smdi: an R package to perform structural missing data investigations on partially observed confounders in real-world evidence studies

Objectives

Partially observed confounder data pose a major challenge in statistical analyses aimed to inform causal inference using electronic health records (EHRs). While analytic approaches such as imputation are available, assumptions on underlying missingness patterns and mechanisms must be verified. We aimed to develop a toolkit to streamline missing data diagnostics to guide choice of analytic approaches based on meeting necessary assumptions.

Materials and methods

We developed the smdi (structural missing data investigations) R package based on results of a previous simulation study which considered structural assumptions of common missing data mechanisms in EHR.

Results

smdi enables users to run principled missing data investigations on partially observed confounders and implement functions to visualize, describe, and infer potential missingness patterns and mechanisms based on observed data.

Conclusions

The smdi R package is freely available on CRAN and can provide valuable insights into underlying missingness patterns and mechanisms and thereby help improve the robustness of real-world evidence studies.

Methodological Challenges and Confounders in Research on the Effects of Ketogenic Diets: A Literature Review of Meta-Analyses

Several meta-analyses have found a positive association between a popular type of "fad diet", ketogenic diets, and their effect on anthropometric and blood parameters. However, the non-specific inclusion criteria for meta-analyses may lead to incorrect conclusions. The aim of this literature review is to highlight the main confounders and methodological pitfalls of meta-analyses on ketogenic diets by inspecting the presence of key inclusion criteria. The PubMed, Embase, and Web of Science databases and the Cochrane Database of Systematic Reviews were searched for meta-analyses. Most meta-analyses did not define the essential parameters of a ketogenic diet (i.e., calories, macronutrient ratio, types of fatty acids, ketone bodies, etc.) as inclusion criteria. Of the 28 included meta-analyses, few addressed collecting real, re-measured nutritional data from the ketogenic diet and control groups in parallel with the pre-designed nutritional data. Most meta-analyses reported positive results in favor of ketogenic diets, which can result in erroneous conclusions considering the numerous methodological pitfalls and confounders. Well-designed clinical trials with comparable results and their meta-analyses are needed. Until then, medical professionals should not recommend ketogenic diets as a form of weight loss when other well-known dietary options have been shown to be healthy and effective.

Upper Airway-Related Symptoms According to Mental Illness and Sleep Disorders among Workers Employed by a Large Non-Profit Organization in the Mountain West Region of the United States

The relationships between selected upper airway-related symptoms (speech disturbances, voice disorders, cough, and breathing abnormalities) and mental illness and sleep disorders have been previously demonstrated. However, these relationships have not been compared in a single study with consideration of potential confounding variables. The current research incorporates a descriptive study design of medical claims data for employees (~21,362 per year 2017-2021) with corporate insurance to evaluate the strength of these relationships, adjusting for demographic variables and other important confounders. The upper airway-related symptoms are each significantly and positively associated with several mental illnesses and sleep disorders, after adjusting for demographic and other potential confounders. The rate of any mental illness is 138% (95% CI 93-195%) higher for speech disturbances, 55% (95% CI 28-88%) higher for voice disorders, 28% (95% CI 22-34%) higher for cough, and 58% (95% CI 50-66%) higher for breathing abnormalities, after adjustment for the confounding variables. Confounding had significant effects on the rate ratios involving cough and breathing abnormalities. The rate of any sleep disorder is 78% (95% CI 34-136%) higher for speech disturbances, 52% (95% CI 21-89%) higher for voice disorders, 34% (95% CI 27-41%) higher for cough, and 172% (95% CI 161-184%) higher for breathing abnormalities, after adjustment for the confounding variables. Confounding had significant effects on each of the upper airway-related symptoms. Rates of mental illness and sleep disorders are positively associated with the number of claims filed for each of the upper airway-related symptoms. The comorbid nature of these conditions should guide clinicians in providing more effective treatment plans that ultimately yield the best outcome for patients.

Causal inference in survival analysis under deterministic missingness of confounders in register data

Long-term register data offer unique opportunities to explore causal effects of treatments on time-to-event outcomes, in well-characterized populations with minimum loss of follow-up. However, the structure of the data may pose methodological challenges. Motivated by the Swedish Renal Registry and estimation of survival differences for renal replacement therapies, we focus on the particular case when an important confounder is not recorded in the early period of the register, so that the entry date to the register deterministically predicts confounder missingness. In addition, an evolving composition of the treatment arms populations, and suspected improved survival outcomes in later periods lead to informative administrative censoring, unless the entry date is appropriately accounted for. We investigate different consequences of these issues on causal effect estimation following multiple imputation of the missing covariate data. We analyse the performance of different combinations of imputation models and estimation methods for the population average survival. We further evaluate the sensitivity of our results to the nature of censoring and misspecification of fitted models. We find that an imputation model including the cumulative baseline hazard, event indicator, covariates and interactions between the cumulative baseline hazard and covariates, followed by regression standardization, leads to the best estimation results overall, in simulations. Standardization has two advantages over inverse probability of treatment weighting here: it can directly account for the informative censoring by including the entry date as a covariate in the outcome model, and allows for straightforward variance computation using readily available software.

