Search

Assessing heterogeneous treatment effects (HTEs) is an essential task in epidemiology. The recent integration of machine learning into causal inference has provided a new, flexible tool for evaluating complex HTEs: causal forest. In a recent paper, Jawadekar et al (Am J Epidemiol. 2023;192(7):1155-1165) introduced this innovative approach and offered practical guidelines for applied users. Building on their work, this commentary provides additional insights and guidance to promote the understanding and application of causal forest in epidemiologic research. We start with conceptual clarifications, differentiating between honesty and cross-fitting, and exploring the interpretation of estimated conditional average treatment effects. We then delve into practical considerations not addressed by Jawadekar et al, including motivations for estimating HTEs, calibration approaches, and ways to leverage causal forest output with examples from simulated data. We conclude by outlining challenges to consider for future advancements and applications of causal forest in epidemiologic research. [ABSTRACT FROM AUTHOR]

In this paper, we present a practical approach for computing the sandwich variance estimator in 2-stage regression model settings. As a motivating example for 2-stage regression, we consider regression calibration, a popular approach for addressing covariate measurement error. The sandwich variance approach has rarely been applied in regression calibration, despite its requiring less computation time than popular resampling approaches for variance estimation, specifically the bootstrap. This is probably because it requires specialized statistical coding. Here we first outline the steps needed to compute the sandwich variance estimator. We then develop a convenient method of computation in R for sandwich variance estimation, which leverages standard regression model outputs and existing R functions and can be applied in the case of a simple random sample or complex survey design. We use a simulation study to compare the sandwich estimator to a resampling variance approach for both settings. Finally, we further compare these 2 variance estimation approaches in data examples from the Women's Health Initiative (1993–2005) and the Hispanic Community Health Study/Study of Latinos (2008–2011). In our simulations, the sandwich variance estimator typically had good numerical performance, but simple Wald bootstrap confidence intervals were unstable or overcovered in certain settings, particularly when there was high correlation between covariates or large measurement error. [ABSTRACT FROM AUTHOR]

In this paper, we consider methods for generating draws of a binary random variable whose expectation conditional on covariates follows a logistic regression model with known covariate coefficients. We examine approximations for finding a "balancing intercept," that is, a value for the intercept of the logistic model that leads to a desired marginal expectation for the binary random variable. We show that a recently proposed analytical approximation can produce inaccurate results, especially when targeting more extreme marginal expectations or when the linear predictor of the regression model has high variance. We then formulate the balancing intercept as a solution to an integral equation, implement a numerical approximation for solving the equation based on Monte Carlo methods, and show that the approximation works well in practice. Our approach to the basic problem of the balancing intercept provides an example of a broadly applicable strategy for formulating and solving problems that arise in the design of simulation studies used to evaluate or teach epidemiologic methods. [ABSTRACT FROM AUTHOR]

Measurement error is pervasive in epidemiologic research. Epidemiologists often assume that mismeasurement of study variables is nondifferential with respect to other analytical variables and then rely on the heuristic that "nondifferential misclassification will bias estimates towards the null." However, there are many exceptions to the heuristic for which bias towards the null cannot be assumed. In this paper, we compile and characterize 7 exceptions to this rule and encourage analysts to take a more critical and nuanced approach to evaluating and discussing bias from nondifferential mismeasurement. [ABSTRACT FROM AUTHOR]

An increasing number of recent studies have suggested that doubly robust estimators with cross-fitting should be used when estimating causal effects with machine learning methods. However, not all existing programs that implement doubly robust estimators support machine learning methods and cross-fitting, or provide estimates on multiplicative scales. To address these needs, we developed AIPW , a software package implementing augmented inverse probability weighting (AIPW) estimation of average causal effects in R (R Foundation for Statistical Computing, Vienna, Austria). Key features of the AIPW package include cross-fitting and flexible covariate adjustment for observational studies and randomized controlled trials (RCTs). In this paper, we use a simulated RCT to illustrate implementation of the AIPW estimator. We also perform a simulation study to evaluate the performance of the AIPW package compared with other doubly robust implementations, including CausalGAM , npcausal , tmle , and tmle3. Our simulation showed that the AIPW package yields performance comparable to that of other programs. Furthermore, we also found that cross-fitting substantively decreases the bias and improves the confidence interval coverage for doubly robust estimators fitted with machine learning algorithms. Our findings suggest that the AIPW package can be a useful tool for estimating average causal effects with machine learning methods in RCTs and observational studies. [ABSTRACT FROM AUTHOR]

