Covariate-adjusted generalized pairwise comparisons in small samples

Semiparametric probabilistic index models allow for the comparison of two groups of observations, whilst adjusting for covariates, thereby fitting nicely within the framework of generalized pairwise comparisons (GPC). As with most regression approaches in this setting, the limited amount of data results in invalid inference as the asymptotic normality assumption is not met. In addition, separation issues might arise when considering small samples. In this article, we show that the parameters of the probabilistic index model can be estimated using generalized estimating equations, for which adjustments exist that lead to estimators of the sandwich variance-covariance matrix with improved finite sample properties and that can deal with bias due to separation. In this way, appropriate inference can be performed as is shown through extensive simulation studies. The known relationships between the probabilistic index and other GPC statistics allow to also provide valid inference for example, the net treatment benefit or the success odds.

Cohort Differences in Physical Health and Disability in the United States and Europe

OBJECTIVES

Declines in mortality have historically been associated with improvements in physical health across generations. While life expectancy in most high-income countries continues to increase, there is evidence that younger generations, particularly in the United States, are less healthy than previous generations at the same age. We compared generational trends in physical health in the United States, England, and continental Europe to explore whether other regions have experienced a similar pattern of worsening health across cohorts.

METHODS

Using data from nationally representative studies of adults aged ≥50 years from the United States (Health and Retirement Study, n = 26,939), England (English Longitudinal Study of Ageing, n = 14,992) and 11 continental European countries (Survey of Health, Ageing and Retirement in Europe, n = 72,595), we estimated differences in the age-adjusted prevalence of self-reported chronic disease and disability and observer-measured health indicators across pseudo-birth cohorts (born <1925, 1925-1935, 1936-1945, 1946-1954, 1955-1959).

RESULTS

Age-adjusted prevalence of doctor-diagnosed chronic disease increased across successive cohorts in all regions. Trends in disability prevalence were more regionally varied. Still, in both the United States and Europe, we observed a structural break in disability trends, with declines observed in prewar cohorts slowing, stalling, or reversing for cohorts born since 1945.

DISCUSSION

In all regions, we found evidence for worsening health across cohorts, particularly for those born since 1945. While more chronic disease in younger cohorts need not necessarily translate to worse quality of life or higher rates of functional limitation, there is some suggestion that worsening chronic disease morbidity may be spilling over into worsening disability.

Long COVID and financial outcomes: evidence from four longitudinal population surveys

BACKGROUND

Long-term sequelae of COVID-19 (long COVID) include muscle weakness, fatigue, breathing difficulties and sleep disturbance over weeks or months. Using UK longitudinal data, we assessed the relationship between long COVID and financial disruption.

METHODS

We estimated associations between long COVID (derived using self-reported length of COVID-19 symptoms) and measures of financial disruption (subjective financial well-being, new benefit claims, changes in household income) by analysing data from four longitudinal population studies, gathered during the first year of the pandemic. We employed modified Poisson regression in a pooled analysis of the four cohorts adjusting for a range of potential confounders, including pre-pandemic (pre-long COVID) factors.

RESULTS

Among the 20 112 observations across four population surveys, 13% reported having COVID-19 with symptoms that impeded their ability to function normally-10.7% had such symptoms for <4 weeks (acute COVID-19), 1.2% had such symptoms for 4-12 weeks (ongoing symptomatic COVID-19) and 0.6% had such symptoms for >12 weeks (post-COVID-19 syndrome). We found that post-COVID-19 syndrome was associated with worse subjective financial well-being (adjusted relative risk ratios (aRRRs)=1.57, 95% CI=1.25, 1.96) and new benefit claims (aRRR=1.79, CI=1.27, 2.53). Associations were broadly similar across sexes and education levels. These results were not meaningfully altered when scaled to represent the population by age.

CONCLUSIONS

Long COVID was associated with financial disruption in the UK. If our findings reflect causal effects, extending employment protection and financial support to people with long COVID may be warranted.

Estimation of the causal effect of sex on neonatal intensive care unit outcomes among very low birth weight infants

OBJECTIVE

Estimate the causal effect of sex on outcomes in the neonatal intensive care unit (NICU) among very low birth weight (VLBW) infants.

STUDY DESIGN

Retrospective cohort study using Vermont Oxford Network data to compare NICU outcomes for VLBW males versus females. Odds ratios (OR) for outcomes that differed significantly by sex were computed using standard unweighted analysis and inverse probability weighted (IPW) analysis to correct for selection bias.

RESULTS

Using standard analysis, males were significantly more likely to die before discharge and experience six other adverse outcomes. From IPW analysis, male sex caused a 56% increase in the odds of death before discharge (OR = 1.56, 95% confidence interval: 1.18-1.94). Standard unweighted results were significantly biased towards increased risk of adverse outcomes for males (p = 0.005) compared to IPW results for which three outcomes were no longer significantly associated with male sex.

CONCLUSION

Standard statistical methods generally overestimate the casual effect of sex among VLBW infants.

Comparing the performance of two-stage residual inclusion methods when using physician's prescribing preference as an instrumental variable: unmeasured confounding and noncollapsibility

Aim: The first objective is to compare the performance of two-stage residual inclusion (2SRI), two-stage least square (2SLS) with the multivariable generalized linear model (GLM) in terms of the reducing unmeasured confounding bias. The second objective is to demonstrate the ability of 2SRI and 2SPS in alleviating unmeasured confounding when noncollapsibility exists. Materials & methods: This study comprises a simulation study and an empirical example from a real-world UK population health dataset (Clinical Practice Research Datalink). The instrumental variable (IV) used is based on physicians' prescribing preferences (defined by prescribing history). Results: The percent bias of 2SRI in terms of treatment effect estimates to be lower than GLM and 2SPS and was less than 15% in most scenarios. Further, 2SRI was found to be robust to mild noncollapsibility with the percent bias less than 50%. As the level of unmeasured confounding increased, the ability to alleviate the noncollapsibility decreased. Strong IVs tended to be more robust to noncollapsibility than weak IVs. Conclusion: 2SRI tends to be less biased than GLM and 2SPS in terms of estimating treatment effect. It can be robust to noncollapsibility in the case of the mild unmeasured confounding effect.

The potential impact fraction of population weight reduction scenarios on non-communicable diseases in Belgium: application of the g-computation approach

BACKGROUND

Overweight is a major risk factor for non-communicable diseases (NCDs) in Europe, affecting almost 60% of all adults. Tackling obesity is therefore a key long-term health challenge and is vital to reduce premature mortality from NCDs. Methodological challenges remain however, to provide actionable evidence on the potential health benefits of population weight reduction interventions. This study aims to use a g-computation approach to assess the impact of hypothetical weight reduction scenarios on NCDs in Belgium in a multi-exposure context.

METHODS

Belgian health interview survey data (2008/2013/2018, n = 27 536) were linked to environmental data at the residential address. A g-computation approach was used to evaluate the potential impact fraction (PIF) of population weight reduction scenarios on four NCDs: diabetes, hypertension, cardiovascular disease (CVD), and musculoskeletal (MSK) disease. Four scenarios were considered: 1) a distribution shift where, for each individual with overweight, a counterfactual weight was drawn from the distribution of individuals with a "normal" BMI 2) a one-unit reduction of the BMI of individuals with overweight, 3) a modification of the BMI of individuals with overweight based on a weight loss of 10%, 4) a reduction of the waist circumference (WC) to half of the height among all people with a WC:height ratio greater than 0.5. Regression models were adjusted for socio-demographic, lifestyle, and environmental factors.

