Inclusion of binary proxy variables in logistic regression improves treatment effect estimation in observational studies in the presence of binary unmeasured confounding variables

We present a simulation study and application that shows inclusion of binary proxy variables related to binary unmeasured confounders improves the estimate of a related treatment effect in binary logistic regression. The simulation study included 60,000 randomly generated parameter scenarios of sample size 10,000 across six different simulation structures. We assessed bias by comparing the probability of finding the expected treatment effect relative to the modeled treatment effect with and without the proxy variable. Inclusion of a proxy variable in the logistic regression model significantly reduced the bias of the treatment or exposure effect when compared to logistic regression without the proxy variable. Including proxy variables in the logistic regression model improves the estimation of the treatment effect at weak, moderate, and strong association with unmeasured confounders and the outcome, treatment, or proxy variables. Comparative advantages held for weakly and strongly collapsible situations, as the number of unmeasured confounders increased, and as the number of proxy variables adjusted for increased.

Software Application Profile: The daggle app-a tool to support learning and teaching the graphical rules of selecting adjustment variables using directed acyclic graphs

MOTIVATION

Directed acyclic graphs (DAGs) are used in epidemiological research to communicate causal assumptions and guide the selection of covariate adjustment sets when estimating causal effects. For any given DAG, a set of graphical rules can be applied to identify minimally sufficient adjustment sets that can be used to adjust for bias due to confounding when estimating the causal effect of an exposure on an outcome. The daggle app is a web-based application that aims to assist in the learning and teaching of adjustment set identification using DAGs.

GENERAL FEATURES

The application offers two modes: tutorial and random. The tutorial mode presents a guided introduction to how common causal structures can be presented using DAGs and how graphical rules can be used to identify minimally sufficient adjustment sets for causal estimation. The random mode tests this understanding by presenting the user with a randomly generated DAG-a daggle. To solve the daggle, users must correctly identify a valid minimally sufficient adjustment set.

IMPLEMENTATION

The daggle app is implemented as an R shiny application using the golem framework. The application builds upon existing R libraries including pcalg to generate reproducible random DAGs, dagitty to identify all valid minimal adjustment sets and ggdag to visualize DAGs.

AVAILABILITY

The daggle app can be accessed online at [http://cbdrh.shinyapps.io/daggle]. The source code is available on GitHub [https://github.com/CBDRH/daggle] and is released under a Creative Commons CC BY-NC-SA 4.0 licence.

Early-Life Exposure to the Chinese Famine of 1959-1961 and Type 2 Diabetes in Adulthood: A Systematic Review and Meta-Analysis

BACKGROUND

The fast-growing literature suggests that the Chinese famine of 1959-1961 drives current and future type 2 diabetes (T2D) epidemics in China. This conclusion may be premature, as many Chinese famine studies have major methodological problems. We examine these problems, demonstrate how they bias the study results, and formulate recommendations to improve the quality of future studies.

METHODS

We searched English and Chinese databases for studies that examined the relationship between prenatal exposure to the Chinese famine and adult T2D from inception to 8 February 2022. We extracted information on T2D cases and study populations of individuals born during the famine (famine births), before the famine (prefamine births), and after the famine (postfamine births). We used random-effects models to compare the odds of T2D in famine births to several control groups, including postfamine births, combined pre- and postfamine births, and prefamine births. We used meta-regressions to examine the impacts of age differences between comparison groups on famine effect estimates and the role of other characteristics, including participant sex, age, and T2D assessments; famine intensity; residence; and publication language. Potential sources of heterogeneity and study quality were also evaluated.

RESULTS

Twenty-three studies met our inclusion criteria. The sample sizes ranged from less than 300 to more than 360,000 participants. All studies defined the famine exposure based on the participants' dates of birth, and 18 studies compared famine births and postfamine births to estimate famine effects on T2D. The famine and postfamine births had an age difference of three years or more in all studies. The estimates of the famine effect varied by the selection of controls. Using postfamine births as controls, the OR for T2D among famine births was 1.50 (95% CI 1.34-1.68); using combined pre- and postfamine births as controls, the OR was 1.12 (95% CI 1.02-1.24); using prefamine births as controls, the OR was 0.89 (95% CI 0.79-1.00). The meta-regressions further showed that the famine effect estimates increased by over 1.05 times with each one-year increase in ignored age differences between famine births and controls. Other newly identified methodological problems included the poorly assessed famine intensity, unsuitable study settings for famine research, and poor confounding adjustment.

