Unbiased Effect Estimates Research Articles

Advancing Interpretable Regression Analysis for Binary Data: A Novel Distributed Algorithm Approach.

Sparse data bias, where there is a lack of sufficient cases, is a common problem in data analysis, particularly when studying rare binary outcomes. Although a two-step meta-analysis approach may be used to lessen the bias by combining the summary statistics to increase the number of cases from multiple studies, this method does not completely eliminate bias in effect estimation. In this paper, we propose a one-shot distributed algorithm for estimating relative risk using a modified Poisson regression for binary data, named ODAP-B. We evaluate the performance of our method through both simulation studies and real-world case analyses of postacute sequelae of SARS-CoV-2 infection in children using data from 184 501 children across eight national academic medical centers. Compared with the meta-analysis method, our method provides closer estimates of the relative risk for all outcomes considered including syndromic and systemic outcomes. Our method is communication-efficient and privacy-preserving, requiring only aggregated data to obtain relatively unbiased effect estimates compared with two-step meta-analysis methods. Overall, ODAP-B is an effective distributed learning algorithm for Poisson regression to study rare binary outcomes. The method provides inference on adjusted relative risk with a robust variance estimator.

Propensity score methods for confounding control in observational studies of therapeutics for COVID-19 infection.

The authors provide a brief overview of different propensity score methods that can be used in observational research studies that lack randomization. Under specific assumptions, these methods result in unbiased estimates of causal effects, but the different ways propensity score are used may require different assumptions and result in estimated treatment effects that can have meaningfully different interpretations. The authors review these issues and consider their implications for studies of therapeutics for COVID-19.

Overcoming Data Gaps in Life Course Epidemiology by Matching Across Cohorts.

Life course epidemiology is hampered by the absence of large studies with exposures and outcomes measured at different life stages in the same individuals. We describe when the effect of an exposure ( A ) on an outcome ( Y ) in a target population is identifiable in a combined ("synthetic") cohort created by pooling an early-life cohort including measures of A with a late-life cohort including measures of Y . We enumerate causal assumptions needed for unbiased effect estimation in the synthetic cohort and illustrate by simulating target populations under four causal models. From each target population, we randomly sampled early- and late-life cohorts and created a synthetic cohort by matching individuals from the two cohorts based on mediators and confounders. We estimated the effect of A on Y in the synthetic cohort, varying matching variables, the match ratio, and the strength of association between matching variables and A . Finally, we compared bias in the synthetic cohort estimates when matching variables did not d-separate A and Y to the bias expected in the original cohort. When the set of matching variables includes all variables d-connecting exposure and outcome (i.e., variables blocking all backdoor and front-door pathways), the synthetic cohort yields unbiased effect estimates. Even when matching variables did not fully account for confounders, the synthetic cohort estimate was sometimes less biased than comparable estimates in the original cohort. Methods based on merging cohorts may hasten the evaluation of early- and mid-life determinants of late-life health but rely on available measures of both confounders and mediators.

Interpreting population- and family-based genome-wide association studies in the presence of confounding.

A central aim of genome-wide association studies (GWASs) is to estimate direct genetic effects: the causal effects on an individual's phenotype of the alleles that they carry. However, estimates of direct effects can be subject to genetic and environmental confounding and can also absorb the "indirect" genetic effects of relatives' genotypes. Recently, an important development in controlling for these confounds has been the use of within-family GWASs, which, because of the randomness of mendelian segregation within pedigrees, are often interpreted as producing unbiased estimates of direct effects. Here, we present a general theoretical analysis of the influence of confounding in standard population-based and within-family GWASs. We show that, contrary to common interpretation, family-based estimates of direct effects can be biased by genetic confounding. In humans, such biases will often be small per-locus, but can be compounded when effect-size estimates are used in polygenic scores (PGSs). We illustrate the influence of genetic confounding on population- and family-based estimates of direct effects using models of assortative mating, population stratification, and stabilizing selection on GWAS traits. We further show how family-based estimates of indirect genetic effects, based on comparisons of parentally transmitted and untransmitted alleles, can suffer substantial genetic confounding. We conclude that, while family-based studies have placed GWAS estimation on a more rigorous footing, they carry subtle issues of interpretation that arise from confounding.