Mesothelin Gene Variants Affect Soluble Mesothelin-Related Protein Levels in the Plasma of Asbestos-Exposed Males and Mesothelioma Patients from Germany

Malignant mesothelioma (MM) is a severe disease mostly caused by asbestos exposure. Today, one of the best available biomarkers is the soluble mesothelin-related protein (SMRP), also known as mesothelin. Recent studies have shown that mesothelin levels are influenced by individual genetic variability. This study aimed to investigate the influence of three mesothelin (MSLN) gene variants (SNPs) in the 5′-untranslated promoter region (5′-UTR), MSLN rs2235503 C > A, rs3764246 A > G, rs3764247 A > C, and one (rs1057147 G > A) in the 3′-untranslated region (3′-UTR) of the MSLN gene on plasma concentrations of mesothelin in 410 asbestos-exposed males without cancer and 43 males with prediagnostic MM (i.e., with MM diagnosed later on) from the prospective MoMar study, as well as 59 males with manifest MM from Germany. The mesothelin concentration differed significantly between the different groups (p < 0.0001), but not between the prediagnostic and manifest MM groups (p = 0.502). Five to eight mutations of the four SNP variants studied were associated with increased mesothelin concentrations (p = 0.001). The highest mesothelin concentrations were observed for homozygous variants of the three promotor SNPs in the 5′-UTR (p < 0.001), and the highest odds ratio for an elevated mesothelin concentration was observed for MSLN rs2235503 C > A. The four studied SNPs had a clear influence on the mesothelin concentration in plasma. Hence, the analysis of these SNPs may help to elucidate the diagnostic background of patients displaying increased mesothelin levels and might help to reduce false-positive results when using mesothelin for MM screening in high-risk groups.

DBSegment: Fast and robust segmentation of deep brain structures considering domain generalization

Segmenting deep brain structures from magnetic resonance images is important for patient diagnosis, surgical planning, and research. Most current state-of-the-art solutions follow a segmentation-by-registration approach, where subject magnetic resonance imaging (MRIs) are mapped to a template with well-defined segmentations. However, registration-based pipelines are time-consuming, thus, limiting their clinical use. This paper uses deep learning to provide a one-step, robust, and efficient deep brain segmentation solution directly in the native space. The method consists of a preprocessing step to conform all MRI images to the same orientation, followed by a convolutional neural network using the nnU-Net framework. We use a total of 14 datasets from both research and clinical collections. Of these, seven were used for training and validation and seven were retained for testing. We trained the network to segment 30 deep brain structures, as well as a brain mask, using labels generated from a registration-based approach. We evaluated the generalizability of the network by performing a leave-one-dataset-out cross-validation, and independent testing on unseen datasets. Furthermore, we assessed cross-domain transportability by evaluating the results separately on different domains. We achieved an average dice score similarity of 0.89 ± 0.04 on the test datasets when compared to the registration-based gold standard. On our test system, the computation time decreased from 43 min for a reference registration-based pipeline to 1.3 min. Our proposed method is fast, robust, and generalizes with high reliability. It can be extended to the segmentation of other brain structures. It is publicly available on GitHub, and as a pip package for convenient usage.

Causation and causal inference in obstetrics-gynecology

Causation and causal inference are of utmost importance in obstetrics and gynecology. In many clinical situations, causal reasoning is involved in etiological explanations, diagnostic considerations, and conversations about prognosis. In this paper, we offer an overview of the philosophical accounts of causation that may not be familiar to, but still be appreciated by, the busy clinician. In our discussion, we do not try to simplify what is a rather complex range of ideas. We begin with an introduction to some important basic ideas, followed by 2 sections on the metaphysical and epistemological aspects of causality, which offer a more detailed discussion of some of its specific philosophical facets, using examples from obstetrical and gynecologic research and practice along the way. We hope our discussion will help deepen the thinking and discourse about causation and causal inference in gynecology and obstetrics.

