Critique 076: Evaluating the association of alcohol intake with cognitive functioning using a Mendelian randomization study design — 22 March 2012
Au Yeung SL, Jiang CQ, Cheng KK, Liu B, Zhang WS, Lam TH, Leung GM, Schooling CM. Evaluation of moderate alcohol use and cognitive function among men using a Mendelian randomization design in the Guangzhou Biobank Cohort Study. Am J Epidemiol 2012; pre-publication release. DOI: 10.1093/aje/kwr462
Observational studies usually show that moderate alcohol use is associated with better cognitive function. Such studies are vulnerable to residual confounding arising from systematic differences between moderate alcohol users and others. A Mendelian randomization study carried out in a suitable population, such as southern Chinese men, in which alcohol use is low to moderate and is influenced by genotype, offers an alternative and superior approach for clarifying the causal effect of moderate alcohol use on cognitive function. The authors used aldehyde dehydrogenase 2 (ALDH2) genotype (AA, GA, or GG) as an instrumental variable in 2-stage least squares analysis to obtain unbiased estimates of the relation of alcohol consumption (measured in alcohol units, 10 g ethanol, per day) with cognitive function, assessed from delayed 10-word recall score (n = 4,707) and Mini-Mental State Examination (MMSE) score (n = 2,284), among men from the Guangzhou Biobank Cohort Study (2003–2008).
ALHD2 genotype was strongly associated with alcohol consumption, with an F statistic of 71.0 in 2-stage least squares analysis. Alcohol consumption was not associated with delayed 10-word recall score (- 0.03 words per alcohol unit, 95% confidence interval: – 0.18, 0.13) or MMSE score (0.06 points per alcohol unit, 95% confidence interval: – 0.22, 0.34). Moderate alcohol use is unlikely to be cognitively protective.
In a previous analysis in a much larger sample from the Guangzhou Biobank Cohort Study, the authors found that women reporting occasional alcohol intake and men reporting occasional or moderate intake had better scores related to cognitive function.1 These outcomes were based on the association of alcohol to delayed 10-word recall score among 28,537 subjects and to Mini-Mental State Examination (MMSE) score in 9,571 subjects.
Specific comments on present paper: The present study focuses on a subset of men in the study for whom DNA was available, permitting the measurements of genotypes affecting ALDH2. The statistical analyses are done appropriately, as were the DNA extractions and genetic analyses. However, it should be pointed out that these analyses are based on subjects who are volunteers, and it is not a population-based study. Alcohol consumption was measured only at base-line, at the same time that cognitive assessments were done (hence a possibility of reverse causality). Ex-drinkers were included with never drinkers as the referent group. The majority of the alcohol in all groups came from rice-wine, with Western table wine being reported for 10.6% of total alcohol intake for occasional drinkers, 16.0% for moderate drinkers, and 0.4% for heavy drinkers (the latter consumed 79% of their alcohol from rice wine and 21% from spirits). Rice wine is devoid of grape polyphenols.
Socio-demographic data by reported alcohol intake showed the expected relations (heavier drinking among subjects with less education and more manual labor, much higher smoking among heavier drinkers, and greater hypertension in the heavier drinkers) that have been typically found in epidemiologic studies of western populations. On the other hand, when comparing the occurrence of lifestyle and biologic factors according to genotype, other than alcohol most factors were unrelated. There was a tendency for the risk of hypertension to be slightly higher for the GG group (not statistically significant), which would be in line with expectations from other studies. On the other hand, the expected lower risk of diabetes in the more alcohol-consuming group was not seen.
In addition to relating ALDH2 genotypes to cognitive function, the investigators also related reported alcohol intake in grams of alcohol to cognition, using the following definitions: never user, occasional user, moderate [drinking weekly and less than 210 g/week (an average of about 3 drinks of 10 grams each per day)], heavy user (>210 g/week), and former user. Reported alcohol consumption varied markedly according to the genetic factors: average reported intake was 0.09 drinks/day for ALDH2 genotype AA, 0.24 drinks/day for those with GA, and 0.90 drinks/day for those subjects with the GG genotype.
One reviewer added: “Genotyping for ADLH2 is a good idea, but the authors should have also included ApoE as a major modulator of brain function and therefore a potential confounder.” Another reviewer agreed: “ApoE is a risk factor for Alzheimer’s disease, and should have been incorporated in the analyses. Excessive alcohol consumption is the trigger for Alzheimer’s disease in the presence of this SNP.”
