ISFAR short note prior to full critique (#278)

Carr S, Bryazka D, McLaughlin SA, Zheng P, Bahadursingh S, Aravkin AY, Hay SI, Lawlor HR, Mullany EC, Murray CJL, Nicholson SI, Rehm J, Roth GA, Sorensen RJD, Lewington S, Gakidou E. A burden of proof study on alcohol consumption and ischemic heart disease. Nat Commun. 2024 May 14;15(1):4082. doi: 10.1038/s41467-024-47632-7. PMID: 38744810.

Carr et al. (2024) re-evaluate the association between alcohol consumption and ischaemic heart disease (IHD). Using the burden of proof meta-analytical framework, the authors conducted an updated systematic review and re-evaluated existing cohort, case-control, and Mendelian Randomization (MR) data. The approach systematically tests and adjusts for common sources of bias defined according to specific criteria. The key statistical tool consists of a flexible meta-regression tool that does not impose a log-linear relationship between the risk and outcome, but instead uses a spline ensemble to model non-linear relationships[1].

The authors summarize their findings as follows: results suggested an inverse association between alcohol and IHD when all conventional observational studies were pooled as well as in evaluating cohort studies only. When only case-control studies were pooled, also alcohol consumption was associated with a decrease in IHD risk up to high levels of consumption. Their analysis of the available MR studies showed no association between genetically predicted alcohol consumption and IHD.

The authors conclude there is a need to advance MR methodologies and emulate randomized trials using large observational databases to obtain more definitive answers to this critical public health question.

This study is interesting in that it uses a more advanced meta-analytical technique involving modelling showing that a J-shaped association exists between alcohol consumption and IHD with a nadir of approximately 23 g alcohol/day.

Various criteria are used to adjust for common sources of bias. One of these is exposure assessment, which was quantified by whether alcohol consumption was recorded once or more than once in conventional observational studies, or with only one or multiple SNPs (genetic variants) in MR studies. Although this improved meta-analytical methodology considers alcohol assessment, it still does not consider nor adjust for important factors like drinking frequency and underreporting as mentioned by the authors.

The study also shows that MR studies cannot find that same association, leading the authors to conclude that MR studies need to improve. In other words, MR studies are not (yet) able to correctly describe a complex association as the alcohol-IHD one. The authors indicate that MR studies do not (sufficiently) allow for non-linear associations such as those for alcohol consumption and IHD. This is probably caused by insufficient genetic variation testing, which does not allow for the study of dose-dependency and drinking pattern. Also, other factors like the motivation to drink or drinking with meals are not captured in the variations studied in MR studies, which are limited both in terms of the number of genes involved in drinking behaviour, as well as the genetic variation in these genes.

This paper essentially puts forward the notion that MR studies are not fit to study the alcohol-IHD association nor the associations between any other disease or total mortality with alcohol consumption.

The authors are not clear, however, on the role of intervention studies. They argue that studies, like the cancelled long-term MACH15 study, would have made a difference. Conversely, they argue that the implementation of a long-term trial would have been fraught with potential issues, such as ethical questions related to alcohol carcinogenicity. This is surprising since Medical Ethics Committees always weigh the scientific benefits against the potential risks for volunteers entering the studies. Medical Ethics Committees also assume the principle of acceptable risk rather than a zero-risk tolerance, which seems to be the current standard for alcohol consumption. Moderate alcohol consumption as proposed in MACH15 study, is associated with a very small increase in cancer risk, if any.

Short-term trials were briefly mentioned in this study, but not included in the overall analysis. This may be a missed opportunity since quite a lot of mechanistic work has been performed by these short-term nutrition interventions. Biomarker changes observed in nutrition interventions have led to estimated contributions in epidemiological associations. This type of analysis has led to the notion that most of the benefit from the alcohol-IHD association could be explained by observable biomarker changes (Mukamal et al., 2005). This means that not only are the conventional observational studies consistent, but also that a mechanism has been elucidated which may fully explain the association observed. In addition, large observational databases may be used to study the changes in disease incidences associated with changes in drinking behaviour as has been analysed for diabetes type 2 (Joosten et al., 2011).

Both approaches could be an emulation of randomized trials using large observational databases.

A full ISFAR critique of this study (#278) will be published in early June 2024.

Joosten, M. M., Chiuve, S. E., Mukama, K. J., Hu, F. B., Hendriks, H. F. J., & Rimm, E. B. (2011). Changes in alcohol consumption and subsequent risk of type 2 diabetes in men. Diabetes, 60(1). https://doi.org/10.2337/db10-1052

Mukamal, K. J., Jensen, M. K., Grønbæk, M., Stampfer, M. J., Manson, J. A. E., Pischon, T., & Rimm, E. B. (2005). Drinking frequency, mediating biomarkers, and risk of myocardial infarction in women and men. Circulation, 112(10), 1406–1413. https://doi.org/10.1161/CIRCULATIONAHA.105.537704

[1] I do not have a strong enough statistical background to understand what this exactly means. It would be nice to receive some explanation on this methodology from a statistician. It seems to adjust or at least weighs results according to prefixed criteria by using modelling techniques rather than a regression analysis.