Why the 'last drinking occasion' approach to measuring alcohol consumption should be avoided

There has been debate about which type of measurement instrument gives the best alcohol consumption estimates. This study used alcohol consumption data for a simulated population to compare the last drinking occasion (LDO) method against the true alcohol consumption. The LDO method requires respondents to indicate the quantity consumed at the LDO, and this information is used to calculate population estimates.

We simulated three hypothetical distributions of alcohol consumption, each with different underlying drinking patterns and each representing an idealised example of the distribution of alcohol consumption in Norway, Russia and France under specified conditions. We investigated the level of agreement between LDO consumption and the within-person average consumption, and the expected bias that arises when parameters of interest such as means, percentiles, exceedance proportions, differences in group means, correlation and regression coefficients, attributable fractions, and estimated disease risk functions, are calculated from LDO consumption.

Severe bias was found when LDO consumption was used to calculate common parameters of interest in alcohol and health research. The bias is caused by poor agreement between consumption at the LDO and the within-person 'true' or average consumption. The asymptotically expected bias is particularly problematic because it is unpredictable in terms of direction and magnitude, it varies across groups and populations, and it is probably impossible to adjust for.

Future data collections should avoid relying on the LDO method as a measure of typical alcohol consumption, despite its ease of administration. [Østhus S, Brunborg GS. Why the 'last drinking occasion' approach to measuring alcohol consumption should be avoided. Drug Alcohol Rev 2015;34:549-58].