Collider Bias Research Articles

Estimating Causal Effects on a Disease Progression Trait Using Bivariate Mendelian Randomisation.

Genome-wide association studies (GWAS) have provided large numbers of genetic markers that can be used as instrumental variables in a Mendelian Randomisation (MR) analysis to assess the causal effect of a risk factor on an outcome. An extension of MR analysis, multivariable MR, has been proposed to handle multiple risk factors. However, adjusting or stratifying the outcome on a variable that is associated with it may induce collider bias. For an outcome that represents progression of a disease, conditioning by selecting only the cases may cause a biased MR estimation of the causal effect of the risk factor of interest on the progression outcome. Recently, we developed instrument effect regression and corrected weighted least squares (CWLS) to adjust for collider bias in observational associations. In this paper, we highlight the importance of adjusting for collider bias in MR with a risk factor of interest and disease progression as the outcome. A generalised version of the instrument effect regression and CWLS adjustment is proposed based on a multivariable MR model. We highlight the assumptions required for this approach and demonstrate its utility for bias reduction. We give an illustrative application to the effect of smoking initiation and smoking cessation on Crohn's disease prognosis, finding no evidence to support a causal effect.

Also long overdue: consideration of collider bias in guidelines and tools for systematic reviews and meta-analyses of observational studies.

Also long overdue: consideration of collider bias in guidelines and tools for systematic reviews and meta-analyses of observational studies.

Correcting for Observation Bias in Cancer Progression Modeling.

Tumor progression is driven by the accumulation of genetic alterations, including both point mutations and copy number changes. Understanding the temporal sequence of these events is crucial for comprehending the disease but is not directly discernible from cross-sectional genomic data. Cancer progression models, including Mutual Hazard Networks (MHNs), aim to reconstruct the dynamics of tumor progression by learning the causal interactions between genetic events based on their co-occurrence patterns in cross-sectional data. Here, we highlight a commonly overlooked bias in cross-sectional datasets that can distort progression modeling. Tumors become clinically detectable when they cause symptoms or are identified through imaging or tests. Detection factors, such as size, inflammation (fever, fatigue), and elevated biochemical markers, are influenced by genomic alterations. Ignoring these effects leads to "conditioning on a collider" bias, where events making the tumor more observable appear anticorrelated, creating false suppressive effects or masking promoting effects among genetic events. We enhance MHNs by incorporating the effects of genetic progression events on the inclusion of a tumor in a dataset, thus correcting for collider bias. We derive an efficient tensor formula for the likelihood function and apply it to two datasets from the MSK-IMPACT study. In colon adenocarcinoma, we observe a significantly higher rate of clinical detection for TP53-positive tumors, while in lung adenocarcinoma, the same is true for EGFR-positive tumors. Compared to classical MHNs, this approach eliminates several spurious suppressive interactions and uncovers multiple promoting effects.

Disentangling discordant vitamin D associations with prostate cancer incidence and fatality in a large, nested case-control study.

Published analyses of prostate cancer nested case-control and survival data in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study cohort suggested that men with higher baseline vitamin D [25(OH)D] concentrations have both (i) increased prostate cancer risk and (ii) decreased prostate cancer-specific fatality. To investigate possible factors responsible for a spurious association with prostate cancer fatality, we reanalysed baseline serum vitamin D associations with prostate cancer risk and prostate cancer-specific fatality in case-control data nested within the ATBC Study (1000 controls and 1000 incident prostate cancer cases). Conditional logistic regression and Cox proportion hazard models were used, respectively, to estimate odds ratios for risk and hazard ratios for prostate cancer-specific fatality, overall and by disease aggressiveness. We replicated these case-control analyses using baseline serum measurements of alpha-tocopherol (vitamin E), beta-carotene and retinol (vitamin A), and used the entire ATBC Study cohort (n = 29 085) to estimate marginal associations between these baseline vitamins and prostate cancer incidence and fatality following blood collection. Vitamin D analyses agreed closely with those originally published, with opposite risk and fatality associations. By contrast, the analyses of alpha-tocopherol, beta-carotene and retinol yielded concordant associations for prostate cancer incidence and prostate cancer-specific fatality. We found evidence of neither artefacts in the nested prostate cancer case-control data set nor detection or collider biases in the fatality analyses. The present findings therefore support a valid inverse (i.e. beneficial) association between vitamin D and prostate cancer-specific survival that warrants further evaluation, including possibly in controlled trials.

