Assessing the Causal Effects of Environmental Tobacco Smoke Exposure: A meta-analytic Mendelian randomisation study

doi:10.21203/rs.3.rs-5285519/v1

Download PDF

Article

Assessing the Causal Effects of Environmental Tobacco Smoke Exposure: A meta-analytic Mendelian randomisation study

https://doi.org/10.21203/rs.3.rs-5285519/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Smoking is a major cause of global morbidity and premature mortality. Exposure to environmental tobacco smoke (ETS; “second-hand” or “passive smoking”) may also contribute to ill health. However, it is difficult to establish causality given problems of confounding and reverse causation. We applied Mendelian randomisation to investigate evidence for causal effects. To instrument ETS exposure we used an index individual’s parent’s or spouse’s genetic liability to smoke, conditional on the index individual’s genetic liability. We then meta-analyse four MR approaches using this. Our findings suggest a causal effect of genetically predicted ETS exposure on lung cancer and chronic obstructive pulmonary disease (p_FDR < 0.001 for both). We did not find evidence supporting an effect on hypertension, depression, coronary heart disease, or stroke (p_FDR = 1.000 for all four non-respiratory outcomes); but this might reflect low statistical power. Overall, these results support public health measures to limit exposure to ETS.

Health sciences/Medical research/Epidemiology

Health sciences/Medical research/Genetics research

Social science/Development studies

We assess the causal effects of environmental tobacco smoking (ETS; “second-hand smoking” or “passive smoking”) using a quasi-experimental method, Mendelian randomisation, which can be more robust to confounding than conventional epidemiological methods.

To proxy ETS exposure we used an index individual’s parent’s or spouses’ genetic liability to smoke, conditional on the index’s genetic liability. We then meta-analyse the effects of different relatives.

Our results imply an effect of ETS exposure on lung cancer and COPD, and therefore provide further support for measures to reduce ETS exposure.

Cigarette smoking is associated with numerous adverse health outcomes, ranging from lung cancer to depression (1–7). Many associations between cigarette smoking and adverse health outcomes have been replicated for environmental tobacco smoke exposure (8–12). Environmental tobacco smoke exposure (ETS, also referred to as “passive smoking” or “second-hand smoking”) occurs when an individual breathes in the smoke from another individual’s cigarette smoke.

Establishing causation for ETS exposure is more challenging than for first-hand cigarette smoking. For example, individuals tend to choose friends with a similar smoking status (13), which may induce selection effects in studies of ETS exposure. In addition to well-described problems of confounding and reverse causation, any observed association with ETS exposure could be attributed to someone selecting friends with a similar smoking status and a direct effect of cigarette smoking. This type of selection bias has been shown to be difficult to account for in observational analyses of socially transmissible phenotypes (14). Many ETS studies have also been identified as at risk of differential recall bias (15). In addition, it is difficult to study the effects of exposure to ETS using many quasi-experimental methods because most interventions targeting ETS (e.g., banning smoking in indoor public spaces) may influence the rate of first-hand smoking (e.g., by reducing the rate that someone might smoke). Studies using other methods are therefore warranted.

Mendelian randomisation (MR) leverages the random allocation of genetic variants at conception, analogous to random allocation in a randomised controlled trial (16, 17), to reduce bias due to confounding. Furthermore, because germline genetic variants are fixed at conception, MR reduces susceptibility to reverse causality. Thus, under specific assumptions, MR can strengthen causal inference. The three core assumptions are that the genetic variants robustly associate with the exposure of interest, that there are no variant-outcome confounders; and that the variants influence the outcome only via the exposure of interest (18).

Traditional MR analyses can be biased by ‘indirect genetic effects’. These occur because variants are inherited from parents, and therefore impact on both the child’s and parent’s phenotype. When the parent’s phenotype impacts on their child’s phenotype, the variants’ association with the child’s phenotype will be partially mediated by the effect that a variant has on the parent’s heritable phenotype (19). Recently, MR has been extended to ‘within-family’ settings to address this and other biases. A typical within-family MR study leverages family data to strengthen the validity of the second MR assumption by ensuring that the distribution of genetic variants is truly random (20).

Here, we use family data to explore the causal effects of ETS exposure among relatives on a range of health outcomes. Variants we have inherited from our parents will predict that parent’s phenotype (Fig. 1). Thus, when only parental phenotypic data is available (e.g., via offspring reported questionnaire), a participant’s own genetic data can be used to proxy their parents’ genotype in a ‘proxy gene-by-environment’ MR design (21). Unlike traditional within-family MR studies, which attempt to remove indirect genetic effects, we leverage them within an MR framework to instrument the phenotypic effect of one relative on another. If the direct effect of the inherited variants on the index individual’s phenotype is not accounted for, the resulting MR estimate will be biased (22, 23). We use a novel multivariable extension of ‘proxy gene-by-environment’ MR to account for this.

We used four approaches: the first two tested the effect of each parent’s smoking on their offsprings health outcome separately (direct effect as displayed in Fig. 1), while the other two approaches examined the effects parents have on each other (not displayed in Fig. 1). Each approach used publicly available data – mostly from the UK Biobank but also available genetic consortia. We focused on six outcomes: lung cancer, chronic obstructive pulmonary disease (COPD), stroke, coronary heart disease, hypertension, and depression. The outcomes were chosen as phenotypes measured in the UK Biobank for both the parental and participant’s generation which have previously been identified as being associated with ETS exposure (8–12). After a risk of bias evaluation, approaches which gave plausibly independent estimates were meta-analysed to provide a pooled estimate of the effect of ETS on each outcome.

Overview of Mendelian randomisation approaches

We applied four approaches that use MR to assess the effect of exposure to a relative’s smoking and therefore ETS exposure (Table 1.A), which we then meta-analyse (Table 1.B). The MR analyses were conducted using a multivariable extension of the ‘proxy gene-by-environment’ MR design (21).

