Recoverability and estimation of causal effects under typical multivariable missingness mechanisms

In the context of missing data, the identifiability or "recoverability" of the average causal effect (ACE) depends not only on the usual causal assumptions but also on missingness assumptions that can be depicted by adding variable-specific missingness indicators to causal diagrams, creating missingness directed acyclic graphs (m-DAGs). Previous research described canonical m-DAGs, representing typical multivariable missingness mechanisms in epidemiological studies, and examined mathematically the recoverability of the ACE in each case. However, this work assumed no effect modification and did not investigate methods for estimation across such scenarios. Here, we extend this research by determining the recoverability of the ACE in settings with effect modification and conducting a simulation study to evaluate the performance of widely used missing data methods when estimating the ACE using correctly specified g-computation. Methods assessed were complete case analysis (CCA) and various implementations of multiple imputation (MI) with varying degrees of compatibility with the outcome model used in g-computation. Simulations were based on an example from the Victorian Adolescent Health Cohort Study (VAHCS), where interest was in estimating the ACE of adolescent cannabis use on mental health in young adulthood. We found that the ACE is recoverable when no incomplete variable (exposure, outcome, or confounder) causes its own missingness, and nonrecoverable otherwise, in simplified versions of 10 canonical m-DAGs that excluded unmeasured common causes of missingness indicators. Despite this nonrecoverability, simulations showed that MI approaches that are compatible with the outcome model in g-computation may enable approximately unbiased estimation across all canonical m-DAGs considered, except when the outcome causes its own missingness or causes the missingness of a variable that causes its own missingness. In the latter settings, researchers may need to consider sensitivity analysis methods incorporating external information (e.g., delta-adjustment methods). The VAHCS case study illustrates the practical implications of these findings.

Characterization of Puberty in an Australian Population-Based Cohort Study

PURPOSE

Current knowledge of the characteristics of puberty beyond age at menarche and thelarche is limited, particularly within population-based cohorts. Secular trends and concerns of the health effects of early puberty reinforce the value of contemporary studies characterizing the timing, tempo, duration, and synchronicity of puberty.

METHODS

The Childhood to Adolescence Transition Study is a unique Australian cohort of individuals followed annually from late childhood to late adolescence, with up to eight assessments of pubertal stage from 9 to 19 years of age (N = 1,183; 636 females). At each assessment, females reported their Tanner Stage of breast and pubic hair development, while males reported on genital/pubic hair development. Nonlinear mixed-effects models characterized pubertal trajectories and were used to derive each individual's estimates of timing, tempo, and synchronicity. Parametric survival models were used to estimate the overall duration of puberty.

RESULTS

Timing of mid-puberty (Tanner Stage 3) ranged from 12.5 to 13.5 years, with females developing approximately 6 months before males. Pubertal tempo (at mid-puberty) was similar across sex (between half and one Tanner Stage per year), but the overall duration of puberty was slightly shorter in males. Most females exhibited asynchronous changes of breast and pubic hair development.

DISCUSSION

Estimates of pubertal timing and tempo are consistent with reports of cohorts from two or more decades ago, suggesting stabilization of certain pubertal characteristics in predominantly White populations. However, our understanding of the duration of puberty and individual differences in pubertal characteristics (e.g., synchronicity of physical changes) remains limited.

Handling missing data when estimating causal effects with Targeted Maximum Likelihood Estimation

Targeted Maximum Likelihood Estimation (TMLE) is increasingly used for doubly robust causal inference, but how missing data should be handled when using TMLE with data-adaptive approaches is unclear. Based on the Victorian Adolescent Health Cohort Study, we conducted a simulation study to evaluate eight missing data methods in this context: complete-case analysis, extended TMLE incorporating outcome-missingness model, missing covariate missing indicator method, five multiple imputation (MI) approaches using parametric or machine-learning models. Six scenarios were considered, varying in exposure/outcome generation models (presence of confounder-confounder interactions) and missingness mechanisms (whether outcome influenced missingness in other variables and presence of interaction/non-linear terms in missingness models). Complete-case analysis and extended TMLE had small biases when outcome did not influence missingness in other variables. Parametric MI without interactions had large bias when exposure/outcome generation models included interactions. Parametric MI including interactions performed best in bias and variance reduction across all settings, except when missingness models included a non-linear term. When choosing a method to handle missing data in the context of TMLE, researchers must consider the missingness mechanism and, for MI, compatibility with the analysis method. In many settings, a parametric MI approach that incorporates interactions and non-linearities is expected to perform well.

Household income supplements in early childhood to reduce inequities in children's development

BACKGROUND

Early childhood interventions have the potential to reduce children's developmental inequities. We aimed to estimate the extent to which household income supplements for lower-income families in early childhood could close the gap in children's developmental outcomes and parental mental health.

METHODS

Data were drawn from a nationally representative birth cohort, the Longitudinal Study of Australian Children (N = 5107), which commenced in 2004 and conducted follow-ups every two years. Exposure was annual household income (0-1 year). Outcomes were children's developmental outcomes, specifically social-emotional, physical functioning, and learning (bottom 15% versus top 85%) at 4-5 years, and an intermediate outcome, parental mental health (poor versus good) at 2-3 years. We modelled hypothetical interventions that provided a fixed-income supplement to lower-income families with a child aged 0-1 year. Considering varying eligibility scenarios and amounts motivated by actual policies in the Australian context, we estimated the risk of poor outcomes for eligible families under no intervention and the hypothetical intervention using marginal structural models. The reduction in risk under intervention relative to no intervention was estimated.

