TY - JOUR T1 - Linkage of survey data with district-level lung cancer registrations: a method of bias reduction in ecological studies JF - Journal of Epidemiology and Community Health JO - J Epidemiol Community Health SP - 1093 LP - 1098 DO - 10.1136/jech.2005.043356 VL - 60 IS - 12 AU - Gillian A Lancaster AU - Mick Green AU - Steven Lane Y1 - 2006/12/01 UR - http://jech.bmj.com/content/60/12/1093.abstract N2 - Objective: To investigate a stratified ecological method for reducing ecological bias in studies that use aggregate data, by incorporating information on individual-level risk factors into the analysis. Design: Cross-sectional study investigating associations between socioeconomic risk factors and lung cancer in the north of England, using 1991 UK Census Small Area Statistics and Sample of Anonymised Records with lung cancer registrations from three regional cancer registries for 1993–6. Setting and patients: 92 local authority districts in the north of England containing over three million people aged 45–74 years. Results: Generally, groups considered more socioeconomically disadvantaged had an increased risk of lung cancer across districts. In the standard ecological analysis, effects for non-car ownership, social class III manual, social class IV/V and socioeconomic inactivity were insignificant, suggesting ecological bias. In the stratified ecological analysis these effects became significant (rate ratio (RR) 2.23, 95% confidence interval (CI) 1.79 to 2.78, p<0.001; RR 1.35, 95% CI 1.04 to 1.74, p = 0.022; RR 2.36, 95% CI 1.86 to 2.99, p<0.001; and RR 0.72, 95% CI 0.53 to 0.98, p = 0.039, respectively), and spuriously large positive effects for the social class III non-manual (RR 20.29) and unemployment groups (RR 147.53) reduced to a more reasonable level (RR 1.92, 95% CI 1.46 to 2.52, p<0.001; and RR 2.36, 95% CI 1.22 to 4.55, p = 0.011, respectively). Conclusions: Stratified ecological analysis incorporating information on individual-level covariates reduced the bias seen in a standard ecological analysis. The method is straightforward to apply and allows the linkage of health data with data from any large-scale complex survey where district of residence is known. ER -