Traditional ‘proxy gene-by-environment’ MR designs use the index individual’s genotype as a proxy of a parent’s genotype, and thereby estimate the parental genetic liability to smoke in settings where parental phenotypic but not genetic data is available. This is possible because children inherit half of each parent’s variants, but these variants also affect the respective parental phenotype. Because this approach uses the offspring’s genetic variants, an advantage of its two-sample MR implementation is that outcome data can be extracted from a conventional genome wide association studies (GWAS), which are much larger than current within-family studies. However, because it uses inherited variants, the design can be biased by the direct effect that a variant has on the index individual’s smoking, and therefore outcome (Fig. 1) (22). We address this with multivariable MR (MVMR) by additionally including the index individual’s genetic liability to smoke in the model. MVMR estimates are interpreted as the direct effect of the exposure conditional on the other phenotype. MVMR has been used to adjust MR estimates for postulated biases, by ensuring that the effect of interest is conditionally independent of a phenotype of concern (24–26). We call this application of MVMR ‘multivariable gene-by-environment MR’. We consider six outcomes (lung cancer, COPD, stroke, coronary heart disease, hypertension, and depression).

In the first approach we perform multivariable gene-by-environment MR by including both maternal genetic liability to smoke and the index individual’s liability to smoke as exposures in the model. This approach explores the effect of maternal genetic liability to smoke on the index individual’s outcomes independent of the offspring’s liability to smoke (Table 1.A1).

In the second approach we include the paternal and offspring’s respective liabilities to smoke as exposure phenotypes. This approach assesses the effects of paternal genetic liability to smoke on the index individual’s outcomes conditional on the index’s liability to smoke (Table 1.A2).

The third and fourth approaches include both maternal genetic liability to smoke and paternal liability to smoke as the exposures in a multivariable gene-by-environment MR model. The third approach investigates paternal outcomes, and assesses as the direct effects of maternal genetic liability to smoke on the paternal outcome, independent of paternal liability to smoke (Table 1.A3). The fourth approach investigates maternal outcomes, and assesses the direct effects of paternal genetic liability to smoke on maternal outcomes conditional on maternal genetic liability to smoke (Table 1.A4).

The last two approaches leveraged GWASs (described in the next section and the Supplementary Methods) for maternal smoking and paternal outcomes to explore the effects that one parent have on each other parent. Here the likely source of bias is assortative mating, where parents tend to partner with people who have a similar smoking status. A naive association between one parent’s smoking and other’s outcomes could represent the combined effects of first-hand smoking, and assortative mating (13). Using similar logic to the first two approaches, assortative mating can be controlled for by adjusting for the other parent’s genetic liability to smoke in a multivariable gene-by-environment MR model.

Data sources

Data on maternal smoking and index individual smoking were used publicly available GWAS summary statistics using a sub-sample of genetically unrelated European ancestry participants in the UKB (n ~ 400,000), which have been described elsewhere (27, 28). Variants associated with paternal smoking were also selected using a multivariate European ancestry UKB GWAS (29). Since this GWAS made strong parametric assumptions (30), variant-paternal smoking associations used in the analysis were estimated in ALSPAC (n = 5,766, see Supplementary Methods and (31, 32) for details).

The index individual outcome GWASs combined the UKB with FinnGen r10 and independent, but demographically similar, samples of the same trait taken from available genetic consortia and meta-analyses. These include the International Lung Cancer Consortium for lung cancer, Psychiatric Genetic Consortium for depression, the European sub-sample of the Global Biobank Meta-Analysis Initiative for COPD, CARDIoGRAMplusC4D for coronary heart disease, and the European ancestry sub-sample of GIGASTROKE for stroke. This resulted in 20,359 cases and controls for 810,746 lung cancer; 58,559 cases and 937,358 controls for COPD; 170,756 cases and 329,443 controls for depression; 1,234,808 cases and 1,308,460 controls for stroke; 239,785 cases and 1,355,114 controls for coronary heart disease; and 242,724 case and 649,418 controls for hypertension. Parental outcome data was also taken from European ancestry UKB GWASs of index individual reported maternal and paternal outcomes (see Supplementary Figure S1). To be consistent with the CARDIoGRAMplusC4D study, we use the term “coronary heart disease” when describing the results of our meta-analysis, although the outcome used in the UK Biobank questionnaire of parental disease outcomes used the less precise term “heart disease”.

The data sources, and number of participants and variants, included in each analysis for the 4 MR approaches are summarised in Supplementary Figure S1. Additional details such as information about the data sources and statistical methods used in the MR approaches can be found in the Supplementary Methods and at the following references (28, 33–41).

Instrument construction

For each of the MR approaches, we created a list of variants that were genome-wide significantly associated with either of the smoking phenotypes used in the analysis, which we then clumped (r² = 0.001, kb = 10,000) after ranking SNPs by the smallest p-value that they had with either exposure. The analysis of maternal smoking on the index individual (Table 1.A1) therefore used the maternal smoking and index individual smoking GWASs which resulted in at most 129 variants. Analogously, the procedure resulted in 127 variants for the second approach (Table 1.A2), and 19 variants in the third and fourth approach (Table 1.A3 and 1.A4). Winner’s Curse in the maternal and index individual smoking GWASs was addressed by using false discovery rate inverse quantile transformation Winner’s Curse correction (42).

We used the TwoSampleMR R package to harmonise the exposure and outcome data (43). Palindromic SNPs were only excluded if TwoSampleMR could not use the allele frequency to infer which strand was positive. Where possible, we additionally imputed LD proxy variants with TwoSampleMR when SNPs were missing in the outcome dataset, using an r2 of 0.8 from the European subsample of the 1000 genomes project.