RESULTS

A single hypothetical supplement of AU$26,000 (equivalent to ∼USD$17,350) provided to lower-income families (below AU$56,137 (∼USD$37,915) per annum) in a child's first year of life demonstrated an absolute reduction of 2.7%, 1.9% and 2.6% in the risk of poor social-emotional, physical functioning and learning outcomes in children, respectively (equivalent to relative reductions of 12%, 10% and 11%, respectively). The absolute reduction in risk of poor mental health in eligible parents was 1.0%, equivalent to a relative reduction of 7%. Benefits were similar across other income thresholds used to assess eligibility (range, AU$73,329-$99,864).

CONCLUSIONS

Household income supplements provided to lower-income families may benefit children's development and parental mental health. This intervention should be considered within a social-ecological approach by stacking complementary interventions to eliminate developmental inequities.

Caries Detection in Primary Teeth Using Intraoral Scanners Featuring Fluorescence: Protocol for a Diagnostic Agreement Study

BACKGROUND

Digital methods that enable early caries identification can streamline data collection in research and optimize dental examinations for young children. Intraoral scanners are devices used for creating 3D models of teeth in dentistry and are being rapidly adopted into clinical workflows. Integrating fluorescence technology into scanner hardware can support early caries detection. However, the performance of caries detection methods using 3D models featuring color and fluorescence in primary teeth is unknown.

OBJECTIVE

This study aims to assess the diagnostic agreement between visual examination (VE), on-screen assessment of 3D models in approximate natural colors with and without fluorescence, and application of an automated caries scoring system to the 3D models with fluorescence for caries detection in primary teeth.

METHODS

The study sample will be drawn from eligible participants in a randomized controlled trial at the Royal Children's Hospital, Melbourne, Australia, where a dental assessment was conducted, including VE using the International Caries Detection and Assessment System (ICDAS) and intraoral scan using the TRIOS 4 (3Shape TRIOS A/S). Participant clinical records will be collected, and all records meeting eligibility criteria will be subject to an on-screen assessment of 3D models by 4 dental practitioners. First, all primary tooth surfaces will be examined for caries based on 3D geometry and color, using a merged ICDAS index. Second, the on-screen assessment of 3D models will include fluorescence, where caries will be classified using a merged ICDAS index that has been modified to incorporate fluorescence criteria. After 4 weeks, all examiners will repeat the on-screen assessment for all 3D models. Finally, an automated caries scoring system will be used to classify caries on primary occlusal surfaces. The agreement in the total number of caries detected per person between methods will be assessed using a Bland-Altman analysis and intraclass correlation coefficients. At a tooth surface level, agreement between methods will be estimated using multilevel models to account for the clustering of dental data.

RESULTS

Automated caries scoring of 3D models was completed as of October 2023, with the publication of results expected by July 2024. On-screen assessment has commenced, with the expected completion of scoring and data analysis by March 2024. Results will be disseminated by the end of 2024.

CONCLUSIONS

The study outcomes may inform new practices that use digital models to facilitate dental assessments. Novel approaches that enable remote dental examination without compromising the accuracy of VE have wide applications in the research environment, clinical practice, and the provision of teledentistry.

TRIAL REGISTRATION

Australian New Zealand Clinical Trials Registry ACTRN12622001237774; https://www.anzctr.org.au/Trial/Registration/TrialReview.aspx?id=384632.

INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID)

DERR1-10.2196/51578.

Confounding-adjustment methods for the causal difference in medians

BACKGROUND

With continuous outcomes, the average causal effect is typically defined using a contrast of expected potential outcomes. However, in the presence of skewed outcome data, the expectation (population mean) may no longer be meaningful. In practice the typical approach is to continue defining the estimand this way or transform the outcome to obtain a more symmetric distribution, although neither approach may be entirely satisfactory. Alternatively the causal effect can be redefined as a contrast of median potential outcomes, yet discussion of confounding-adjustment methods to estimate the causal difference in medians is limited. In this study we described and compared confounding-adjustment methods to address this gap.

METHODS

The methods considered were multivariable quantile regression, an inverse probability weighted (IPW) estimator, weighted quantile regression (another form of IPW) and two little-known implementations of g-computation for this problem. Methods were evaluated within a simulation study under varying degrees of skewness in the outcome and applied to an empirical study using data from the Longitudinal Study of Australian Children.

RESULTS

Simulation results indicated the IPW estimator, weighted quantile regression and g-computation implementations minimised bias across all settings when the relevant models were correctly specified, with g-computation additionally minimising the variance. Multivariable quantile regression, which relies on a constant-effect assumption, consistently yielded biased results. Application to the empirical study illustrated the practical value of these methods.

CONCLUSION

The presented methods provide appealing avenues for estimating the causal difference in medians.

On the use of multiple imputation to address data missing by design as well as unintended missing data in case-cohort studies with a binary endpoint

BACKGROUND

Case-cohort studies are conducted within cohort studies, with the defining feature that collection of exposure data is limited to a subset of the cohort, leading to a large proportion of missing data by design. Standard analysis uses inverse probability weighting (IPW) to address this intended missing data, but little research has been conducted into how best to perform analysis when there is also unintended missingness. Multiple imputation (MI) has become a default standard for handling unintended missingness and is typically used in combination with IPW to handle the intended missingness due to the case-control sampling. Alternatively, MI could be used to handle both the intended and unintended missingness. While the performance of an MI-only approach has been investigated in the context of a case-cohort study with a time-to-event outcome, it is unclear how this approach performs with a binary outcome.