INTERPRETATION

The current estimates of a positive relationship between prenatal exposure to the Chinese famine and adult T2D are mainly driven by uncontrolled age differences between famine births and postfamine births. Studies with more rigorous methods, including age-balanced controls and robust famine intensity measures, are needed to quantify to what extent the famine exposure is related to current T2D patterns in China.

Short interpregnancy interval and pregnancy outcomes: How important is the timing of confounding variable ascertainment?

BACKGROUND

Estimation of causal effects of short interpregnancy interval on pregnancy outcomes may be confounded by time-varying factors. These confounders should be ascertained at or before delivery of the first ("index") pregnancy, but are often only measured at the subsequent pregnancy.

OBJECTIVES

To quantify bias induced by adjusting for time-varying confounders ascertained at the subsequent (rather than the index) pregnancy in estimated effects of short interpregnancy interval on pregnancy outcomes.

METHODS

We analysed linked records for births in British Columbia, Canada, 2004-2014, to women with ≥2 singleton pregnancies (n = 121 151). We used log binomial regression to compare short (<6, 6-11, 12-17 months) to 18-23-month reference intervals for 5 outcomes: perinatal mortality (stillbirth and neonatal death); small for gestational age (SGA) birth and preterm delivery (all, early, spontaneous). We calculated per cent differences between adjusted risk ratios (aRR) from two models with maternal age, low socio-economic status, body mass index, and smoking ascertained in the index pregnancy and the subsequent pregnancy. We considered relative per cent differences <5% minimal, 5%-9% modest, and ≥10% substantial.

RESULTS

Adjustment for confounders measured at the subsequent pregnancy introduced modest bias towards the null for perinatal mortality aRRs for <6-month interpregnancy intervals [-9.7%, 95% confidence interval [CI] -15.3, -6.2). SGA aRRs were minimally biased towards the null (-1.1%, 95% CI -2.6, 0.8) for <6-month intervals. While early preterm delivery aRRs were substantially biased towards the null (-10.4%, 95% CI -14.0, -6.6) for <6-month interpregnancy intervals, bias was minimal for <6-month intervals for all preterm deliveries (-0.6%, 95% CI -2.0, 0.8) and spontaneous preterm deliveries (-1.3%, 95% CI -3.1, 0.1). For all outcomes, bias was attenuated and minimal for 6-11-month and 12-17-month interpregnancy intervals.

CONCLUSION

These findings suggest that maternally linked pregnancy data may not be needed for appropriate confounder adjustment when studying the effects of short interpregnancy interval on pregnancy outcomes.

A Systematic Review of Methods Used for Confounding Adjustment in Observational Economic Evaluations in Cardiology Conducted between 2013 and 2017

Background. Observational economic evaluations (i.e., economic evaluations in which treatment allocation is not randomized) are prone to confounding bias. Prior reviews published in 2013 have shown that adjusting for confounding is poorly done, if done at all. Although these reviews raised awareness on the issues, it is unclear if their results improved the methodological quality of future work. We therefore aimed to investigate whether and how confounding was accounted for in recently published observational economic evaluations in the field of cardiology. Methods. We performed a systematic review of PubMed, Embase, Cochrane Library, Web of Science, and PsycInfo databases using a set of Medical Subject Headings and keywords covering topics in "observational economic evaluations in health within humans" and "cardiovascular diseases." Any study published in either English or French between January 1, 2013, and December 31, 2017, addressing our search criteria was eligible for inclusion in our review. Our protocol was registered with PROSPERO (CRD42018112391). Results. Forty-two (0.6%) out of 7523 unique citations met our inclusion criteria. Fewer than half of the selected studies adjusted for confounding (n = 19 [45.2%]). Of those that adjusted for confounding, propensity score matching (n = 8 [42.1%]) and other matching-based approaches were favored (n = 8 [42.1%]). Our results also highlighted that most authors who adjusted for confounding rarely justified their methodological choices. Conclusion. Our results indicate that adjustment for confounding is often ignored when conducting an observational economic evaluation. Continued knowledge translation efforts aimed at improving researchers' knowledge regarding confounding bias and methods aimed at addressing this issue are required and should be supported by journal editors.