Central corneal thickness and the risk of primary open-angle glaucoma: a Mendelian randomisation mediation analysis

BackgroundThe association of central corneal thickness (CCT) with primary open-angle glaucoma (POAG) remains uncertain. Although several observational studies assessing this relationship have reported an inverse association between CCT and POAG,...

How collider bias affects the relationship between skin color and heart attack using directed acyclic graphs, propensity scores, and stepwise approaches

BackgroundStatistical methods are essential in epidemiology research, but they can generate erroneous estimates when selecting variables based only on statistical criteria. The use of directed acyclic graphs (DAG) helps to understand the causal relationships between variables and to avoid biases.ObjectiveCompare the estimate of the effect of skin color on heart attack obtained from three data analysis techniques: a stepwise approach based on statistical criteria, a propensity score technique, and a graphical approach based on causal criteria.MethodsPopulation-based cross-sectional study using data from the second National Health and Nutrition Examination Survey (NHANES). The exposure variable was skin color (black or non-black) and the outcome was heart attack (yes or no). Multivariable logistic regressions were carried out using the stepwise, propensity score techniques and the DAG-based approach to identify the association between the variables. In the stepwise technique, all variables potentially related to the outcome were included in the model and a forward or backward algorithm was used. The propensity score was applied, estimating the probability of exposure based on the covariates and helping to create balanced groups for comparison. Different possible causal models were developed between the variables in the DAG-based approach, identifying confounding, mediation, and collision factors. The models were created considering self-rated health as a confounding or collider variable, and the modeling results were verified.ResultsA total of 10,351 adults were evaluated, the majority female (52.1%), aged 20 to 39 years (48.5%), and with non-black skin color (90.4%). The prevalence of heart attacks was 3.0%, and 17% rated their health as fair or poor. Using different modeling techniques, no association was found between skin color and heart attack (p > 0.05), except when self-rated health, a collider variable, was included in the stepwise models. In this case, there was an inverse and biased association between the two variables, indicating a collision bias (stepwise-backward-OR 0.48; 95%CI 0.33–0.70; stepwise-forward-OR 0.64; 95%CI 0.44–0.94).ConclusionSkin color was not associated with heart attack when controlling for appropriate confounding factors. However, when adjusting for self-rated health in stepwise techniques, a colliding variable, there was an inverse and distorted association between the two variables, indicating a collider bias. The DAG-based approach and propensity score can avoid this bias by correctly identifying confounding factors and colliders.

Application of information from external data to correct for collider bias in a Covid-19 hospitalised cohort

BackgroundThroughout the Covid-19 pandemic, researchers have made use of electronic health records to research this disease in a rapidly evolving environment of questions and discoveries. These studies are prone to collider bias as they restrict the population of Covid-19 patients to only those with severe disease. Inverse probability weighting is typically used to correct for this bias but requires information from the unrestricted population. Using electronic health records from a South London NHS trust, this work demonstrates a method to correct for collider bias using externally sourced data while examining the relationship between minority ethnicities and poor Covid-19 outcomes.MethodsThe probability of inclusion within the observed hospitalised cohort was modelled based on estimates from published national data. The model described the relationship between patient ethnicity, hospitalisation, and death due to Covid-19 – a relationship suggested to be susceptible to collider bias. The obtained probabilities (as applied to the observed patient cohort) were used as inverse probability weights in survival analysis examining ethnicity (and covariates) as a risk factor for death due to Covid-19.ResultsWithin the observed cohort, unweighted analysis of survival suggested a reduced risk of death in those of Black ethnicity – differing from the published literature. Applying inverse probability weights to this analysis amended this aberrant result to one more compatible with the literature. This effect was consistent when the analysis was applied to patients within only the first wave of Covid-19 and across two waves of Covid-19 and was robust against adjustments to the modelled relationship between hospitalisation, patient ethnicity, and death due to Covid-19 made as part of a sensitivity analysis.ConclusionsIn conclusion, this analysis demonstrates the feasibility of using external publications to correct for collider bias (or other forms of selection bias) induced by the restriction of a population to a hospitalised cohort using an example from the recent Covid-19 pandemic.

