Selection bias is unlikely to fully explain the protective effect of childhood adiposity on breast cancer risk

Triangulation of evidence to examine selection bias in lifecourse Mendelian randomization studies: an example using early life adiposity and breast cancer | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Triangulation of evidence to examine selection bias in lifecourse Mendelian randomization studies: an example using early life adiposity and breast cancer View ORCID Profile Grace M. Power , View ORCID Profile Eleanor Sanderson , Apostolos Gkatzionis , Tom G. Richardson , Kate Tilling , George Davey Smith , View ORCID Profile Gibran Hemani doi: https://doi.org/10.1101/2025.09.10.25335479 Grace M. Power 1 MRC Integrative Epidemiology Unit, University of Bristol , Bristol, UK 2 Population Health Sciences, Bristol Medical School, University of Bristol , UK 3 Institute for Molecular Bioscience, The University of Queensland , Brisbane, Queensland, Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Grace M. Power For correspondence: grace.power{at}bristol.ac.uk Eleanor Sanderson 1 MRC Integrative Epidemiology Unit, University of Bristol , Bristol, UK 2 Population Health Sciences, Bristol Medical School, University of Bristol , UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Eleanor Sanderson Apostolos Gkatzionis 1 MRC Integrative Epidemiology Unit, University of Bristol , Bristol, UK 2 Population Health Sciences, Bristol Medical School, University of Bristol , UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Tom G. Richardson 1 MRC Integrative Epidemiology Unit, University of Bristol , Bristol, UK 2 Population Health Sciences, Bristol Medical School, University of Bristol , UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kate Tilling 1 MRC Integrative Epidemiology Unit, University of Bristol , Bristol, UK 2 Population Health Sciences, Bristol Medical School, University of Bristol , UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site George Davey Smith 1 MRC Integrative Epidemiology Unit, University of Bristol , Bristol, UK 2 Population Health Sciences, Bristol Medical School, University of Bristol , UK 4 NIHR Bristol Biomedical Research Centre Bristol, University Hospitals Bristol and Weston NHS Foundation Trust, University of Bristol , Bristol, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Gibran Hemani 1 MRC Integrative Epidemiology Unit, University of Bristol , Bristol, UK 2 Population Health Sciences, Bristol Medical School, University of Bristol , UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Gibran Hemani Abstract Full Text Info/History Metrics Supplementary material Data/Code Preview PDF Abstract Background Higher adiposity in early life has consistently been associated with a reduced risk of breast cancer in later life, with Mendelian randomization (MR) studies supporting a potential causal effect. However, concerns have been raised that selection bias, particularly collider stratification due to selective participation or survival, may induce spurious protective effects. Methods We used a triangulation framework combining empirical analyses and simulations to evaluate whether selection-induced bias could plausibly explain the inverse effect of early life body size on breast cancer risk. First, we re-examined proxy-genotype MR analyses and conducted family-based simulations to assess whether attenuation in relative-based estimates could arise without selection bias. Second, we performed multivariable MR analyses of parental survival to evaluate survival-related selection mechanisms. Third, we conducted extensive simulations under a null causal model to quantify the magnitude of bias introduced by selection under a range of plausible and extreme scenarios, including interaction-driven selection. Results Attenuation in proxy-genotype MR estimates was reproduced in simulations without selection bias, indicating that this pattern does not provide evidence for selection bias. Multivariable MR analyses of parental survival indicated that survival differences are primarily driven by adulthood, not childhood, adiposity, providing little support for survival-related selection acting through childhood body size. In simulation analyses, additive selection produced minimal bias, while interaction-driven selection generated increasing distortion; however, even under extreme scenarios, the magnitude of bias was insufficient to replicate the observed protective effect. When selection operated through adulthood body size, bias was confined largely to adulthood estimates. Across all scenarios, the joint pattern of univariable and multivariable MR findings was not reproduced under selection alone. Conclusions Although selection bias can influence MR estimates, our findings suggest that plausible selection mechanisms are unlikely to fully explain the observed inverse effect of early life