RESULTS

The first scenario resulted in preventing a proportion of cases ranging from 32.3% for diabetes to 6% for MSK diseases. The second scenario prevented a proportion of cases ranging from 4.5% for diabetes to 0.8% for MSK diseases. The third scenario prevented a proportion of cases, ranging from 13.6% for diabetes to 2.4% for MSK diseases and the fourth scenario prevented a proportion of cases ranging from 36.4% for diabetes to 7.1% for MSK diseases.

CONCLUSION

Implementing weight reduction scenarios among individuals with excess weight could lead to a substantial and statistically significant decrease in the prevalence of diabetes, hypertension, cardiovascular disease (CVD), and musculoskeletal (MSK) diseases in Belgium. The g-computation approach to assess PIF of interventions represents a straightforward approach for drawing causal inferences from observational data while providing useful information for policy makers.

Generational differences in physical health and disability in the United States and Europe

Objectives

Declines in mortality have typically been associated with improvements in physical health across generations. While life expectancy in most high-income countries continues to increase, there is evidence that younger generations, particularly in the United States (US), are less healthy than previous generations at the same age. We compared generational trends in physical health in the US, England, and continental Europe to explore whether other regions have experienced a similar pattern of worsening health across cohorts.

Methods

Using data from nationally representative studies of adults aged ≥50 years from the US (Health and Retirement Study, n=26,939), England (English Longitudinal Study of Ageing, n=14,992) and 11 continental European countries (Survey of Health, Ageing and Retirement in Europe, n=72,595), we estimated differences in the age-adjusted prevalence of self-reported chronic disease and disability and observer-measured health indicators across pseudo-birth cohorts (born <1925, 1925-1935, 1936-1945, 1946-1954, 1955-1959).

Results

Age-adjusted prevalence of doctor-diagnosed chronic disease increased across cohorts in all regions. Trends in disability prevalence were more regionally varied. Still, in both the US and Europe, we observed a structural break in disability trends, with declines observed in pre-war cohorts slowing, stalling, or reversing for cohorts born since 1945.

Discussion

In all regions, we found evidence for worsening health across cohorts, particularly for those born since 1945. While more chronic disease in younger cohorts need not necessarily translate to worse quality of life or higher rates of functional limitation, there is some suggestion that worsening chronic disease morbidity may be spilling over into worsening disability.

Evaluating hybrid controls methodology in early-phase oncology trials: A simulation study based on the MORPHEUS-UC trial

Phase Ib/II oncology trials, despite their small sample sizes, aim to provide information for optimal internal company decision-making concerning novel drug development. Hybrid controls (a combination of the current control arm and controls from one or more sources of historical trial data [HTD]) can be used to increase statistical precision. Here we assess combining two sources of Roche HTD to construct a hybrid control in targeted therapy for decision-making via an extensive simulation study. Our simulations are based on the real data of one of the experimental arms and the control arm of the MORPHEUS-UC Phase Ib/II study and two Roche HTD for atezolizumab monotherapy. We consider potential complications such as model misspecification, unmeasured confounding, different sample sizes of current treatment groups, and heterogeneity among the three trials. We evaluate two frequentist methods (with both Cox and Weibull accelerated failure time [AFT] models) and three different commensurate priors in Bayesian dynamic borrowing (with a Weibull AFT model), and modifications within each of those, when estimating the effect of treatment on survival outcomes and measures of effect such as marginal hazard ratios. We assess the performance of these methods in different settings and the potential of generalizations to supplement decisions in early-phase oncology trials. The results show that the proposed joint frequentist methods and noninformative priors within Bayesian dynamic borrowing with no adjustment on covariates are preferred, especially when treatment effects across the three trials are heterogeneous. For generalization of hybrid control methods in such settings, we recommend more simulation studies.

Causal relationship between gastro-esophageal reflux disease and risk of lung cancer: insights from multivariable Mendelian randomization and mediation analysis

AIM

A recent study has reported that anti-reflux surgery reduced the risk of lung cancer. However, the exact causal association between gastro-esophageal reflux disease (GORD) and lung cancer remains obscure. Therefore, we conducted a multivariable and network Mendelian randomization (MR) study to explore this potential association and mediation effect.

METHODS

Independent single nucleotide polymorphisms (SNPs) strongly associated with GORD were selected as instrumental variables (IVs) from the corresponding genome-wide association studies (GWAS). The summary statistics were obtained from the largest GORD GWAS meta-analysis of 367 441 (78 707 cases) European individuals, and the summary statistics of lung cancer and pathological subtypes came from International Lung Cancer Consortium (ILCCO) and FinnGen databases. Univariable and multivariable MR analyses were performed to investigate and verify the causal relationship between genetically predicted GORD and lung cancer. Network MR analysis was conducted to reveal the mediating role of GORD between smoking initiation and lung cancer.

RESULTS

The univariable MR analysis demonstrated that GORD was associated with an increased risk of total lung cancer in both ILCCO [inverse variance weighted (IVW): odds ratio (OR) = 1.37, 95% confidence interval (CI) 1.16-1.62, P = 1.70E-04] and FinnGen database (IVW: OR = 1.25, 95% confidence interval CI 1.03-1.52, P = 2.27E-02). The consistent results were observed after adjusting the potential confounders [smoking traits, body mass index (BMI) and type 2 diabetes] in multivariable MR analyses. In subtype analyses, GORD was associated with lung adenocarcinoma (IVW: OR = 1.27, 95% CI 1.02-1.59, P = 3.48E-02) and lung squamous cell carcinomas (IVW: OR = 1.50, 95% CI 1.22-1.86, P = 1.52E-04). Moreover, GORD mediated 32.43% (95% CI 14.18-49.82%) and 25.00% (95% CI 3.13-50.00%) of the smoking initiation effects on lung cancer risk in the ILCCO and FinnGen databases, respectively.

CONCLUSION

This study provides credible evidence that genetically predicted GORD was significantly associated with an increased risk of total lung cancer, lung adenocarcinoma and lung squamous cell carcinomas. Furthermore, our results suggest GORD is involved in the mechanism of smoking initiation-induced lung cancer.

Accuracy of physicians' intuitive risk estimation in the diagnostic management of pulmonary embolism: an individual patient data meta-analysis

BACKGROUND

In patients clinically suspected of having pulmonary embolism (PE), physicians often rely on intuitive estimation ("gestalt") of PE presence. Although shown to be predictive, gestalt is criticized for its assumed variation across physicians and lack of standardization.

OBJECTIVES

To assess the diagnostic accuracy of gestalt in the diagnosis of PE and gain insight into its possible variation.

METHODS

We performed an individual patient data meta-analysis including patients suspected of having PE. The primary outcome was diagnostic accuracy of gestalt for the diagnosis of PE, quantified as risk ratio (RR) between gestalt and PE based on 2-stage random-effect log-binomial meta-analysis regression as well as gestalts' sensitivity and specificity. The variability of these measures was explored across different health care settings, publication period, PE prevalence, patient subgroups (sex, heart failure, chronic lung disease, and items of the Wells score other than gestalt), and age.

RESULTS

We analyzed 20 770 patients suspected of having PE from 16 original studies. The prevalence of PE in patients with and without a positive gestalt was 28.8% vs 9.1%, respectively. The overall RR was 3.02 (95% CI, 2.35-3.87), and the overall sensitivity and specificity were 74% (95% CI, 68%-79%) and 61% (95% CI, 53%-68%), respectively. Although variation was observed across individual studies (I2, 90.63%), the diagnostic accuracy was consistent across all subgroups and health care settings.