Sensitivity analyses for the MR approaches

We ran three sensitivity analyses for the MR approaches. First, as a positive control, we ran a univariable MR analysis of the index individual’s own cigarette smoking on their own outcomes. This also demonstrates that there was sufficient statistical power to detect an effect on the outcomes. Second, we used hair colour as a negative control outcome for residual population structure. Natural hair colour is known to vary by population sub-group in the UK but should not be causally related to any of the outcomes or smoking status. Therefore, if the any of the genetic instruments used here (i.e., for the index individual’s smoking, maternal smoking, or paternal smoking) associate with hair colour, then this will most likely be due to residual population structure (44). Third, the UK Biobank genotyped participants (recruited into the UK BiLEVE study) at the extreme and middle of the distributions for smoking and lung function were genotyped using a different genotyping chip as compared with the rest of the sample. This could potentially introduce confounding into genotype-phenotype associations, and previous guidance suggests adjusting for the genotyping chip in analyses (27). Although our previous research indicates that there is little bias induced in MR estimates of first-hand smoking (45), we contrast MR estimates using chip adjusted and non-chip-adjusted GWASs for lung-function related phenotypes.

Meta-analysis of MR approaches

A literal interpretation of our MR estimates requires strong, and implausible, assumptions, such as that ETS exposure does not vary across the life course (46). One can alternatively treat MR as a test of a causal null hypothesis (47). In this context the approaches test the joint null hypothesis that exposure to another individual’s genetic liability to smoke does not increase an index individual’s disease risk independent of the index individual’s genetic liability to smoke. It has been similarly argued that meta-analyses of randomised controlled trial should be used to "determine whether or not some type of treatment - tested in a wide range of trials - produces any effect”, rather than “provide exact quantitative estimates” (48). Power to reject the joint null hypothesis can therefore be increased by a quantitative meta-analysis of the four MR estimates.

For such a meta-analysis to be valid we must assume that the estimates are independent of each other. In the context of a two-sample MR analysis, this requires that either the instruments are uncorrelated, the genetically predicted exposures are conditionally independent (by adjustment in the MVMR model), or the outcome samples are independent. The first and second are analogous to a factorial experiment in which people are independently randomised to two separate interventions with the same intended therapeutic effect (e.g., two different drugs that lower blood pressure). The second is equivalent to the assumption of no participant overlap made in traditional meta-analyses of randomised controlled trials.

Supplementary Table S1 presents six pairwise comparisons of each of the MR approaches and the rationale for the independence, or lack of independence, between them. Independence was only questionable for the pair with the two approaches assessing parental smoking and the index individual outcomes. This is because assortative mating and the social transmissibility of smoking mean that one parent’s liability to smoke might predict the other parent’s liability. We accounted for potential within-pair correlations by meta-analysing the unweighted average of the effect estimates and 95% CI of potentially correlated pairs (49, 50). Of the remaining 5 pairs, 2 had both non-overlapping outcome GWASs and independent primary instruments, while 3 only had non-overlapping outcome GWASs.

Since our aim is to test the joint null hypotheses, we follow Peto and others and use a common effect meta-analysis (of the IVW MR estimates) as our primary analyses (51, 52). The summary measure of the association in meta-analyses was the log odds in the outcome per genetically predicted standard deviation increase in exposure to ETS.

Sensitivity and additional analyses for the meta-analysis

Heterogeneity was assessed using the I² and τ ² statistics, and we used the random-effects model as a secondary estimator. We also used a leave-one-out sensitivity analysis to explore the potential effect of, and robustness to, outliers in the primary analysis. Finally, we control for multiple testing across the 6 outcomes using the Benjamini and Hochberg correction (53).

Certainty assessment for the meta-analysis

Uncertainty was evaluated for each outcome using the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) approach for the primary estimate (i.e., common effect meta-analysis) (54, 55). Since our estimates are from theoretically low risk of bias natural experiments, we follow Kim and colleagues and start all meta-analyses at high certainty and then downgrade them where appropriate (56). To aid with the certainty evaluation we developed a bespoke risk of bias tool described in the Supplementary Methods. Although we otherwise adhere to the GRADE guidelines (57), judged imprecision, heterogeneity, and publication bias are inherently somewhat subjective.

Results of individual MR approaches

The results of the MR approaches using the IVW estimator are presented in Fig. 2, and provided in full in Supplementary Table S2. A risk of bias evaluation for each approach stratified by outcome can be found in Supplementary Table S3 and the Supplementary Results. Although the estimates across weak instrument robust estimators tended to be consistent, they had very wide 95% confidence intervals and conditional F statistics were all low. All studies were therefore placed at unclear risk of bias because of potential weak instrument bias.

Results of the meta-analyses

The forest plots for the meta-analyses of results are presented in Fig. 2 with units scaled to the effect of a genetically predicted standard deviation increase in ETS exposure on the log odds of the outcome. We found evidence – in our primary meta-analyses – that greater genetically-predicted exposure to ETS associates with an increase in the odds of developing lung cancer (odds per SD = 4.14 [95% CI: 2.03 to 8.33, p_FDR < 0.001, I² = 61%]) and chronic obstructive pulmonary disease [COPD] (odds per SD = 2.53 [95% CI: 1.65 to 3.90, p_FDR < 0.001, I² = 73%]). We did not find clear evidence of association for other outcomes (heart disease = 0.94 [95% CI: 0.74 to 1.21, I² = 0%, p_crude = 0.652, p_FDR = 1], stroke 0.89 [95% CI: 0.68 to 1.16, I² = 0%, p_crude = 0.396, p_FDR = 1], hypertension = 0.87 [95% CI: 0.66 to 1.14, I² = 0%, p_crude = 0.319, p_FDR = 1], depression = 1.17 [95% CI: 0.81 to 1.70, I² = 0%, p_crude = 0.388, p_FDR = 1]).

Results of sensitivity and additional analyses

Our positive control analysis found evidence of an association between index individual smoking and all index individual outcomes in a univariable MR analysis (Supplementary Table S2). Neither the hair colour negative control analysis nor the comparison between the chip adjusted GWASs and no chip adjusted GWAS MR estimates indicated evidence of bias (Supplementary Table S4 and Supplementary Table S5 respectively).