adiposity on breast cancer risk. These results support a causal interpretation of the protective effect and highlight the value of triangulating evidence across complementary approaches when evaluating bias in lifecourse MR. Background Higher levels of adiposity in early life have consistently been observed to be inversely associated with breast cancer risk later in life across multiple lines of evidence [ 1 - 12 ]. Mendelian randomization (MR) studies, which use germline genetic variants as instrumental variables, strengthen the case for a causal interpretation by mitigating confounding and reverse causation [ 13 ]. In univariable MR analyses, both higher earlier life and higher mid-to-late adulthood body size have been estimated to exert protective causal effects on breast cancer risk [ 1 - 6 ]. In lifecourse MR applications, causal effects of the same trait measured at different life stages can be separated using a multivariable MR framework [ 14 , 15 ]. Within this framework, the apparent protective effect of mid-to-late adulthood body size attenuates, whereas the direct effect of early life body size persists after adjustment [ 1 - 6 ]. These findings have prompted debate that selection processes could induce spurious protective effects under a null causal model [ 16 , 17 ]. Existing critiques of the childhood adiposity-breast cancer association are framed primarily in terms of collider stratification bias [ 18 ], whereby conditioning on selection influenced by both childhood adiposity (or its causes) and breast cancer risk (or its causes) can induce a non-causal association, even if the true causal effect is null. Accordingly, here we use the term selection bias to refer specifically to this collider-driven mechanism, while recognising that selection bias can arise through other forms of selection or stratification [ 19 , 20 ]. The magnitude of such collider-induced bias depends on the underlying causal structure and the presence of interactions influencing selection, and may be underestimated if these are ignored [ 21 , 22 ]. Bias arising from selection on an effect modifier (i.e., when study participation depends on a variable that changes the size of the exposure effect), which can occur even in the absence of collider structures [ 23 ], is not considered here, as it cannot operate under the sharp null and is therefore not relevant to this evaluation. Since early methodological discussions of selection bias in epidemiology, it has been emphasised that the possibility of selection bias should not be a barrier to future research but should instead motivate careful evaluation of whether plausible selection mechanisms are sufficient to explain observed findings [ 19 ]. Rather than assuming that selection bias necessarily invalidates lifecourse MR results, this commentary uses triangulation across several complementary approaches to assess the role selection-induced collider stratification is likely to play in this setting. Data sources for empirical analyses are described in Note S1; Table S1. 1. Proxy-genotype Mendelian randomization (MR) analyses Work using proxy-genotype MR, in which offspring genotype is used to proxy parental genotype, reported weaker associations of childhood body size with breast cancer risk in first-degree relatives than with an individual’s own breast cancer risk [ 16 , 17 ]. This attenuation was interpreted as evidence that the inverse association observed for own breast cancer may be driven by selection or survival bias, on the basis that family-history outcomes capture cases occurring before recruitment. However, such attenuation is expected under Mendelian inheritance and measurement dilution when offspring genotype is used as an imperfect proxy for parental genotype [ 11 ]. We re-performed the summary-based MR analysis to obtain an estimated influence of childhood body size on own breast cancer, as well as the proxy-MR estimates of own childhood body size on sibling’s breast cancer and mother’s breast cancer. We then conducted a family-based simulation designed to have the same causal effect of childhood body size on own breast cancer for mothers, siblings and probands, obtaining self-MR and proxy-MR estimates within the simulation framework. No selection bias was introduced in the simulation framework. We show that the observed pattern of estimates for self, sibling, and mother designs is reproduced within the basic proxy-MR framework and does not require selection bias to explain it (Figure S1). 