Paul Norman & Harry Taylor

Transcript

1. Geography of UK Biobank: considerations for avoiding bias in measures of association Paul Norman School of Geography, University of Leeds Harry Taylor School of Social Sciences, University of Manchester Where going? • Intro to Biobank • Model framework • Results • Hearing aid use: Biobank • General health: Biobank • General health: HSE • Discussion / conclusion
2. UK Biobank • Recruitment: Invitation via letter sent to NHS patients living within 10 mile radius of one of 23 recruitment centres • Sample: 502,671 people aged 40-69 collected 2006-2010 • Questionnaire: Self-report completion • Testing: Physical measurements, provision of blood, urine and saliva samples • Benefits: Very large sample, comprehensive set of measurements • Issues: – Sample not representative: older, more white, female and affluent (Fry et al. 2017) – Geographically-clustered (urban bias) – Some measures only collected part-way through and some not available for certain collection centres
3. Data and model framework Hearing aid use as outcome • Different accessibility? Urban – rural differences? – e.g. Missed appointments due to transport issues (Hamilton & Gourlay 2002; TfA 2014) • The effect of deprivation on hearing aid use modelled using UK Biobank data – “Do you use a hearing aid most of the time?” Yes / No / Prefer not to answer – Relationship between Townsend deprivation and hearing aid use modelled controlling for age, sex, ethnicity, education and employment status – Predicted probabilities generated for hearing aid use across the observed Townsend scores Deprivation Health outcome Urban / rural home location
4. Hearing aid use • Deprivation was associated with an increase in hearing aid use from least to most deprived areas in urban (>10k population) areas • In rural/small town areas, the association between deprivation and hearing aid use was stronger Predicted probabilities for hearing aid use, UK Biobank Townsend deprivation index (at recruitment) Proportion of hearing aid use
5. General health using Biobank • Equivalent models were run for the general health outcome in UK Biobank • “In general how would you rate your overall health?” Excellent/Good/Fair/Poor/Do not know/Prefer not to answer • In rural/town areas, deprivation steeper gradient for general health than in urban areas Proportion reporting “good” or “excellent” general health Townsend deprivation index (at recruitment) Predicted probabilities for general health, UK Biobank
6. General health using Health Survey for England • Equivalent models were run for general health outcomes in the Health Survey for England (HSE); a representative sample • “How is your health in general? Would you say it was …” (Very good/good, fair, very bad/bad) • Cf Biobank: Urban – rural differences less pronounced in HSE & urban gradient steeper than rural Predicted probabilities for general health, Health Survey for England 2009-2017 Proportion reporting “good” or “very good” general health IMD (deprivation) score
7. Discussion • Effect of deprivation on hearing aid use much stronger in more rural areas • General health in the Biobank seems to be affected more by deprivation cf HSE • Data source equivalence: • Different sociodemographic profiles (e.g. age, Biobank sample less deprived) • Different measures of health scale & deprivation (IMD vs Townsend) • Studies subject to self-selection of respondents • Health Survey: interest in health • Biobank ‘healthy volunteer’ effect especially if travelling far • Considerations of sample study size • HSE has no significant difference but has much smaller sample size than Biobank • UK Biobank has very large numbers - we shouldn't let the significance lead us to conclude uncritically
8. Conclusion • Relationships between outcomes and predictors in UK Biobank risk being inaccurate, unless geography is taken into account • Implications of the sampling and data collection methods should be understood and explained as part of any research using this data source
9. References Fry et al. (2017) Comparison of sociodemographic and health-related characteristics of UK Biobank participants with the general population; American Journal of Epidemiology, 21 June 2017. Hamilton, K., Gourlay, M. (2002) Missed hospital appointments and transport, University of East London Transport for All (2014) Sick of Waiting, available at: https://www.transportforall.org.uk/files/reportsowonlineversionnonaccessiblelq.pdf