CONCLUSION

A positive gestalt was associated with a 3-fold increased risk of PE in suspected patients. Although variation was observed across studies, the RR of gestalt was similar across prespecified subgroups and health care settings, exemplifying its diagnostic value for all patients suspected of having PE.

Case-control matching on confounders revisited

Matching by a confounder in a case-control study nearly always produces a control-selection bias that mixes with the confounding to produce a net bias. Previous theoretical work has assumed that control for a single confounder, the matching factor, is sufficient to remove all the confounding and that the confounder-exposure, confounder-outcome and exposure-outcome associations are monotonic. Under these conditions: (a) The net bias is toward the null if the exposure affects the outcome and nil if it does not. (b) If the confounding is away from the null, the selection bias is toward the null. (c) If the confounding is toward the null, the selection bias can be in any direction or even nil. If more than one confounder needs to be controlled to remove all the confounding, the net bias from matching by one of them can be away from the null, whether the exposure affects the outcome or not. An influential heuristic, that matching controls to cases by a variable associated with exposure always brings the marginal exposure distributions of the case and control groups closer together, turns out to be faulty. The implications of matching by confounders in case-control studies are less straightforward than previously thought. Suggestions are offered for advancing the methodologic literature on this topic.

Is COPD mortality in South China causally linked to the long-term PM1 exposure? Evidence from a large community-based cohort

BACKGROUND

Long-term ambient particulate matter (PM) exposure has been found associated with chronic obstructive pulmonary disease (COPD) mortality in an increasing body of research. However, limited evidence was available on the potential causal links between PM1 and COPD mortality, especially in highly exposed areas.

OBJECTIVES

To examine the COPD mortality risk following long-term ambient PM1 exposure in south China.

METHODS

The cohort included 580,757 participants recruited during 2009-2015. Satellite-based annual concentrations of PM1 were estimated at a spatial resolution of 1 km × 1 km and assigned to each participant based on their residential addresses. We analyzed the potential causal links between time-varying PM1 exposure and COPD mortality using marginal structural cox models within causal frameworks. Stratified analyses were also performed to identify the potential susceptible groups.

RESULTS

The annual average PM1 concentration continuously decreased over time. After adjusting for confounders, each 1 μg/m3 increase in PM1 concentration corresponded to an 8.1 % (95% confidence interval: 6.4-9.9 %) increment in the risk of COPD mortality. The impact of PM1 was more pronounced among the elderly and those with low exercise frequency, with a 1.9-6.9 % higher risk than their counterparts. We further observed a 0.1-9.7 % greater risk among those who lived in lower greenness settings. Additionally, we observed higher effect estimates in participants with long-term low PM1 exposure compared to the general population.

CONCLUSIONS

COPD mortality risk significantly increased following long term ambient PM1 exposure, particularly among groups with certain demographics or long-term low exposure.

Potential causal links between long-term ambient particulate matter exposure and cerebrovascular mortality: Insights from a large cohort in southern China

Cohort studies conducted in North America and Europe have linked cerebrovascular mortality to long-term exposure to particulate matter (PM). However, limited evidence from large cohorts in high-exposure areas and the traditional approach of association assessment may cause residual confounding issues. In this study, we aimed to investigate the causal links between cerebrovascular mortality and long-term exposure to PM_2.5, PM₁₀, and PM_2.5-10 in an ongoing cohort study with 580,757 participants in southern China. Using satellite-based estimates of PM concentration at a 1-km² spatial resolution, we assigned exposure levels to each participant and used the marginal structural Cox model to assess the association between PM exposure and cerebrovascular mortality while accounting for time-varying covariates. We also explored the potential modification effects of sociodemographic and behavioral factors on the PM-health associations. Adjusted hazard ratios (HR) for overall cerebrovascular mortality were 1.041 (95% confidence interval (CI): 1.034-1.049) and 1.032 (95% CI: 1.026-1.038) for each 1 μg/m³ increase in PM_2.5, and PM₁₀, respectively. Similar trends were observed in the mortality risk from stroke and ischemic stroke, with HRs ranging from 1.040 to 1.069 and 1.025 to 1.052, respectively, across 2 p.m. exposures. The impact of PM exposure was generally more apparent among women, participants with primary school diplomas and below, and the subgroup under low-exposure. Multiple sensitivity analyses confirmed the robustness of the results. In conclusion, this sizable prospective cohort study hypothesizes causal links between long-term PM exposure and cerebrovascular mortality, particularly among vulnerable participants, supporting the rationale for reducing PM concentration in China to reduce cerebrovascular mortality.

Association between psychological distress trajectories from adolescence to midlife and mental health during the pandemic: evidence from two British birth cohorts

BACKGROUND

This paper examined whether distinct life-course trajectories of psychological distress from adolescence to midlife were associated with poorer mental health outcomes during the pandemic.

METHODS

We present a secondary analysis of two nationally representative British birth cohorts, the 1958 National Child Development Study (NCDS) and 1970 British Cohort Study (BCS70). We used latent variable mixture models to identify pre-pandemic longitudinal trajectories of psychological distress and a modified Poisson model with robust standard errors to estimate associations with psychological distress, life satisfaction and loneliness at different points during the pandemic.

RESULTS

Our analysis identified five distinct pre-pandemic trajectories of psychological distress in both cohorts. All trajectories with prior symptoms of psychological distress irrespective of age of onset, severity and chronicity were associated with a greater relative risk of poorer mental health outcomes during the pandemic and the probability of poorer mental health associated with psychological distress trajectories remained fairly constant. The relationship was not fully attenuated when most recent pre-pandemic psychological distress and other midlife factors were controlled for.

CONCLUSIONS

Whilst life-course trajectories with any prior symptoms of psychological distress put individuals at greater risk of poor mental health outcomes during the pandemic, those with chronic and more recent occurrences were at highest risk. In addition, prior poor mental health during the adult life-course may mean individuals are less resilient to shocks, such as pandemics. Our findings show the importance of considering heterogeneous mental health trajectories across the life-course in the general population in addition to population average trends.

Parent- and teacher-reported associations from adolescent bifactor models of psychopathology: an outcome-wide association study of 26 outcomes in mid-life

BACKGROUND

Adolescent mental health problems have lasting impacts on health and social functioning later in life. Evidence to date mostly comes from studies of specific diagnostic categories/dimensions, but hierarchical models can elucidate associations with general as well as specific dimensions of psychopathology. We provide evidence on long-term outcomes of general and specific dimensions of adolescent psychopathology using both parent and teacher reports.

METHODS

Parents and teachers from the Isle of Wight study completed Rutter behaviour scales when participants were 14-15 years old (n = 2,275), assessing conduct, emotional and hyperactivity problems. Metric-invariant bifactor models for parents and teachers were used to test domain-specific and domain-general associations with 26 self-reported psychosocial outcomes at mid-life (age 44-45 years, n = 1,423). Analyses examined the individual and joint contributions of parent and teacher reports of adolescent psychopathology. All analyses were adjusted for covariates (gender, IQ and family social class) and weighted to adjust for the probability of nonresponse.

RESULTS

Parent- and teacher-reported general factors of psychopathology (GFP) were associated with 15 and 12 outcomes, respectively, across the socioeconomic, relationship, health and personality domains, along with an index of social exclusion. Nine outcomes were associated with both parent- and teacher-reported GFP, with no differences in the strength of the associations across reporters. Teacher-reported specific factors (conduct, emotional and hyperactivity) were associated with 21 outcomes, and parent-reported specific factors were associated with seven. Five outcomes were associated with the same specific factors from both reporters; only one showed reporter differences in the strength of the associations.