2. Evaluating survival-related selection via parental survival analyses A further line of critique has suggested that the apparent protective effect of childhood adiposity on breast cancer could arise through survival-related selection, whereby childhood body size influences mortality prior to recruitment [ 16 , 17 ]. However, the genetic influences on childhood and adulthood body size are correlated, so univariable estimates may be misleading. We conducted multivariable MR analyses using UK Biobank data described in Note S1, applying the same childhood and adulthood body size instruments as those used by Richardson et al. [ 1 ] to assess whether childhood body size exerts a direct effect on survival outcomes after accounting for adulthood body size. Parental survival reflects inherited differences in survival and therefore provides a test of whether survival-related selection acting through these instruments could generate protective effect estimates of the type observed for breast cancer. Using MR analyses of parental survival, with offspring genotype used as a proxy for parental genotype, we found that higher adulthood body size exerted a robust adverse effect on survival in both univariable and multivariable models, whereas the effect estimate for childhood body size observed in univariable analyses fully attenuated after adjustment for adulthood body size (Note S2). These results indicate that survival differences are driven primarily by adulthood rather than childhood adiposity. This suggests that survival-related selection acting through childhood body size is unlikely to explain the estimated protective effect on breast cancer risk. Full results are provided in Table S2. This interpretation is further supported by consistent observational findings which show similar patterns and are unlikely to be materially influenced by selection bias [ 7 , 12 ]. 3. Evaluating selection-induced collider stratification Selection bias mechanisms can theoretically generate any relationship, and there is particular concern in settings such as the UK Biobank, where participation is selective (5.5% response rate) and recruitment occurs in mid-to-late adulthood [ 24 ]. However, whether such processes could plausibly generate the magnitude and pattern of effects observed in lifecourse MR is the key question, and we conducted simulations to evaluate this. Because of the scale and two-sample design of the empirical lifecourse MR analyses, reproducing selection mechanisms that perfectly match observed parameters is not easily tractable. Instead, we use simulations to evaluate whether collider-based selection processes, operating with or without interaction, could reproduce the magnitude and pattern of lifecourse MR estimates reported for childhood body size and breast cancer (Table S3). Data were generated under a sharp null model, assuming no true causal effect of either childhood or adulthood body size on breast cancer risk. The analyses are intended to be illustrative rather than definitive, with full methodological details and results provided in the Supplementary Material. To inform the specification of these simulations, we first examined the influence of confounding in conventional observational analyses (Note S3). In minimally adjusted logistic regression analyses of UK Biobank data, higher childhood body size remained inversely associated with breast cancer risk after adjustment for adulthood body size and age. The magnitude of this association was smaller than that observed in the MR analyses (Note S4), consistent with attenuation of observational estimates arising from measurement error in retrospectively reported childhood body size, which instrumental variable approaches are less susceptible to under standard assumptions.. However residual confounding could still be making a contribution to the observational association and accordingly, we include an unmeasured confounder in the simulation framework to reflect this empirically informed structure. 3a. Magnitude of bias that could arise under simulated selection mechanisms To assess the strength of selection required to reproduce the magnitude of the protective lifecourse MR estimate under a null causal model, we conducted simulations in which childhood and adulthood body size were modelled as latent continuous traits instrumented by correlated genetic risk scores, with a direct tracking path from childhood to adulthood to capture lifecourse persistence (Note S5). Breast cancer status was generated independently of body size, such that the true causal effects of childhood and adulthood body size were set to zero; any estimated effects therefore reflect bias introduced by selection. Selection into the analytic sample was modelled as a function of childhood body size and breast cancer status, with and without interaction ( Figure 1 ). We examined 27 scenarios spanning no selection, additive selection, and interaction-driven selection, with parameters varied from null to magnitudes exceeding those typically observed empirically, in order to assess upper-bound collider bias (Tables S5 and S6). Lifecourse MR analyses were performed in samples corresponding to the number of female UK Biobank participants with complete data (n = 246,511), with 500 simulation replicates per setting. Download