Confounding: Linking Evidence to Practice

A confounder is a variable that influences the relationship between an exposure and an outcome. This means that a difference in outcome is not only explained by the factor that we think explains it (the exposure), but also (completely or partly) by another factor (the confounder). Confounding underpins the hackneyed expression, "Correlation does not imply causation." Your assessment of a study's risk of confounding should inform how much confidence you have in the results. J Orthop Sports Phys Ther 2021;51(8):412-413. doi:10.2519/jospt.2021.0702.

Discovery of a Generalization Gap of Convolutional Neural Networks on COVID-19 X-Rays Classification

A number of recent papers have shown experimental evidence that suggests it is possible to build highly accurate deep neural network models to detect COVID-19 from chest X-ray images. In this paper, we show that good generalization to unseen sources has not been achieved. Experiments with richer data sets than have previously been used show models have high accuracy on seen sources, but poor accuracy on unseen sources. The reason for the disparity is that the convolutional neural network model, which learns features, can focus on differences in X-ray machines or in positioning within the machines, for example. Any feature that a person would clearly rule out is called a confounding feature. Some of the models were trained on COVID-19 image data taken from publications, which may be different than raw images. Some data sets were of pediatric cases with pneumonia where COVID-19 chest X-rays are almost exclusively from adults, so lung size becomes a spurious feature that can be exploited. In this work, we have eliminated many confounding features by working with as close to raw data as possible. Still, deep learned models may leverage source specific confounders to differentiate COVID-19 from pneumonia preventing generalizing to new data sources (i.e. external sites). Our models have achieved an AUC of 1.00 on seen data sources but in the worst case only scored an AUC of 0.38 on unseen ones. This indicates that such models need further assessment/development before they can be broadly clinically deployed. An example of fine-tuning to improve performance at a new site is given.

The Confounder-Mediator Dilemma: Should We Control for Obesity to Estimate the Effect of Perfluoroalkyl Substances on Health Outcomes?

Confounding adjustment is important for observational studies to derive valid effect estimates for inference. Despite the theoretical advancement of confounding selection procedure, it is often challenging to distinguish between confounders and mediators due to the lack of information about the time-ordering and latency of each variable in the data. This is also the case for the studies of perfluoroalkyl substances (PFAS), a group of synthetic chemicals used in industry and consumer products that are persistent and have endocrine-disrupting properties on health outcomes. In this article, we used directed acyclic graphs to describe potential biases introduced by adjusting for or stratifying by the measure of obesity as an intermediate variable in PFAS exposure analyses. We compared results with or without adjusting for body mass index in two cross-sectional data analyses: (1) PFAS levels and maternal thyroid function during early pregnancy using the Danish National Birth Cohort and (2) PFAS levels and cardiovascular disease in adults using the National Health and Nutrition Examination Survey. In these examples, we showed that the potential heterogeneity observed in stratified analyses by overweight or obese status needs to be interpreted cautiously considering collider stratification bias. This article highlights the complexity of seemingly simple adjustment or stratification analyses, and the need for careful consideration of the confounding and/or mediating role of obesity in PFAS studies.

Accounting for Expected Adjusted Effect

The point that adjustment for confounders do not always guarantee protection against spurious findings and type 1-errors has been made before. The present simulation study indicates that for traditional regression methods, this risk is accentuated by a large sample size, low reliability in the measurement of the confounder, and high reliability in the measurement of the predictor and the outcome. However, this risk might be attenuated by calculating the expected adjusted effect, or the required reliability in the measurement of the possible confounder, with equations presented in the present paper.

A Lesson in Standardization - Subtle Aspects of the Processing of Samples Can Greatly Affect Dogs' Learning

Training new medical odors presents challenges in procuring sufficient target samples, and suitably matched controls. Organizations are often forced to choose between using fewer samples and risking dogs learning individuals or using differently sourced samples. Even when aiming to standardize all aspects of collection, processing, storage and presentation, this risks there being subtle differences which dogs use to discriminate, leading to artificially high performance, not replicable when novel samples are presented. We describe lessons learnt during early training of dogs to detect prostate cancer from urine. Initially, six dogs were trained to discriminate between hospital-sourced target and externally-sourced controls believed to be processed and stored the same way. Dogs performed well: mean sensitivity 93.5% (92.2–94.5) and specificity 87.9% (78.2–91.9). When training progressed to include hospital-sourced controls, dogs greatly decreased in specificity 67.3% (43.2–83.3). Alerted to a potential issue, we carried out a methodical, investigation. We presented new strategically chosen samples to the dogs and conducted a logistic regression analysis to ascertain which factor most affected specificity. We discovered the two sets of samples varied in a critical aspect, hospital-processed samples were tested by dipping the urinalysis stick into the sample, whilst for externally sourced samples a small amount of urine was poured onto the stick. Dogs had learnt to distinguish target aided by the odor of this stick. This highlights the importance of considering every aspect of sample processing even when using urine, often believed to be less susceptible to contamination than media like breath.