MRBEE: A bias-corrected multivariable Mendelian randomization method

Mendelian randomization (MR) is an instrumental variable approach used to infer causal relationships between exposures and outcomes, which is becoming increasingly popular because of its ability to handle summary statistics from genome-wide association studies (GWAS). However, existing MR approaches often suffer the bias from weak instrumental variables, horizontal pleiotropy and sample overlap. We introduce MRBEE (MR using Bias corrected Estimating Equation), a multivariable MR method capable of simultaneously removing weak instrument and sample overlap bias and identifying horizontal pleiotropy. Our extensive simulations and real data analyses reveal that MRBEE provides nearly unbiased estimates of causal effects, well-controlled type I error rates and higher power than comparably robust methods and is computationally efficient. Our real data analyses result in consistent causal effect estimates and offer valuable guidance for conducting multivariable MR studies, elucidating the roles of pleiotropy, and identifying total 42 horizontal pleiotropic loci missed previously that are associated with myopia, schizophrenia, and coronary artery disease.

Mechanism of nearshore sediment-facilitated oil transport: New insights from causal inference analysis

A quantitative understanding of spilled oil transport in a nearshore environment is challenging due to the complex physicochemical processes in aqueous conditions. The physicochemical processes involved in oil sinking mainly include oil dispersion, sediment settling, and oil-sediment interaction. For the first time, this work attempts to address the sinking mechanism in petroleum contaminant transport using structural causal models based on observed data. The effects of nearshore salinity distribution from the estuary to the ocean on those three processes are examined. The causal inference reveals sediment settling is the crucial process for oil sinking. Salinity indirectly affects oil sinking by promoting sediment settling rather than directly affecting oil-sediment interaction. The increase of salinity from 0‰ to 35‰ provides a natural enhancement for sediment settling. Notably, unbiased causal effect estimates demonstrate the strongest positive causal effect on the settling efficiency of sediments is posed by increasing oil dispersion effectiveness, with a normalized value of 1.023. The highest strength of the causal relationship between oil dispersion and sediment settling highlights the importance of the dispersing characteristics of spilled oil to sediment-facilitated oil transport. The employed logic, a data-driven method, will shed light on adopting advanced causal inference tools to unravel the complicated contaminants’ transport.

Polygenic score: An anchor holding the whole life course.

Polygenic score: An anchor holding the whole life course.

The Effect of Metformin on Vestibular Schwannoma Growth: A Systematic Review and Meta-analysis.

To systematically review and evaluate metformin's potential impact on vestibular schwannoma (VS) growth. PubMed, Cochrane Library, and Embase. A retrospective cohort study was performed on sporadic VS patients undergoing initial observation who had at least two magnetic resonance imaging studies. Patients were stratified by metformin use during the observation period. Primary endpoint was VS growth, defined as at least a 2 mm increase in diameter. Survival free of tumor growth was evaluated between groups. Systematic review and meta-analysis were performed to produce a pooled odds ratio [OR]. Study heterogeneity was assessed and post-hoc power analysis was performed. A total of 123 patients were included, of which 17% were taking metformin. Median patient age was 56.6 years (range, 25.1-84.5). There were no statistically significant differences between the groups. Survival analysis did not demonstrate a statistically significant difference in time to VS growth between groups (hazard ratio=0.61, 95% confidence interval [CI]=0.29-1.29). Furthermore, logistic regression analysis did not demonstrate a statistically significant difference between groups in the odds of VS growth (OR=0.46, 95% CI=0.17-1.27). Systematic review identified 3 studies. Meta-analysis suggested that metformin reduces the odds of developing VS growth (pooled OR=0.45, 95% CI=0.29-0.71). Studies demonstrated low between-study heterogeneity. Power analysis demonstrated a sample size of 220 patients with equal randomization would be required to prospectively identify a true difference with 80% power. Metformin use may reduce the odds of VS growth. A randomized trial would be ideal to identify an unbiased estimate of metformin's effect on VS growth. Laryngoscope, 133:2066-2072, 2023.

Breaking the winner’s curse in Mendelian randomization: Rerandomized inverse variance weighted estimator

Developments in genome-wide association studies and the increasing availability of summary genetic association data have made the application of two-sample Mendelian Randomization (MR) with summary data increasingly popular. Conventional two-sample MR methods often employ the same sample for selecting relevant genetic variants and for constructing final causal estimates. Such a practice often leads to biased causal effect estimates due to the well-known “winner’s curse” phenomenon. To address this fundamental challenge, we first examine its consequence on causal effect estimation both theoretically and empirically. We then propose a novel framework that systematically breaks the winner’s curse, leading to unbiased association effect estimates for the selected genetic variants. Building upon the proposed framework, we introduce a novel rerandomized inverse variance weighted estimator that is consistent when selection and parameter estimation are conducted on the same sample. Under appropriate conditions, we show that the proposed RIVW estimator for the causal effect converges to a normal distribution asymptotically and its variance can be well estimated. We illustrate the finite-sample performance of our approach through Monte Carlo experiments and two empirical examples.