figure Open in new tab Figure 1. Schematic of the selection mechanism in simulations. Childhood ( Z 1 ) and adulthood ( Z 2 ) genetic risk scores (GRS) influence latent childhood ( X 1 ) and adulthood ( X 2 ) body size, respectively, with a genetic correlation (dotted line) between Z 1 and Z 2 . Childhood body size ( X 1 ) and breast cancer status (Y) influence selection into the analytic sample ( S ), representing either differential survival prior to recruitment or differential participation among individuals alive at recruitment. An unmeasured confounder ( U ) influences childhood body size, adulthood body size, and breast cancer risk, allowing assessment of whether conditioning on S induces or amplifies collider bias through non-causal pathways. The strength of confounding was varied across two levels: Mild, where U was weakly associated with these variables, and Strong, where U had stronger associations with them. Empirical estimates of disease-related selection into UK Biobank informed the lower range of interaction parameters considered (Note S6; Figure S2; Table S7). For breast cancer in women, cases are slightly over-represented in UK Biobank relative to the general population, with modest age-related interaction. Parameter values in the simulations were therefore extended beyond these empirical magnitudes, including configurations implying substantial case under-representation, in order to evaluate whether extreme selection structures could generate MR effect estimates comparable to those observed. The simulations thus represent a stress test of the selection hypothesis rather than an attempt to reproduce the precise selection mechanism operating in UK Biobank. Consistent with the multivariable MR analyses of parental survival, which indicated that survival-related differences are primarily associated with adulthood rather than childhood adiposity, we conducted a sensitivity analysis in which selection was specified to operate through adulthood body size and breast cancer status, with childhood selection fixed at zero. In the primary simulations where selection depended on childhood body size and breast cancer status ( Figure 1 ), additive selection produced little distortion of childhood MR estimates (Table S8). Interaction-dependent selection induced progressively more negative childhood estimates as interaction magnitude increased ( Figure 2A ), and the presence of confounding amplified this distortion. However, even under the most extreme combinations of interaction and confounding evaluated, simulated childhood estimates remained less extreme than the empirical lifecourse MR estimate (logOR =-0.53). Download figure Open in new tab Figure 2. Simulated bias under under varying selection and unmeasured confounding scenarios. A. Mean log(OR) MR estimates for childhood (solid) and adult (dashed) body size on breast cancer across 500 replicates are shown under increasing strengths of unmeasured confounding (U) affecting both body size and breast cancer risk. Horizontal panels represent different confounding strengths (no, mild, strong), and vertical panels represent increasing levels of interaction-driven selection. Error bars indicate the 2.5th-97.5th percentile range. The red dashed line marks the observed childhood MR estimate (logOR =-0.53). B. UpSet plot showing how often simulation scenarios produce results that overlap with empirical results across four MR estimates. None of the simulations gave rise to childhood and adulthood effect estimates matching for both univariable and multivariable methods. When selection was instead specified to operate through adulthood body size, bias was expressed primarily in adulthood MR estimates, while childhood estimates remained centred near the null across scenarios (Figure S4). This contrast indicates that bias tracks the exposure driving selection. Given that survival-related differences appear to be driven predominantly by adulthood adiposity, a structure in which selection operates primarily through adulthood would be expected to distort adulthood estimates rather than generate a spurious protective childhood effect of the magnitude observed. Results from simulations without confounding are presented as a baseline in Figure 2A , with coverage of the empirical MR estimate under selection alone shown in Figure S3. 