CONCLUSIONS

These findings confirm the relevance of the GFP and the utility of teacher as well as parent reports of adolescent mental health in predicting psychosocial outcomes later in the life course.

The hepato-ovarian axis: genetic evidence for a causal association between non-alcoholic fatty liver disease and polycystic ovary syndrome

BACKGROUND

Recent studies found associations between non-alcoholic fatty liver disease (NAFLD) and polycystic ovary syndrome (PCOS), but the causal nature of this association is still uncertain.

METHODS

We performed a bidirectional two-sample Mendelian randomization (MR) analysis to test for the causal association between NAFLD and PCOS using data from a large-scale biopsy-confirmed NAFLD genome-wide association study (GWAS) (1483 cases and 17,781 controls) and PCOS GWAS (10,074 cases and 103,164 controls) in European ancestries. Data from glycemic-related traits GWAS (in up to 200,622 individuals) and sex hormones GWAS (in 189,473 women) in the UK Biobank (UKB) were used in the MR mediation analysis to assess potential mediating roles of these molecules in the causal pathway between NAFLD and PCOS. Replication analysis was conducted using two independent datasets from NAFLD and PCOS GWASs in the UKB and a meta-analysis of data from FinnGen and the Estonian Biobank, respectively. A linkage disequilibrium score regression was conducted to assess genetic correlations between NAFLD, PCOS, glycemic-related traits, and sex hormones using full summary statistics.

RESULTS

Individuals with higher genetic liability to NAFLD were more likely to develop PCOS (OR per one-unit log odds increase in NAFLD: 1.10, 95% CI: 1.02-1.18; P = 0.013). Indirect causal effects of NAFLD on PCOS via fasting insulin only (OR: 1.02, 95% CI: 1.01-1.03; P = 0.004) and further a suggestive indirect causal effect via fasting insulin in concert with androgen levels were revealed in MR mediation analyses. However, the conditional F statistics of NAFLD and fasting insulin were less than 10, suggesting likely weak instrument bias in the MVMR and MR mediation analyses.

CONCLUSIONS

Our study suggests that genetically predicted NAFLD was associated with a higher risk of developing PCOS but less evidence for vice versa. Fasting insulin and sex hormones might mediate the link between NAFLD and PCOS.

MR-BOIL: Causal inference in one-sample Mendelian randomization for binary outcome with integrated likelihood method

Mendelian randomization is a statistical method for inferring the causal relationship between exposures and outcomes using an economics-derived instrumental variable approach. The research results are relatively complete when both exposures and outcomes are continuous variables. However, due to the noncollapsing nature of the logistic model, the existing methods inherited from the linear model for exploring binary outcome cannot take the effect of confounding factors into account, which leads to biased estimate of the causal effect. In this article, we propose an integrated likelihood method MR-BOIL to investigate causal relationships for binary outcomes by treating confounders as latent variables in one-sample Mendelian randomization. Under the assumption of a joint normal distribution of the confounders, we use expectation maximization algorithm to estimate the causal effect. Extensive simulations demonstrate that the estimator of MR-BOIL is asymptotically unbiased and that our method improves statistical power without inflating type I error rate. We then apply this method to analyze the data from Atherosclerosis Risk in Communications Study. The results show that MR-BOIL can better identify plausible causal relationships with high reliability, compared with the unreliable results of existing methods. MR-BOIL is implemented in R and the corresponding R code is provided for free download.

Investigating confounding in network-based test-negative design influenza vaccine effectiveness studies-Experience from the DRIVE project

BACKGROUND

Establishing a large study network to conduct influenza vaccine effectiveness (IVE) studies while collecting appropriate variables to account for potential bias is important; the most relevant variables should be prioritized. We explored the impact of potential confounders on IVE in the DRIVE multi-country network of sites conducting test-negative design (TND) studies.

METHODS

We constructed a directed acyclic graph (DAG) to map the relationship between influenza vaccination, medically attended influenza infection, confounders, and other variables. Additionally, we used the Development of Robust and Innovative Vaccines Effectiveness (DRIVE) data from the 2018/2019 and 2019/2020 seasons to explore the effect of covariate adjustment on IVE estimates. The reference model was adjusted for age, sex, calendar time, and season. The covariates studied were presence of at least one, two, or three chronic diseases; presence of six specific chronic diseases; and prior healthcare use. Analyses were conducted by site and subsequently pooled.

RESULTS

The following variables were included in the DAG: age, sex, time within influenza season and year, health status and comorbidities, study site, health-care-seeking behavior, contact patterns and social precautionary behavior, socioeconomic status, and pre-existing immunity. Across all age groups and settings, only adjustment for lung disease in older adults in the primary care setting resulted in a relative change of the IVE point estimate >10%.

CONCLUSION

Our study supports a parsimonious approach to confounder adjustment in TND studies, limited to adjusting for age, sex, and calendar time. Practical implications are that necessitating fewer variables lowers the threshold for enrollment of sites in IVE studies and simplifies the pooling of data from different IVE studies or study networks.

The UK Coronavirus Job Retention Scheme and smoking, alcohol consumption and vaping during the COVID-19 pandemic: evidence from eight longitudinal population surveys

BACKGROUND

Employment disruptions can impact smoking and alcohol consumption. During the COVID-19 pandemic, many countries implemented furlough schemes to prevent job loss. We examine how furlough was associated with smoking, vaping and alcohol consumption in the UK.

METHODS

Data from 27,841 participants in eight UK adult longitudinal surveys were analysed. Participants self-reported employment status and current smoking, current vaping and alcohol consumption (>4 days/week or 5+ drinks per typical occasion) both before and during the early stages of the pandemic (April-July 2020). Risk ratios were estimated within each study using modified Poisson regression, adjusting for a range of potential confounders, including pre-pandemic behaviour. Findings were synthesised using random effects meta-analysis.

RESULTS

Compared to stable employment and after adjustment for pre-pandemic characteristics, furlough was not associated with smoking (ARR = 1.05; 95% CI: 0.95-1.16; I²: 10%), vaping (ARR = 0.89; 95% CI: 0.74-1.08; I²: 0%) or drinking (ARR = 1.03; 95% CI: 0.94-1.13; I²: 48%). There were similar findings for no longer being employed, and stable unemployment, though this varied by sex: stable unemployment was associated with smoking for women (ARR = 1.35; 95% CI: 1.00-1.82; I²: 47%) but not men (0.84; 95% CI: 0.67-1.05; I²: 0%). No longer being employed was associated with vaping among women (ARR = 2.74; 95% CI: 1.59-4.72; I²: 0%) but not men (ARR = 1.25; 95% CI: 0.83-1.87; I²: 0%).

CONCLUSIONS

We found no clear evidence of furlough or unemployment having adverse impacts on smoking, vaping or drinking behaviours during the early stages of the COVID-19 pandemic in the UK. Differences in risk compared to those who remained employed were largely explained by pre-pandemic characteristics.

The UK Coronavirus Job Retention Scheme and diet, physical activity, and sleep during the COVID-19 pandemic: evidence from eight longitudinal population surveys

BACKGROUND

In March 2020, the UK implemented the Coronavirus Job Retention Scheme (furlough) to minimise job losses. Our aim was to investigate associations between furlough and diet, physical activity, and sleep during the early stages of the COVID-19 pandemic.

