Can large but highly selected population surveys provide valid information on the descriptive epidemiology and associations of common health conditions? An analysis of UK Biobank data on chronic pain

INTRODUCTION:  Large-scale epidemiological studies to investigate genetic and environmental influences on disease often have very low participation rates. We aimed to determine the extent of selection bias within one such study, UK Biobank. The focus of the investigation was chronic pain and the extent to which this study replicated information on prevalence and associated factors determined from previous studies.

METHODS: Persons aged 40-69 years across Great Britain, were invited to take part in UK Biobank between 2006 and 2010. 9.2 million invitations were sent and 5.5% participated. Participants completed questionnaires on health and lifestyle and were asked about pain which “interfered with usual activities during the past month” and if present, site(s) and duration. Pain was considered chronic if present for >3 months. Exposures measured included psychosocial factors (visiting a GP for “nerves, anxiety or depression”, the number of episodes of depression, adverse life-events in the previous two years). These factors have been strongly associated with pain in previous studies. Prevalence rates were directly-adjusted for the UK population. 

RESULTS:  498,071 participants provided valid answers. Prevalence of chronic pain was 42.9% (99% CI 42.7-43.1). Prevalence was higher in females (45.7% vs 40.0%) and the most common sites were: back ( prevalence 17.4%), knee (15.7%), shoulder/neck (15.4%) and headache (9.7%). Prevalence increased with lower income but only varied to a small extent by age-group. Those who had consulted a GP for depression were more likely to report chronic pain (Relative Risk (RR) 1.36 99% CI 1.34-1.37) and risk  increased with increasing number of episodes (P for trend <0.001). There was a strong relationship between chronic pain and number of recent adverse life events in the past two years (>3 vs 0 events RR 1.70, 1.67-1.74).

CONCLUSIONS: Despite very low participation, UK Biobank has produced estimates of chronic prevalence very close to those reported from smaller, higher participation, chronic pain studies. This suggests, at least for common symptoms, that there is low selection bias. It has also replicated previously demonstrated associations with psychosocial factors but produced more robust estimates.