The leave-one-out analysis indicated that excluding the ‘Maternal smoking on paternal outcomes’ approach results in smaller estimates (which are only indicatively significant for COPD) for the effect of ETS on lung cancer and COPD (Supplementary Figure S2).

Certainty of evidence in meta-analysis

The GRADE evaluation of the certainty of evidence for our meta-analyses can be found in Table 2. We downgrade all outcomes to moderate certainty due to unclear risk of bias from conditional weak instruments. In addition, lung cancer and COPD were downgraded to low certainty due to heterogeneity in the leave-one-out analyses.

We leveraged genetic data in a familial MR framework to investigate the long-term effects of ETS on lung cancer, COPD, stroke, heart disease, hypertension, and depression. After accounting for multiple testing, we find evidence of an association between second-hand smoking and both lung cancer and COPD. According to the GRADE approach, these should be interpreted with low certainty because they are driven by the strong effect of maternal smoking on maternal outcomes. However, (with moderate certainty) we did not find evidence of an association with heart disease, stroke, hypertension, or depression.

Comparison with existing research

Since both first-hand smoking and ETS involve breathing tobacco smoke, it is reasonable to expect the harms of first-hand smoking to generalise, to some extent, to ETS. The measures introduced by the UK government to reduce or ban smoking in public places were, nevertheless, controversial when introduced (58). Our results, on the one hand, lend support to these measures by providing evidence for causal harmful effect of ETS. Consistent with previous studies, our results support an effect of ETS on lung cancer (9, 59) and COPD (60–62). Moreover, Dietrich and colleagues did not observe an association between ETS exposure and blood pressure (63).

The Global Burden of Disease estimated that ETS accounts for around 1.3 million deaths annually, which makes it the 13th leading non-aggregated cause of death (64). It has been claimed that previous estimates of ETS exposure and its attributable risk are, however, either spurious or overestimated (65, 66). The effects of ETS on a per-cigarette basis should be smaller than that of directly inhaled smoke, in particular because ETS is dispersed in the air and therefore substantially diluted. There thus may be less power to detect the effects of ETS. Likewise, we have only explored effects on disease diagnosis. ETS might have sub-threshold effects on non-respiratory outcomes, such as a small reduction in depressive symptoms, which are insufficient to result in substantial greater risk of disease diagnoses. While this could be addressed by assessing underlying continuous measures like blood pressure, data on these outcomes is not available for parents. Previous meta-analyses of observational studies comparing people’s reported ETS exposure to outcome risk, however, have found a significant dose response relationship between time exposed to ETS and risk of depression, stroke, and heart disease (11, 67–73). It is possible that the absence of an association with non-respiratory outcomes observed in the present study reflects a greater robustness to residual confounding, selection bias, and reverse causation.

Two caveats to our results need to be considered. Firstly, the estimates for first-hand smoking on the non-respiratory outcomes were much smaller than the estimates on lung cancer or COPD (Supplementary Table S2). Given the wide 95% confidence intervals in our ETS analyses, it is possible that our study may have been underpowered to detect smaller effects of ETS on these outcomes, and thus we cannot rule out the possibility of a causal link. Secondly, risk of lung cancer from smoke inhalation is cumulative over the life course (74). However, there may be a recovery process for other outcomes, like depression (75), so that once people have stopped smoking for a prolonged period of time their risk starts to converge with never smokers. This could mask a short-term risk increase given a large lag between ETS exposure and outcome measurement.

Strengths and limitations

Multivariable proxy gene-by-environment MR has several theoretical advantages over a conventional observational design. Because biological relatives do not choose each other, the parent-children MR analyses should be more robust to potential selection biases (76). In addition, because genes are fixed at conception, genetically predicted smoking cannot be impacted by the amount of smoking in one’s environment. Thus, our MR estimates should be more robust to reverse causation than an observational study. In addition, many pleiotropic effects occur through variants having multiple biological functions. Since these pleiotropic effects would be mediated by the index individual’s genotype, adjusting for the index individual’s genetic liability to smoke might protect against biological (i.e., horizontal) pleiotropy. It is also unclear which socially mediated (pleiotropic) phenotypes could cause COPD and lung cancer in other people intendent of smoking.

However, like all approaches to causal inference, our MR analysis makes strong assumptions which are not verifiable. Thus, multivariable proxy-MR analyses cannot provide definitive evidence for or against a causal effect. Since it provides a radically different type of evidence to other designs, it is most useful within a triangulation framework. The use of summary level data necessitates making parametric assumptions like linearity and no effect modification. MVMR also requires that we have conditionally strong instruments for the phenotypes. The low conditional F statistics (Supplementary Table S2) are a major factor driving the low certainty in our study results. Since variant-phenotype associations are likely to be similar between parents and their offspring for all biologically plausible phenotypes, conditional weak instrument bias is likely to be a major caveat for any plausible application of the multivariable proxy-MR design. A stronger alternative would be to test if parental non-inherited smoking variants associate with offspring outcomes using individual-level data.

MR approaches are generally less statistically powerful than a traditional observational analysis. We were able to somewhat mitigate this issue by meta-analysing independent MR approaches. This resulted in between 75,000 to 1,358,000 cases, and over a million controls, for each of our outcomes. This is a far larger scale of evidence than would be feasible using stronger designs like an MR analysis using non-inherited variants. Since there are no existing power calculators for multivariable-MR studies we used first-hand smoking as a positive control. However, we cannot exclude the possibility of low power to detect effects,

An additional caveat is the large apparent effect of ETS on lung cancer and COPD, which our meta-analyses (implausibly) imply are only somewhat smaller per standard deviation (SD) than the effects of first-hand smoking. One explanation is that a measured SD exposure to ETS represents less variability in smoke inhalation than a SD of first-hand smoke. However, this cannot explain why inflated estimates are only observed for the maternal-smoking on paternal-outcome analyses for lung cancer and COPD. These outcomes generally develop in later life, and follow a dose-response relationship with smoke exposure. The generation of parents of UKB participants should have larger effects because the parents will be older and exposed to more of each other’s smoke than their offspring. However, other outcomes that also have dose-response relationships with smoking are not inflated. A final explanation is outcome-related differential recall bias: individuals whose parents had lung cancer or COPD might more accurately remember if their mother smoked when they were children (15). Assuming our instruments are otherwise valid, this will inflate estimates under the alternative, but not result in false positives (26). Thus, while we do not think the large estimates represent false positive findings, our MR estimates do not represent the potential quantitative impact of a shift in smoking behaviour in applied practice.

