Bias in estimates of alcohol use among older people: selection effects due to design, health, and cohort replacement

Susanne Kelfve, Kozma Ahacic

There is a growing awareness of the need to include the oldest age groups in the epidemiological monitoring of alcohol consumption. This poses a number of challenges and this study sets out to examine the possible selection effects due to survey design, health status, and cohort replacement on estimates of alcohol use among the oldest old. Analyses were based on three repeated cross-sectional interview surveys from 1992, 2002 and 2011, with relatively high response rates (86 %). The samples were nationally representative of the Swedish population aged 77+ (total n = 2022). Current alcohol use was assessed by the question "How often do you drink alcoholic beverages, such as wine, beer or spirits?" Alcohol use was examined in relation to survey design (response rate, use of proxy interviews and telephone interviews), health (institutional living, limitations with Activities of Daily Living and mobility problems) and birth cohort (in relation to age and period). Two outcomes were studied using binary and ordered logistic regression; use of alcohol and frequency of use among alcohol users. Higher estimates of alcohol use, as well as more frequent use, were associated with lower response rates, not using proxy interviews and exclusion of institutionalized respondents. When adjusted for health, none of these factors related to the survey design were significant. Moreover, the increase in alcohol use during the period was fully explained by cohort replacement. This cohort effect was also at least partially confounded by survey design and health effects. Results were similar for both outcomes. Survey non-participation in old age is likely to be associated with poor health and low alcohol consumption. Failure to include institutionalized respondents or those who are difficult to recruit is likely to lead to an overestimation of alcohol consumption, whereas basing prevalence on older data, at least in Sweden, is likely to underestimate the alcohol use of the oldest old. Trends in alcohol consumption in old age are highly sensitive for cohort effects. When analysing age-period-cohort effects, it is important to be aware of these health and design issues as they may lead to incorrect conclusions.