Quantifying the bias due to observed individual confounders in causal treatment effect estimates

It is often of interest to use observational data to estimate the causal effect of a target exposure or treatment on an outcome. When estimating the treatment effect, it is essential to appropriately adjust for selection bias due to observed confounders using, for example, propensity score weighting. Selection bias due to confounders occurs when individuals who are treated are substantially different from those who are untreated with respect to covariates that are also associated with the outcome. A comparison of the unadjusted, naive treatment effect estimate with the propensity score adjusted treatment effect estimate provides an estimate of the selection bias due to these observed confounders. In this article, we propose methods to identify the observed covariate that explains the largest proportion of the estimated selection bias. Identification of the most influential observed covariate or covariates is important in resource-sensitive settings where the number of covariates obtained from individuals needs to be minimized due to cost and/or patient burden and in settings where this covariate can provide actionable information to healthcare agencies, providers, and stakeholders. We propose straightforward parametric and nonparametric procedures to examine the role of observed covariates and quantify the proportion of the observed selection bias explained by each covariate. We demonstrate good finite sample performance of our proposed estimates using a simulation study and use our procedures to identify the most influential covariates that explain the observed selection bias in estimating the causal effect of alcohol use on progression of Huntington's disease, a rare neurological disease.

Doubly robust estimation and causal inference for recurrent event data

Many longitudinal databases record the occurrence of recurrent events over time. In this article, we propose a new method to estimate the average causal effect of a binary treatment for recurrent event data in the presence of confounders. We propose a doubly robust semiparametric estimator based on a weighted version of the Nelson-Aalen estimator and a conditional regression estimator under an assumed semiparametric multiplicative rate model for recurrent event data. We show that the proposed doubly robust estimator is consistent and asymptotically normal. In addition, a model diagnostic plot of residuals is presented to assess the adequacy of our proposed semiparametric model. We then evaluate the finite sample behavior of the proposed estimators under a number of simulation scenarios. Finally, we illustrate the proposed methodology via a database of circus artist injuries.

Direction Dependence Analysis in the Presence of Confounders: Applications to Linear Mediation Models Using Observational Data

Statistical methods to identify mis-specifications of linear regression models with respect to the direction of dependence (i.e. whether x→y or y→x better approximates the data-generating mechanism) have received considerable attention. Direction dependence analysis (DDA) constitutes such a statistical tool and makes use of higher-moment information of variables to derive statements concerning directional model mis-specifications in observational data. Previous studies on direction of dependence mainly focused on statistical inference and guidelines for the selection from the two directionally competing candidate models (x→y versus y→x) while assuming the absence of unobserved common causes. The present study describes properties of DDA when confounders are present and extends existing DDA methodology by incorporating the confounder model as a possible explanation. We show that all three explanatory models can be uniquely identified under standard DDA assumptions. Further, we discuss the proposed approach in the context of testing competing mediation models and evaluate an organizational model proposing a mediational relation between school leadership and student achievement via school safety using observational data from an urban school district. Overall, DDA provides strong empirical support that school safety has indeed a causal effect on student achievement but suggests that important confounders are present in the school leadership-safety relation.

Misinterpretations of Guidelines Leading to Incorrect Diagnosis of Brain Death: A Case Report and Discussion

Guidelines describe the process necessary for the diagnosis of brain death. We present a case of a 3-month-old former 36-week-gestation infant after a prolonged out-of-hospital cardiac arrest of 37 minutes who was clinically diagnosed as brain dead at 120 hours after the event. Unusual findings included a normal slightly sunken anterior fontanelle, normal cerebral blood flow perfusion scan at 73 hours after the event, only localized parieto-temporal edema on the latest computed tomographic (CT) scan of the brain at 48 hours after the event, and discussion of whether nonconvulsive seizures could have confounded the examination for brain death. In light of these unusual findings, we discuss and highlight what may be common misinterpretations of brain death guidelines that led to the mistaken diagnosis of death (as opposed to severe neurologic injury) in this child.

An efficient integrative resampling method for gene-trait association analysis

Genetic association studies are popular for identifying genetic variants, such as single nucleotide polymorphisms (SNPs), that are associated with complex traits. Statistical tests are commonly performed one SNP at a time with an assumed mode of inheritance such as recessive, additive, or dominant genetic model. Such analysis can result in inadequate power when the employed model deviates from the underlying true genetic model. We propose an integrative association test procedure under a generalized linear model framework to flexibly model the data from the above three common genetic models and beyond. A computationally efficient resampling procedure is adopted to estimate the null distribution of the proposed test statistic. Simulation results show that our methods maintain the Type I error rate irrespective of the existence of confounding covariates and achieve adequate power compared to the methods with the true genetic model. The new methods are applied to two genetic studies on the resistance of severe malaria and sarcoidosis.

