Persons who previously had been hospitalized for alcohol-related reasons were more likely than others to become non-responders in population health surveys, but adjusting for this particular source of non-response bias made little difference to the estimated prevalence of hazardous use or of abstinence.
The results were in line with previous studies [6–16], demonstrating higher risks (odds ratios up to 2.0) of becoming non-responders among excessive alcohol users. These studies have used alcohol consumption rather than hospitalization as a predictor of non-response, and had varying definitions of consumption levels and/or excessive drinking. In general, they have examined different sub-populations of all non-responders. By contrast, the current study analysed the entire samples, corresponding to a 100% response rate. It therefore also included the persons least likely to be reached by representative surveys. Examining non-random samples of non-responders can be misleading, which possibly explains the sometimes contradictory findings of previous studies. Our study indicates somewhat larger ORs than those reported in previous studies, which might suggest that non-response related to alcohol use is underestimated in designs based on earlier or later responses. But it might also reflect people with severe alcohol problems being more likely to become non-responders than people with less of a problem.
The current study also considered the hypothesis that previous hospitalization predicts current alcohol use. Only the opposite relationship has been shown previously . Further analysis and discussion of this particular result may be found in an accompanying article . In this context, alcohol-related hospitalization can reasonably be seen as a marker of deviant alcohol habits.
During the decades investigated, fewer than one in fifty adults received alcohol-related hospital care. Thus, only a small proportion of the population suffered this particular adverse consequence of their drinking behaviour. A history of alcohol-related morbidity, indicated here by hospitalization, suggests a life style of abuse and deviance. To our knowledge, the extent to which such a population is reached by general representative population surveys has not been examined previously. This is also the first study to present hospitalization rates over a period of time as long as ten years.
Although the bias caused by a small sub-population of non-responders may be rather modest, the true alcohol-related reporting bias may be larger than the adjustment for the non-response hospitalization rate in our study suggests. Only a small proportion of the population with alcohol problems receive hospital care for alcohol-related causes; most persons with alcohol-use disorders go untreated or undetected by the health care system . Also, few studies have examined the relationship between alcohol-related hospitalization and alcohol use . Thus, it is not clear whether people who receive alcohol-related hospital care resemble the larger population of people with alcohol problems with respect to reporting behaviour. It is, therefore, also unclear whether our results on those who have been hospitalized generalize to all people with alcohol-use disorders or possibly to an even larger population, such as hazardous users. Future studies may wish to examine hospitalization, alongside abstinence and hazardous use, as predictors of attrition. If hazardous users are also twice as likely to become non-responders than others, the prevalence of hazardous use may be seriously underestimated.
Non-response is likely to be related to factors other than hospitalization, such as gender, age, social class, etc., some of which are also likely to covary with alcohol use. A limitation to the study was that it was restricted solely to examining the bias from a different hospitalization rate among non-responders. Yet, there are other biases that probably affect estimated prevalence rates.
Self-rated alcohol consumption figures capture about half the population’s actual consumption . Like other studies of self-rated alcohol consumption, this study rests on the assumption that underrating in self-reports is proportionately distributed but otherwise random. But it remains unclear whether this is true. Underreporting may differ by sub-group; for example, heavy consumers may underreport to a greater degree.
The assumption of similar alcohol use among non-responders and responders for the different strata of hospitalization is questionable. But it was mainly because of the small proportion of the population hospitalized (just 1.7%) that the effects on estimated prevalence of hazardous use and of abstinence were insignificant. Even if hazardous alcohol use was more widespread among non-responders than among responders, the effects on prevalence would still have been small.
In this longitudinal study, consent to link response data with registry data was not sought from responders until follow-up in 2007, which meant that the hospitalization rate for non-responders between 2002 and 2007 was not available. It was, therefore, not possible to differentiate between non-response and attrition in the analyses. Although attrition is likely to contribute to selection bias, researchers usually have some information to go on, e.g., from baseline measures, and are therefore in a position to analyse it. Further, attrition bias can be mitigated because there are usually variables available that can be used to treat the missing data.