METHODS

We analysed data on 25,092 participants aged 16-66 years from eight UK longitudinal studies. Changes in employment, including being furloughed, were based on employment status before and during the first lockdown. Health behaviours included fruit and vegetable consumption, physical activity, and sleep. Study-specific estimates obtained using modified Poisson regression, adjusting for socio-demographic characteristics and pre-pandemic health and health behaviours, were statistically pooled using random effects meta-analysis. Associations were also stratified by sex, age, and education.

RESULTS

Across studies, between 8 and 25% of participants were furloughed. Compared to those who remained working, furloughed workers were slightly less likely to be physically inactive (RR = 0.85; [95% CI 0.75-0.97]; I ² = 59%) and did not differ overall with respect to low fruit and vegetable consumption or atypical sleep, although findings for sleep were heterogenous (I ² = 85%). In stratified analyses, furlough was associated with lower fruit and vegetable consumption among males (RR = 1.11; [1.01-1.22]; I ² = 0%) but not females (RR = 0.84; [0.68-1.04]; I ² = 65%). Considering changes in quantity, furloughed workers were more likely than those who remained working to report increases in fruit and vegetable consumption, exercise, and hours of sleep.

CONCLUSIONS

Those furloughed exhibited similar health behaviours to those who remained in employment during the initial stages of the pandemic. There was little evidence to suggest that adoption of such social protection policies in the post-pandemic recovery period and during future economic crises had adverse effects on population health behaviours.

Controversy and Debate : Questionable utility of the relative risk in clinical research: Paper 4 :Odds Ratios are far from "portable" - A call to use realistic models for effect variation in meta-analysis

OBJECTIVE

Recently Doi et al. argued that risk ratios should be replaced with odds ratios in clinical research. We disagreed, and empirically documented the lack of portability of odds ratios, while Doi et al. defended their position. In this response we highlight important errors in their position.

STUDY DESIGN AND SETTING

We counter Doi et al.'s arguments by further examining the correlations of odds ratios, and risk ratios, with baseline risks in 20,198 meta-analyses from the Cochrane Database of Systematic Reviews.

RESULTS

Doi et al.'s claim that odds ratios are portable is invalid because 1) their reasoning is circular: they assume a model under which the odds ratio is constant and show that under such a model the odds ratio is portable; 2) the method they advocate to convert odds ratios to risk ratios is biased; 3) their empirical example is readily-refuted by counter-examples of meta-analyses in which the risk ratio is portable but the odds ratio isn't; and 4) they fail to consider the causal determinants of meta-analytic inclusion criteria: Doi et al. mistakenly claim that variation in odds ratios with different baseline risks in meta-analyses is due to collider bias. Empirical comparison between the correlations of odds ratios, and risk ratios, with baseline risks show that the portability of odds ratios and risk ratios varies across settings.

CONCLUSION

The suggestion to replace risk ratios with odds ratios is based on circular reasoning and a confusion of mathematical and empirical results. It is especially misleading for meta-analyses and clinical guidance. Neither the odds ratio nor the risk ratio is universally portable. To address this lack of portability, we reinforce our suggestion to report variation in effect measures conditioning on modifying factors such as baseline risk; understanding such variation is essential to patient-centered practice.

Comparison of Conditional and Marginal Models in Assessing a Child Nutrition Intervention in Armenia

INTRODUCTION

Child malnutrition is a major issue in conflict zones. Evidence-based interventions and their thorough evaluation could help to eliminate malnutrition. We aimed to assess the causal effect of a community-based multidisciplinary nutrition program for children in a chronic conflict zone near the northeastern border of Armenia on two main outcomes: stunting and anemia. We further compared the interpretations and public health relevance of the obtained effect estimates.

METHODS

In 2016, the study measured hemoglobin and anthropometric measures and collected data from the children's caregivers. We used propensity score matching analyses, inverse probability weighting, and overlap weighting methods to examine the average treatment effects among treated population (ATT), and among population with overlapping weights (ATO).

RESULTS

The ATT for stunting among children who participated in the intervention program estimated by propensity score matching analyses (PSM-ATT) was (1.95; 95%CI 1.15-3.28). Nevertheless, children who took part in the program had a lower risk of anemia (0.28; 95%CI 0.19-0.42). The ATT, estimated by inverse probability weighting (IPTW-ATT), was slightly lower for stunting (1.82; 95%CI 1.16-2.86) while similar for anemia (0.33; 95%CI 0.23-0.46) compared to PSM-ATT. Compared to the IPTW-ATT and PSM-ATT the ATO was lower for stunting (1.75; 95%CI 1.14-2.68) and similar for anemia (0.31; 95%CI 0.22-0.43).

DISCUSSION

Marginal models could be used in similar quasi-experimental settings to identify the causal effect of interventions in specific populations of interest. Nonetheless, these methods do not eliminate threats to internal validity. Thorough study design and accurate data collection are necessary to improve the efficiency of marginal models.

Statistical Mediation Analysis for Models with a Binary Mediator and a Binary Outcome: the Differences Between Causal and Traditional Mediation Analysis

Mediation analysis is an important statistical method in prevention research, as it can be used to determine effective intervention components. Traditional mediation analysis defines direct and indirect effects in terms of linear regression coefficients. It is unclear how these traditional effects are estimated in settings with binary variables. An important recent methodological advancement in the mediation analysis literature is the development of the causal mediation analysis framework. Causal mediation analysis defines causal effects as the difference between two potential outcomes. These definitions can be applied to any mediation model to estimate natural direct and indirect effects, including models with binary variables and an exposure-mediator interaction. This paper aims to clarify the similarities and differences between the causal and traditional effect estimates for mediation models with a binary mediator and a binary outcome. Causal and traditional mediation analyses were applied to an empirical example to demonstrate these similarities and differences. Causal and traditional mediation analysis provided similar controlled direct effect estimates, but different estimates of the natural direct effects, natural indirect effects, and total effect. Traditional mediation analysis methods do not generalize well to mediation models with binary variables, while the natural effect definitions can be applied to any mediation model. Causal mediation analysis is therefore the preferred method for the analysis of mediation models with binary variables.

Impact of prognostic factor distributions on mortality disparities for socioeconomically disadvantaged cancer patients

PURPOSE

We aimed to assess whether differences in the distributions of prognostic factors explain reported mortality disparities between urban safety-net and Surveillance, Epidemiology, and End Results (SEER) cancer populations.

METHODS

We used data from SEER and a safety-net cancer center in Texas. Eligible patients were adults aged ≤64 years and diagnosed with first primary female breast, colorectal, or lung cancer between 2008 and 2016. We estimated crude and adjusted risk differences (RD) in 3- and 5-year all-cause mortality (1- and 3-year for lung cancer), where adjustment was based on entropy balancing weights that standardized the distribution of sociodemographic and tumor characteristics between the two populations.

RESULTS

Our study populations comprised 1914 safety-net patients and 389,709 SEER patients. For breast cancer, the crude 3- and 5-year mortality RDs between safety-net and SEER populations were 7.7% (95% confidence limits [CL]: 4.3%, 11%) and 11% (95% CL: 6.7%, 16%). Adjustment for measured prognostic factors reduced the mortality RDs (3-year adjusted RD = 0.049%, 95% CL: -2.6%, 2.6%; 5-year adjusted RD = 5.6%, 95% CL: -0.83%, 12%). We observed similar patterns for colorectal and lung cancer albeit less magnitude.

CONCLUSIONS

Sociodemographic and tumor characteristics may largely explain early mortality disparities between safety-net and SEER populations but not late mortality disparities.

