Critique 194: Long-term alcohol use and mortality among Swedish women – 22 November 2016
Licaj I, Sandin S, Skeie G, Adami H-O, Roswall N, Weiderpass E. Alcohol consumption over time and mortality in the Swedish Women’s Lifestyle and Health cohort. BMJ Open 2016;6:e012862. doi:10.1136/ bmjopen-2016-012862
Background: Alcohol consumption is steadily increasing in high-income countries but the harm and possible net benefits of light-to-moderate drinking remain controversial. We prospectively investigated the association between time-varying alcohol consumption and overall and cause-specific mortality among middle-aged women.
Methods: Among 48 249 women at baseline (33 404 at follow-up) in the prospective Swedish Women’s Lifestyle and Health cohort, age 30–49 years at baseline, we used repeated information on alcohol consumption and combined this method with multiple imputation in order to maximise the number of participants and deaths included in the analyses. Multivariable Cox regression models were used to calculate HRs for overall and cause-specific mortality.
Results: During >900 000 person/years, a total of 2100 deaths were recorded through Swedish registries. The median alcohol consumption increased from 2.3 g/day in 1991/1992 (baseline) to 4.7 g/day in 2004 (follow-up). Compared with light drinkers (0.1–1.5 g/day), a null association was observed for all categories of alcohol consumption with the exception of never drinkers. The HR comparing never with light drinkers was 1.46 (95% CI 1.22 to 1.74). There was a statistically significant negative trend between increasing alcohol consumption and cardiovascular and ischaemic heart diseases mortality. The results were similar when women with prevalent conditions were excluded.
Conclusions: In conclusion, in a cohort of young women, light alcohol consumption was protective for cardiovascular and ischaemic heart disease mortality but not for cancer and overall mortality.
This paper is based on a large group of relatively young Swedish women, a very good cohort for evaluating the long-term effects of alcohol on disease. The subjects were recruited randomly from the population, and excellent follow-up data were available for virtually all subjects. The authors had self-reported estimates of alcohol intake on two occasions, approximately 12 years apart, and attempted to judge the effects of changes in alcohol intake (stopping, starting, increases, decreases) on the risk of mortality over a total of more than 20 years from baseline.
Overall, the median reported intake of alcohol at the second assessment (in 2003-2004) of 4.7 g/day was considerably higher than at the first assessment (in 1991-1992) of 2.3 g/day, and the proportion of self-reported abstainers decreased between the two assessments. This was interpreted as a marked increase in alcohol consumption. Reviewer Waterhouse suggested, however, that “one factor the authors seem to have overlooked is the possibility that under-reporting of alcohol consumption might have declined by the second survey, as the stigma against women drinking might have abated.”
The methods used for the analysis were generally very appropriate, and well-done. The use of multiple imputation for certain missing variables is probably reasonable, as are the other assumptions of the authors (with a few exceptions, as noted below). Forum member Djoussé considered this to be “a reasonable study focused on light-drinking in women. The study had a limited number of events for many outcomes and therefore lacked adequate statistical power to detect small yet meaningful effects. The number of variables controlled for is of concern given the limited number of events.” Others commented that the authors reported results separately for relatively narrow ranges of alcohol intake (none, 0.1-1.49, 1.5-4.9, 5-9.9, 10-14.9, 15+ g/day), which makes the numbers of subjects in many groups quite small. Given that the total mortality risks for most groups of drinkers were similar, it would have been interesting to see the effects of alcohol consumption (versus no consumption) using a broader definition for “moderate” drinking.
Reviewer Thelle noted: “The Swedes have access to incidence and morbidity data. Such data would help in understanding why people changed drinking habits. One may assume that someone surviving an acute CVD event might change their intake and the subsequent analysis is likely to give biased results. The researchers should be encouraged to re-analyze the data using available morbidity data.”