A final limitation pertains to the generalisability of our results. Since our study mostly uses data from the UKB, which has a relatively middle to older aged British population (mean date of birth = 1952), our effect estimates may not be applicable to younger generations or non-European populations. However, because of the similar biological mechanism, we believe that it is reasonable to expect the harmful effect of ETS exposure on the risk of lung cancer and COPD to qualitatively generalise across age, ethnicity, and sex – even if the precise numerical estimates do not (48).

We have meta-analysed a series of MR approaches to assess the causal effect of ETS exposure on lung cancer, COPD, stroke, heart disease, hypertension, and depression. Our findings suggest that there is a causal relationship between exposure to ETS and the development of lung cancer and COPD. This strengthens the evidence supporting the use of public-health interventions to reduce people’s exposure to ETS. However, we did not find evidence of a causal effect of ETS on heart disease, stroke, depression, or hypertension. Our study may have been underpowered to detect smaller effects on these outcomes, and thus we cannot definitively rule out the possibility of a causal link. Until it becomes possible to conduct a better powered MR study, the null associations observed between ETS and non-pulmonary conditions should be interpreted with caution.

Funding

BW is funded by an Economic and Social Research Council (ESRC) South West Doctoral Training Partnership (SWDTP) 1+3 PhD Studentship Award (ES/P000630/1) and the Wellcome Trust (225790/Z/22/Z). The research was supported by the United Kingdom Research and Innovation Medical Research Council (MC_UU_00032/07and MC_UU_00040/01). For the purpose of open access, the authors have applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising.

The UK Medical Research Council and Wellcome (Grant ref: 217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC. This publication is the work of the authors who will serve as guarantors for the contents of this paper.

Author contributions

BW and designed and implemented the study under the supervision of DG, HMS, SB, and MRM. JP performed a code review. All authors contributed to the writing of the manuscript.

Data and data sharing

The data that was used in the primary analyses of this study is publicly available from the MRC-IEU OpenGWAS platform. The R code used in the manuscript, and GWASs created for this project, are available from https://doi.org/10.17605/OSF.IO/BKYXT. This work was carried out using the computational facilities of the Advanced Computing Research Centre, University of Bristol (http://www.bris.ac.uk/acrc/).

This project was conducted using UK Biobank application no. 15825. UK Biobank was established by the Wellcome Trust medical charity, Medical Research Council, Department of Health, Scottish Government and the Northwest Regional Development Agency. It has also had funding from the Welsh Government, British Heart Foundation, Cancer Research UK and Diabetes UK. UK Biobank is supported by the National Health Service (NHS). UK Biobank is open to bona fide researchers anywhere in the world.

The data used in this study is available via application directly to ALSPAC (proposal number B4180). Access is subject to an approved proposal and payment of a data access fee. Proposals can be submitted via the study website (http://www.bristol.ac.uk/alspac/).

Acknowledgments

This work was carried out using the computational facilities of the Advanced Computing Research Centre, University of Bristol - http://www.bris.ac.uk/acrc/. We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. We also want to acknowledge the participants and investigators of the FinnGen and UK Biobank studies.

Ethics Approval statement

The UKB received ethics approval from the North West Multi-Centre Research Ethics Committee (REC reference 11/NW/0382). All participants provided written informed consent to participate in the study. Data from the UKB are fully anonymised.

Ethical approval for the study was obtained from the ALSPAC Ethics and Law Committee and the Local Research Ethics Committees. Consent for biological samples from ALSPAC has been collected in accordance with the Human Tissue Act (2004). Informed consent for the use of data collected via questionnaires and clinics was obtained from participants following the recommendations of the ALSPAC Ethics and Law Committee at the time.

Conflicts of interests

The authors declare no conflicts of interest.