In this paper, we evaluate 11 measures of inequality, d (p 1, p 2), between 2 proportions p 1 and p 2, some of which are new to the health disparities literature. These measures are selected because they are continuous, nonnegative, equal to 0 if and only if | p 1 − p 2| = 0, and maximal when | p 1 − p 2| = 1. They are also symmetrical [ d (p 1, p 2) =  d (p 2, p 1)] and complement-invariant [ d (p 1, p 2) =  d (1 − p 2, 1 − p 1)]. To study intermeasure agreement, 5 of the 11 measures, including the absolute difference, are retained, because they remain finite and are maximal if and only if | p 1 − p 2| = 1. Even when the 2 proportions are assumed to be drawn at random from a shared distribution—interpreted as the absence of an avoidable difference—the expected value of d (p 1, p 2) depends on the shape of the distribution (and the choice of d) and can be quite large. To allow for direct comparisons among measures, we propose a standard measurement unit akin to a z score. For skewed underlying beta distributions, 4 of the 5 retained measures, once standardized, offer more conservative assessments of the magnitude of inequality than the absolute difference. We conclude that, even for measures that share the highlighted mathematical properties, magnitude comparisons are most usefully assessed relative to an elicited or estimated underlying distribution for the 2 proportions. [ABSTRACT FROM AUTHOR]

Epidemiologists often compare the observed number of deaths in a cohort with the expected number of deaths, obtained by multiplying person-time accrued in the cohort by mortality rates for a reference population (ideally, a reference that represents the mortality rate in the cohort in the absence of exposure). However, if exposure is hazardous (or salutary), this calculation will not consistently estimate the number of deaths expected in the absence of exposure because exposure will have affected the distribution of person-time observed in the study cohort. While problems with interpretation of this standard calculation of expected counts were discussed more than 2 decades ago, these discussions had little impact on epidemiologic practice. The logic of counterfactuals may help clarify this topic as we revisit these issues. In this paper, we describe a simple way to consistently estimate the expected number of deaths in such settings, and we illustrate the approach using data from a cohort study of mortality among underground miners. [ABSTRACT FROM AUTHOR]

Recently, many new approaches, study designs, and statistical and analytical methods have emerged for studying gene-environment interactions (GxEs) in large-scale studies of human populations. There are opportunities in this field, particularly with respect to the incorporation of -omics and next-generation sequencing data and continual improvement in measures of environmental exposures implicated in complex disease outcomes. In a workshop called "Current Challenges and New Opportunities for Gene-Environment Interaction Studies of Complex Diseases," held October 17-18, 2014, by the National Institute of Environmental Health Sciences and the National Cancer Institute in conjunction with the annual American Society of Human Genetics meeting, participants explored new approaches and tools that have been developed in recent years for GxE discovery. This paper highlights current and critical issues and themes in GxE research that need additional consideration, including the improved data analytical methods, environmental exposure assessment, and incorporation of functional data and annotations. [ABSTRACT FROM AUTHOR]

The analysis of gene-environment interaction (GxE) may hold the key for further understanding the etiology of many complex traits. The current availability of high-volume genetic data, the wide range in types of environmental data that can be measured, and the formation of consortiums of multiple studies provide new opportunities to identify GxE but also new analytical challenges. In this article, we summarize several statistical approaches that can be used to test for GxE in a genome-wide association study. These include traditional models of GxE in a case-control or quantitative trait study as well as alternative approaches that can provide substantially greater power. The latest methods for analyzing GxE with gene sets and with data in a consortium setting are summarized, as are issues that arise due to the complexity of environmental data. We provide some speculation on why detecting GxE in a genome-wide association study has thus far been difficult. We conclude with a description of software programs that can be used to implement most of the methods described in the paper. [ABSTRACT FROM AUTHOR]