Title: Is OR "portable" in meta-analysis? Time to consider bivariate generalized linear mixed model

OBJECTIVES

 A recent paper by Doi et al. advocated completely replacing the relative risk (RR) with the odds ratio (OR) as the effect measure in clinical trials and meta-analyses with binary outcomes. Besides some practical advantages of RR over OR, Doi et al.'s key assumption that the OR is "portable" in the meta-analysis, i.e., study-specific ORs are likely not correlated with baseline risks, was not well justified. Study designs and settings: We summarized Spearman's rank correlation coefficient between study-specific ORs and baseline risks in 40,243 meta-analyses from the Cochrane Database of Systematic Reviews (CDSR).

RESULTS

Study-specific ORs tend to be higher in studies with lower baseline risks of disease for most meta-analyses in CDSR. Using an actual meta-analysis example, we demonstrate that there is a strong negative correlation between OR (RR or RD) with the baseline risk and the conditional effects notably vary with baseline risks.

CONCLUSIONS

 Replacing RR or RD with OR is currently unadvisable in clinical trials and meta-analyses. It is possible that no effect measure is "portable" in a meta-analysis. In addition to the overall (or marginal) effect, we suggest presenting the conditional effect based on the baseline risk using a bivariate generalized linear mixed model.

Prevalence and early-life determinants of mid-life multimorbidity: evidence from the 1970 British birth cohort

BACKGROUND

We sought to: [1] estimate the prevalence of multimorbidity at age 46-48 in the 1970 British Cohort Study-a nationally representative sample in mid-life; and [2] examine the association between early-life characteristics and mid-life multimorbidity.

METHOD

A prospective longitudinal birth cohort of a community-based sample from the 1970 British Cohort Study (BCS70). Participants included all surviving children born in mainland Britain in a single week in April 1970; the analytical sample included those with valid data at age 46-48 (n = 7951; 2016-2018). The main outcome was multimorbidity, which was operationalised as a binary indicator of two or more long-term health conditions where at least one of these conditions was of physical health. It also included symptom complexes (e.g., chronic pain), sensory impairments, and alcohol problems.

RESULTS

Prevalence of mid-life multimorbidity was 33.8% at age 46-48. Those with fathers from unskilled social occupational class (vs professional) at birth had 43% higher risk of mid-life multimorbidity (risk ratio = 1.43, 95% confidence interval 1.15 to 1.77). After accounting for potential child and family confounding, an additional kilogram of birthweight was associated with 10% reduced risk of multimorbidity (risk ratio = 0.90, 95% confidence interval 0.84 to 0.96); a decrease of one body mass index point at age 10 was associated with 3% lower risk (risk ratio = 1.03, 95% confidence interval 1.01 to 1.05); one standard deviation higher cognitive ability score at age 10 corresponded to 4% lower risk (risk ratio = 0.96, 95% confidence interval 0.91 to 1.00); an increase of one internalising problem at age 16 was equated with 4% higher risk (risk ratio = 1.04, 95% confidence interval 1.00 to 1.08) and of one externalising problem at age 16 with 6% higher risk (risk ratio = 1.06, 1.03 to 1.09).

CONCLUSION

Prevalence of multimorbidity was high in mid-life (33.8% at age 46-48) in Britain. Potentially modifiable early-life exposures, including early-life social circumstances, cognitive, physical and emotional development, were associated with elevated risk of mid-life multimorbidity.

Noncollapsibility and its role in quantifying confounding bias in logistic regression

BACKGROUND

Confounding bias is a common concern in epidemiological research. Its presence is often determined by comparing exposure effects between univariable- and multivariable regression models, using an arbitrary threshold of a 10% difference to indicate confounding bias. However, many clinical researchers are not aware that the use of this change-in-estimate criterion may lead to wrong conclusions when applied to logistic regression coefficients. This is due to a statistical phenomenon called noncollapsibility, which manifests itself in logistic regression models. This paper aims to clarify the role of noncollapsibility in logistic regression and to provide guidance in determining the presence of confounding bias.

METHODS

A Monte Carlo simulation study was designed to uncover patterns of confounding bias and noncollapsibility effects in logistic regression. An empirical data example was used to illustrate the inability of the change-in-estimate criterion to distinguish confounding bias from noncollapsibility effects.

RESULTS

The simulation study showed that, depending on the sign and magnitude of the confounding bias and the noncollapsibility effect, the difference between the effect estimates from univariable- and multivariable regression models may underestimate or overestimate the magnitude of the confounding bias. Because of the noncollapsibility effect, multivariable regression analysis and inverse probability weighting provided different but valid estimates of the confounder-adjusted exposure effect. In our data example, confounding bias was underestimated by the change in estimate due to the presence of a noncollapsibility effect.

CONCLUSION

In logistic regression, the difference between the univariable- and multivariable effect estimate might not only reflect confounding bias but also a noncollapsibility effect. Ideally, the set of confounders is determined at the study design phase and based on subject matter knowledge. To quantify confounding bias, one could compare the unadjusted exposure effect estimate and the estimate from an inverse probability weighted model.

Noncollapsibility, confounding, and sparse-data bias. Part 1: The oddities of odds

To prevent statistical misinterpretations, it has long been advised to focus on estimation instead of statistical testing. This sound advice brings with it the need to choose the outcome and effect measures on which to focus. Measures based on odds or their logarithms have often been promoted due to their pleasing statistical properties, but have an undesirable property for risk summarization and communication: Noncollapsibility, defined as a failure of the measure when taken on a group to equal a simple average of the measure when taken on the group's members or subgroups. The present note illustrates this problem with a basic numeric example involving the odds, which is not collapsible when the odds vary across individuals and are not low in all subgroups. Its sequel will illustrate how this problem is amplified in odds ratios and logistic regression.

Noncollapsibility, confounding, and sparse-data bias. Part 2: What should researchers make of persistent controversies about the odds ratio?

A previous note illustrated how the odds of an outcome has an undesirable property for risk summarization and communication: Noncollapsibility, defined as a failure of a group measure to represent a simple average of the measure over individuals or subgroups. The present sequel discusses how odds ratios amplify odds noncollapsibility and provides a basic numeric illustration of how noncollapsibility differs from confounding of effects (with which it is often confused). It also draws a connection of noncollapsibility to sparse-data bias in logistic, log-linear, and proportional-hazards regression.

Mendelian randomisation for mediation analysis: current methods and challenges for implementation

Mediation analysis seeks to explain the pathway(s) through which an exposure affects an outcome. Traditional, non-instrumental variable methods for mediation analysis experience a number of methodological difficulties, including bias due to confounding between an exposure, mediator and outcome and measurement error. Mendelian randomisation (MR) can be used to improve causal inference for mediation analysis. We describe two approaches that can be used for estimating mediation analysis with MR: multivariable MR (MVMR) and two-step MR. We outline the approaches and provide code to demonstrate how they can be used in mediation analysis. We review issues that can affect analyses, including confounding, measurement error, weak instrument bias, interactions between exposures and mediators and analysis of multiple mediators. Description of the methods is supplemented by simulated and real data examples. Although MR relies on large sample sizes and strong assumptions, such as having strong instruments and no horizontally pleiotropic pathways, our simulations demonstrate that these methods are unaffected by confounders of the exposure or mediator and the outcome and non-differential measurement error of the exposure or mediator. Both MVMR and two-step MR can be implemented in both individual-level MR and summary data MR. MR mediation methods require different assumptions to be made, compared with non-instrumental variable mediation methods. Where these assumptions are more plausible, MR can be used to improve causal inference in mediation analysis.