3b. Recapitulating univariable and multivariable MR patterns under simulated selection mechanisms Another way to examine the potential biasing role of selection mechanisms is to examine how bias-induced marginal effect estimates in univariable MR get adjusted in multivariable MR. In this case we are examining if there are strong assocations for childhood and adulthood body size on breast cancer induced by selection bias in univariable MR, in which only the adulthood body size effect gets attenuated in multivariable MR. To examine this we extended the previous simulation study to also perform multivariable MR. Therefore each simulation produces two univariable MR estimates and two multivariable MR estimates. In addition to simulating no true effect of adult or childhood body size on cancer, we also included scenarios involving a moderate influence of adult body size on cancer. A ‘match’ in the simulation is when an effect estimate overlaps with a target estimate. For example a 10% match would relate to the simulation estimate residing within the 10% confidence interval of the target estimate. Our interest is in understanding how likely it is to have all four estimates ‘matching’ simply due to selection bias. For the purpose of illustration we used the empirical results from Richardson et al. (Table S3) as target estimates for the simulation, though we again note that a true sensitivity analysis would aim to fix simulations to match observed parameters [ 25 ], which we do not achieve here. Across 781,250 simulation scenarios, none yielded all four estimates consistent with the empirical targets (defined as satisfying the matching criteria above; Figure 2B ). Discussion Selection bias is a recognised concern in conventional epidemiology and MR, particularly in large cohort studies with selective participation and recruitment in mid-to-late adulthood. Such concerns warrant careful evaluation of whether plausible selection mechanisms could meaningfully distort effect estimates, rather than presuming that observed associations are artefactual. Using triangulation across complementary lines of evidence (Table S9), we examined whether selection-induced collider stratification could plausibly explain the inverse effect of childhood body size on breast cancer risk reported in lifecourse MR and conventional epidemiology studies. An important line of evidence comes from a prospective childhood cohort in which BMI was measured directly in school health examinations and participants were followed through linkage to national cancer registries, which demonstrates similar protective associations [ 7 , 10 ]. Because exposure was recorded in childhood and outcomes were ascertained through population-based registry linkage rather than voluntary adult recruitment, these findings are less susceptible to participation mechanisms specific to adult cohort studies. Genetically informed analyses provide further context. Attenuation observed in proxy-genotype MR analyses of breast cancer in relatives is consistent with Mendelian inheritance and measurement dilution and can be reproduced in family-based simulations without introducing selection bias. MR analyses of parental survival indicate that survival-related differences associated with adiposity are primarily attributable to adulthood rather than childhood body size, limiting support for a childhood-driven survival selection explanation of the breast cancer finding. Simulation analyses conducted under a sharp null model assessed the magnitude of bias that could arise under explicit selection structures. Additive selection on childhood body size and breast cancer status produced little distortion of childhood MR estimates. Interaction parameters were varied beyond typical empirical magnitudes to approximate upper-bound selection effects; even under these scenarios, and with confounding included, simulated childhood estimates remained smaller than the empirical effect. When selection was instead specified to operate through adulthood body size, bias was expressed primarily in adulthood MR estimates, while childhood estimates remained close to the null. This pattern, consistent with the parental longevity analyses indicating survival-related differences are driven predominantly by adulthood adiposity, suggests that adulthood-driven selection does not account for the protective childhood MR association with breast cancer. Across extended simulations allowing more complex interaction structures, the joint empirical pattern of univariable and multivariable MR estimates was not reproduced under selection alone. Although the simulations were not parameterised to reproduce the exact selection structure of UK Biobank, parameters were intentionally varied beyond empirically typical magnitudes to evaluate whether extreme selection configurations could generate effects comparable to those observed. Taken together, these findings indicate that selection mechanisms, while capable of inducing bias, do not adequately account for the magnitude and structure of the observed protective childhood MR effect for breast cancer. Alternative explanations for this inverse effect have been proposed. In particular, higher adiposity during early life has been linked to differences in circulating sex hormone levels and lower mammographic breast density in adulthood, each of which has been implicated in breast cancer risk [ 5 , 6 ]. Ongoing work aims to further clarify these pathways