Concerns about the cognitive function tests used: A Forum reviewer had another concern: “My main problem with this paper is the use of less than ideal cognitive tests. In this paper the authors used the MMSE, whilst the Montreal Cognitive Assessment (MoCA) test would have been more stringent. The MoCA test validation study2 has shown the MoCA to be a promising tool for detecting Mild Cognitive Impairment (MCI) and Early Alzheimer’s disease compared with the Mini-Mental State Examination (MMSE). Furthermore, the MoCA assesses multiple cognitive domains; it may be a useful cognitive screening tool for several neurological diseases that affect younger populations, such as Parkinson’s disease. Furthermore the results in the present study may have been corrected with the Wechsler Adult Intelligence Scale (WAIS) in order to bring socio-economic and educational factors into account.”
Use of Mendelian randomization design for studying human diseases: In using a Mendelian randomization design (a form of instrumental variable or IV technique) for studying alcohol consumption, scientists assume that genetic factors determine the degree to which people consume alcohol. Such an approach leaves no room for the effects of environmental exposures, which are generally considered to be important determinants of alcohol use, especially abuse. In this study, while the ALDH2 genotype related to reported alcohol consumption, the variance in alcohol use explained by ALDH2 was only 3%, suggesting that the use of the ALDH2 genotype was not a strong determinant of alcohol consumption.
Johnson et al3 have published an excellent paper comparing research designs for determining causality of an exposure, with information especially relevant to Mendelian randomization. A simple way to describe the IV estimate on the risk difference between alcohol consumption and cognitive function can be written as the following: Risk difference = (effect of gene on cognitive function)/(correlation between gene and alcohol consumption). In a randomized trial, the association between assignment (i.e., instrument variable, IV) and outcome is simply the intent to treat estimate. The correlation between assignment and treatment that is actually received would be an estimate for compliance. A strong IV should yield a correlation between IV (e.g., gene) and treatment (e.g., alcohol consumption) toward 1 (or -1), whereas a weak IV will yield a correlation to be closer to zero. If any of the IV assumptions have been violated, a weak IV may magnify any bias from residual unmeasured confounding that is factored into the numerator.3
In the present study, key comparisons between the association between reported alcohol consumption and the word-recall test, and between ALDH2 genotypes and the word-recall test, were very different. With all subjects included, there was no significant relation between either observed alcohol drinking or genotyped groups and the two measures of cognitive function. With heavy alcohol drinkers excluded, the only significant finding was slightly higher word-recall scores associated with increasing alcohol: using multivariable linear regression analysis, the beta for the association with delayed word recall was +0.12 for observed alcohol intake (CI 0.3,0.20). In contrast, the association with the genotype-classified subjects was -0.12 (CI -0.77, 0.53). These opposite influences are difficult to explain if both approaches are serving as reasonable measurements of alcohol consumption.
For results from the MMSE, the Mendelian approach suggested a weak positive effect on cognition, as did observational data; however, neither estimate was statistically significant. As stated by the authors, the discrepancies between the two approaches in this study, or between the results in this study and previous studies, “could be due to failure to adjust fully for confounding factors, such as socioeconomic position, lifestyle, health status, or cognitive potential, which are cognitively protective and difficult to assess comprehensively.”
Is Mendelian randomization the “Holy Grail?” The authors of the seminal paper on study designs for evaluating causality,3 described above, point out that the Mendelian randomization approach is markedly weakened if there is not a strong correlation with the outcome of interest. (In the present paper, ALDH2 genotypes explained only 3% of the variance in alcohol intake.) These study design authors add: “Unfortunately, several studies have shown that weak instruments may lead not only to larger standard errors in treatment estimates but may, in fact, lead to estimates that have larger bias than OLS” (ordinary least squares regression) or other multivariable regression analyses based on observational studies.3
Regardless of whether the results of these analyses turn out to reflect the “true” association of alcohol with cognitive functioning in China, there would be problems in relating the results to western cultures. The usual beverage in this study was rice-wine (which does not contain the polyphenols present in grape wines) and the pattern of drinking is reportedly quite different in China. Hence, applicability of the results to western populations could be difficult. As stated by one reviewer, “The cautions regarding confounding expressed by the authors are valid, but I doubt the transferability of the indigenous characteristics of southern China to the world at large, particularly those viewed through the lens of aldehyde dehydrogenase enzyme genetic variability.”