Khuder SA, Milz S, Jordan T, Price J, Silvestri K, Butler P. The impact of a smoking ban on hospital admissions for coronary heart disease. Preventive Medicine. 2007;45(1):3–8.
Doll R, Hill AB. Lung Cancer and Other Causes of Death in Relation to Smoking. Br Med J. 1956;2(5001):1071–81.
Narkiewicz K, Kjeldsen SE, Hedner T. Is smoking a causative factor of hypertension? Blood Pressure. 2005;14(2):69–71.
C. LOWELL F, FRANKLIN W, L. MICHELSON A, W. SCHILLER I. CHRONIC OBSTRUCTIVE PULMONARY EMPHYSEMA: A DISEASE OF SMOKERS*. Annals of Internal Medicine [Internet]. 2008 Dec 1 [cited 2022 Jun 22]; Available from: https://www.acpjournals.org/doi/10.7326/0003-4819-45-2-268
Shah RS, Cole JW. Smoking and stroke: the more you smoke the more you stroke. Expert Review of Cardiovascular Therapy. 2010;8(7):917–32.
Chaiton MO, Cohen JE, O’Loughlin J, Rehm J. A systematic review of longitudinal studies on the association between depression and smoking in adolescents. BMC Public Health. 2009;9(1):356.
Jee SH, Foong AW, Hur NW, Samet JM. Smoking and Risk for Diabetes Incidence and Mortality in Korean Men and Women. Diabetes Care. 2010;33(12):2567–72.
Bernabe-Ortiz A, Carrillo-Larco RM. Second-hand smoking, hypertension and cardiovascular risk: findings from Peru. BMC Cardiovascular Disorders. 2021;21(1):576.
Besaratinia A, Pfeifer GP. Second-hand smoke and human lung cancer. The Lancet Oncology. 2008;9(7):657–66.
Patten SB, Williams JVA, Lavorato DH, Woolf B, Wang JL, Bulloch AGM, et al. Major depression and secondhand smoke exposure. Journal of Affective Disorders. 2018;225:260–4.
Fischer F, Kraemer A. Meta-analysis of the association between second-hand smoke exposure and ischaemic heart diseases, COPD and stroke. BMC Public Health. 2015;15(1):1202.
Glantz SA, Parmley WW. Passive smoking and heart disease. Epidemiology, physiology, and biochemistry. Circulation. 1991;83(1):1–12.
Agrawal A, Heath AC, Grant JD, Pergadia ML, Statham DJ, Bucholz KK, et al. Assortative mating for cigarette smoking and for alcohol consumption in female Australian twins and their spouses. Behav Genet. 2006;36(4):553–66.
Cohen-Cole E, Fletcher JM. Detecting implausible social network effects in acne, height, and headaches: longitudinal analysis. BMJ. 2008;337:a2533.
Levois M, Switzer P. Differential Exposure Misclassification in Case-Control Studies of Environmental Tobacco Smoke and Lung Cancer. Journal of Clinical Epidemiology. 1998;51(1):37–54.
Davies NM, Holmes MV, Smith GD. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. BMJ. 2018;362:k601.
Skrivankova VW, Richmond RC, Woolf BAR, Davies NM, Swanson SA, VanderWeele TJ, et al. Strengthening the reporting of observational studies in epidemiology using mendelian randomisation (STROBE-MR): explanation and elaboration. BMJ. 2021;375:n2233.
Sanderson E, Glymour MM, Holmes MV, Kang H, Morrison J, Munafò MR, et al. Mendelian randomization. Nat Rev Methods Primers. 2022;2(1):1–21.
Howe LJ, Nivard MG, Morris TT, Hansen AF, Rasheed H, Cho Y, et al. Within-sibship genome-wide association analyses decrease bias in estimates of direct genetic effects. Nat Genet. 2022;54(5):581–92.
Davies NM, Howe LJ, Brumpton B, Havdahl A, Evans DM, Davey Smith G. Within family Mendelian randomization studies. Human Molecular Genetics. 2019;28(R2):R170–9.
Yang Q, Millard LAC, Davey Smith G. Proxy gene-by-environment Mendelian randomization study confirms a causal effect of maternal smoking on offspring birthweight, but little evidence of long-term influences on offspring health. International Journal of Epidemiology. 2020;49(4):1207–18.
Hwang LD, Evans DM. Commentary: Proxy gene-by-environment Mendelian randomization for assessing causal effects of maternal exposures on offspring outcomes. International Journal of Epidemiology. 2020;49(4):1218–20.
Havdahl A, Hughes AM, Sanderson E, Ask H, Cheesman R, Reichborn-Kjennerud T, et al. Intergenerational effects of parental educational attainment on parenting and childhood educational outcomes: Evidence from MoBa using within-family Mendelian randomization [Internet]. medRxiv; 2023 [cited 2023 Apr 26]. p. 2023.02.22.23285699. Available from: https://www.medrxiv.org/content/10.1101/2023.02.22.23285699v2
Schooling CM, Lopez PM, Yang Z, Zhao JV, Au Yeung SL, Huang JV. Use of Multivariable Mendelian Randomization to Address Biases Due to Competing Risk Before Recruitment. Frontiers in Genetics [Internet]. 2021 [cited 2022 Apr 6];11. Available from: https://www.frontiersin.org/article/10.3389/fgene.2020.610852
Sanderson E, Spiller W, Bowden J. Testing and correcting for weak and pleiotropic instruments in two-sample multivariable Mendelian randomization. Statistics in Medicine. 2021;40(25):5434–52.
Woolf B, Karhunen V, Yarmolinsky J, Tilling K, Gill D. Re-evaluating the robustness of Mendelian randomisation to measurement error [Internet]. medRxiv; 2022 [cited 2022 Oct 5]. p. 2022.10.02.22280617. Available from: https://www.medrxiv.org/content/10.1101/2022.10.02.22280617v1
Mitchell R, Hemani G, Dudding T, Corbin L, Harrison S, Paternoster L. UK Biobank Genetic Data: MRC-IEU Quality Control, version 2 [Internet]. [cited 2022 Jun 22]. Available from: https://research-information.bris.ac.uk/en/datasets/uk-biobank-genetic-data-mrc-ieu-quality-control-version-2
Wootton RE, Richmond RC, Stuijfzand BG, Lawn RB, Sallis HM, Taylor GMJ, et al. Evidence for causal effects of lifetime smoking on risk for depression and schizophrenia: a Mendelian randomisation study. Psychol Med. 2020;50(14):2435–43.
Woolf B, Sallis H, Munafo M, Gill D. Deriving GWAS summary estimates for paternal smoking in UK Biobank: A GWAS by subtraction [Internet]. OSF Preprints; 2023 [cited 2023 Mar 27]. Available from: https://osf.io/9yeu4/
Evans D, Smith G, Moen GH. Woolf et als GWAS by subtraction is not useful for cross-generational Mendelian randomization studies. 2023.
Boyd A, Golding J, Macleod J, Lawlor DA, Fraser A, Henderson J, et al. Cohort Profile: the ’children of the 90s’--the index offspring of the Avon Longitudinal Study of Parents and Children. Int J Epidemiol. 2013;42(1):111–27.
Fraser A, Macdonald-Wallis C, Tilling K, Boyd A, Golding J, Davey Smith G, et al. Cohort Profile: the Avon Longitudinal Study of Parents and Children: ALSPAC mothers cohort. Int J Epidemiol. 2013;42(1):97–110.
Howard DM, Adams MJ, Clarke TK, Hafferty JD, Gibson J, Shirali M, et al. Genome-wide meta-analysis of depression identifies 102 independent variants and highlights the importance of the prefrontal brain regions.: Supplementary Information [Internet]. Genetics; 2018 Oct [cited 2019 Apr 17]. Available from: http://biorxiv.org/lookup/doi/10.1101/433367
Kurki MI, Karjalainen J, Palta P, Sipilä TP, Kristiansson K, Donner KM, et al. FinnGen provides genetic insights from a well-phenotyped isolated population. Nature. 2023;613(7944):508–18.
Kimberley Burrows PH. data.bris. 2021 [cited 2022 Jun 28]. Genome-wide Association Study of Cancer Risk in UK Biobank. Available from: https://data.bris.ac.uk/data/dataset/aed0u12w0ede20olb0m77p4b9
Wang Y, McKay JD, Rafnar T, Wang Z, Timofeeva MN, Broderick P, et al. Rare variants of large effect in BRCA2 and CHEK2 affect risk of lung cancer. Nat Genet. 2014;46(7):736–41.