There is no complete solution for the problem of abuse of statistics, but methodological training needs to cover cognitive biases and other psychosocial factors affecting inferences. The present paper discusses 3 common cognitive distortions: 1) dichotomania, the compulsion to perceive quantities as dichotomous even when dichotomization is unnecessary and misleading, as in inferences based on whether a P value is "statistically significant"; 2) nullism, the tendency to privilege the hypothesis of no difference or no effect when there is no scientific basis for doing so, as when testing only the null hypothesis; and 3) statistical reification, treating hypothetical data distributions and statistical models as if they reflect known physical laws rather than speculative assumptions for thought experiments. As commonly misused, null-hypothesis significance testing combines these cognitive problems to produce highly distorted interpretation and reporting of study results. Interval estimation has so far proven to be an inadequate solution because it involves dichotomization, an avenue for nullism. Sensitivity and bias analyses have been proposed to address reproducibility problems (Am J Epidemiol. 2017;186(6):646--647); these methods can indeed address reification, but they can also introduce new distortions via misleading specifications for bias parameters. P values can be reframed to lessen distortions by presenting them without reference to a cutoff, providing them for relevant alternatives to the null, and recognizing their dependence on all assumptions used in their computation; they nonetheless require rescaling for measuring evidence. I conclude that methodological development and training should go beyond coverage of mechanistic biases (e.g., confounding, selection bias, measurement error) to cover distortions of conclusions produced by statistical methods and psychosocial forces. [ABSTRACT FROM AUTHOR]

In this article, I provide a perspective on the tobacco epidemic and epidemiology, describing the impact of the tobacco-caused disease epidemic on the field of epidemiology. Although there is an enormous body of epidemiologic evidence on the associations of smoking with health, little systematic attention has been given to how decades of research have affected epidemiology and its practice. I address the many advances that resulted from epidemiologic research on smoking and health, such as demonstration of the utility of observational designs and important parameters (the odds ratio and the population attributable risk), guidelines for causal inference, and systematic review approaches. I also cover unintended and adverse consequences for the field, including the strategy of doubt creation and the recruitment of epidemiologists by the tobacco industry to serve its mission. The paradigm of evidence-based action for addressing noncommunicable diseases began with the need to address the epidemic of tobacco-caused disease, an imperative for action documented by epidemiologic research. [ABSTRACT FROM AUTHOR]

In the accompanying article (Am J Epidemiol. 2017;186(6):646-647), Dr. Timothy Lash makes a forceful case that the problems with reproducibility in science stem from our "culture" of null hypothesis significance testing. He notes that when attention is selectively given to statistically significant findings, the estimated effects will be systematically biased away from the null. Here I revisit the recent history of genetic epidemiology and argue for retaining statistical testing as an important part of the tool kit. Particularly when many factors are considered in an agnostic way, in what Lash calls "innovative" research, investigators need a selection strategy to identify which findings are most likely to be genuine, and hence worthy of further study. [ABSTRACT FROM AUTHOR]

We explored the use of the E value to gauge the robustness of fields of epidemiologic inquiry to unmeasured confounding. We surveyed nutritional and air pollution studies that found statistically significant associations between exposures and incident outcomes. For 100 studies in each field, we extracted adjusted relative effect estimates and associated confidence intervals. We inverted estimates where necessary so that all effects were greater than 1. We calculated E values for both the effect estimate and the lower limit of the 95% confidence interval. Nutritional studies were smaller than air pollution studies (median participants per study, 40,652 vs. 72,460). More than 90% of nutritional studies categorized the exposure, whereas 89% of air pollution studies analyzed the exposure as a continuous variable. The median relative effect was 1.33 in nutrition and 1.16 in air pollution. The corresponding median E values for the estimates were 2.00 and 1.59, respectively. E values for the 95% confidence intervals had median values of 1.39 and 1.26, respectively. Little to moderate unmeasured confounding could explain away most observed associations. The E value is necessarily larger for smaller studies that reach statistical significance, making cross-field comparison difficult. The E value for the 95% confidence interval might be a more useful measure in reports of epidemiologic observational studies. [ABSTRACT FROM AUTHOR]

With incomplete data, the "missing at random" (MAR) assumption is widely understood to enable unbiased estimation with appropriate methods. While the need to assess the plausibility of MAR and to perform sensitivity analyses considering "missing not at random" (MNAR) scenarios has been emphasized, the practical difficulty of these tasks is rarely acknowledged. With multivariable missingness, what MAR means is difficult to grasp, and in many MNAR scenarios unbiased estimation is possible using methods commonly associated with MAR. Directed acyclic graphs (DAGs) have been proposed as an alternative framework for specifying practically accessible assumptions beyond the MAR-MNAR dichotomy. However, there is currently no general algorithm for deciding how to handle the missing data given a specific DAG. Here we construct "canonical" DAGs capturing typical missingness mechanisms in epidemiologic studies with incomplete data on exposure, outcome, and confounding factors. For each DAG, we determine whether common target parameters are "recoverable," meaning that they can be expressed as functions of the available data distribution and thus estimated consistently, or whether sensitivity analyses are necessary. We investigate the performance of available-case and multiple-imputation procedures. Using data from waves 1–3 of the Longitudinal Study of Australian Children (2004–2008), we illustrate how our findings can guide the treatment of missing data in point-exposure studies. [ABSTRACT FROM AUTHOR]