Automated data-adaptive analytics for electronic healthcare data to study causal treatment effects

BACKGROUND

Decision makers in health care increasingly rely on nonrandomized database analyses to assess the effectiveness, safety, and value of medical products. Health care data scientists use data-adaptive approaches that automatically optimize confounding control to study causal treatment effects. This article summarizes relevant experiences and extensions.

METHODS

The literature was reviewed on the uses of high-dimensional propensity score (HDPS) and related approaches for health care database analyses, including methodological articles on their performance and improvement. Articles were grouped into applications, comparative performance studies, and statistical simulation experiments.

RESULTS

The HDPS algorithm has been referenced frequently with a variety of clinical applications and data sources from around the world. The appeal of HDPS for database research rests in 1) its superior performance in situations of unobserved confounding through proxy adjustment, 2) its predictable efficiency in extracting confounding information from a given data source, 3) its ability to automate estimation of causal treatment effects to the extent achievable in a given data source, and 4) its independence of data source and coding system. Extensions of the HDPS approach have focused on improving variable selection when exposure is sparse, using free text information and time-varying confounding adjustment.

CONCLUSION

Semiautomated and optimized confounding adjustment in health care database analyses has proven successful across a wide range of settings. Machine-learning extensions further automate its use in estimating causal treatment effects across a range of data scenarios.

A comparison of confounding adjustment methods with an application to early life determinants of childhood obesity

We implemented six confounding adjustment methods: (1) covariate-adjusted regression, (2) propensity score (PS) regression, (3) PS stratification, (4) PS matching with two calipers, (5) inverse probability weighting and (6) doubly robust estimation to examine the associations between the body mass index (BMI) z-score at 3 years and two separate dichotomous exposure measures: exclusive breastfeeding v. formula only (n=437) and cesarean section v. vaginal delivery (n=1236). Data were drawn from a prospective pre-birth cohort study, Project Viva. The goal is to demonstrate the necessity and usefulness, and approaches for multiple confounding adjustment methods to analyze observational data. Unadjusted (univariate) and covariate-adjusted linear regression associations of breastfeeding with BMI z-score were -0.33 (95% CI -0.53, -0.13) and -0.24 (-0.46, -0.02), respectively. The other approaches resulted in smaller n (204-276) because of poor overlap of covariates, but CIs were of similar width except for inverse probability weighting (75% wider) and PS matching with a wider caliper (76% wider). Point estimates ranged widely, however, from -0.01 to -0.38. For cesarean section, because of better covariate overlap, the covariate-adjusted regression estimate (0.20) was remarkably robust to all adjustment methods, and the widths of the 95% CIs differed less than in the breastfeeding example. Choice of covariate adjustment method can matter. Lack of overlap in covariate structure between exposed and unexposed participants in observational studies can lead to erroneous covariate-adjusted estimates and confidence intervals. We recommend inspecting covariate overlap and using multiple confounding adjustment methods. Similar results bring reassurance. Contradictory results suggest issues with either the data or the analytic method.

A comparison of confounding adjustment methods for assessment of asthma controller medication effectiveness

We compared the impact of 3 confounding adjustment procedures-covariate-adjusted regression, propensity score regression, and high-dimensional propensity score regression-to assess the effects of selected asthma controller medication use (leukotriene antagonists and inhaled corticosteroids) on the following 4 asthma-related adverse outcomes: emergency department visits, hospitalizations, oral corticosteroid use, and the composite outcome of these. We examined a cohort of 24,680 new users who were 4-17 years of age at the incident dispensing from the Population-Based Effectiveness in Asthma and Lung Diseases (PEAL) Network of 5 commercial health plans and TennCare, the Tennessee Medicaid program, during the period January 1, 2004, to December 31, 2010. The 3 methods yielded similar results, indicating that pediatric patients treated with leukotriene antagonists were no more likely than those treated with inhaled corticosteroids to experience adverse outcomes. Children in the TennCare population who had a diagnosis of allergic rhinitis and who then initiated the use of leukotriene antagonists were less likely to experience an asthma-related emergency department visit. A plausible explanation is that our data set is large enough that the 2 advanced propensity score-based analyses do not have advantages over the traditional covariate-adjusted regression approach. We provide important observations on how to correctly apply the methods in observational data analysis and suggest statistical research areas that need more work to guide implementation.