We agree with the concluding remarks of the authors of the present paper: “Causality should be thoroughly verified in a variety of settings using different kinds of evidence, including experimental or genetic studies, rather than relying on simple observations in a particular setting.” We strongly encourage attempts at using the Mendelian design in further studies. On the other hand, we appreciate the conclusions of a recent summary of good clinical practices by Johnson et al3 that pointed out potential problems with Mendelian randomization. That major report3 concluded that such Mendelian randomization sounds good, but it is not the “Holy Grail.”
References from Forum Comments
1. Au Yeung SL, Jiang C, Zhang W, et al. Moderate alcohol use and cognitive function in the Guangzhou Biobank Cohort Study. Ann Epidemiol 2010;20:873–882.
2. Nasreddine ZS, Phillips NA, Bédirian V, Charbonneau S, Whitehead V, Collin I, Cummings JL, Chertkow H. The Montreal Cognitive Assessment (MoCA): A Brief Screening Tool For Mild Cognitive Impairment. Journal of the American Geriatrics Society 2005;53:695-699.
3. Johnson ML, Crown W. Martin BC. Dormuth CR, Siebert U. Good Research Practices for Comparative Effectiveness Research:Analytic Methods to Improve Causal Inference from Nonrandomized Studies of Treatment Effects Using Secondary Data Sources: The ISPOR Good Research Practices for Retrospective Database Analysis Task Force Report—Part III. Value in Health 2009;12:1062-1073. © 2009, International Society for Pharmacoeconomics and Outcomes Research (ISPOR) 1098-3015/09/1062 1062–1073.
Many observational cohort studies have shown that moderate alcohol use is associated with better cognitive function. However, since such studies are vulnerable to residual confounding by other lifestyle and physiologic factors, the authors conducted a Mendelian randomization study, using aldehyde dehydrogenase 2 (ALDH2) genotype (AA, GA, or GG) as an instrumental variable in 2-stage least squares analysis. Cognitive function was assessed from delayed 10-word recall score (n = 4,707) and Mini-Mental State Examination (MMSE) score (n = 2,284) among men from the Guangzhou Biobank Cohort Study (2003–2008). The authors had previously reported an association between reported alcohol intake and cognitive function from a larger group of subjects from the same study finding that women reporting occasional alcohol intake and men reporting occasional or moderate intake had better scores related to cognitive function than did abstainers.1
In the present Mendelian study, the authors found no significant association between groups defined by the ALDH2 genotype (as an “unbiased” estimate of alcohol consumption) and the two measures of cognitive functioning. A problem with the present analysis is that ALDH2 genotypes explained only 3% of the variance in reported alcohol intake, which weakens the conclusions of the authors. Further, differences in the predominant type of beverage consumed (rice wine), and probably marked differences in drinking patterns between these subjects and Europeans and Americans, make it difficult to know what the implications of this study are for western industrialized societies.
We agree with the authors that “Causality should be thoroughly verified in a variety of settings using different kinds of evidence, including experimental or genetic studies, rather than relying on simple observations in a particular setting.” We strongly support future attempts at using Mendelian randomization studies, hopefully using better instruments for estimating alcohol intake. On the other hand, as stated by recent evaluations of various study designs for determining causality,3 we appreciate that Mendelian randomization sounds good, but it is not the “Holy Grail.”
* * *
Comments on this paper were provided by the following members of the International Scientific Forum on Alcohol Research:
Yuqing Zhang, MD, DSc, Epidemiology, Boston University School of Medicine, Boston, MA, USA
David Vauzour, PhD, Senior Research Associate, Department of Nutrition, Norwich Medical School, University of East Anglia, Norwich, UK
Harvey Finkel, MD, Hematology/Oncology, Boston University Medical Center, Boston, MA, USA
David Van Velden, MD, Dept. of Pathology, Stellenbosch University, Stellenbosch, South Africa
R. Curtis Ellison, MD, Section of Preventive Medicine & Epidemiology, Boston University School of Medicine, Boston, MA, USA