Forum member Zhang had other comments: “In general, when assessing the relation of alcohol consumption to cause-specific mortality (not all-cause mortality), one should consider competing risk. For example, when evaluating the association between alcohol consumption and risk of alcohol-related cancer, one should consider the competing risk of death from cardiovascular disease. The competing risk, if not dealt with properly, may generate biased effect estimates. For instance, beneficial effect from alcohol on coronary mortality may selectively protect the persons susceptible to a particular type of cancer; this could make a null association between alcohol consumption and risk of that type of cancer become positive, or a weak association become stronger. This was well described many years ago by Schatzkin and Slud.”
Effects of choice of “referent group”: While the vast majority of well-done cohort epidemiologic studies have shown lower mortality from cardiovascular disease and lower total mortality among moderate drinkers, the specific relations shown in any study will be affected by the group of subjects (e.g., never drinkers, current non-drinkers, occasional drinkers, light drinkers) selected as the referent group. In the present analyses, the authors compared non-drinkers and other drinkers with subjects reporting “light” drinking, defined as drinking 0.1 to 1.49 g/day; subjects falling into this category on both assessments reported an average of 1.7 g/day, the equivalent of about 1 typical drink/week. The authors report higher mortality among non-drinkers than for all other groups, but no single category of drinkers showed a significant difference from their light-drinking reference group. They conclude that “light alcohol consumption was protective for cardiovascular and ischemic heart disease mortality but not for cancer and overall mortality.”
The problem of “under-reporting” of alcohol intake: Reviewer Ellison noted: “In observational studies, there is always the problem with under-reporting of alcohol intake. In the present study, some of the ‘light’ drinkers may have actually consumed more. This would tend to decrease any differences between the referent group and the next group of drinkers (1.50-4.9 g/day), or even those reporting 5.0-9.9 g/day. That could help explain why, despite the strikingly lower RRs for CVD and IHD for all drinkers when compared with the referent group shown in the paper, the inverse relation of alcohol with CVD was not statistically significant for any single group considered alone. Collapsing all truly ‘moderate’ drinkers into a single group (say, 5-14.9 g/day) may well have led to a statistically significant inverse association, not only for CVD but also for total mortality (which is the usual finding in cohort studies).
“Further, with approximately 10% of subjects reporting no alcohol consumption on both assessments (about 12 years apart), the number appears to be adequate for the authors to also have reported their results using non-drinkers as the referent group, being compared with all moderate alcohol consumers.”
Forum member Skovenborg also had comments regarding the alcohol consumption reported in this paper: “The median alcohol consumption of the cohort of Swedish women in 1991/1992 was reported by the authors to be 2.3 g/day = 1 litre of pure alcohol per person. At that time, the annual consumption based on sales was 7.8 litres of pure alcohol per person. According to Håkan Leifman, reliable surveys estimated the level of unrecorded alcohol consumption in that period to be 1.2 – 1.6 litres of pure alcohol/person/year. Accordingly, the real annual consumption may have been 9.0 – 9.4 litres of pure alcohol per person in Sweden.
“In 2004, the median alcohol consumption of the cohort of Swedish women was reported in this paper to be 4.7 g/day = 2.1 litres of pure alcohol per person. At that time, the annual consumption based on sales was 10.3 litres of pure alcohol per person. According to Leifman, reliable surveys estimated the level of unrecorded alcohol consumption in that period to 2.0 – 2.2 litres of pure alcohol/person/year. Accordingly the real annual consumption may have been 12.3 – 12.5 litres of pure alcohol per person. The large differences between the level of self-reported alcohol consumption and the mean annual consumption level of the population estimated including unrecorded consumption raises the suspicion of a serious underreporting bias.” Forum member Svilaas agreed with Skovenborg, and added: “These data are supported by my impression of changes of alcohol consumption in Norway, which is different from that in (southern) Europe.”
Forum member Skovenborg commented further on the reported alcohol intake: “Answers to questions about drinking pattern were not included in the study report. It begs the question of how to describe a drinker with a consumption of 0.1 g/day = 4 drinks per year (who would have been included in the “light” drinking referent category in this study). Is the participant enjoying a drink every 3 months or, more likely, will she have a few drinks in the Christmas – New Year season and be abstaining for the remainder of the year? What is the plausibility of a biological effect of such a minute dose of alcohol? The self-reported alcohol consumption reported by 71.1% of the female drinkers was 0.1 – 4.9 g/day.”