Modeling multiple dependent variables in meta-analysis of single-case experimental design using multilevel modeling.

Although meta-analyses of single-case experimental design (SCED) often include multiple types of dependent variables (DVs), multiple DVs are rarely considered within models in the analysis. Baek et al. (Journal of Experimental Education, 90(4), 934-961, 2022) identified several statistical issues that arise when researchers fail to model multiple DVs in meta-analyses of SCED data. However, the degree to which non-modeling of multiple DVs impacts the results of the meta-analysis of SCED has not been fully examined. In this simulation study, we have systematically investigated the impact of non-modeling of multiple DVs when analyzing meta SCED data using multilevel modeling. The result demonstrates that modeling multiple DVs has advantages over the non-modeling option for meta-analysis of SCED. Modeling multiple DVs enables the determination of precise effects from different DVs in addition to the unbiased and accurate average effect and accurate estimates and inferences for the error variances at the study level as well as the observation level. The current study also reveals potential factors (i.e., the number of DVs, degree of heterogeneity in the level-1 error variances and autocorrelation, and presence of the moderator effect) that impact the precision and accuracy of the variance parameters.

Interaction-based Mendelian randomization with measured and unmeasured gene-by-covariate interactions.

Studies leveraging gene-environment (GxE) interactions within Mendelian randomization (MR) analyses have prompted the emergence of two similar methodologies: MR-GxE and MR-GENIUS. Such methods are attractive in allowing for pleiotropic bias to be corrected when using individual instruments. Specifically, MR-GxE requires an interaction to be explicitly identified, while MR-GENIUS does not. We critically examine the assumptions of MR-GxE and MR-GENIUS in the absence of a pre-defined covariate, and propose sensitivity analyses to evaluate their performance. Finally, we explore the effect of body mass index (BMI) upon systolic blood pressure (SBP) using data from the UK Biobank, finding evidence of a positive effect of BMI on SBP. We find both approaches share similar assumptions, though differences between the approaches lend themselves to differing research settings. Where a suitable gene-by-covariate interaction is observed MR-GxE can produce unbiased causal effect estimates. MR-GENIUS can circumvent the need to identify interactions, but as a consequence relies on either the MR-GxE assumptions holding globally, or additional information with respect to the distribution of pleiotropic effects in the absence of an explicitly defined interaction covariate.

Using adopted individuals to partition indirect maternal genetic effects into prenatal and postnatal effects on offspring phenotypes.

Maternal genetic effects can be defined as the effect of a mother's genotype on the phenotype of her offspring, independent of the offspring's genotype. Maternal genetic effects can act via the intrauterine environment during pregnancy and/or via the postnatal environment. In this manuscript, we present a simple extension to the basic adoption design that uses structural equation modelling (SEM) to partition maternal genetic effects into prenatal and postnatal effects. We examine the power, utility and type I error rate of our model using simulations and asymptotic power calculations. We apply our model to polygenic scores of educational attainment and birth weight associated variants, in up to 5,178 adopted singletons, 943 trios, 2687 mother-offspring pairs, 712 father-offspring pairs and 347,980 singletons from the UK Biobank. Our results show the expected pattern of maternal genetic effects on offspring birth weight, but unexpectedly large prenatal maternal genetic effects on offspring educational attainment. Sensitivity and simulation analyses suggest this result may be at least partially due to adopted individuals in the UK Biobank being raised by their biological relatives. We show that accurate modelling of these sorts of cryptic relationships is sufficient to bring type I error rate under control and produce asymptotically unbiased estimates of prenatal and postnatal maternal genetic effects. We conclude that there would be considerable value in following up adopted individuals in the UK Biobank to determine whether they were raised by their biological relatives, and if so, to precisely ascertain the nature of these relationships. These adopted individuals could then be incorporated into informative statistical genetics models like the one described in our manuscript to further elucidate the genetic architecture of complex traits and diseases.

Methods of analysis for survival outcomes with time-updated mediators, with application to longitudinal disease registry data.

Mediation analysis is a useful tool to illuminate the mechanisms through which an exposure affects an outcome but statistical challenges exist with time-to-event outcomes and longitudinal observational data. Natural direct and indirect effects cannot be identified when there are exposure-induced confounders of the mediator-outcome relationship. Previous measurements of a repeatedly-measured mediator may themselves confound the relationship between the mediator and the outcome. To overcome these obstacles, two recent methods have been proposed, one based on path-specific effects and one based on an additive hazards model and the concept of exposure splitting. We investigate these techniques, focusing on their application to observational datasets. We apply both methods to an analysis of the UK Cystic Fibrosis Registry dataset to identify how much of the relationship between onset of cystic fibrosis-related diabetes and subsequent survival acts through pulmonary function. Statistical properties of the methods are investigated using simulation. Both methods produce unbiased estimates of indirect and direct effects in scenarios consistent with their stated assumptions but, if the data are measured infrequently, estimates may be biased. Findings are used to highlight considerations in the interpretation of the observational data analysis.