In the last few years, stakeholders in the scientific community have raised alarms about a perceived lack of reproducibility of scientific results. In reaction, guidelines for journals have been promulgated and grant applicants have been asked to address the rigor and reproducibility of their proposed projects. Neither solution addresses a primary culprit, which is the culture of null hypothesis significance testing that dominates statistical analysis and inference. In an innovative research enterprise, selection of results for further evaluation based on null hypothesis significance testing is doomed to yield a low proportion of reproducible results and a high proportion of effects that are initially overestimated. In addition, the culture of null hypothesis significance testing discourages quantitative adjustments to account for systematic errors and quantitative incorporation of prior information. These strategies would otherwise improve reproducibility and have not been previously proposed in the widely cited literature on this topic. Without discarding the culture of null hypothesis significance testing and implementing these alternative methods for statistical analysis and inference, all other strategies for improving reproducibility will yield marginal gains at best. [ABSTRACT FROM AUTHOR]

Conventional measures of causality (which compare risks between exposed and unexposed individuals) do not factor in the population-scale dynamics of infectious disease transmission. We used mathematical models of 2 childhood infections (respiratory syncytial virus and rotavirus) to illustrate this problem. These models incorporated 3 causal pathways whereby malnutrition could act to increase the incidence of severe infection: increasing the proportion of infected children who develop severe infection, increasing the children's susceptibility to infection, and increasing infectiousness. For risk factors that increased the proportion of infected children who developed severe infection, the population attributable fraction (PAF) calculated conventionally was the same as the PAF calculated directly from the models. However, for risk factors that increased transmission (by either increasing susceptibility to infection or increasing infectiousness), the PAF calculated directly from the models was much larger than that predicted by the conventional PAF calculation. The models also showed that even when conventional studies find no association between a risk factor and an outcome, risk factors that increase transmission can still have a large impact on disease burden. For a complete picture of infectious disease causality, transmission effects must be incorporated into causal models. [ABSTRACT FROM AUTHOR]

Epidemiology is concerned with determining the distribution and causes of disease. Throughout its history, epidemiology has drawn upon statistical ideas and methods to achieve its aims. Because of the exponential growth in our capacity to measure and analyze data on the underlying processes that define each person's state of health, there is an emerging opportunity for population-based epidemiologic studies to influence health decisions made by individuals in ways that take into account the individuals' characteristics, circumstances, and preferences. We refer to this endeavor as "individualized health." The present article comprises 2 sections. In the first, we describe how graphical, longitudinal, and hierarchical models can inform the project of individualized health. We propose a simple graphical model for informing individual health decisions using population-based data. In the second, we review selected topics in causal inference that we believe to be particularly useful for individualized health. Epidemiology and biostatistics were 2 of the 4 founding departments in the world's first graduate school of public health at Johns Hopkins University, the centennial of which we honor. This survey of a small part of the literature is intended to demonstrate that the 2 fields remain just as inextricably linked today as they were 100 years ago. [ABSTRACT FROM AUTHOR]

Ungrouping binned data can be desirable for many reasons: Bins can be too coarse to allow for accurate analysis; comparisons can be hindered when different grouping approaches are used in different histograms; and the last interval is often wide and open-ended and, thus, covers a lot of information in the tail area. Age group-specific disease incidence rates and abridged life tables are examples of binned data. We propose a versatile method for ungrouping histograms that assumes that only the underlying distribution is smooth. Because of this modest assumption, the approach is suitable for most applications. The method is based on the composite link model, with a penalty added to ensure the smoothness of the target distribution. Estimates are obtained by maximizing a penalized likelihood. This maximization is performed efficiently by a version of the iteratively reweighted least-squares algorithm. Optimal values of the smoothing parameter are chosen by minimizing Akaike’s Information Criterion.We demonstrate the performance of this method in a simulation study and provide several examples that illustrate the approach. Wide, open-ended intervals can be handled properly. The method can be extended to the estimation of rates when both the event counts and the exposures to risk are grouped. [ABSTRACT FROM AUTHOR]