Missing at random assumption made more plausible: evidence from the 1958 British birth cohort

OBJECTIVE

Non-response is unavoidable in longitudinal surveys. The consequences are lower statistical power and the potential for bias. We implemented a systematic data-driven approach to identify predictors of non-response in the National Child Development Study (NCDS; 1958 British birth cohort). Such variables can help make the missing at random assumption more plausible, which has implications for the handling of missing data STUDY DESIGN AND SETTING: We identified predictors of non-response using data from the 11 sweeps (birth to age 55) of the NCDS (n = 17,415), employing parametric regressions and the LASSO for variable selection.

RESULTS

Disadvantaged socio-economic background in childhood, worse mental health and lower cognitive ability in early life, and lack of civic and social participation in adulthood were consistently associated with non-response. Using this information, along with other data from NCDS, we were able to replicate the "population distribution" of educational attainment and marital status (derived from external data), and the original distributions of key early life characteristics.

CONCLUSION

The identified predictors of non-response have the potential to improve the plausibility of the missing at random assumption. They can be straightforwardly used as "auxiliary variables" in analyses with principled methods to reduce bias due to missing data.

Association of Early-Life Mental Health With Biomarkers in Midlife and Premature Mortality: Evidence From the 1958 British Birth Cohort

Importance

Early-life mental health is known to be associated with socioeconomic adversity and psychological distress in adulthood, but less is known about potential associations with biomarkers and mortality.

Objective

To investigate the association between early-life mental health trajectories with biomarkers in midlife and premature mortality.

Design, Setting, and Participants

This study used data from the British National Child Development Study, a population-based birth cohort. The initial sample of 17 415 individuals consisted of all infants born in Great Britain in a single week in 1958. Analysis began Feburary 2017 and ended May 2020.

Main Outcomes and Measures

Biomarkers collected at age 44 to 45 years were fibrinogen, C-reactive protein, glycated hemoglobin, high- and low-density lipoprotein cholesterol, forced expiratory volume, blood pressure, and waist-to-hip ratio. Information on all-cause mortality was available up to age 58 years and cause-specific mortality up to age 50 years.

Results

Biomarkers were analyzed from 9377 participants (age, 44-45 years, 4712 female [50.3%]) and mortality data from 15 067 participants (age, 58 years; 7379 female [49.0%]). A 4-group longitudinal typology of early-life conduct problems and affective symptoms was identified: (1) stable low, (2) teacher-identified adolescent onset, (3) moderate, and (4) stable high. Compared with the stable-low group, the stable-high (B, 2.308; 95% CI, 0.213-4.404) and teacher-identified adolescent-onset (B, 2.114; 95% CI, 0.725-3.503) groups had less favorable levels of fibrinogen in middle age. Effect modification was observed by sex (P < .005) such that compared with the stable-low group, differences in high-density lipoprotein and abdominal obesity were only observed in female individuals. The stable-high and teacher-identified adolescent-onset groups had elevated risk for all-cause mortality (hazard ratio, 1.878; 95% CI, 1.501-2.350 and hazard ratio, 1.412; 95% CI, 1.174-1.698). Psychopathology-associated mortality was higher in both groups but unintentional injuries-associated mortality only in the stable-high group.

Conclusions and Relevance

Experiencing affective symptoms and conduct problems in childhood and adolescence is associated with less favorable levels of biomarkers 28 years later and elevated risk of premature mortality. These findings, if causal and generalizable to younger cohorts, may imply that effective interventions on early-life mental health have the potential to shift the distribution of risk and improve population health.

Is OR "portable" in meta-analysis? Time to consider bivariate generalized linear mixed model

OBJECTIVES

A recent paper by Doi et al. advocated completely replacing the relative risk (RR) with the odds ratio (OR) as the effect measure used to report the association between a treatment and a binary outcome in clinical trials and meta-analyses. Besides some practical advantages of RR over OR and the well-known issue of the OR being non-collapsible, Doi et al.'s key assumption that the OR is "portable" in the meta-analysis, i.e., study-specific ORs are likely not correlated with baseline risks, was not well justified. Study designs and settings: We summarized the Spearman's rank correlation coefficient between study-specific OR and the baseline risk in 40,243 meta-analyses from the Cochrane Database of Systematic Reviews (CDSR).

RESULTS

Study-specific ORs are negatively correlated with baseline risk of disease (i.e., higher ORs tend to be observed in studies with lower baseline risks of disease) for most meta-analyses in CDSR. Using a meta-analysis comparing the effect of oral sumatriptan (100 mg) versus placebo on mitigating the acute headache at 2 hours after drug administration, we demonstrate that there is a strong negative correlation between OR (RR or RD) with the baseline risk and the conditional effects notably vary with baseline risks.

CONCLUSIONS

Replacing RR or RD with OR is currently unadvisable in clinical trials and meta-analyses. It is possible that no effect measure is "portable" in a meta-analysis. In cases where portability of the effect measure is challenging to satisfy, we suggest presenting the conditional effect based on the baseline risk using a bivariate generalized linear mixed model. The bivariate generalized linear mixed model can be used to account for correlation between the effect measure and baseline disease risk. Furthermore, in addition to the overall (or marginal) effect, we recommend that investigators also report the effects conditioning on the baseline risk.

Postmenopausal Androgen Metabolism and Endometrial Cancer Risk in the Women's Health Initiative Observational Study

Background

After menopause, several androgens continue to be produced primarily by the adrenal glands; these can be converted into estrogens via aromatization or into androgen metabolites. It is unclear if androgens are associated with endometrial cancer risk independently of their being precursors to estrogens or if alternative metabolic pathways influence risk.

Methods

We measured prediagnostic serum concentrations of 12 androgens and their metabolites using highly sensitive liquid chromatography-tandem mass spectrometry assays in a nested case-control study of postmenopausal women from the Women's Health Initiative Observational Study (313 endometrial cancer case subjects, 354 matched control subjects). Estrogens were previously assayed. We used conditional logistic regression to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for endometrial cancer with adjustment for confounders.

Results

Compared to the lowest concentrations, the highest levels of adrenal androgens were associated with increased endometrial cancer risk: dehydroepiandrosterone (5th vs 1st quintile: OR = 1.85, 95% CI = 1.06 to 3.25), androstenedione (OR = 2.36, 95% CI = 1.34 to 4.16), and testosterone (OR = 1.91, 95% CI = 1.12 to 3.24). Downstream androgen metabolites were not associated with endometrial cancer. Although increased risks for the parent androgens were still suggested after adjustment for unconjugated estradiol, the associations attenuated, and with the exception of androstenedione, were no longer statistically significant. We also evaluated ratios of estrogens relative to their androgenic precursors; both higher unconjugated estrone:androstenedione and higher unconjugated estradiol:testosterone were associated with increased endometrial cancer risk.

Conclusions

We identified increased risks for endometrial cancer with the highest levels of adrenal androgens and high levels of estrogens relative to these androgens. As adrenal androgens can be aromatized to estrogens, this suggests androgens likely influence endometrial carcinogenesis via estrogen metabolism.

Health administrative data enrichment using cohort information: Comparative evaluation of methods by simulation and application to real data

Background

Studies using health administrative databases (HAD) may lead to biased results since information on potential confounders is often missing. Methods that integrate confounder data from cohort studies, such as multivariate imputation by chained equations (MICE) and two-stage calibration (TSC), aim to reduce confounding bias. We provide new insights into their behavior under different deviations from representativeness of the cohort.