Beyond mass: detecting secondary halo properties with galaxy-galaxy lensing

ABSTRACTSecondary halo properties beyond mass, such as the mass accretion rate (MAR), concentration, and the half mass scale, are essential in understanding the formation of large-scale structure and dark matter haloes. In this paper, we study the impact of secondary halo properties on the galaxy-galaxy lensing observable, ΔΣ. We build an emulator trained on N-body simulations to model ΔΣ and quantify the impact of different secondary parameters on the ΔΣ profile. We focus on the impact of MAR on ΔΣ. We show that a 3σ detection of variations in MAR at fixed halo mass could be achieved with the Hyper Suprime Cam survey assuming no baryonic effects and a proxy for MAR with scatter <1.5. We show that the full radial profile of ΔΣ depends on secondary properties at fixed halo mass. Consequently, an emulator that can perform full shape fitting yields better than two times improvement upon the constraints on MAR than only using the outer part of the halo. Finally, we highlight that miscentring and MAR impact the radial profile of ΔΣ in a similar fashion, implying that miscentring and MAR need to be modelled jointly for unbiased estimates of both effects. We show that present-day lensing data sets have the statistical capability to place constraints on halo MAR within our assumptions. Our analysis opens up new possibilities for observationally measuring the assembly history of the dark matter haloes that host galaxies and clusters.

Accounting for confounding by time, early intervention adoption, and time-varying effect modification in the design and analysis of stepped-wedge designs: application to a proposed study design to reduce opioid-related mortality

BackgroundBeginning in 2019, stepped-wedge designs (SWDs) were being used in the investigation of interventions to reduce opioid-related deaths in communities across the United States. However, these interventions are competing with external factors such as newly initiated public policies limiting opioid prescriptions, media awareness campaigns, and the COVID-19 pandemic. Furthermore, control communities may prematurely adopt components of the intervention as they become available. The presence of time-varying external factors that impact study outcomes is a well-known limitation of SWDs; common approaches to adjusting for them make use of a mixed effects modeling framework. However, these models have several shortcomings when external factors differentially impact intervention and control clusters.MethodsWe discuss limitations of commonly used mixed effects models in the context of proposed SWDs to investigate interventions intended to reduce opioid-related mortality, and propose extensions of these models to address these limitations. We conduct an extensive simulation study of anticipated data from SWD trials targeting the current opioid epidemic in order to examine the performance of these models in the presence of external factors. We consider confounding by time, premature adoption of intervention components, and time-varying effect modification— in which external factors differentially impact intervention and control clusters.ResultsIn the presence of confounding by time, commonly used mixed effects models yield unbiased intervention effect estimates, but can have inflated Type 1 error and result in under coverage of confidence intervals. These models yield biased intervention effect estimates when premature intervention adoption or effect modification are present. In such scenarios, models incorporating fixed intervention-by-time interactions with an unstructured covariance for intervention-by-cluster-by-time random effects result in unbiased intervention effect estimates, reach nominal confidence interval coverage, and preserve Type 1 error.ConclusionsMixed effects models can adjust for different combinations of external factors through correct specification of fixed and random time effects. Since model choice has considerable impact on validity of results and study power, careful consideration must be given to how these external factors impact study endpoints and what estimands are most appropriate in the presence of such factors.

CAN QUASI‐EXPERIMENTAL EVALUATIONS THAT RELY ON STATE LONGITUDINAL DATA SYSTEMS REPLICATE EXPERIMENTAL RESULTS?

AbstractDo quasi‐experimental (QE) studies conducted with baseline covariates that are typically available in the longitudinal administrative state databases yield unbiased effect estimates? This paper conducts a within‐study comparison (WSC) study that compares experimental impacts of early college high school (ECHS) attendance with QE impacts drawn from the state and locales. We find that (1) QE models for outcomes with natural (matching) pretests replicated the randomized benchmarks quite well; (2) the replication bias is not sensitive to type of propensity score model or method; and (3) imposing locational restrictions (i.e., local matching) on the comparison students––specifically choosing them from among non‐treatment students who came from the same feeder middle schools as the treatment students––does not decrease the QE bias; on the contrary, it performed worse than the models that did not impose this restriction for most outcomes. The first two findings are generally consistent with other education WSCs while the third one is not, suggesting that in cases where selection may be driven by individual‐level factors, such as this one, local matching may yield biased treatment effect estimates by greatly reducing the pool of potential comparison units and distorting balance on unobservable confounders while prioritizing balance on observable factors.