Disentangling the causal effects of education and participation bias on Alzheimer's disease using Mendelian Randomization.

People with university degrees have a lower incidence of Alzheimer's Disease (AD). However, the relationship between education and AD could be due to selection, collider, and ascertainment biases, such as lower familiarity with cognitive testing or the fact that those with degrees are more likely to participate in research. Here, we use two-sample Mendelian randomization (MR) to investigate the causal relationships between education, participation, and AD. We used genome-wide association study (GWAS) summary statistics for educational attainment, three different measures of participation, AD (clinically diagnosed AD), and AD/ADRD (clinical diagnosis and family history of AD and related dementias). Independent genome-wide significant single nucleotide polymorphisms (SNPs) were extracted from the exposure summary statistics and harmonized with the outcome SNPs. Fixed-effects inverse variance weighted meta-analysis was the primary MR method; Cochran's Q statistic and MR Egger intercept were used to test for heterogeneity and pleiotropy, and Radial-MR was used to identify outliers. Sensitivity analyses included MR Egger, Weighted Median, and Weighted mode. Bidirectional analyses were used to test if AD pathology affects participation and multivariable MR (MVMR) assessed independent exposure-outcome effects. Educational attainment reduced the risk of AD (OR IVW 95% CI= 0.70 [0.63, 0.79], p = 8e-10), and the results were robust based on sensitivity analyses. However, education increased the risk of AD/ADRD, though the results were not robust to sensitivity analyses (OR IVW 95% CI= 1.09 [1.02, 1.15], p = 0.006). Participation in MHQ reduced the odds of AD (OR IVW 95% CI= 0.325 [0.128, 0.326], p = 0.01). When adjusting for participation in MVMR, education remained associated with a reduced risk of AD (OR IVW 95% CI= 0.76 [0.62, 0.92], p = 0.006). Univariate MR analyses indicated that education and participation reduced the risk of AD. However, MR also suggested that education increased the risk of AD/ADRD, highlighting the inconsistencies between clinical and proxy diagnoses of AD, as proxy-AD may be affected by selection, collider, or ascertainment bias. MVMR indicated that participation is unlikely to explain the effect of education on AD identified in MR, and the protective effect of educational attainment may be due to other biological mechanisms, such as cognitive reserve.

O-089 Association of pituitary suppression method with euploid blastocyst yield in preimplantation genetic testing for aneuploidy (PGT-A) cycles

Abstract Study question Do progestin primed ovarian stimulation (PPOS) cycles have lower euploidy yield compared to pituitary suppression with gonadotropin releasing hormone (GnRH) antagonists? Summary answer The choice of pituitary suppression method has no bearing on euploid retrieval per blastocyst tested. What is known already Progestin primed ovarian stimulation cycles are an alternative to GnRH antagonists when fresh embryo transfer is not a priority. While the two approaches have been shown to be similar in terms of conventional stimulation outcomes such as mature oocyte retrieval, and blastulation rates, there is still some lingering controversy regarding cumulative pregnancy rates and the effects of progestins on embryo genetic constitution. Study design, size, duration This was a multicenter retrospective cohort study including data from two tertiary-level treatment centers. Participants/materials, setting, methods Infertile women undergoing ovarian stimulation with either progestin primed or GnRH antagonist pituitary suppression with the intention of transfer after PGT-A embryo testing with next-generation sequencing. The main outcome parameter was euploid embryo count per blastocyst tested and was modeled with mixed-effects log-binomial regression models. Results were adjusted for patient age and laboratory performing the PGT-A tests while they were not adjusted for embryo quality to avoid collider bias. Main results and the role of chance A total of 1425 PGT-A tested blastocysts from 383 cycles of 365 patients were included in the analysis. 243 cycles (63.4%) received GnRH antagonists for pituitary suppression while 140 (36.5%) received progestins. The overall euploidy rates per blastocyst tested in the GnRH antagonist and PPOS groups were 38.8% vs. 31.1%, 30.2% vs. 29.4%, 27.0% vs. 25.2%, and 13.8% vs. 14.5% for the age categories <35, 35-37, 38-40, and >40, respectively. After adjusting for the effect of female age at testing and the laboratory performing the PGT-A test, the type of pituitary suppression was not significantly associated with euploid retrieval per blastocyst tested (adjusted risk ratio: 0.94, 95% CI: 0.70 - 1.28, P = 0.701). After testing, 191 women received 233 embryo transfers. Human chorionic gonadotropin (hCG) positivity per transferred embryo was not significantly different between progestin primed and GnRH antagonist cycles (RR: 0.89, 95% CI: 0.71 – 1.08, P = 0.268). Limitations, reasons for caution This is a retrospective study with data from multiple centers with a non-standardized selection of blastocysts to be tested. In addition, the sample size and retrospective nature of the data may have biased the estimates. Wider implications of the findings The choice of pituitary suppression method does not appear to have a significant effect on euploid embryo retrieval rates or hCG positivity per transfer. The choice of medication can be based on the needs of the individual case without concern for suboptimal PGT-A outcomes. Trial registration number Not Applicable

