Monthly variations in self-report of time-specified and typical alcohol use: The Nord-Trøndelag Health Study (HUNT3)

Monthly variations in reporting of recent alcohol use

Studies from European countries and the United States have found a tendency of alcohol use to peak in December and during the summer months [23-29]. The reporting of recent alcohol use (last 4 weeks) in the present study are in line with these results, as our results indicates equal to or higher effects of seasonal variations compared to previous studies. For example, in the present study the odds ratio of reporting any alcohol consumption last 4 weeks in January was 1.42 while two previous studies reported odds ratios of 1.17 [27] and 1.22 [26]. The highest number who reported alcohol use in the last 4 weeks in the HUNT3 data was found in January, when over 80% of the sample reported that they had consumed alcohol the previous month. There was also an increased odds of reporting alcohol use in the last 4 when asked in April, August and September – reflecting increased alcohol use during Easter and in the Summer months. Thus, there is likely to be seasonal variations in alcohol use also in Norway.

Monthly variations in reporting of alcohol use the last 12 months and typical alcohol use

In contrast to the findings on recent alcohol use, when people were asked to report on whether they used alcohol weekly in the last 12 months, or whether they engaged in typical weekly binge drinking, the number of people answering affirmative on these questions were found to be lowest in January and highest in the summer months. These findings are indicating that monthly variations in actual alcohol consumption in some degree affect the reporting of more long-term alcohol consumption. The lower reporting in January is somewhat contrary to the hypothesis that reporting on alcohol use is based on recall of recent drinking behavior, but is in line with a previous study which showed that reports of drinking in the past 12 months were lower when assessed in January compared to December [31]. These findings could perhaps be explained by people considering their alcohol use during Christmas as atypical, and thus leaves the Christmas indulgence out of the equation when asked to estimate their typical alcohol use. This may not be the case during the summer months, when people may have engaged in higher levels of alcohol use over an extended period of time. As a consequence, the behavior may not stand out in their mind as atypical or exceptional.

Further, in contrast to the alcohol sale numbers, the level of alcohol use last 12 months (weekly drinking) and typical alcohol use (weekly binge drinking and number of alcoholic beverages) was also reported to be lower when this was assessed in December. An explanation for this finding may be that HUNT3 data collection in December was conducted during the first three weeks of the month, and thus before the Christmas festive season began.

In relation to binge drinking, the pattern of more monthly variation for women compared with men was also found in a previous study [25]. This finding could indicate that female binge-drinking is more situation-specific, while male drinking is more habitual [25]. If so, self-reported binge drinking among men may be less prone to time-related bias than among women. Regardless, it should be noted that monthly variations in self-reported binge drinking was present for men as well.

Implications of the results

The major difference between the present study and previous studies in this area is that while the previous studies employed defined reference periods, ranging from the current day [25], or previous week [25], two-weeks [30], month [26-28] or 6 months [25] before the assessment, the present study also included questions on typical alcohol use, without any specific reference period. Thus, while previous studies have primarily focused on assessing seasonal variations in alcohol use per se, the present study also examined whether the anticipated seasonal variations in alcohol use may affect reporting on measures assumed to minimize or eliminate time-dependent bias, such as QF or GF measures [4,12,15-17]. In 2005, Heeb and Gmel claimed that short reference periods “may be plagued with errors related to clustering effects of the day of the interview” (p.1015 [14]). Longer reference periods are generally believed to reduce the time-related effects associated with atypical drinking episodes and variations in alcohol consumption patterns affected by cultural or seasonal aspects, and thus obtain more representative average values of long-term and stable patterns of alcohol use. Our results indicate that this may not be the case, and that regardless of reference period, self-reported measures tend to assess current or recent rather than typical alcohol use in the population. This notion is further strengthened by the registered sales data obtained from external sources. Reports of alcohol use seem to fluctuate between the months of the year even when the respondents are asked about their alcohol use the last 4 weeks, the last 12 months, or their typical alcohol use.

The reliability of the reporting of typical or 12 months alcohol use depends on both the respondents’ ability to remember past drinking behavior, and on their ability to estimate the average use across this time span. This is a rather demanding task, and the memory of recent alcohol use episodes may stand out more vividly than more distant ones, and affect the reporting. The monthly variations in reporting of long-term and typical alcohol use is thus probably due to responders giving answers based on their recent alcohol habits, as these comes easier to mind than more distant or typical alcohol use [20,21]. Thus, responders tend to be quite imprecise in their estimates of alcohol consumption when a longer or an unspecified time-frame is introduced [23].