METHODS

We conducted a simulation study to assess and compare the performance of approaches using only MI, only IPW, and a combination of MI and IPW, for handling intended and unintended missingness in the case-cohort setting. We also applied the approaches to a case study.

RESULTS

Our results show that the combined approach is approximately unbiased for estimation of the exposure effect when the sample size is large, and was the least biased with small sample sizes, while MI-only and IPW-only exhibited larger biases in both sample size settings.

CONCLUSIONS

These findings suggest that a combined MI/IPW approach should be preferred to handle intended and unintended missing data in case-cohort studies with binary outcomes.

The potential of intervening on childhood adversity to reduce socioeconomic inequities in body mass index and inflammation among Australian and UK children: A causal mediation analysis

BACKGROUND

Lower maternal education is associated with higher body mass index (BMI) and higher chronic inflammation in offspring. Childhood adversity potentially mediates these associations. We examined the extent to which addressing childhood adversity could reduce socioeconomic inequities in these outcomes.

METHODS

We analysed data from two early-life longitudinal cohorts: the Longitudinal Study of Australian Children (LSAC; n=1873) and the UK Avon Longitudinal Study of Parents and Children (ALSPAC; n=7085).

EXPOSURE

low/medium (below university degree) versus high maternal education, as a key indicator of family socioeconomic position (0-1 year).

OUTCOMES

BMI and log-transformed glycoprotein acetyls (GlycA) (LSAC: 11-12 years; ALSPAC: 15.5 years). Mediator: multiple adversities (≥2/<2) indicated by family violence, mental illness, substance abuse and harsh parenting (LSAC: 2-11 years; ALSPAC: 1-12 years). A causal mediation analysis was conducted.

RESULTS

Low/medium maternal education was associated with up to 1.03 kg/m2 higher BMI (95% CI: 0.95 to 1.10) and up to 1.69% higher GlycA (95% CI: 1.68 to 1.71) compared with high maternal education, adjusting for confounders. Causal mediation analysis estimated that decreasing the levels of multiple adversities in children with low/medium maternal education to be like their high maternal education peers could reduce BMI inequalities by up to 1.8% and up to 3.3% in GlycA.

CONCLUSIONS

Our findings in both cohorts suggest that slight reductions in socioeconomic inequities in children's BMI and inflammation could be achieved by addressing childhood adversities. Public health and social policy efforts should help those affected by childhood adversity, but also consider underlying socioeconomic conditions that drive health inequities.

Development of the TrAnsparent ReportinG of observational studies Emulating a Target trial (TARGET) guideline

BACKGROUND

Observational studies are increasingly used to inform health decision-making when randomised trials are not feasible, ethical or timely. The target trial approach provides a framework to help minimise common biases in observational studies that aim to estimate the causal effect of interventions. Incomplete reporting of studies using the target trial framework limits the ability for clinicians, researchers, patients and other decision-makers to appraise, synthesise and interpret findings to inform clinical and public health practice and policy. This paper describes the methods that we will use to develop the TrAnsparent ReportinG of observational studies Emulating a Target trial (TARGET) reporting guideline.

METHODS/DESIGN

The TARGET reporting guideline will be developed in five stages following recommended guidance. The first stage will identify target trial reporting practices by systematically reviewing published studies that explicitly emulated a target trial. The second stage will identify and refine items to be considered for inclusion in the TARGET guideline by consulting content experts using sequential online surveys. The third stage will prioritise and consolidate key items to be included in the TARGET guideline at an in-person consensus meeting of TARGET investigators. The fourth stage will produce and pilot-test both the TARGET guideline and explanation and elaboration document with relevant stakeholders. The fifth stage will disseminate the TARGET guideline and resources via journals, conferences and courses.

ETHICS AND DISSEMINATION

Ethical approval for the survey has been attained (HC220536). The TARGET guideline will be disseminated widely in partnership with stakeholders to maximise adoption and improve reporting of these studies.

Reporting of Observational Studies Explicitly Aiming to Emulate Randomized Trials: A Systematic Review

Importance

Observational (nonexperimental) studies that aim to emulate a randomized trial (ie, the target trial) are increasingly informing medical and policy decision-making, but it is unclear how these studies are reported in the literature. Consistent reporting is essential for quality appraisal, evidence synthesis, and translation of evidence to policy and practice.

Objective

To assess the reporting of observational studies that explicitly aimed to emulate a target trial.

Evidence Review

We searched Medline, Embase, PsycINFO, and Web of Science for observational studies published between March 2012 and October 2022 that explicitly aimed to emulate a target trial of a health or medical intervention. Two reviewers double-screened and -extracted data on study characteristics, key predefined components of the target trial protocol and its emulation (eligibility criteria, treatment strategies, treatment assignment, outcome[s], follow-up, causal contrast[s], and analysis plan), and other items related to the target trial emulation.

Findings

A total of 200 studies that explicitly aimed to emulate a target trial were included. These studies included 26 subfields of medicine, and 168 (84%) were published from January 2020 to October 2022. The aim to emulate a target trial was explicit in 70 study titles (35%). Forty-three studies (22%) reported use of a published reporting guideline (eg, Strengthening the Reporting of Observational Studies in Epidemiology). Eighty-five studies (43%) did not describe all key items of how the target trial was emulated and 113 (57%) did not describe the protocol of the target trial and its emulation.