Independent Effects of Blood Pressure on Intraocular Pressure and Retinal Ganglion Cell Degeneration: A Mendelian Randomization Study.

To investigate the causal effect of elevated blood pressure on primary open-angle glaucoma (POAG) and POAG endophenotypes. Two-sample Mendelian randomization (MR) was performed to investigate the causal effect of elevated systolic blood pressure (SBP) (N = 757,601) and diastolic blood pressure (DBP) (N = 757,601) on intraocular pressure (IOP) (N = 139,555), macular retinal nerve fiber layer (mRNFL) thickness (N = 33,129), ganglion cell complex (GCC) thickness (N = 33,129), vertical cup-to-disc ratio (VCDR) (N = 111,724), and POAG liability (Ncases = 16,677, Ncontrols = 199,580). The primary analysis was conducted using the inverse-variance weighted approach. Sensitivity analyses were performed to investigate robustness to horizontal pleiotropy, winner's curse, and collider bias. Multivariable MR was performed to investigate whether any effect of blood pressure on retinal ganglion cell degeneration was mediated through increased IOP. Increased genetically predicted SBP and DBP associated with an increase in IOP (0.17 mm Hg [95% CI = 0.11 to 0.24] per 10 mm Hg higher SBP, P = 5.18 × 10-7, and 0.17 mm Hg [95% CI = 0.05 to 0.28 mm Hg] per 10 mm Hg higher DBP, P = 0.004). Increased genetically predicted SBP associated with a thinner GCC (0.04µm [95% CI = -0.07 to -0.01µm], P = 0.018) and a thinner mRNFL (0.04µm [95% CI = -0.07 to -0.01µm], P = 0.004), an effect that arises independently of IOP according to our mediation analysis. Neither SBP nor DBP associated with VCDR or POAG liability. These findings support a causal effect of elevated blood pressure on retinal ganglion cell degeneration that does not require intermediary changes in IOP. Targeted blood pressure control may help preserve vision by lowering IOP and, independently, by preventing retinal ganglion cell degeneration, including in individuals with a normal IOP.

Causal relationship between gut microbiota and functional outcomes after ischemic stroke: A comprehensive Mendelian randomization study

AimsTo investigate the association of the genetic predisposition of specific gut microbiotas with the clinical outcome of ischemic stroke. MethodsWe leveraged publicly available genome-wide association study (GWAS) data to perform Mendelian randomization (MR) analysis. The gut microbiota-related GWAS data from 18,340 individuals from the international consortium MiBioGen was used. The summary data for functional outcomes after ischemic stroke was obtained from the Genetics of Ischemic Stroke Functional Outcome (GISCOME) network meta-analysis. The primary outcomes were judged by the modified Rankin Scale (mRS). The principal analyses were conducted using the inverse-variance weighted (IVW) MR method. The Cochran's Q test, weighted median, MR-Egger regression, leave-one-SNP-out analysis, MR-Pleiotropy Residual Sum, and Outlier methods were adopted as sensitivity analyses. Furthermore, we performed bi-directional MR analysis and the MR Steiger directionality test to examine the direction of the causal relations. ResultsThe results demonstrated that the genetic predisposition of genus Lactococcus, genus Ruminococcaceae NK4A214 group, family Peptostreptococcaceae, and genus Odoribacter was positively associated with favorable functional outcome after ischemic stroke. Genus Collinsella, genus Ruminococcaceae UCG005, genus Akkermansia, genus Eubacterium oxidoreducens group, and family Verrucomicrobiaceae were identified to be associated with worse functional outcomes after ischemic stroke. Our results showed no evidence of heterogeneity, directional pleiotropic effects, or collider bias, and the sensitivity of our analysis was acceptable. ConclusionThe genetic predisposition of different gut microbiotas was associated with the clinical outcome of ischemic stroke. Microbiota adjustment was a promising method to improve the clinical outcome of ischemic stroke.