Elsworth B, Lyon M, Alexander T, Liu Y, Matthews P, Hallett J, et al. The MRC IEU OpenGWAS data infrastructure [Internet]. bioRxiv; 2020 [cited 2022 Mar 30]. p. 2020.08.10.244293. Available from: https://www.biorxiv.org/content/10.1101/2020.08.10.244293v1
Ruth Mitchell E. data.bris. 2019 [cited 2022 Mar 12]. MRC IEU UK Biobank GWAS pipeline version 2. Available from: https://data.bris.ac.uk/data/dataset/pnoat8cxo0u52p6ynfaekeigi
Aragam KG, Jiang T, Goel A, Kanoni S, Wolford BN, Atri DS, et al. Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants. Nat Genet. 2022;54(12):1803–15.
Zhou W, Kanai M, Wu KHH, Rasheed H, Tsuo K, Hirbo JB, et al. Global Biobank Meta-analysis Initiative: Powering genetic discovery across human disease. Cell Genom. 2022;2(10):100192.
Mishra A, Malik R, Hachiya T, Jürgenson T, Namba S, Posner DC, et al. Stroke genetics informs drug discovery and risk prediction across ancestries. Nature. 2022;611(7934):115–23.
Bigdeli TB, Lee D, Riley BP, Vladimirov V, Fanous AH, Kendler KS, et al. FIQT: a simple, powerful method to accurately estimate effect sizes in genome scans [Internet]. bioRxiv; 2015 [cited 2022 Mar 12]. p. 019299. Available from: https://www.biorxiv.org/content/10.1101/019299v1
Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, et al. The MR-Base platform supports systematic causal inference across the human phenome. eLife. 2018;7:e34408.
Sanderson E, Richardson TG, Hemani G, Davey Smith G. The use of negative control outcomes in Mendelian randomization to detect potential population stratification. International Journal of Epidemiology. 2021;50(4):1350–61.
Woolf B, Gill D, Sallis H, Munafo M. The UK BiLEVE and Mendelian randomisation: Using multivariable instrumental variables to address “damned if you, dammed if you don’t” adjustment problems. [Internet]. medRxiv; 2022 [cited 2022 Oct 28]. p. 2022.10.25.22281084. Available from: https://www.medrxiv.org/content/10.1101/2022.10.25.22281084v1
Labrecque JA, Swanson SA. Interpretation and Potential Biases of Mendelian Randomization Estimates With Time-Varying Exposures. Am J Epidemiol. 2019;188(1):231–8.
VanderWeele TJ, Tchetgen Tchetgen EJ, Cornelis M, Kraft P. Methodological challenges in mendelian randomization. Epidemiology. 2014;25(3):427–35.
Peto R. Why do we need systematic overviews of randomized trials? Statist Med. 1987;6(3):233–40.
Harrer M, Cuijpers P, Furukawa TA, Ebert DD. Doing Meta-Analysis in R [Internet]. Chapman & Hall/CRC; 2021 [cited 2022 Apr 23]. Available from: https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/
Cheung MWL. A Guide to Conducting a Meta-Analysis with Non-Independent Effect Sizes. Neuropsychol Rev. 2019;29(4):387–96.
Peto R, Awasthi S, Read S, Clark S, Bundy D. Vitamin A supplementation in Indian children – Authors’ reply. The Lancet. 2013;382(9892):594–6.
Senn S. The Many Modes of Meta. Drug Information Journal. 2000;34(2):535–49.
Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society: Series B (Methodological). 1995;57(1):289–300.
Guyatt GH, Oxman AD, Kunz R, Brozek J, Alonso-Coello P, Rind D, et al. GRADE guidelines 6. Rating the quality of evidence–imprecision. J Clin Epidemiol. 2011;64(12):1283–93.
Zhang Y, Akl EA, Schünemann HJ. Using systematic reviews in guideline development: the GRADE approach. Res Synth Methods. 2018;
Kim MS, Kim WJ, Khera AV, Kim JY, Yon DK, Lee SW, et al. Association between adiposity and cardiovascular outcomes: an umbrella review and meta-analysis of observational and Mendelian randomization studies. Eur Heart J. 2021;42(34):3388–403.
Iorio A, Spencer FA, Falavigna M, Alba C, Lang E, Burnand B, et al. Use of GRADE for assessment of evidence about prognosis: rating confidence in estimates of event rates in broad categories of patients. BMJ. 2015;350:h870.
Policy Navigator [Internet]. [cited 2022 Nov 3]. Smoking ban in England. Available from: https://navigator.health.org.uk/theme/smoking-ban-england
Hori M, Tanaka H, Wakai K, Sasazuki S, Katanoda K. Secondhand smoke exposure and risk of lung cancer in Japan: a systematic review and meta-analysis of epidemiologic studies. Japanese Journal of Clinical Oncology. 2016;46(10):942–51.
Pando-Sandoval A, Ruano-Ravina A, Candal-Pedreira C, Rodríguez-García C, Represas-Represas C, Golpe R, et al. Risk factors for chronic obstructive pulmonary disease in never-smokers: A systematic review. The Clinical Respiratory Journal. 2022;16(4):261–75.
Putcha N, Barr RG, Han MK, Woodruff PG, Bleecker ER, Kanner RE, et al. Understanding the impact of second-hand smoke exposure on clinical outcomes in participants with COPD in the SPIROMICS cohort. Thorax. 2016;71(5):411–20.
Lee PN, Forey BA, Coombs KJ, Hamling JS, Thornton AJ. Epidemiological evidence relating environmental smoke to COPD in lifelong non-smokers: a systematic review. F1000Res. 2020;7:146.
Felber Dietrich D, Schwartz J, Schindler C, Gaspoz JM, Barthélémy JC, Tschopp JM, et al. Effects of passive smoking on heart rate variability, heart rate and blood pressure: an observational study§. International Journal of Epidemiology. 2007;36(4):834–40.
Institute for Health Metrics and Evaluation [Internet]. 2020 [cited 2022 Nov 3]. Secondhand smoke — Level 3 risk. Available from: https://www.healthdata.org/results/gbd_summaries/2019/secondhand-smoke-level-3-risk
Peres J. No Clear Link Between Passive Smoking and Lung Cancer. JNCI: Journal of the National Cancer Institute. 2013;105(24):1844–6.
Sir Richard Peto Presentation April 2016 [Internet]. 2016 [cited 2022 Nov 3]. Available from: https://www.youtube.com/watch?v=egr7oyKWEOM
Barnoya J, Glantz SA. Cardiovascular Effects of Secondhand Smoke. Circulation. 2005;111(20):2684–98.
Ye X, Li L, Gao Y, Zhou S, Yang Y, Chen S. Dose–response relations between second-hand smoke exposure and depressive symptoms among middle-aged women. Psychiatry Research. 2015;229(1):533–8.
Zhang D, Liu Y, Cheng C, Wang Y, Xue Y, Li W, et al. Dose-related effect of secondhand smoke on cardiovascular disease in nonsmokers: Systematic review and meta-analysis. International Journal of Hygiene and Environmental Health. 2020;228:113546.
Han C, Liu Y, Gong X, Ye X, Zhou J. Relationship between Secondhand Smoke Exposure and Depressive Symptoms: A Systematic Review and Dose–Response Meta-Analysis. International Journal of Environmental Research and Public Health. 2019;16(8):1356.
Christian F, Kim Y. Association Between Place of Secondhand Smoke Exposure and Depression Among Nonsmoking Children and Adolescents: A Systematic Review. Asia Pac J Public Health. 2022;34(6–7):616–26.
Oono IP, Mackay DF, Pell JP. Meta-analysis of the association between secondhand smoke exposure and stroke. Journal of Public Health. 2011;33(4):496–502.
Pan B, Jin X, Jun L, Qiu S, Zheng Q, Pan M. The relationship between smoking and stroke. Medicine (Baltimore). 2019;98(12):e14872.
Armitage P, Doll R. The age distribution of cancer and a multi-stage theory of carcinogenesis. Br J Cancer. 2004;91(12):1983–9.
Change in mental health after smoking cessation: systematic review and meta-analysis | The BMJ [Internet]. [cited 2024 Oct 14]. Available from: https://www.bmj.com/content/348/bmj.g1151
Cawley J, Han E, Kim J (June), Norton EC. Testing for Peer Effects Using Genetic Data [Internet]. National Bureau of Economic Research; 2017 [cited 2022 Sep 12]. (Working Paper Series). Available from: https://www.nber.org/papers/w23719