Conclusion and Relevance

In this systematic review of 200 studies that explicitly aimed to emulate a target trial, reporting of how the target trial was emulated was inconsistent. A reporting guideline for studies explicitly aiming to emulate a target trial may improve the reporting of the target trial protocols and other aspects of these emulation attempts.

Assumptions and analysis planning in studies with missing data in multiple variables: moving beyond the MCAR/MAR/MNAR classification

Researchers faced with incomplete data are encouraged to consider whether their data are 'missing completely at random' (MCAR), 'missing at random' (MAR) or 'missing not at random' (MNAR) when planning their analysis. However, there are two major problems with this classification as originally defined by Rubin in the 1970s. First, when there are missing data in multiple variables, the plausibility of the MAR assumption is difficult to assess using substantive knowledge and is more stringent than is generally appreciated. Second, although MCAR and MAR are sufficient conditions for consistent estimation with specific methods, they are not necessary conditions and therefore this categorization does not directly determine the best approach for handling the missing data in an analysis. How best to handle missing data depends on the assumed causal relationships between variables and their missingness, and what these relationships imply in terms of the 'recoverability' of the target estimand (the population parameter that encodes the answer to the underlying research question). Recoverability is defined as whether the estimand can be consistently estimated from the patterns and associations in the observed data without needing to invoke external information on the extent to which the distribution of missing values might differ from that of observed values. In this manuscript we outline an approach for deciding which method to use to handle multivariable missing data in an analysis, using directed acyclic graphs to depict missingness assumptions and determining the implications in terms of recoverability of the target estimand.

Addressing Child Mental Health Inequities Through Parental Mental Health and Preschool Attendance

BACKGROUND

Prevention is key to reducing socioeconomic inequities in children's mental health problems, especially given limited availability and accessibility of services. We investigated the potential to reduce inequities for disadvantaged children by improving parental mental health and preschool attendance in early childhood.

METHODS

Data from the nationally representative birth cohort, Longitudinal Study of Australian Children (N = 5107, commenced in 2004), were used to examine the impact of socioeconomic disadvantage (0-1 year) on children's mental health problems (10-11 years). Using an interventional effects approach, we estimated the extent to which inequities could be reduced by improving disadvantaged children's parental mental health (4-5 years) and their preschool attendance (4-5 years).

RESULTS

Disadvantaged children had a higher prevalence of elevated mental health symptoms (32.8%) compared with their nondisadvantaged peers (18.7%): confounder-adjusted difference in prevalence is 11.6% (95% confidence interval: 7.7% to 15.4%). Improving disadvantaged children's parental mental health and their preschool attendance to the level of their nondisadvantaged peers could reduce 6.5% and 0.3% of socioeconomic differences in children's mental health problems, respectively (equivalent to 0.8% and 0.04% absolute reductions). If these interventions were delivered in combination, a 10.8% (95% confidence interval: 6.9% to 14.7%) higher prevalence of elevated symptoms would remain for disadvantaged children.

CONCLUSIONS

Targeted policy interventions that improve parental mental health and preschool attendance for disadvantaged children are potential opportunities to reduce socioeconomic inequities in children's mental health problems. Such interventions should be considered within a broader, sustained, and multipronged approach that includes addressing socioeconomic disadvantage itself.

Parental personality and early life ecology: a prospective cohort study from preconception to postpartum

Personality reliably predicts life outcomes ranging from social and material resources to mental health and interpersonal capacities. However, little is known about the potential intergenerational impact of parent personality prior to offspring conception on family resources and child development across the first thousand days of life. We analysed data from the Victorian Intergenerational Health Cohort Study (665 parents, 1030 infants; est. 1992), a two-generation study with prospective assessment of preconception background factors in parental adolescence, preconception personality traits in young adulthood (agreeableness, conscientiousness, emotional stability, extraversion, and openness), and multiple parental resources and infant characteristics in pregnancy and after the birth of their child. After adjusting for pre-exposure confounders, both maternal and paternal preconception personality traits were associated with numerous parental resources and attributes in pregnancy and postpartum, as well as with infant biobehavioural characteristics. Effect sizes ranged from small to moderate when considering parent personality traits as continuous exposures, and from small to large when considering personality traits as binary exposures. Young adult personality, well before offspring conception, is associated with the perinatal household social and financial context, parental mental health, parenting style and self-efficacy, and temperamental characteristics of offspring. These are pivotal aspects of early life development that ultimately predict a child's long-term health and development.

Should multiple imputation be stratified by exposure group when estimating causal effects via outcome regression in observational studies?

BACKGROUND

Despite recent advances in causal inference methods, outcome regression remains the most widely used approach for estimating causal effects in epidemiological studies with a single-point exposure and outcome. Missing data are common in these studies, and complete-case analysis (CCA) and multiple imputation (MI) are two frequently used methods for handling them. In randomised controlled trials (RCTs), it has been shown that MI should be conducted separately by treatment group. In observational studies, causal inference is now understood as the task of emulating an RCT, which raises the question of whether MI should be conducted by exposure group in such studies.

METHODS

We addressed this question by evaluating the performance of seven methods for handling missing data when estimating causal effects with outcome regression. We conducted an extensive simulation study based on an illustrative case study from the Victorian Adolescent Health Cohort Study, assessing a range of scenarios, including seven outcome generation models with exposure-confounder interactions of differing strength.

