Commentary: “No Genetic Causality between Tobacco Smoking and Venous Thromboembolism: A Two-Sample Mendelian Randomization Study”

Mendelian randomization (MR) is a powerful method for causal inference that uses genetic variants as instrumental variables (IVs) and has gained popularity with the increase in genome-wide association studies (GWAS) in recent years. Recently, Du et al conducted an MR analysis to explore the causal relationship between smoking and venous thromboembolism (VTE), including pulmonary embolism (PE) and deep vein thrombosis (DVT), in European populations.[1] Interestingly, the authors found no causal association between the two, which contrasts sharply with a previous MR study that indicated that genetic susceptibility to smoking is associated with an increased incidence of DVT and PE.[2] Upon careful review, the inconsistencies in the results can be attributed to several limitations, including an incomplete consideration of smoking phenotypes, sample overlap, and the use of outdated databases. Consequently, the conclusions of this study should be interpreted with caution.

In the MR framework, obtaining a robust causal association requires strict adherence to three fundamental assumptions, as well as consideration of biases caused by weak IVs, sample overlap, and inadequate statistical power. A major issue in this study is the failure to account for sample overlap. Both smoking and DVT data were sourced from the UK Biobank (UKB), with sample overlap rates reaching as high as 92 and 99%, respectively. Such high overlap rates can increase the risk of type I errors and the winner's curse, adversely affecting the accuracy of causal inference.[3] Therefore, conducting additional sensitivity analyses or using nonoverlapping cohorts is necessary. Recently, a new method designed to address sample overlap—MRLap,[4] which uses cross-trait linkage disequilibrium score regression intercepts for correction—has shown good fit in simulations with 5 to 95% overlap.

A second major issue in this study is the choice of data, including the use of outdated datasets and an incomplete consideration of the exposure phenotype. The study utilized the 2021 release (R5) of the FinnGen consortium for DVT and PE data, despite the availability of the more recent R9 version at the time. Employing the latest datasets could provide a greater number of genetic variants, enhance the strength of the IVs, and include more accurate environmental and demographic information, thereby improving the accuracy and reliability of causal inference. Notably, the impact of using outdated data and resulting discrepancies has been validated in other MR analyses.[5] [6] The exposure phenotypes in the study were sourced from the UKB and included only current and past smoking status, which significantly lacks in capturing the multidimensional characteristics of smoking behavior. A more comprehensive approach would be to initially consider smoking phenotypes provided by the GWAS and Sequencing Consortium of Alcohol and Nicotine use (GSCAN) and supplemented by the UKB, which would include but not be limited to smoking intensity, frequency, duration, and cessation behaviors.[7] These aspects are crucial for assessing the impact of smoking on health ([Table 1]). Therefore, the authors' assertion that there is no causal link between smoking and VTE risk is premature and one-sided.

Table 1 Indication of suitable smoking phenotypes based on available high-quality evidence

Phenotype	Definition/diagnostics	Ref.	Consortium	Ancestry	Participants
Pack-years	Total number of pack-years smoked in adulthood	36402876	UKB	EUR	142,387 individuals
CigDay (past and current)	Cigarettes smoked per day by current smokers only or current and former smokers combined	30643251	GSCAN	EUR	337,334 individuals
CigDay (current)	Number of cigarettes currently smoked daily	36402876	UKB	EUR	33,229 individuals
Lifetime smoking	Measures the cumulative effect on smoking on health outcomes by encompassing the initiation, duration, intensity, and time since stopping smoking	31689377	UKB	EUR	462,690 individuals
SmkInit	An individual has ever smoked regularly (yes/no)	30643251	GSCAN	EUR	1,232,091 individuals

Abbreviations: CigDay, cigarettes per day; EUR, European; GSCAN, GWAS and Sequencing Consortium of Alcohol and Nicotine use; SmkInit, smoking initiation; UKB, UK biobank.

Statistical power is considered one of the main challenges in MR studies, as most genetic variants predict only a small fraction of the phenotypic variation.[8] A third major issue in this study is the neglect of power calculations. A recent study highlights the need for caution when excluding single-nucleotide polymorphisms (SNPs) to control for horizontal pleiotropy; if SNPs associated with confounding factors are crucial for the phenotype under investigation, their exclusion could inadvertently “reduce noise blindly,” thereby weakening the detection capability and increasing the risk of type I errors.[9] While we appreciate the authors' efforts to address the independence assumption, the importance of statistical power must also be recognized. Moreover, traditional sensitivity analyses only address uncorrelated horizontal pleiotropy, whereas a new method—Causal Analysis Using Summary Effect Estimates (CAUSE) analysis—takes both correlated and uncorrelated horizontal pleiotropy into account, providing a valuable complement in supporting causal evidence.[10]

In summary, we appreciate the efforts made by Du et al in exploring the causal relationship between smoking and VTE and its subtypes. However, obtaining robust evidence of causality requires a meticulous research design and attention to detail. Future studies should therefore consider reanalyzing this topic. Beyond traditional MR analyses, the inclusion of prospective cohort studies and within-family MR designs could facilitate “triangulating,” providing more compelling evidence for causal effects.

Ethical Approval Statement

Each investigation incorporated within the GWAS framework was sanctioned by the relevant ethical review panels. Please see the original article citing the GWAS article for the specific approval documents. No ethics approval was necessary for this research.

Data Availability Statement

The data sources have been shown in detail in the original article.

^* These authors are co-first authors.

Publication History

Received: 01 May 2024

Accepted: 19 May 2024

Article published online:
06 June 2024

• Reference

• 1 Du H-C, Zheng Y-F, Shen M-Q, Deng BY. No genetic causality between tobacco smoking and venous thromboembolism: a two-sample Mendelian randomization study. Thromb Haemost 2024; (e-pub ahead of print) DOI: 10.1055/s-0044-1781425.
• 2 Larsson SC, Mason AM, Bäck M. et al; Million Veteran Program. Genetic predisposition to smoking in relation to 14 cardiovascular diseases. Eur Heart J 2020; 41 (35) 3304-3310
• 3 Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two-sample Mendelian randomization. Genet Epidemiol 2016; 40 (07) 597-608
• 4 Mounier N, Kutalik Z. Bias correction for inverse variance weighting Mendelian randomization. Genet Epidemiol 2023; 47 (04) 314-331
• 5 Chen L, Wei Y, Hu M, Liu Y, Zheng X. Psoriasis may increase the risk of idiopathic pulmonary fibrosis: a two-sample Mendelian randomization study. Respir Res 2024; 25 (01) 101
• 6 Zhou Q, Jin X, Li H. et al. Cholesterol and low-density lipoprotein as a cause of psoriasis: Results from bidirectional Mendelian randomization. J Eur Acad Dermatol Venereol 2024; 38 (04) 710-718
• 7 Carrasquilla GD, García-Ureña M, Romero-Lado MJ, Kilpeläinen TO. Estimating causality between smoking and abdominal obesity by Mendelian randomization. Addiction 2024; 119 (06) 1024-1034
• 8 Brion M-JA, Shakhbazov K, Visscher PM. Calculating statistical power in Mendelian randomization studies. Int J Epidemiol 2013; 42 (05) 1497-1501
• 9 Cho Y, Haycock PC, Sanderson E. et al. Exploiting horizontal pleiotropy to search for causal pathways within a Mendelian randomization framework. Nat Commun 2020; 11 (01) 1010
• 10 Morrison J, Knoblauch N, Marcus JH, Stephens M, He X. Mendelian randomization accounting for correlated and uncorrelated pleiotropic effects using genome-wide summary statistics. Nat Genet 2020; 52 (07) 740-747