Mother’s age at marriage and gender-differential in child schooling: Evidence from Pakistan

In this study, we provide empirical evidence in support of intergenerational effects of women’s marriage age on girl-specific child education outcomes using a nationally representative household survey of 24,809 households from Pakistan. Our key findings are four-fold: First, we find that mother’s age at marriage has no girl-specific impact on the education of children of school going age in terms of school enrolment, type of school (public/private) or education expenditure. However, there is a positive effect of delaying mother’s marriage on girls in terms of relative grade progression and primary school attainment, implying that late-marrying mothers are more likely to transfer their human capital advantages/disadvantages to the daughters. Second, the impact does not depend on whether the child is firstborn or later-order, youngest child or with no siblings. Sex of the previous child does not affect the relationship either. Third, the beneficial impact of mother’s marriage age on girls’ education is visible only in the cohort of women who got married in 2000 or later. Fourth, father’s marriage age is significantly associated with an increase in daughters’ school enrolment and primary completion. The impact of smaller spousal age difference is also positive. These findings are robust to the use of empirical strategies and specifications that address potential endogeneity, collider bias, recall bias, sample selection and confounding factors. These findings suggest that delay in women’s marriage is helping to narrow down gender disparities in education. The findings underscore the need to promote social and behavioural changes that encourage later marriages.

Identification and correction for collider bias in a genome-wide association study of diabetes-related heart failure

Type 2 diabetes (T2D) is a major risk factor for heart failure (HF) and has elevated incidence among individuals with HF. Since genetics and HF can independently influence T2D, collider bias may occur when T2D (i.e., collider) is controlled for by design or analysis. Thus, we conducted a genome-wide association study (GWAS) of diabetes-related HF with correction for collider bias. We first performed a GWAS of HF to identify genetic instrumental variables (GIVs) for HF and to enable bidirectional Mendelian randomization (MR) analysis between T2D and HF. We identified 61 genomic loci, significantly associated with all-cause HF in 114,275 individuals with HF and over 1.5 million controls of European ancestry. Using a two-sample bidirectional MR approach with 59 and 82 GIVs for HF and T2D, respectively, we estimated that T2D increased HF risk (odds ratio [OR] 1.07, 95% confidence interval [CI] 1.04-1.10), while HF also increased T2D risk (OR 1.60, 95% CI 1.36-1.88). Then we performed a GWAS of diabetes-related HF corrected for collider bias due to the study design of index cases. After removing the spurious association of TCF7L2 locus due to collider bias, we identified two genome-wide significant loci close to PITX2 (chromosome 4) and CDKN2B-AS1 (chromosome 9) associated with diabetes-related HF in the Million Veteran Program and replicated the associations in the UK Biobank. Our MR findings provide strong evidence that HF increases T2D risk. As a result, collider bias leads to spurious genetic associations of diabetes-related HF, which can be effectively corrected to identify true positive loci.

Collider Bias Assessment in Colombian Indigenous Wiwa and Kogui Populations with Chronic Gastroenteric Disorder of Likely Infectious Etiology Suggests Complex Microbial Interactions Rather Than Clear Assignments of Etiological Relevance.