Table 1 is available in the Supplementary Files section.

Table 2: GRADE evaluation of the certainty in the evidence for our six outcomes.

GRADE domain	Lung cancer and Chronic obstructive pulmonary disease	Stroke, Cardiovascular disease, Hypertension, Depression
Overall GRADE evaluation	Low certainty	Moderate certainty
Heterogeneity and inconsistency	Heterogeneity in the leave-one-out analysis (downgrade).	No heterogeneity in the primary meta-analysis or leave-one-out (no downgrade).
Risk of bias	A full risk of bias evaluation can be found in Supplementary Table S2. All designs were downgraded to unclear risk of bias due to very low conditional F-statistics. Although the weak-instrument robust estimators had consistent estimates, they also tended to have wide confidence intervals. A downgrade to unclear risk of bias was therefore deemed appropriate to reflect the uncertainty in the estimates (downgrade).
Imprecision	There are no currently existing power calculations for multivariable MR studies. We instead used the positive control analysis of first-hand smoking as a benchmark. This should ensure sufficient power to detect effects at least as large as first-hand smoking, although there may still be low power to detect much smaller effects (no downgrade).
Indirectness	All studies estimate the effect of ETS in a population of adult Europeans (no downgrade).
Publication bias	Publication bias occurs from research not included in a literature search. Since our meta-analysis was not conducted after a literate review, but to combine primary analyses, publication bias should not occur (no downgrade).

There is NO Competing Interest.

Download PDF

Version 1

posted

You are reading this latest preprint version

Assessing the Causal Effects of Environmental Tobacco Smoke Exposure: A meta-analytic Mendelian randomisation study

Status:

Version 1

Abstract

Figures

Key Message

Introduction

Methods

Overview of Mendelian randomisation approaches

Data sources

Instrument construction

Sensitivity analyses for the MR approaches

Meta-analysis of MR approaches

Sensitivity and additional analyses for the meta-analysis

Certainty assessment for the meta-analysis

Results

Results of individual MR approaches

Results of the meta-analyses

Results of sensitivity and additional analyses

Certainty of evidence in meta-analysis

Discussion

Comparison with existing research

Strengths and limitations

Conclusion

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1