Adjusting for Principal Components of Molecular Phenotypes Induces Replicating False Positives

High-throughput measurements of molecular phenotypes provide an unprecedented opportunity to model cellular processes and their impact on disease. These highly structured datasets are usually strongly confounded, creating false positives and reducing power. This has motivated many approaches based on principal components analysis (PCA) to estimate and correct for confounders, which have become indispensable elements of association tests between molecular phenotypes and both genetic and nongenetic factors. Here, we show that these correction approaches induce a bias, and that it persists for large sample sizes and replicates out-of-sample. We prove this theoretically for PCA by deriving an analytic, deterministic, and intuitive bias approximation. We assess other methods with realistic simulations, which show that perturbing any of several basic parameters can cause false positive rate (FPR) inflation. Our experiments show the bias depends on covariate and confounder sparsity, effect sizes, and their correlation. Surprisingly, when the covariate and confounder have [Formula: see text], standard two-step methods all have [Formula: see text]-fold FPR inflation. Our analysis informs best practices for confounder correction in genomic studies, and suggests many false discoveries have been made and replicated in some differential expression analyses.

Is exposure to personal music players a confounder in adolescent mobile phone use and hearing health studies?

Objective

This study was performed to determine whether exposure to personal music players (PMPs) in the immediate morning prior to hearing testing confounds the association between mobile phone use and hearing thresholds of adolescents.

Design

In this cohort study of cognitive function in year 7 students (median age 13 years, range 11–14), information regarding the weekly use of mobile phones and the use of PMPs was assessed by a questionnaire. Pure-tone audiometry was used to establish hearing thresholds for all participants.

Results

Among a cohort of 317 adolescents (60.9% females), 130 were unexposed to PMP use while 33 were exposed to PMP use in the morning prior to hearing testing. No statistically significant difference in hearing threshold shifts was found between adolescents who were and were not exposed to PMP use prior to hearing testing. Likewise, the difference in the use of mobile phones according to the PMP use status was not statistically significant.

Conclusion

Exposure to PMPs prior to hearing testing did not introduce confounding in the present study of mobile phone use and hearing loss among adolescents.

Confounding, causality, and confusion: the role of intermediate variables in interpreting observational studies in obstetrics

Prospective and retrospective cohorts and case-control studies are some of the most important study designs in epidemiology because, under certain assumptions, they can mimic a randomized trial when done well. These assumptions include, but are not limited to, properly accounting for 2 important sources of bias: confounding and selection bias. While not adjusting the causal association for an intermediate variable will yield an unbiased estimate of the exposure-outcome's total causal effect, it is often that obstetricians will want to adjust for an intermediate variable to assess if the intermediate is the underlying driver of the association. Such a practice must be weighed in light of the underlying research question and whether such an adjustment is necessary should be carefully considered. Gestational age is, by far, the most commonly encountered variable in obstetrics that is often mislabeled as a confounder when, in fact, it may be an intermediate. If, indeed, gestational age is an intermediate but if mistakenly labeled as a confounding variable and consequently adjusted in an analysis, the conclusions can be unexpected. The implications of this overadjustment of an intermediate as though it were a confounder can render an otherwise persuasive study downright meaningless. This commentary provides an exposition of confounding bias, collider stratification, and selection biases, with applications in obstetrics and perinatal epidemiology.

Confounding factors affecting the National Institutes of Health (NIH) chronic Graft-Versus-Host Disease Organ-Specific Score and global severity

The 2005 NIH chronic GVHD (cGVHD) organ severity is based on the assessment of current status regardless of whether abnormalities are due to GVHD. The score assignment does not require knowledge of past manifestations, attribution or whether cGVHD is still active. The aim of this study is to describe confounding factors affecting organ scores in patients with cGVHD. The study included 189 consecutive cGVHD patients evaluated at our center in 2013. Providers completed the NIH 0-3 organ-specific scoring evaluation with two questions added for each organ to identify abnormalities that were (i) not attributed to cGVHD or (ii) attributed to cGVHD plus other causes. Abnormalities attributed to causes other than GVHD were recorded. Eighty (14%) abnormalities were not attributed to cGVHD in at least one organ, and 41 (7%) abnormalities were attributed to cGVHD plus other causes in at least one organ. A total of 436 (78%) abnormalities were attributed only to cGVHD. Abnormalities not attributed to cGVHD were observed most frequently in the lung, gastrointestinal tract and skin. Most common abnormalities included pre-transplant condition, sequelae from GVHD, deconditioning, infections and medications. Our results support the 2014 NIH consensus recommendation to consider attribution when scoring organ abnormalities.