RESULTS

The simulation results showed that MI by exposure group led to the least bias when the size of the smallest exposure group was relatively large, followed by MI approaches that included the exposure-confounder interactions.

CONCLUSIONS

The findings from our simulation study, which was designed based on a real case study, suggest that current practice for the conduct of MI in causal inference may need to shift to stratifying by exposure group where feasible, or otherwise including exposure-confounder interactions in the imputation model.

Handling of missing data with multiple imputation in observational studies that address causal questions: protocol for a scoping review

INTRODUCTION

Observational studies in health-related research often aim to answer causal questions. Missing data are common in these studies and often occur in multiple variables, such as the exposure, outcome and/or variables used to control for confounding. The standard classification of missing data as missing completely at random, missing at random (MAR) or missing not at random does not allow for a clear assessment of missingness assumptions when missingness arises in more than one variable. This presents challenges for selecting an analytic approach and determining when a sensitivity analysis under plausible alternative missing data assumptions is required. This is particularly pertinent with multiple imputation (MI), which is often justified by assuming data are MAR. The objective of this scoping review is to examine the use of MI in observational studies that address causal questions, with a focus on if and how (a) missingness assumptions are expressed and assessed, (b) missingness assumptions are used to justify the choice of a complete case analysis and/or MI for handling missing data and (c) sensitivity analyses under alternative plausible assumptions about the missingness mechanism are conducted.

METHODS AND ANALYSIS

We will review observational studies that aim to answer causal questions and use MI, published between January 2019 and December 2021 in five top general epidemiology journals. Studies will be identified using a full text search for the term 'multiple imputation' and then assessed for eligibility. Information extracted will include details about the study characteristics, missing data, missingness assumptions and MI implementation. Data will be summarised using descriptive statistics.

ETHICS AND DISSEMINATION

Ethics approval is not required for this review because data will be collected only from published studies. The results will be disseminated through a peer reviewed publication and conference presentations.

TRIAL REGISTRATION NUMBER

This protocol is registered on figshare (https://doi.org/10.6084/m9.figshare.20010497.v1).

Modelling timing and tempo of adrenarche in a prospective cohort study

To better understand how health risk processes are linked to adrenarche, measures of adrenarcheal timing and tempo are needed. Our objective was to describe and classify adrenal trajectories, in terms of timing and tempo, in a population of children transitioning to adolescence with repeated measurements of salivary dehydroepiandrosterone (DHEA), DHEA-sulphate, and testosterone. We analysed data from the Childhood to Adolescence Transition Study (CATS), a longitudinal study of 1239 participants, recruited at 8-9 years old and followed up annually. Saliva samples were assayed for adrenal hormones. Linear mixed-effect models with subject-specific random intercepts and slopes were used to model longitudinal hormone trajectories by sex and derive measures of adrenarcheal timing and tempo. The median values for all hormones were higher at each consecutive study wave for both sexes, and higher for females than males. For all hormones, between-individual variation in hormone levels at age 9 (timing) was moderately large and similar for females and males. Between-individual variation in hormone progression over time (tempo) was of moderate magnitude compared with the population average age-slope, which itself was small compared with overall hormone level at each age. This suggests that between-individual variation in tempo was less important for modelling hormone trajectories. Between-individual variation in timing was more important for determining relative adrenal hormonal level in childhood than tempo. This finding suggests that adrenal hormonal levels at age 8-9 years can be used to predict relative levels in early adolescence (up to 13 years).

Evaluation of approaches for accommodating interactions and non-linear terms in multiple imputation of incomplete three-level data

Three-level data structures arising from repeated measures on individuals clustered within larger units are common in health research studies. Missing data are prominent in such studies and are often handled via multiple imputation (MI). Although several MI approaches can be used to account for the three-level structure, including adaptations to single- and two-level approaches, when the substantive analysis model includes interactions or quadratic effects, these too need to be accommodated in the imputation model. In such analyses, substantive model compatible (SMC) MI has shown great promise in the context of single-level data. Although there have been recent developments in multilevel SMC MI, to date only one approach that explicitly handles incomplete three-level data is available. Alternatively, researchers can use pragmatic adaptations to single- and two-level MI approaches, or two-level SMC-MI approaches. We describe the available approaches and evaluate them via simulations in the context of three three-level random effects analysis models involving an interaction between the incomplete time-varying exposure and time, an interaction between the time-varying exposure and an incomplete time-fixed confounder, or a quadratic effect of the exposure. Results showed that all approaches considered performed well in terms of bias and precision when the target analysis involved an interaction with time, but the three-level SMC MI approach performed best when the target analysis involved an interaction between the time-varying exposure and an incomplete time-fixed confounder, or a quadratic effect of the exposure. We illustrate the methods using data from the Childhood to Adolescence Transition Study.

Learning outcomes in primary school children with emotional problems: a prospective cohort study

BACKGROUND

Academic difficulties are common in adolescents with mental health problems. Although earlier childhood emotional problems, characterised by heightened anxiety and depressive symptoms are common forerunners to adolescent mental health problems, the degree to which mental health problems in childhood may contribute independently to academic difficulties has been little explored.

METHODS

Data were drawn from a prospective cohort study of students in Melbourne, Australia (N = 1239). Data were linked with a standardised national assessment of academic performance at baseline (9 years) and wave three (11 years). Depressive and anxiety symptoms were assessed at baseline and wave two (10 years). Regression analyses estimated the association between emotional problems (9 and/or 10 years) and academic performance at 11 years, adjusting for baseline academic performance, sex, age and socioeconomic status, and hyperactivity/inattention symptoms.