Reviewer Mattivi also had concerns about alcohol assessment: “Self-report questionnaire of alcohol is the exposure assessment tool of choice for large-scale epidemiologic studies such as this study. But there is serious concern about the reliability of the self-reported data on alcohol consumption, not combined with calibration sub-studies that involve more expensive reference instruments/methods. Such reported values from self-report questionnaire are subject to substantial error, both systematic and random, that can profoundly affect the results. In my view, two self-report questionnaires at about 12 years of distance would indicate the need to evaluate, at least on a reasonable sub-sample of the participants, the consumption via multiple 24-hour dietary recalls and, again at least on a sub-sample, to use some of the existing biomarkers capable of detecting recent heavy or prolonged drinking (as described by Petersen and by Würtz et al).” Other Forum members noted that for studies that are based on data collected many years previously, such applications are not possible. However, for ongoing studies, it would be very important to utilize such methodology, or even use methods such as those described by Klatsky et al for identifying under-reporting, whenever possible.
Lack of adjustment for pattern of drinking or type of beverage: There were criticisms by Forum members about the authors not taking the pattern of drinking into consideration. The authors had data on binge drinking and whether or not alcohol was usually consumed with meals, but chose not to include these data in their analyses. Given that drinking pattern plays a large role in determining favorable or unfavorable health effects of alcohol, the inclusion of such information may have better delineated the true effects of alcohol. Also, the authors “chose not to separately study the effects of different alcohol beverages.” Many studies indicate differences in effect according to the type of beverage, especially additional favorable effects from wine. Such differences are thought to be, at least in part, related to non-alcoholic constituents in wine and, to some extent, in beer.
Forum member de Gaetano noted: “I agree that meanigful comparisons should be made with lifelong abstainers, and that the potential effects on their results from under-reporting of alcohol intake should have been discussed. Moreover, I disagree with their decision not to report data on different beverages, especially in a country where spirits seem to be consumed more frequently than beer or wine.”
Assessing effects of change in level of alcohol intake: The authors attempted to judge how reported change in alcohol intake between the two assessments, about 12 years apart, affected mortality. It is interesting that, despite small numbers in some categories, the risk of essentially all types of deaths showed increases among subjects who decreased or stopped drinking between the two assessments. It should be noted that the specific cause why someone stopped drinking was not known; worsening health remains as a potential reason for someone to change his/her drinking habits. (Thus, the reason for decreasing alcohol, not necessarily the lowering of alcohol intake per se, could affect later mortality.)
When calculating the effects of change in alcohol intake, the investigators (as is typically done) adjusted for each subject’s baseline intake. (This, for example, attempts to take into account if a heavy drinker decreases or stops alcohol intake when compared with a regular light-to-moderate consumer who makes such a change; the health effects could be quite different.) It is unclear from other research whether adjusting for the baseline value of alcohol intake may or may not introduce bias.
References from Forum critique
Klatsky AL, Udaltsova N, Li Y, Baer D, Tran HN, Friedman GD. Moderate alcohol intake and cancer: the role of underreporting. Cancer Causes Control 2014;25:693–699. DOI 10.1007/s10552-014-0372-8..
Leifman H. Estimations of unrecorded alcohol consumption levels and trends in 14 European countries. Nordisk Alkohol- & Narkotikatidsskrift 2001;18:54-70.
Peterson K. Biomarkers for Alcohol Use and Abuse: A Summary. Alcohol Research & Health 2004/2005;28:30-37.
Schatzkin A, Slud E. Competing risks bias arising from an omitted risk factor. Am J Epidemiol 1989;129:850-856.
Würtz P, Cook S, Wang Q, Tiainen M, Tynkkynen T, Kangas AJ, et al. Metabolic profiling of alcohol consumption in 9778 young adults. Int J Epidemiol 2016; pre-publication: doi: 0.1093/ije/dyw175.
In a follow-up analysis of almost 50,000 young women, aged 30-49 at baseline, in the Swedish Women’s Lifestyle and Health cohort, the authors used self-reported information on alcohol consumption on two occasions, 12 years apart, to estimate the effects of alcohol on overall and cause-specific mortality. There were 2,100 deaths during follow up. Effects of alcohol on the two occasions, and changes between the two assessments among 33,000 women with available data, were related to mortality, as assessed from virtually complete national records.