Bariatric Surgery and Emergency Department Visits and Hospitalizations for Heart Failure Exacerbation: Population-Based, Self-Controlled Series

BACKGROUND

The United States is battling obesity and heart failure (HF) epidemics. Although studies have suggested relationships between obesity and HF morbidity, little is known regarding the effects of substantial weight reduction in obese patients with HF.

OBJECTIVES

This study investigated whether bariatric surgery is associated with a decreased rate of HF exacerbation.

METHODS

We performed a self-controlled case series study of obese patients with HF who underwent bariatric surgery, using the population-based emergency department (ED) and inpatient sample in California, Florida, and Nebraska. Primary outcome was ED visit or hospitalization for HF exacerbation from 2005 to 2011. We used conditional logistic regression to compare the outcome event rate during sequential 12-month periods, using pre-surgery months 13 to 24 as the reference period.

RESULTS

We identified 524 patients with HF who underwent bariatric surgery. During the reference period, 16.2% of patients had an ED visit or hospitalization for HF exacerbation. The rate remained unchanged in the subsequent 12-month pre-surgery period (15.3%; p = 0.67). In the first 12-month period after bariatric surgery, we observed a nonsignificantly reduced rate (12.0%; p = 0.052). However, the rate was significantly lower in the subsequent 13 to 24 months after bariatric surgery (9.9%; adjusted odds ratio: 0.57; p = 0.003). By contrast, there was no significant reduction in the rate of HF exacerbation among obese patients who underwent nonbariatric surgery (i.e., cholecystectomy, hysterectomy).

CONCLUSIONS

Our findings indicate that bariatric surgery is associated with a decline in the rate of HF exacerbation requiring ED evaluation or hospitalization among obese patients with HF.

Assessing Omitted Confounder Bias in Multilevel Mediation Models

To draw valid inference about an indirect effect in a mediation model, there must be no omitted confounders. No omitted confounders means that there are no common causes of hypothesized causal relationships. When the no-omitted-confounder assumption is violated, inference about indirect effects can be severely biased and the results potentially misleading. Despite the increasing attention to address confounder bias in single-level mediation, this topic has received little attention in the growing area of multilevel mediation analysis. A formidable challenge is that the no-omitted-confounder assumption is untestable. To address this challenge, we first analytically examined the biasing effects of potential violations of this critical assumption in a two-level mediation model with random intercepts and slopes, in which all the variables are measured at Level 1. Our analytic results show that omitting a Level 1 confounder can yield misleading results about key quantities of interest, such as Level 1 and Level 2 indirect effects. Second, we proposed a sensitivity analysis technique to assess the extent to which potential violation of the no-omitted-confounder assumption might invalidate or alter the conclusions about the indirect effects observed. We illustrated the methods using an empirical study and provided computer code so that researchers can implement the methods discussed.

Evaluation in medical education: A topical review of target parameters, data collection tools and confounding factors

Background and objective: Evaluation is an integral part of education in German medical schools. According to the quality standards set by the German Society for Evaluation, evaluation tools must provide an accurate and fair appraisal of teaching quality. Thus, data collection tools must be highly reliable and valid. This review summarises the current literature on evaluation of medical education with regard to the possible dimensions of teaching quality, the psychometric properties of survey instruments and potential confounding factors.

Methods: We searched Pubmed, PsycINFO and PSYNDEX for literature on evaluation in medical education and included studies published up until June 30, 2011 as well as articles identified in the “grey literature”. Results are presented as a narrative review.

Results: We identified four dimensions of teaching quality: structure, process, teacher characteristics, and outcome. Student ratings are predominantly used to address the first three dimensions, and a number of reliable tools are available for this purpose. However, potential confounders of student ratings pose a threat to the validity of these instruments. Outcome is usually operationalised in terms of student performance on examinations, but methodological problems may limit the usability of these data for evaluation purposes. In addition, not all examinations at German medical schools meet current quality standards.

Conclusion: The choice of tools for evaluating medical education should be guided by the dimension that is targeted by the evaluation. Likewise, evaluation results can only be interpreted within the context of the construct addressed by the data collection tool that was used as well as its specific confounding factors.

Causal diagrams and multivariate analysis III: confound it!