and the extent to which early-life adiposity may influence breast tissue development and hormonal exposures across the lifecourse. Data Availability All code used to generate the simulations and reproduce the results reported here is publicly available at https://github.com/gracemarionpower/selection-bias-simulations/ . The data underlying the findings can be derived directly from these simulations. Sources of funding GMP, ES, AG, KT, GH, and GDS were supported by the Integrative Epidemiology Unit which receives funding from the UK Medical Research Council and the University of Bristol (MC_UU_00032/01 for GMP, ES, GH and GDS; MC_UU_00032/02 for AG and KT). GMP was additionally supported by the University of Bristol Cancer research fund for this work. ES is supported by the MRC (UKRI077). GDS conducts research at the NIHR Biomedical Research Centre at the University Hospitals Bristol NHS Foundation Trust and the University of Bristol. Research Ethics and Informed Consent The UK Biobank study have obtained ethics approval from the Research Ethics Committee (REC; approval number: 11/NW/0382). The UK Biobank study have obtained informed consent from all participants enrolled in UK Biobank. Empirical estimates were derived using data from the UK Biobank (app #81499). The remaining aspects of this study were based entirely on simulated data and did not involve human participants directly. Conflicts of interest TGR is a full-time employee of GlaxoSmithKline outside of this research. All other authors declare no conflict of interest. Data and computing code availability All code used to generate the simulations and reproduce the results reported here is publicly available at https://github.com/gracemarionpower/selection-bias-simulations/ . Individual-level data used to derive these results can be obtained with an approved application to the UK Biobank study ( http://www.ukbiobank.ac.uk ). The data underlying the simulated findings can be derived directly from these simulations. Footnotes ↵ * Joint last authors Revision of the manuscript structure to better align with a commentary-style format. References 1. ↵ Richardson , T.G. , et al. , Use of genetic variation to separate the effects of early and later life adiposity on disease risk: mendelian randomisation study . BMJ , 2020 . 369 : p. m1203 . OpenUrl Abstract / FREE Full Text 2. Hao , Y. , et al. , Reassessing the causal role of obesity in breast cancer susceptibility: a comprehensive multivariable Mendelian randomization investigating the distribution and timing of exposure . Int J Epidemiol , 2023 . 52 ( 1 ): p. 58 – 70 . OpenUrl PubMed 3. Power , G.M. , et al. , Lifecourse genome-wide association study meta-analysis refines the critical life stages for adiposity’s influence on breast cancer risk . Science Advances , 2026 . 12 ( 4 ): p. eady0374 . OpenUrl PubMed 4. Power , G.M. , et al. , A structural mean modelling Mendelian randomization approach to investigate the lifecourse effect of adiposity: applied and methodological considerations . American Journal of Epidemiology , 2025 . 5. ↵ Vabistsevits , M. , et al. , Deciphering how early life adiposity influences breast cancer risk using Mendelian randomization . Commun Biol , 2022 . 5 ( 1 ): p. 337 . OpenUrl PubMed 6. ↵ Vabistsevits , M. , et al. , Mammographic density mediates the protective effect of early-life body size on breast cancer risk . Nature Communications , 2024 . 15 ( 1 ): p. 4021 . OpenUrl PubMed 7. ↵ Jensen , B.W. , et al. , Childhood body mass index trajectories, adult-onset type 2 diabetes, and obesity-related cancers . J Natl Cancer Inst , 2023 . 115 ( 1 ): p. 43 – 51 . OpenUrl PubMed 8. Llewellyn , A. , et al. , Childhood obesity as a predictor of morbidity in adulthood: a systematic review and meta-analysis . Obesity reviews , 2016 . 17 ( 1 ): p. 56 – 67 . OpenUrl PubMed 9. Yang , T.O. , et al. , Body size in early life and the risk of postmenopausal breast cancer . BMC cancer , 2022 . 22 ( 1 ): p. 232 . OpenUrl PubMed 10. ↵ Andersen , Z.J. , et al. , Birth weight, childhood body mass index, and height in relation to mammographic density and breast cancer: a register-based cohort study . Breast Cancer Research , 2014 . 16 ( 1 ): p. R4 . OpenUrl CrossRef PubMed 11. ↵ Richardson , T.G. , et al. , Leveraging family history data to disentangle time-varying effects on disease risk using lifecourse mendelian randomization . European Journal of Epidemiology , 2023 . 38 ( 7 ): p. 765 – 769 . OpenUrl PubMed 12. ↵ Mandic , M. , et al. , Association of childhood-to-adulthood body size change with cancer risk: UK Biobank prospective cohort . BMC Med , 2025 . 23 ( 1 ): p. 268 . OpenUrl PubMed 13. ↵ Richmond , R.C. and G. Davey Smith , Mendelian Randomization: Concepts and Scope . Cold Spring Harb Perspect Med , 2022 . 12 ( 1 ). 14. ↵ Sanderson , E. , et al. , Estimation of causal effects of a time-varying exposure at multiple time points through multivariable mendelian randomization . PLoS Genet , 2022 . 