Modeling Parent-Specific Genetic Nurture in Families with Missing Parental Genotypes: Application to Birthweight and BMI.

Disaggregation and estimation of genetic effects from offspring and parents has long been of interest to statistical geneticists. Recently, technical and methodological advances have made the genome-wide and loci-specific estimation of direct offspring and parental genetic nurture effects more possible. However, unbiased estimation using these methods requires datasets where both parents and at least one child have been genotyped, which are relatively scarce. Our group has recently developed a method and accompanying software (IMPISH; Hwang et al. in PLoS Genet 16:e1009154, 2020) which is able to impute missing parental genotypes from observed data on sibships and estimate their effects on an offspring phenotype conditional on the effects of genetic transmission. However, this method is unable to disentangle maternal and paternal effects, which may differ in magnitude and direction. Here, we introduce an extension to the original IMPISH routine which takes advantage of all available nuclear families to impute parent-specific missing genotypes and obtain asymptotically unbiased estimates of genetic effects on offspring phenotypes. We apply this this method to data from related individuals in the UK Biobank, showing concordance with previous estimates of maternal genetic effects on offspring birthweight. We also conduct the first GWAS jointly estimating offspring-, maternal-, and paternal-specific genetic effects on body-mass index.

Statistical Modeling of Sensitive Period Effects Using the Structured Life Course Modeling Approach (SLCMA).

Sensitive periods are times during development when life experiences can have a greater impact on outcomes than at other periods during the life course. However, a dearth of sophisticated methods for studying time-dependent exposure-outcome relationships means that sensitive periods are often overlooked in research studies in favor of more simplistic and easier-to-use hypotheses such as ever being exposed, or the effect of an exposure accumulated over time. The structured life course modeling approach (SLCMA; pronounced "slick-mah") allows researchers to model complex life course hypotheses, such as sensitive periods, to determine which hypothesis best explains the amount of variation between a repeated exposure and an outcome. The SLCMA makes use of the least angle regression (LARS) variable selection technique, a type of least absolute shrinkage and selection operator (LASSO) estimation procedure, to yield a parsimonious model for the exposure-outcome relationship of interest. The results of the LARS procedure are complemented with a post-selection inference method, called selective inference, which provides unbiased effect estimates, confidence intervals, and p-values for the final explanatory model. In this chapter, we provide a brief overview of the genesis of this sensitive period modeling approach and provide a didactic step-by-step user's guide to implement the SLCMA in sensitive- period research. R code to complete the SLCMA is available on our GitHub page at: https://github.com/thedunnlab/SLCMA-pipeline .

Longitudinal Studies 2: Modeling Data Using Multivariate Analysis.

Statistical models are used to study the relationship between exposure and disease while accounting for the potential role of other factors' impact upon outcomes. This adjustment is useful to obtain unbiased estimates of true effects or to predict future outcomes. Statistical models include a systematic and an error component. The systematic component explains the variability of the response variable as a function of the predictors and is summarized in the effect estimates (model coefficients). The error element of the model represents the variability in the data unexplained by the model and is used to build measures of precisions around the point estimates (Confidence Intervals).

Calibration of Heterogeneous Treatment Effects in Random Experiments

Using machine learning to estimate the heterogeneity of treatment effects (HTE) in randomized experiments is common when firms and digital platforms seek to understand how individuals differ in their responses to a policy. However, will the average effect from the HTE model align with simple subgroup average effect estimates from random experiments? Using a large-scale random experiment on Facebook, we observe a substantial discrepancy between machine learning-based treatment effect estimates and the difference-in-means estimator from the same random experiment. We propose to use the quantile-quantile plot to diagnose whether there exists bias in the model-based estimates. To correct for the bias, we provide a model-agnostic method, in the vein of \citet{platt1999probabilistic} technique used in supervised learning, to calibrate black-box estimates of HTEs to known unbiased average effect estimates, ensuring that sign and magnitude approximate experimental benchmarks. Our method requires no additional data beyond what's necessary for estimating HTEs, and it can be scaled to arbitrarily large datasets. Our technique enables the use of stacking to ensemble estimates from multiple HTE models based on their out-of-sample estimates, improving performance. We demonstrate the effectiveness of our method in a random experiment run on Facebook. We also use extensive synthetic simulations to demonstrate two mechanisms of the bias and show the effectiveness of our method. The issue observed by our study and the proposed diagnostic and correction approach have strong implications for and broad applications in the IS communities and for online platforms.