This commentary concludes my series concerning inclusion of variables in multivariate analyses. We take up the issues of confounding and effect modification and summarize the work we have thus far done. Finally, we provide a rough algorithm to help guide us through the maze of possibilities that we have outlined.

Causal diagrams and multivariate analysis II: precision work

In this Investigators' Corner, I continue my discussion of when and why we researchers should include variables in multivariate regression. My examination focuses on studies comparing treatment groups and situations for which we can either exclude variables from multivariate analyses or include them for reasons of precision.

Causal diagrams and multivariate analysis I: a quiver full of arrows

How do we know which variables we should include in our multivariate analyses? What role does each variable play in our understanding of the analysis? In this article I begin a discussion of these issues and describe 2 different types of studies for which this problem must be handled in different ways.

Adjustment for smoking does not alter the FOXO3A association with longevity

Human longevity is a multifactorial phenotype influenced by both genetic and environmental factors. Despite its heritability of 25-32 %, the genetic background of longevity is as yet largely unexplained. Apart from APOE status, variation in the FOXO3A gene is the only confirmed genetic contributor to survival into old age. On the other hand, FOXO3A activity is known to be downregulated in various cancers, and the gene was recently identified as a novel deletion hotspot in human lung adenocarcinoma. In view of the strong association between smoking and lung cancer, we set out to explore whether smoking modifies the known association between FOXO3A variation and longevity. To this end, we conducted a case-control study in two different populations, drawing upon extensive collections of old-aged individuals and younger controls available to us (1,613 German centenarians/nonagenarians and 1,104 controls; 1,088 Danish nonagenarians and 736 controls). In the German sample, 21 single nucleotide polymorphisms (SNPs) from the FOXO3A gene region were genotyped, whereas 15 FOXO3A SNPs were analyzed in the Danish sample. Eight SNPs were typed in both populations. Logistic regression analysis revealed that adjustment for smoking does not systematically alter the association between FOXO3A variation and longevity in neither population. Our analysis therefore suggests that the said association is not largely due to the confounding effects of lung cancer.

Causal inference, probability theory, and graphical insights

Causal inference from observational studies is a fundamental topic in biostatistics. The causal graph literature typically views probability theory as insufficient to express causal concepts in observational studies. In contrast, the view here is that probability theory is a desirable and sufficient basis for many topics in causal inference for the following two reasons. First, probability theory is generally more flexible than causal graphs: Besides explaining such causal graph topics as M-bias (adjusting for a collider) and bias amplification and attenuation (when adjusting for instrumental variable), probability theory is also the foundation of the paired availability design for historical controls, which does not fit into a causal graph framework. Second, probability theory is the basis for insightful graphical displays including the BK-Plot for understanding Simpson's paradox with a binary confounder, the BK2-Plot for understanding bias amplification and attenuation in the presence of an unobserved binary confounder, and the PAD-Plot for understanding the principal stratification component of the paired availability design.

Inverse probability weighted estimating equations for randomized trials in transfusion medicine

Thrombocytopenia is a condition characterized by extremely low platelet counts, which puts patients at elevated risk of morbidity and mortality because of bleeding. Trials in transfusion medicine are routinely designed to assess the effect of experimental platelet products on patients' platelet counts. In such trials, patients may receive multiple platelet transfusions over a predefined period of treatment, and a response is available from each such administration. The resulting data comprised multiple responses per patient, and although it is natural to want to use this data in testing for treatment effects, naive analyses of the multiple responses can yield biased estimates of the probability of response and associated treatment effects. These biases arise because only subsets of the patients randomized contribute response data on the second and subsequent administrations of therapy and the balance between treatment groups with respect to potential confounding factors is lost. We discuss the design and analysis issues involved in this setting and make recommendations for the design of future platelet transfusion trials.

On the definition of a confounder

The causal inference literature has provided a clear formal definition of confounding expressed in terms of counterfactual independence. The causal inference literature has not, however, produced a clear formal definition of a confounder, as it has given priority to the concept of confounding over that of a confounder. We consider a number of candidate definitions arising from various more informal statements made in the literature. We consider the properties satisfied by each candidate definition, principally focusing on (i) whether under the candidate definition control for all "confounders" suffices to control for "confounding" and (ii) whether each confounder in some context helps eliminate or reduce confounding bias. Several of the candidate definitions do not have these two properties. Only one candidate definition of those considered satisfies both properties. We propose that a "confounder" be defined as a pre-exposure covariate C for which there exists a set of other covariates X such that effect of the exposure on the outcome is unconfounded conditional on (X, C) but such that for no proper subset of (X, C) is the effect of the exposure on the outcome unconfounded given the subset. A variable that helps reduce bias but not eliminate bias we propose referring to as a "surrogate confounder."