18 ( 7 ): p. e1010290 . OpenUrl CrossRef PubMed 15. ↵ Power , G.M. , et al. , Methodological approaches, challenges, and opportunities in the application of Mendelian randomisation to lifecourse epidemiology: A systematic literature review . Eur J Epidemiol , 2024 . 39 ( 5 ): p. 501 – 520 . OpenUrl CrossRef PubMed 16. ↵ Schooling , C.M. , K. Fei , and J.V. Zhao , Selection bias as an explanation for the observed protective association of childhood adiposity with breast cancer . J Clin Epidemiol , 2023 . 164 : p. 104 – 111 . OpenUrl PubMed 17. ↵ Schooling , C.M. , K. Fei , and M.B. Terry , Reassessing the causal role of early-life adiposity in breast cancer: could the apparent inverse associations be a manifestation of survival bias? International Journal of Epidemiology , 2023 . 52 ( 4 ): p. 1292 – 1293 . OpenUrl PubMed 18. ↵ Banack , H.R. , et al. , Collider Stratification Bias I: Principles and Structure . American Journal of Epidemiology , 2023 . 193 ( 2 ): p. 238 – 240 . OpenUrl 19. ↵ Greenland , S. , Response and follow-up bias in cohort studies . Am J Epidemiol , 1977 . 106 ( 3 ): p. 184 – 7 . OpenUrl PubMed Web of Science 20. ↵ Lu , H. , et al. , Toward a Clearer Definition of Selection Bias When Estimating Causal Effects . Epidemiology , 2022 . 33 ( 5 ): p. 699 – 706 . OpenUrl CrossRef PubMed 21. ↵ Banack , H.R. , et al. , Collider stratification bias II: magnitude of bias . Am J Epidemiol , 2025 . 194 ( 5 ): p. 1149 – 1151 . OpenUrl PubMed 22. ↵ Gkatzionis , A. , et al. , Relationship between collider bias and interactions on the log-additive scale . Stat Methods Med Res , 2025 : p. 9622802241306860. 23. ↵ Hernán , M.A. , Invited Commentary: Selection Bias Without Colliders . Am J Epidemiol , 2017 . 185 ( 11 ): p. 1048 – 1050 . OpenUrl CrossRef PubMed 24. ↵ Batty , G.D. , et al. , Comparison of risk factor associations in UK Biobank against representative, general population based studies with conventional response rates: prospective cohort study and individual participant meta-analysis . Bmj , 2020 . 368 : p. m131 . OpenUrl Abstract / FREE Full Text 25. ↵ Hemani , G. , et al. , Sensitivity analyses gain relevance by fixing parameters observable during the empirical analyses . Genet Epidemiol , 2023 . 47 ( 6 ): p. 461 – 462 . OpenUrl PubMed View the discussion thread. Back to top Previous Next Posted March 17, 2026. Download PDF Supplementary Material Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Triangulation of evidence to examine selection bias in lifecourse Mendelian randomization studies: an example using early life adiposity and breast cancer Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Triangulation of evidence to examine selection bias in lifecourse Mendelian randomization studies: an example using early life adiposity and breast cancer Grace M. Power , Eleanor Sanderson , Apostolos Gkatzionis , Tom G. Richardson , Kate Tilling , George Davey Smith , Gibran Hemani medRxiv 2025.09.10.25335479; doi: https://doi.org/10.1101/2025.09.10.25335479 Share This Article: Copy Citation Tools Triangulation of evidence to examine selection bias in lifecourse Mendelian randomization studies: an example using early life adiposity and breast cancer Grace M. Power , Eleanor Sanderson , Apostolos Gkatzionis , Tom G. Richardson , Kate Tilling , George Davey Smith , Gibran Hemani medRxiv 2025.09.10.25335479; doi: https://doi.org/10.1101/2025.09.10.25335479 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Epidemiology Subject Areas All Articles Addiction Medicine (568) Allergy and Immunology (863) Anesthesia (300) Cardiovascular Medicine (4435) Dentistry and Oral Medicine (444) Dermatology (382) Emergency Medicine (608) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1509) Epidemiology (15228) Forensic Medicine (30) Gastroenterology (1124) Genetic and Genomic Medicine (6599) Geriatric Medicine (668) Health Economics (997) Health Informatics (4536) Health Policy (1368) Health Systems and Quality Improvement (1613) Hematology (540) HIV/AIDS (1264) Infectious Diseases (except HIV/AIDS) (15916) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (146) Nephrology (667) Neurology (6599) Nursing (346) Nutrition (998) Obstetrics and Gynecology (1144) Occupational and Environmental Health (957) Oncology (3332) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (663) Pediatrics (1693) Pharmacology and Therapeutics (691) Primary Care Research (711) Psychiatry and Clinical Psychology (5447) Public and Global Health (9231) Radiology and Imaging (2198) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1196) Rheumatology (593) Sexual and Reproductive Health (712) Sports Medicine (530) Surgery (712) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a005be9b9ebf58f4',t:'MTc3OTU1Njk2NQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();