Using “light” drinkers (0.1 – 1.49 grams of alcohol/day) as the referent group, the authors report an inverse association between greater amounts of alcohol consumption and mortality from cardiovascular disease. They report increased risk of cardiovascular and total mortality for abstainers. Despite the large number of total subjects, the authors report results separately for relatively narrow ranges of alcohol intake (none, 0.1-1.49, 1.5-4.9, 5-9.9, 10-14.9, 15+ g/day), which makes the numbers of subjects in many groups quite small. Given that the total mortality risks for most groups of drinkers were similar, it would have been interesting to see the effects of alcohol consumption (versus no consumption) using a broader definition of “moderate” drinking.
Forum members in general thought that this was a well-done study, using an excellent source of national records available in Sweden for determining mortality. They had concerns that the under-reporting of alcohol intake was not addressed; there is a high probability that some of the supposedly “light” drinkers may have actually consumed more alcohol, which would tend to decrease any differences between the referent group and the next groups of drinkers. This could help explain why, despite the strikingly lower risk ratios for cardiovascular disease for all drinkers when compared with the referent group, the inverse relation of alcohol was not statistically significant for any single group considered alone.
The Forum was also concerned that while the investigators had data on binge drinking and whether or not alcohol was usually consumed with meals, they chose not to include these data in their analyses. Given that drinking pattern plays a large role in determining favorable or unfavorable health effects of alcohol, the inclusion of such information may have better delineated the true effects of alcohol. Also, the authors “chose not to separately study the effects of different alcohol beverages;” reporting effects separately for beer, wine, and spirits could have provided key data on the effects of alcohol consumption.
The attempt of the authors to judge the effects of changes in alcohol intake during the study is laudable. Importantly, data on the reason that a woman may decide to increase or decrease her intake is not known; if due to the development of a serious disease, it may be the disease and not the change in alcohol intake that relates to subsequent mortality. However, given the inherent problems in assessing change, the results of this study are consistent with other studies that have suggested an increase in mortality risk for moderate drinkers who stop their alcohol consumption.
* * * * * * *
Comments on this critique by the International Scientific Forum on Alcohol Research were provided by the following members:
Giovanni de Gaetano, MD, PhD, Department of Epidemiology and Prevention, IRCCS Istituto Neurologico Mediterraneo NEUROMED, Pozzilli, Italy
Luc Djoussé, MD, DSc, Dept. of Medicine, Division of Aging, Brigham & Women’s Hospital and Harvard Medical School, Boston, MA, USA
R. Curtis Ellison, MD, Professor of Medicine & Public Health, Boston University School of Medicine, Boston, MA, USA
Harvey Finkel, MD, Hematology/Oncology, Boston University Medical Center, Boston, MA, USA
Fulvio Mattivi, MSc, Head of the Department of Food Quality and Nutrition, Research and Innovation Centre, Fondazione Edmund Mach, in San Michele all’Adige, Italy
Erik Skovenborg, MD, specialized in family medicine, member of the Scandinavian Medical Alcohol Board, Aarhus, Denmark
Creina Stockley, PhD, MSc Clinical Pharmacology, MBA; Health and Regulatory Information Manager, Australian Wine Research Institute, Glen Osmond, South Australia, Australia
Arne Svilaas, MD, PhD, general practice and lipidology, Oslo University Hospital, Oslo, Norway
Dag S. Thelle, MD, PhD, Department of Biostatistics, Institute of Basic Medical Sciences, University of Oslo, Norway; Section for Epidemiology and Social Medicine, Sahlgrenska Academy, University of Gothenburg, Sweden
David Van Velden, MD, Dept. of Pathology, Stellenbosch University, Stellenbosch, South Africa
Andrew L. Waterhouse, PhD, Department of Viticulture and Enology, University of California, Davis, USA
Yuqing Zhang, MD, DSc, Clinical Epidemiology, Boston University School of Medicine, Boston, MA, USA