Are type 2 diabetes patients who self-monitor blood glucose special? The role of confounders in the observational ROSSO study

BACKGROUND

In the German multicenter, retrospective cohort study (ROSSO), those patients with type 2 diabetes who performed self-monitoring of blood glucose (SMBG) had a better long-term clinical outcome. We analyzed whether confounders accounted for the lower rate of clinical events in the SMBG cohort.

METHODS

ROSSO followed 3268 persons from diagnosis of type 2 diabetes for a mean of 6.5 years. Data were retrieved from patient files of randomly contacted primary care practices.

RESULTS

In total, more than 60 potential confounders were documented, including nondisease-associated parameters such as patient's health insurance, marital status, habitation, and characteristics of diabetes centers. There were only modest differences for these parameters between groups with versus without SMBG, and multiple adjustments did not weaken the association of SMBG use with better outcome (odds ratio 0.65, 95% confidence interval 0.53-0.81, p < .001). This was also true for subgroups of patients defined by type of antidiabetes treatment. Propensity score analysis confirmed the association of SMBG use with outcome. Using key baseline parameters, 813 matching pairs of patients were identified. The analysis again showed a better long-term outcome in the SMBG group (hazard ratio 0.67 p = .004).

CONCLUSION

An influence of nonrecognized confounders on better outcome in the SMBG group is rendered improbable by similar results obtained with adjustments for disease-associated or disease-independent parameters, by the analysis of patient subgroups, by propensity score analysis and by performing a matched-pair analysis. The higher flexibility in pharmacological antidiabetes treatment regimens in the SMBG cohort suggests a different attitude of treating physicians and patients in association with SMBG.

The conundrum of unmeasured confounding: Comment on: "Can some of the detrimental neurodevelopmental effects attributed to lead be due to pesticides? by Brian Gulson"

The problem described by Dr. Brian Gulson - confounding by unmeasured exposures to pesticides - is only the most recent in a series of potential confounders cited to explain the observed effect of lead on children's intellectual abilities or behavioral problems. Despite the persistent problem of unmeasured confounders, there are several lines of evidence implicating lead as a toxicant at blood lead levels <10 microg/dL. First, in striking contrast with pesticides, there is considerable evidence from numerous studies linking low-level lead exposure with cognitive deficits and behavioral problems, even after controlling for a variety of potential confounders. Second, the consistency of evidence from diverse cohorts and distinct, if not always directly measured potential confounders - enhances our confidence that the lead effect observed at blood lead levels <10 microg/dL is not attributable to unmeasured confounders. Third, in our reanalysis of the Rochester Lead Study, the inclusion of parent-reported mouthing behaviors and breastfeeding status did not attenuate the effect of lead exposure on children's intellectual function. Finally, although we can never entirely dismiss unmeasured confounding in observational studies, we can rely on experimental studies of lead-exposed animals to confirm that lead is a toxicant. Thus, while we must remain vigilant for unmeasured or poorly measured confounders, it is crucial to balance the endless search for confounders with the evidence of toxicity and the need to take action to protect public health. The alternative, to perpetually permit children to be exposed to lead and other emerging toxicants, is both absurd and unacceptable.

Arsenic cancer risk confounder in southwest Taiwan data set

Quantitative analysis for the risk of human cancer from the ingestion of inorganic arsenic has been based on the reported cancer mortality experience in the blackfoot disease (BFD) -endemic area of southwest Taiwan. Linear regression analysis shows that arsenic as the sole etiologic factor accounts for only 21% of the variance in the village standardized mortality ratios for bladder and lung cancer. A previous study had reported the influence of confounders (township, BFD prevalence, and artesian well dependency) qualitatively, but they have not been introduced into a quantitative assessment. In this six-township study, only three townships (2, 4, and 6) showed a significant positive dose-response relationship with arsenic exposure. The other three townships (0, 3, and 5) demonstrated significant bladder and lung cancer risks that were independent of arsenic exposure. The data for bladder and lung cancer mortality for townships 2, 4, and 6 fit an inverse linear regression model (p < 0.001) with an estimated threshold at 151 microg/L (95% confidence interval, 42 to 229 microg/L) . Such a model is consistent with epidemiologic and toxicologic literature for bladder cancer. Exploration of the southwest Taiwan cancer mortality data set has clarified the dose-response relationship with arsenic exposure by separating out township as a confounding factor. Key words: arsenic, blackfoot disease, bladder cancer, cancer risk, confounder, dose-response relationship, southwest Taiwan, threshold model.