Methods

We conducted an extensive simulation study to assess the performance of these two methods under different deviations from representativeness of the cohort. We illustrate these approaches by studying the association between benzodiazepine use and fractures in the elderly using the general sample of French health insurance beneficiaries (EGB) as main database and two French cohorts (Paquid and 3C) as validation samples.

Results

When the cohort was representative from the same population as the HAD, the two methods are unbiased. TSC was more efficient and faster but its variance could be slightly underestimated when confounders were non-Gaussian. If the cohort was a subsample of the HAD (internal validation) with the probability of the subject being included in the cohort depending on both exposure and outcome, MICE was unbiased while TSC was biased. The two methods appeared biased when the inclusion probability in the cohort depended on unobserved confounders.

Conclusion

When choosing the most appropriate method, epidemiologists should consider the origin of the cohort (internal or external validation) as well as the (anticipated or observed) selection biases of the validation sample.

Retrospectively reported childhood adversity is associated with asthma and chronic bronchitis, independent of mental health

Several researchers have raised the concern that the cross-sectional association of retrospectively reported childhood adversity with self-reported onset of asthma and chronic bronchitis in adulthood may be confounded, as well as mediated by an individual's mental health. The aim of this study was to assess the effect of retrospectively reported childhood adversity on self-reported onset of asthma and chronic bronchitis in adulthood, independent of potential confounding and mediating variables (including respondent's mental health). We used data collected in 2007-2008 within the framework of the Tromsø Study (N = 12,981), a representative study of adult men and women in Norway. The associations of childhood adversity with asthma and chronic bronchitis were assessed with Poisson regression models. Relative risks (RR) and 95% confidence intervals (CI) were estimated with bias-corrected bootstrapping. Childhood adversity was associated with a 9% increased risk of asthma (RR = 1.09, 95% CI: 1.02, 1.16) and a 14% increased risk chronic bronchitis (RR = 1.14, 95% CI: 1.03, 1.26) in adulthood, independent of age, sex, parental history of psychiatric problems/asthma/dementia, education, smoking, social support, and respondent's mental health. Controlling for indicators of respondent's mental health reduced the strength of associations of childhood adversity with asthma and chronic bronchitis; however, the associations were still present in the same direction (p < .05). These findings suggest that the association of retrospectively reported childhood adversity with asthma and chronic bronchitis is independent of respondent's mental health. We recommend controlling for indicators of the respondent's mental health to assess an unbiased association of retrospectively measured childhood adversity with self-reported asthma and chronic bronchitis.

Targeted Maximum Likelihood Estimation for Pharmacoepidemiologic Research

Supplemental Digital Content is available in the text.

Background:

Targeted maximum likelihood estimation has been proposed for estimating marginal causal effects, and is robust to misspecification of either the treatment or outcome model. However, due perhaps to its novelty, targeted maximum likelihood estimation has not been widely used in pharmacoepidemiology. The objective of this study was to demonstrate targeted maximum likelihood estimation in a pharmacoepidemiological study with a high-dimensional covariate space, to incorporate the use of high-dimensional propensity scores into this method, and to compare the results to those of inverse probability weighting.

Methods:

We implemented the targeted maximum likelihood estimation procedure in a single-point exposure study of the use of statins and the 1-year risk of all-cause mortality postmyocardial infarction using data from the UK Clinical Practice Research Datalink. A range of known potential confounders were considered, and empirical covariates were selected using the high-dimensional propensity scores algorithm. We estimated odds ratios using targeted maximum likelihood estimation and inverse probability weighting with a variety of covariate selection strategies.

Results:

Through a real example, we demonstrated the double robustness of targeted maximum likelihood estimation. We showed that results with this method and inverse probability weighting differed when a large number of covariates were included in the treatment model.

Conclusions:

Targeted maximum likelihood can be used in high-dimensional covariate settings. In high-dimensional covariate settings, differences in results between targeted maximum likelihood and inverse probability weighted estimation are likely due to sensitivity to (near) positivity violations. Further investigations are needed to gain better understanding of the advantages and limitations of this method in pharmacoepidemiological studies.

The alarming problems of confounding equivalence using logistic regression models in the perspective of causal diagrams

Background

Confounders can produce spurious associations between exposure and outcome in observational studies. For majority of epidemiologists, adjusting for confounders using logistic regression model is their habitual method, though it has some problems in accuracy and precision. It is, therefore, important to highlight the problems of logistic regression and search the alternative method.

Methods

Four causal diagram models were defined to summarize confounding equivalence. Both theoretical proofs and simulation studies were performed to verify whether conditioning on different confounding equivalence sets had the same bias-reducing potential and then to select the optimum adjusting strategy, in which logistic regression model and inverse probability weighting based marginal structural model (IPW-based-MSM) were compared. The “do-calculus” was used to calculate the true causal effect of exposure on outcome, then the bias and standard error were used to evaluate the performances of different strategies.

Results

Adjusting for different sets of confounding equivalence, as judged by identical Markov boundaries, produced different bias-reducing potential in the logistic regression model. For the sets satisfied G-admissibility, adjusting for the set including all the confounders reduced the equivalent bias to the one containing the parent nodes of the outcome, while the bias after adjusting for the parent nodes of exposure was not equivalent to them. In addition, all causal effect estimations through logistic regression were biased, although the estimation after adjusting for the parent nodes of exposure was nearest to the true causal effect. However, conditioning on different confounding equivalence sets had the same bias-reducing potential under IPW-based-MSM. Compared with logistic regression, the IPW-based-MSM could obtain unbiased causal effect estimation when the adjusted confounders satisfied G-admissibility and the optimal strategy was to adjust for the parent nodes of outcome, which obtained the highest precision.

Conclusions

All adjustment strategies through logistic regression were biased for causal effect estimation, while IPW-based-MSM could always obtain unbiased estimation when the adjusted set satisfied G-admissibility. Thus, IPW-based-MSM was recommended to adjust for confounders set.

Electronic supplementary material

The online version of this article (10.1186/s12874-017-0449-7) contains supplementary material, which is available to authorized users.

Performance of the marginal structural cox model for estimating individual and joined effects of treatments given in combination

BACKGROUND

The Marginal Structural Cox Model (Cox-MSM), an alternative approach to handle time-dependent confounder, was introduced for survival analysis and applied to estimate the joint causal effect of two time-dependent nonrandomized treatments on survival among HIV-positive subjects. Nevertheless, Cox-MSM performance in the case of multiple treatments has not been fully explored under different degree of time-dependent confounding for treatments or in case of interaction between treatments. We aimed to evaluate and compare the performance of the marginal structural Cox model (Cox-MSM) to the standard Cox model in estimating the treatment effect in the case of multiple treatments under different scenarios of time-dependent confounding and when an interaction between treatment effects is present.

METHODS

We specified a Cox-MSM with two treatments including an interaction term for situations where an adverse event might be caused by two treatments taken simultaneously but not by each treatment taken alone. We simulated longitudinal data with two treatments and a time-dependent confounder affected by one or the two treatments. To fit the Cox-MSM, we used the inverse probability weighting method. We illustrated the method to evaluate the specific effect of protease inhibitors combined (or not) to other antiretroviral medications on the anal cancer risk in HIV-infected individuals, with CD4 cell count as time-dependent confounder.

RESULTS

Overall, Cox-MSM performed better than the standard Cox model. Furthermore, we showed that estimates were unbiased when an interaction term was included in the model.

CONCLUSION

Cox-MSM may be used for accurately estimating causal individual and joined treatment effects from a combination therapy in presence of time-dependent confounding provided that an interaction term is estimated.