Multiple microbial detections in stool samples of indigenous individuals suffering from chronic gastroenteric disorder of a likely infectious origin, characterized by recurring diarrhea of variable intensity, in the rural north-east of Colombia are common findings, making the assignment of etiological relevance to individual pathogens challenging. In a population of 773 indigenous people from either the tribe Wiwa or Kogui, collider bias analysis was conducted comprising 32 assessed microorganisms including 10 bacteria (Aeromonas spp., Campylobacter spp., enteroaggregative Escherichia coli (EAEC), enteropathogenic Escherichia coli (EPEC), enterotoxigenic Escherichia coli (ETEC), Salmonella spp., Shiga toxin-producing Escherichia coli (STEC), Shigella spp./enteroinvasive Escherichia coli (EIEC), Tropheryma whipplei and Yersinia spp.), 11 protozoa (Blastocystis spp., Cryptosporidium spp., Cyclospora spp., Dientamoeba fragilis, Entamoeba coli, Entamoeba bangladeshi/dispar/histolytica/moshkovskii complex, Entamoeba histolytica, Endolimax nana, Giardia duodenalis, Iodamoeba buetschlii and Pentatrichomonas hominis), 8 helminths (Ascaris spp., Enterobius vermicularis, Hymenolepis spp., Necator americanus, Schistosoma spp., Strongyloides spp., Taenia spp. and Trichuris spp.), microsporidia (Encephalocytozoon spp.) and fungal elements (microscopically observed conidia and pseudoconidia). The main results indicated that negative associations potentially pointing towards collider bias were infrequent events (n = 14), while positive associations indicating increased likelihood of co-occurrence of microorganisms quantitatively dominated (n = 88). Microorganisms showing the most frequent negative associations were EPEC (n = 6) and Blastocystis spp. (n = 3), while positive associations were most common for Trichuris spp. (n = 16), Dientamoeba fragilis (n = 15), Shigella spp./EIEC (n = 12), Ascaris spp. (n = 11) and Blastocystis spp. (n = 10). Of note, positive associations quantitively dominated for Blastocystis spp. In conclusion, collider bias assessment did not allow clear-cut assignment of etiological relevance for detected enteric microorganisms within the assessed Colombian indigenous population. Instead, the results suggested complex microbial interactions with potential summative effects. Future studies applying alternative biostatistical approaches should be considered to further delineate respective interactions.

Collider bias correction for multiple covariates in GWAS using robust multivariable Mendelian randomization.

Genome-wide association studies (GWAS) have identified many genetic loci associated with complex traits and diseases in the past 20 years. Multiple heritable covariates may be added into GWAS regression models to estimate direct effects of genetic variants on a focal trait, or to improve the power by accounting for environmental effects and other sources of trait variations. When one or more covariates are causally affected by both genetic variants and hidden confounders, adjusting for them in GWAS will produce biased estimation of SNP effects, known as collider bias. Several approaches have been developed to correct collider bias through estimating the bias by Mendelian randomization (MR). However, these methods work for only one covariate, some of which utilize MR methods with relatively strong assumptions, both of which may not hold in practice. In this paper, we extend the bias-correction approaches in two aspects: first we derive an analytical expression for the collider bias in the presence of multiple covariates, then we propose estimating the bias using a robust multivariable MR (MVMR) method based on constrained maximum likelihood (called MVMR-cML), allowing the presence of invalid instrumental variables (IVs) and correlated pleiotropy. We also established the estimation consistency and asymptotic normality of the new bias-corrected estimator. We conducted simulations to show that all methods mitigated collider bias under various scenarios. In real data analyses, we applied the methods to two GWAS examples, the first a GWAS of waist-hip ratio with adjustment for only one covariate, body-mass index (BMI), and the second a GWAS of BMI adjusting metabolomic principle components as multiple covariates, illustrating the effectiveness of bias correction.

Correcting for collider effects and sample selection bias in psychological research.

Colliders, variables that serve as a common outcome of an independent and dependent variable, pose a major challenge in psychological research. Collider variables can induce bias in the estimation of a population relationship of interest when (a) the composition of a research sample is restricted by scores on a collider variable or (b) researchers adjust for a collider variable in their statistical analyses, as they might do for confounder variables. Both cases interfere with the accuracy and generalizability of statistical results. Despite their importance, however, collider effects remain relatively unknown in psychology. This tutorial article summarizes both the conceptual and the mathematical foundation for collider effects and their relevance to psychological research, and then proposes a method to correct for collider bias in cases of restrictive sample selection based on Thorndike's Case III adjustment (1982). Two simulation studies demonstrated Thorndike's correction as a viable solution for correcting collider bias in research studies, even when restriction on the collider variable was extreme and the selected sample size was as low as N = 100. Bias and relative bias results are reported to evaluate how well the correction equation approximates targeted population correlations under a variety of parameter conditions. We illustrate the application of the correction method to a hypothetical study of intelligence and conscientiousness, discuss the applicability of the method to more complex statistical models as a means of detection for collider bias, and provide code for researchers to apply to their own research. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Collider bias in strategy and management research: An illustration using women CEO's effect on other women's career outcomes