RESULTS

Students with depressive symptoms at 9 years of age had lost nearly 4 months of numeracy learning two years later after controlling for baseline academic performance and confounders. Results were similar for anxiety symptoms. Regardless of when depressive symptoms occurred there were consistent associations with poorer numeracy performance at 11 years. The association of depressive symptoms with reading performance was weaker than for numeracy if they were present at wave two. Persistent anxiety symptoms across two waves led to nearly a 4 month loss of numeracy learning at 11 years, but the difference was not meaningful for reading. Findings were similar when including hyperactivity/inattention symptoms.

CONCLUSIONS

Childhood anxiety and depression are not only forerunners of later mental health problems but predict academic achievement. Partnerships between education and health systems have the potential to not only improve childhood emotional problems but also improve learning.

Better together: Advancing life course research through multi-cohort analytic approaches

Longitudinal cohorts can provide timely and cost-efficient evidence about the best points of health service and preventive interventions over the life course. Working systematically across cohorts has the potential to further exploit these valuable data assets, such as by improving the precision of estimates, enhancing (or appropriately reducing) confidence in the replicability of findings, and investigating interrelated questions within a broader theoretical model. In this conceptual review, we explore the opportunities and challenges presented by multi-cohort approaches in life course research. Specifically, we: 1) describe key motivations for multi-cohort work and the analytic approaches that are commonly used in each case; 2) flag some of the scientific and pragmatic challenges that arise when adopting these approaches; and 3) outline emerging directions for multi-cohort work in life course research. Harnessing their potential while thoughtfully considering limitations of multi-cohort approaches can contribute to the robust and granular evidence base needed to promote health and wellbeing over the life span.

Impact of early intervention on the population prevalence of common mental disorders: 20-year prospective study

BACKGROUND

The potential for early interventions to reduce the later prevalence of common mental disorders (CMD) first experienced in adolescence is unclear.

AIMS

To examine the course of CMD and evaluate the extent to which the prevalence of CMD could be reduced by preventing adolescent CMD, or by intervening to change four young adult processes, between the ages of 20 and 29 years, that could be mediating the link between adolescent and adult disorder.

METHOD

This was a prospective cohort study of 1923 Australian participants assessed repeatedly from adolescence (wave 1, mean age 14 years) to adulthood (wave 10, mean age 35 years). Causal mediation analysis was undertaken to evaluate the extent to which the prevalence of CMD at age 35 years in those with adolescent CMD could be reduced by either preventing adolescent CMD, or by intervening on four young adult mediating processes: the occurrence of young adult CMD, frequent cannabis use, parenting a child by age 24 years, and engagement in higher education and employment.

RESULTS

At age 35, 19.2% of participants reported CMD; a quarter of these participants experienced CMD during both adolescence and young adulthood. In total, 49% of those with CMD during both adolescence and young adulthood went on to report CMD at age 35 years. Preventing adolescent CMD reduced the population prevalence at age 35 years by 3.9%. Intervening on all four young adult processes among those with adolescent CMD, reduced this prevalence by 1.6%.

CONCLUSIONS

In this Australian cohort, a large proportion of adolescent CMD resolved by adulthood, and by age 35 years, the largest proportion of CMD emerged among individuals without prior CMD. Time-limited, early intervention in those with earlier adolescent disorder is unlikely to substantially reduce the prevalence of CMD in midlife.

Quantifying cause-related mortality in Australia, incorporating multiple causes: observed patterns, trends and practical considerations

BACKGROUND

Mortality statistics using a single underlying cause of death (UC) are key health indicators. Rising multimorbidity and chronic disease mean that deaths increasingly involve multiple conditions. However, additional causes reported on death certificates are rarely integrated into mortality indicators, partly due to complexities in data and methods. This study aimed to assess trends and patterns in cause-related mortality in Australia, integrating multiple causes (MC) of death.

METHODS

Deaths (n = 1 773 399) in Australia (2006-17) were mapped to 136 ICD-10-based groups and MC indicators applied. Age-standardized cause-related rates (deaths/100 000) based on the UC (ASRUC) were compared with rates based on any mention of the cause (ASRAM) using rate ratios (RR = ASRAM/ASRUC) and to rates based on weighting multiple contributing causes (ASRW).

RESULTS

Deaths involved on average 3.4 causes in 2017; the percentage with >4 causes increased from 20.9 (2006) to 24.4 (2017). Ischaemic heart disease (ASRUC = 73.3, ASRAM = 135.8, ASRW = 63.5), dementia (ASRUC = 51.1, ASRAM = 98.1, ASRW = 52.1) and cerebrovascular diseases (ASRUC = 39.9, ASRAM = 76.7, ASRW = 33.5) ranked as leading causes by all methods. Causes with high RR included hypertension (ASRUC = 2.2, RR = 35.5), atrial fibrillation (ASRUC = 8.0, RR = 6.5) and diabetes (ASRUC = 18.5, RR = 3.5); the corresponding ASRW were 12.5, 12.6 and 24.0, respectively. Renal failure, atrial fibrillation and hypertension ranked among the 10 leading causes by ASRAM and ASRW but not by ASRUC. Practical considerations in working with MC data are discussed.

