Measurement error as an explanation for the alcohol harm paradox: analysis of eight cohort studies.

Abstract

Despite reporting lower levels of alcohol consumption, people with lower socio-economic status (SES) experience greater alcohol-related harm. Whether differential biases in the measurement of alcohol use could explain this apparent paradox is unknown. Using alcohol biomarkers to account for measurement error, we examined whether differential exposure to alcohol could explain the socio-economic differences in alcohol mortality. Participants from eight representative health surveys (n = 52164, mean age 47.7 years) were linked to mortality data and followed up until December 2016. The primary outcome was alcohol-attributable mortality. We used income and education as proxies for SES. Exposures include self-reported alcohol use and four alcohol biomarkers [serum gamma-glutamyl transferase (available in all surveys), carbohydrate-deficient transferrin, alanine aminotransferase and aspartate aminotransferase (available in subsamples)]. We used shared frailty Cox proportional hazards to account for survey heterogeneity. During a mean follow-up of 20.3 years, totalling 1 056 844 person-years, there were 828 alcohol-attributable deaths. Lower SES was associated with higher alcohol mortality despite reporting lower alcohol use. Alcohol biomarkers were associated with alcohol mortality and improved the predictive ability when used in conjunction with self-reported alcohol use. Alcohol biomarkers explained a very small fraction of the socio-economic differences in alcohol mortality, since hazard ratios either slightly attenuated (percent attenuation range 1.0-12.1%) or increased. Using alcohol biomarkers in addition to self-reported alcohol use did not explain the socio-economic differences in alcohol mortality. Differential bias in the measurement of alcohol use is not a likely explanation for the alcohol-harm paradox.