Bias in consent to health data linkage: evidence from a UK cross-sectional survey
Dr. Lynsey Patterson (CoEfPHNI)
12.00, Tuesday 9 July. The Clubhouse, level 4
Introduction
The linkage of data from representative population surveys to routine administrative data sources is a powerful and efficient way of turning a cross-sectional study into a longitudinal one that obviates responder burden and bias, and simplifies issues of temporality. However, linkage depends on individual consent and while the factors influencing survey response are well described much less is known about factors influencing consent for record linkage. The aims of this study were to examine the demographic, socioeconomic, health and lifestyle factors associated with consent to linkage to personal medical records. The study is based on wave 1 of Understanding Society which was conducted between January 2009 – December 2010 and had an overall household response rate of 57.6% and an individual response rate of 81.8%.
Methods
All survey respondents were provided with an information leaflet about the linkage of health data and were asked to read and sign a consent form (main outcome). The analysis herein included individuals aged 16-74 years (n=43,709). Univariate and multivariate multi-level logistic regression models were used to investigate the association, for a range of demographic, lifestyle and health variables, with consent to linkage.
Results
Overall, 68.6% of respondents agreed to data linkage. Consent to linkage was lower at older ages, but there was no variation according to sex, marital status or socioeconomic status; those of white ethnicity were 60% more likely to consent to linkage compared to non-white ethnicity (Adjusted Odds Ratio (AOR) 1.60 95% CIs 1.50, 1.71). Poorer mental and physical health was associated with higher consent rates, for example AOR 1.16 (95% CI 1.10, 1.23) for those with a limiting long standing illness. Respondents from Scotland were 21% more likely than those from England to consent to linkage (AOR 1.21 95% CI 1.09, 1.34) whilst individuals from Northern Ireland were 45% less likely (AOR 0.55 95% CI 0.49, 0.62).
Conclusion
There are important differences in the factors influencing response to an initial survey and subsequent consent to data linkage. An interesting finding is the regional variation in consent rates, particularly the higher uptake in Scotland. More research is needed to understand if these variations reflect general mistrust, lack of understanding, or factors related to interviewer perception.
The linkage of data from representative population surveys to routine administrative data sources is a powerful and efficient way of turning a cross-sectional study into a longitudinal one that obviates responder burden and bias, and simplifies issues of temporality. However, linkage depends on individual consent and while the factors influencing survey response are well described much less is known about factors influencing consent for record linkage. The aims of this study were to examine the demographic, socioeconomic, health and lifestyle factors associated with consent to linkage to personal medical records. The study is based on wave 1 of Understanding Society which was conducted between January 2009 – December 2010 and had an overall household response rate of 57.6% and an individual response rate of 81.8%.
Methods
All survey respondents were provided with an information leaflet about the linkage of health data and were asked to read and sign a consent form (main outcome). The analysis herein included individuals aged 16-74 years (n=43,709). Univariate and multivariate multi-level logistic regression models were used to investigate the association, for a range of demographic, lifestyle and health variables, with consent to linkage.
Results
Overall, 68.6% of respondents agreed to data linkage. Consent to linkage was lower at older ages, but there was no variation according to sex, marital status or socioeconomic status; those of white ethnicity were 60% more likely to consent to linkage compared to non-white ethnicity (Adjusted Odds Ratio (AOR) 1.60 95% CIs 1.50, 1.71). Poorer mental and physical health was associated with higher consent rates, for example AOR 1.16 (95% CI 1.10, 1.23) for those with a limiting long standing illness. Respondents from Scotland were 21% more likely than those from England to consent to linkage (AOR 1.21 95% CI 1.09, 1.34) whilst individuals from Northern Ireland were 45% less likely (AOR 0.55 95% CI 0.49, 0.62).
Conclusion
There are important differences in the factors influencing response to an initial survey and subsequent consent to data linkage. An interesting finding is the regional variation in consent rates, particularly the higher uptake in Scotland. More research is needed to understand if these variations reflect general mistrust, lack of understanding, or factors related to interviewer perception.