CONCLUSIONS

Despite the similarities in leading causes under the three methods, with integration of MC several preventable diseases emerged as leading causes. MC analyses offer a richer additional perspective for population health monitoring and policy development.

Multiple imputation approaches for handling incomplete three-level data with time-varying cluster-memberships

Three-level data arising from repeated measures on individuals clustered within higher-level units are common in medical research. A complexity arises when individuals change clusters over time, resulting in a cross-classified data structure. Missing values in these studies are commonly handled via multiple imputation (MI). If the three-level, cross-classified structure is modeled in the analysis, it also needs to be accommodated in the imputation model to ensure valid results. While incomplete three-level data can be handled using various approaches within MI, the performance of these in the cross-classified data setting remains unclear. We conducted simulations under a range of scenarios to compare these approaches in the context of an acute-effects cross-classified random effects substantive model, which models the time-varying cluster membership via simple additive random effects. The simulation study was based on a case study in a longitudinal cohort of students clustered within schools. We evaluated methods that ignore the time-varying cluster memberships by taking the first or most common cluster for each individual; pragmatic extensions of single- and two-level MI approaches within the joint modeling (JM) and the fully conditional specification (FCS) frameworks, using dummy indicators (DI) and/or imputing repeated measures in wide format to account for the cross-classified structure; and a three-level FCS MI approach developed specifically for cross-classified data. Results indicated that the FCS implementations performed well in terms of bias and precision while JM approaches performed poorly. Under both frameworks approaches using the DI extension should be used with caution in the presence of sparse data.

Association Between Earlier Introduction of Peanut and Prevalence of Peanut Allergy in Infants in Australia

IMPORTANCE

Randomized clinical trials showed that earlier peanut introduction can prevent peanut allergy in select high-risk populations. This led to changes in infant feeding guidelines in 2016 to recommend early peanut introduction for all infants to reduce the risk of peanut allergy.

OBJECTIVE

To measure the change in population prevalence of peanut allergy in infants after the introduction of these new guidelines and evaluate the association between early peanut introduction and peanut allergy.

DESIGN

Two population-based cross-sectional samples of infants aged 12 months were recruited 10 years apart using the same sampling frame and methods to allow comparison of changes over time. Infants were recruited from immunization centers around Melbourne, Australia. Infants attending their 12-month immunization visit were eligible to participate (eligible age range, 11-15 months), regardless of history of peanut exposure or allergy history.

EXPOSURES

Questionnaires collected data on demographics, food allergy risk factors, peanut introduction, and reactions.

MAIN OUTCOME AND MEASURES

All infants underwent skin prick tests to peanut and those with positive results underwent oral food challenges. Prevalence estimates were standardized to account for changes in population demographics over time.

RESULTS

This study included 7209 infants (1933 in 2018-2019 and 5276 in 2007-2011). Of the participants in the older vs more recent cohort, 51.8% vs 50.8% were male; median (IQR) ages were 12.5 (12.2-13.0) months vs 12.4 (12.2-12.9) months. There was an increase in infants of East Asian ancestry over time (16.5% in 2018-2019 vs 10.5% in 2007-2011), which is a food allergy risk factor. After standardizing for infant ancestry and other demographics changes, peanut allergy prevalence was 2.6% (95% CI, 1.8%-3.4%) in 2018-2019, compared with 3.1% in 2007-2011 (difference, -0.5% [95% CI, -1.4% to 0.4%]; P = .26). Earlier age of peanut introduction was significantly associated with a lower risk of peanut allergy among infants of Australian ancestry in 2018-2019 (age 12 months compared with age 6 months or younger: adjusted odds ratio, 0.08 [05% CI, 0.02-0.36]; age 12 months compared with 7 to less than 10 months: adjusted odds ratio, 0.09 [95% CI, 0.02-0.53]), but not significant among infants of East Asian ancestry (P for interaction = .002).

CONCLUSIONS AND RELEVANCE

In cross-sectional analyses, introduction of a guideline recommending early peanut introduction in Australia was not associated with a statistically significant lower or higher prevalence of peanut allergy across the population.

Emulating a target trial of intensive nurse home visiting in the policy-relevant population using linked administrative data

BACKGROUND

Populations willing to participate in randomized trials may not correspond well to policy-relevant target populations. Evidence of effectiveness that is complementary to randomized trials may be obtained by combining the 'target trial' causal inference framework with whole-of-population linked administrative data.

METHODS

We demonstrate this approach in an evaluation of the South Australian Family Home Visiting Program, a nurse home visiting programme targeting socially disadvantaged families. Using de-identified data from 2004-10 in the ethics-approved Better Evidence Better Outcomes Linked Data (BEBOLD) platform, we characterized the policy-relevant population and emulated a trial evaluating effects on child developmental vulnerability at 5 years (n = 4160) and academic achievement at 9 years (n = 6370). Linkage to seven health, welfare and education data sources allowed adjustment for 29 confounders using Targeted Maximum Likelihood Estimation (TMLE) with SuperLearner. Sensitivity analyses assessed robustness to analytical choices.

RESULTS

We demonstrated how the target trial framework may be used with linked administrative data to generate evidence for an intervention as it is delivered in practice in the community in the policy-relevant target population, and considering effects on outcomes years down the track. The target trial lens also aided in understanding and limiting the increased measurement, confounding and selection bias risks arising with such data. Substantively, we did not find robust evidence of a meaningful beneficial intervention effect.