Previous studies that have examined the issue of seasonality in alcohol consumption conclude that studies that do not take seasonal variation into account when planning the study or interpreting the results may be biased [23-30]. Monthly variations in reported alcohol use are likely to produce bias in studies that aims to estimate the general incidence and prevalence of alcohol use in the population, and in studies that employ aggregated estimates of alcohol use. As an example from our analyses the mean prevalence of weekly binge drinking during the winter season (December, January and February) is 2.9% versus 4.2% during the summer months (June, July and August). Depending on the month or season the data is collected, the estimate generated may thus be either an over-estimate or an under-estimate of the average alcohol use during a year. Monthly variations in reporting of alcohol use may also have an impact of effect estimates, For instance, people who report weekly binge drinking behavior in a month where this is relatively rare in the population (i.e. January) may differ from people who report weekly binge drinking in a month where this is more common (i.e. August). Thus, a person who may be considered a case in August may be a control in January, and this may affect the effect estimates.

Already in 1978, Fitzgerald and colleagues recommended caution when data based on measures beyond time-specific estimates of consumption, for instance projections of annual consumption or trends in annual consumption, were interpreted [23]. The validity of aggregated data on alcohol consumption based on self-reports may be particularly biased in studies with short data-collection frames, and caution is required when data on alcohol consumption collected in a short time-frame are compared with other data. Thus Fitzgerald further suggested that “interviews should be conducted at comparable times of the year” (p. 883 [23]). Other authors have also given support to this recommendation, for instance concluded Lemmens and Knibbe that “comparisons of survey data on drinking are more or less invalid if the respective time-frames do not correspond” [25].

Our results indicate that even though both the time and the geographic location are different in our study compared to the previous studies, and even though we mainly focused on reports of typical alcohol use or alcohol use in the last 12 months, these concerns are still relevant. Framing of questions of alcohol consumption with the use of prolonged or typical reference periods do not seem to eliminate seasonal variations in reporting of alcohol consumption.

Strengths and limitations

The present study holds several strengths: Firstly, the data comes from a large well-defined population-based study, ensuring a sufficient sample size to investigate monthly variations in self-reported alcohol consumption as well as potential gender-differences. Secondly, the data collection had a time-frame of almost 2 years (ranging from October 2006 to June 2008) making it less likely that any monthly variation was due to some anomaly for a given calendar year. Thirdly, the present study included several different measures of alcohol use, with both specified and un-specified reference periods. As a result we were able to investigate the potential monthly variation in self-reported alcohol use across different types of question formulations with different reference periods. There is now general agreement that one should include questions on both frequency of drinking events, quantity of alcohol units consumed per drinking event and more irregular episodes of heavy drinking when alcohol consumption is assessed [4,12,15-17], and all these measures were included in the present study.

Despite these strengths, the study also holds some notable limitations: Firstly, when considering generalizability of the results, the geographical location of the study setting should be taken into account. The county of Nord-Trøndelag consists of mainly rural areas, and is located in Mid-Norway. The high latitude of the setting means that there are large differences in hours of sunlight and temperature between summer and winter [38]. These contextual factors may have an impact on both actual alcohol consumption and on self-report patterns, although the official sales statistics of Vinmonopolet shows a similar monthly variation in Nord-Trøndelag as for Norway as a whole for the 2006 and 2007 (data not shown). Previous studies employing wave two of the HUNT-study (HUNT2) have not found any seasonal variation in self-reported insomnia, time in bed or prevalence of case-level anxiety as measured by the Hospital Anxiety and Depression Scale (HADS) [39,40]. A seasonal variation for prevalence of case-level depression has, however, been reported [39]. There is thus little reason to believe that our findings are reflections of a general seasonal variation in self-report patterns, but rather that they either reflect variations in actual alcohol habits during the year or self-report patterns specifically related to questions regarding alcohol consumption. Secondly, a limited number of individuals participated in July, and the results from this month must therefore be interpreted with caution. Despite this, the patterns seen in July were in accordance with the other summer months. Thirdly, although several different questions regarding alcohol consumption were included, it would be interesting to compare the monthly variations seen in this study with other techniques of data collection on alcohol consumption through self-report, such as a records in an alcohol diary, or measures of more harmful alcohol use patterns, such as problem drinking and alcohol use disorders. Fourthly, although beyond the scope of the present paper, it would have been advantageous to assess the impact the monthly variations observed with respect to an outcome from an external data source, such as life-time clinically diagnosed alcohol use disorder. This would have enabled an investigation of whether seasonal variation in the reporting of alcohol use leads to a differential association with important outcomes. Fifthly, the information obtained regarding registered sales does not include tax-free sales and home-produced alcoholic beverages. There is, however, little reason to assume that the lack of such information invalidates the pattern shown in Figure 3. Finally, nonparticipation is always a challenge in epidemiological studies, and although data from HUNT3 in general are considered representative for the general population of Norway, selection bias cannot be excluded [32]. In particular, nonparticipation in health surveys is found to be higher among individuals with alcohol problems [41,42]. It is thus likely that individuals with a high and stable alcohol consumption pattern did not participate in the present study.