AbstractResearch SummaryCollider bias can cause spurious correlations when researchers condition on a variable that is caused by—or shares a common cause with—both the outcome and the exposure variable. Despite its threat to inference, empirical research in strategy and management has largely overlooked the issue of collider bias. We distinguish colliders from other threats to identification and estimation and illustrate its importance with replications of published work suggesting that having a woman CEO reduces the career outcomes (compensation and representation) of other women executives. After accounting for collider bias, we find no evidence that women CEOs damage the career outcomes of other women in their organizations. We close by providing generalizable approaches to identify and mitigate the risk of collider bias in applied research.Managerial SummaryCollider bias is a type of statistical problem that can generate misleading results in empirical research. Although research in strategy and management has given substantial attention to other types of statistical problems, the issue of collider bias has not received sufficient scrutiny. We illustrate this point with replications of published work suggesting that having a woman CEO reduces the career outcomes of other women executives. After accounting for collider bias, we find no evidence that women CEOs damage the career outcomes of other women in their organizations. We provide advice for detecting and addressing collider bias in empirical research.

A data-adaptive method for investigating effect heterogeneity with high-dimensional covariates in Mendelian randomization

BackgroundMendelian randomization is a popular method for causal inference with observational data that uses genetic variants as instrumental variables. Similarly to a randomized trial, a standard Mendelian randomization analysis estimates the population-averaged effect of an exposure on an outcome. Dividing the population into subgroups can reveal effect heterogeneity to inform who would most benefit from intervention on the exposure. However, as covariates are measured post-“randomization”, naive stratification typically induces collider bias in stratum-specific estimates.MethodWe extend a previously proposed stratification method (the “doubly-ranked method”) to form strata based on a single covariate, and introduce a data-adaptive random forest method to calculate stratum-specific estimates that are robust to collider bias based on a high-dimensional covariate set. We also propose measures based on the Q statistic to assess heterogeneity between stratum-specific estimates (to understand whether estimates are more variable than expected due to chance alone) and variable importance (to identify the key drivers of effect heterogeneity).ResultWe show that the effect of body mass index (BMI) on lung function is heterogeneous, depending most strongly on hip circumference and weight. While for most individuals, the predicted effect of increasing BMI on lung function is negative, it is positive for some individuals and strongly negative for others.ConclusionOur data-adaptive approach allows for the exploration of effect heterogeneity in the relationship between an exposure and an outcome within a Mendelian randomization framework. This can yield valuable insights into disease aetiology and help identify specific groups of individuals who would derive the greatest benefit from targeted interventions on the exposure.

Investigating the association between genetically proxied circulating levels of immune checkpoint proteins and cancer survival: protocol for a Mendelian randomisation analysis

IntroductionCompared with the traditional drug development pathway, investigating alternative uses for existing drugs (ie, drug repurposing) requires substantially less time, cost and resources. Immune checkpoint inhibitors are licensed for the...

Ranking the causal impact of recommendations under collider bias in k -spots recommender systems

The first objective of recommender systems is to provide personalized recommendations for each user. However, personalization may not be its only use. Past recommendations can be further analyzed to gain global insights into users’ behavior with respect to recommended items. Such insights can help to answer design-related questions such as which items’ recommendations are the most impactful in terms of users’ utility, which type of recommendations are the most followed ones, which items could be dropped from the catalog, or which recommendations are under-performing compared to what one would expect. In order to answer those questions, we need to rank item recommendations’ performances in terms of their causal impact on some user-related outcome measures. Unfortunately, in previous work leveraging causal inference for recommendation systems, the attention is fully focused on correcting confounding bias and not on the collider bias. This bias is particularly relevant in the recommender context, where multiple items are simultaneously recommended. Indeed, when there is a fixed number of available spots (i.e., k -spots) and recommendations need to be provided at each session, we argue that it is not possible to estimate the causal impacts of recommendations but only the differences between them. Therefore, in this paper, we provide an unbiased estimator of the differences in the impacts of items’ recommendations, that work for any outcome of interest, and any type of recommender system as long as it has some degree of randomization. We apply our results both in a simulated environment and in a real-world offline environment leveraging logged data for recommended items in a digital healthcare app.