study 2: We performed cross-sectional comparisons, estimating odds ratios (OR) (logistic regression) for association between eczema and mental illness (anxiety, depression), defining mental illness using: 1) primary care data; 2) self-reported previous diagnosis; and 3) responses to PHQ-9/GAD-7 (depression/anxiety scores) questionnaires.
Results: Study 1 (n=11,866): Results produced shortly pending data linkage verification.
Study 2 (n=230,047): Agreement between data sources was generally poor. We found associations between eczema and mental illness defining mental illness in UK Biobank (e.g., eczema -> depression OR 1.36 [95% CI1.27-1.45]) and linked EHR data (OR 1.56 [95%CI 1.50-1.62]).
Conclusions: UK Biobank and linked primary care EHRs appear to capture different eczema, depression, and anxiety definitions. However, we found associations in both sources, using primary care diagnosis, self-report or screening questionnaires. Together, our results highlight the importance of carefully choosing disease definitions in EHR studies.