CONCLUSIONS

This approach could be a valuable avenue for generating high-quality, policy-relevant evidence that is complementary to trials, particularly when the target populations are multiply disadvantaged and less likely to participate in trials.

Evaluation of multiple imputation approaches for handling missing covariate information in a case-cohort study with a binary outcome

Background

In case-cohort studies a random subcohort is selected from the inception cohort and acts as the sample of controls for several outcome investigations. Analysis is conducted using only the cases and the subcohort, with inverse probability weighting (IPW) used to account for the unequal sampling probabilities resulting from the study design. Like all epidemiological studies, case-cohort studies are susceptible to missing data. Multiple imputation (MI) has become increasingly popular for addressing missing data in epidemiological studies. It is currently unclear how best to incorporate the weights from a case-cohort analysis in MI procedures used to address missing covariate data.

Method

A simulation study was conducted with missingness in two covariates, motivated by a case study within the Barwon Infant Study. MI methods considered were: using the outcome, a proxy for weights in the simple case-cohort design considered, as a predictor in the imputation model, with and without exposure and covariate interactions; imputing separately within each weight category; and using a weighted imputation model. These methods were compared to a complete case analysis (CCA) within the context of a standard IPW analysis model estimating either the risk or odds ratio. The strength of associations, missing data mechanism, proportion of observations with incomplete covariate data, and subcohort selection probability varied across the simulation scenarios. Methods were also applied to the case study.

Results

There was similar performance in terms of relative bias and precision with all MI methods across the scenarios considered, with expected improvements compared with the CCA. Slight underestimation of the standard error was seen throughout but the nominal level of coverage (95%) was generally achieved. All MI methods showed a similar increase in precision as the subcohort selection probability increased, irrespective of the scenario. A similar pattern of results was seen in the case study.

Conclusions

How weights were incorporated into the imputation model had minimal effect on the performance of MI; this may be due to case-cohort studies only having two weight categories. In this context, inclusion of the outcome in the imputation model was sufficient to account for the unequal sampling probabilities in the analysis model.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12874-021-01495-4.

Longitudinal prediction of periconception alcohol use: a 20-year prospective cohort study across adolescence, young adulthood and pregnancy

BACKGROUND AND AIMS

Alcohol consumption is common in adolescence and young adulthood and may continue into pregnancy, posing serious risk to early fetal development. We examine the frequency of periconception alcohol use (prior to pregnancy awareness) and the extent to which adolescent and young adult alcohol use prospectively predict periconception use.

DESIGN

A longitudinal, population-based study.

SETTING

Victoria, Australia.

PARTICIPANTS

A total of 289 women in trimester three of pregnancy (age 29-35 years; 388 pregnancies).

MEASURES

The main exposures were binge [≥ 4.0 standard drinks (SDs)/day] and frequent (≥ 3 days/week) drinking in adolescence (mean age = 14.9-17.4 years) and young adulthood (mean age 20.7-29.1 years). Outcomes were frequency (≥ 3 days/week, ≥ monthly, never) and quantity (≥ 4.0 SDs, ≥ 0.5 and < 4.0 SDs, none) of periconception drinking.

FINDINGS

Alcohol use was common in young adulthood prior to pregnancy (72%) and in the early weeks of pregnancy (76%). The proportions drinking on most days and binge drinking were similar at both points. Reflecting a high degree of continuity in alcohol use behaviours, most women who drank periconceptionally had an earlier history of frequent (77%) and/or binge (85%) drinking throughout the adolescent or young adult years. Young adult binge drinking prospectively predicted periconception drinking quantity [odds ratio (OR) = 3.7, 95% confidence interval (CI) = 1.9-7.4], compared with women with no prior history. Similarly, frequent young adult drinking prospectively predicted frequent periconception drinking (OR = 30.7, 95% CI = 12.3-76.7).

CONCLUSIONS

Women who engage in risky (i.e. frequent and binge) drinking in their adolescent and young adult years are more likely to report risky drinking in early pregnancy prior to pregnancy recognition than women with no prior history of risky drinking.

Competing risks analysis with missing cause-of-failure-penalized likelihood estimation of cause-specific Cox models

Competing risks models are attractive tools to analyze time-to-event data where several causes of an event are competing. However, a complexity may arise when, for instance, some subjects experience the event of interest but the causes are not known. Assuming that unknown causes of events are missing at random, we developed a novel constrained maximum penalized likelihood method for fitting semi-parametric cause-specific Cox regression models. Here, penalty functions were used to smooth the baseline hazards. An appealing feature of this approach is that all the relevant estimands in competing risks models are estimated including cause-specific hazard ratios, cause-specific baseline hazards, and cumulative incidence functions. Asymptotic results for these estimators were also developed, allowing for direct inferences. The proposed method was compared with some existing methods through a simulation study. A real data example was analyzed using the new method to evaluate the association of age at diagnosis with melanoma-death and non-melanoma-death in patients diagnosed with thin melanoma (tumour thickness ≤1.0  mm). An R function for our proposed method is currently available on GitHub and will be included in the R package "survivalMPL" at CRAN.

Multiple imputation of semi-continuous exposure variables that are categorized for analysis

Semi-continuous variables are characterized by a point mass at one value and a continuous range of values for remaining observations. An example is alcohol consumption quantity, with a spike of zeros representing non-drinkers and positive values for drinkers. If multiple imputation is used to handle missing values for semi-continuous variables, it is unclear how this should be implemented within the standard approaches of fully conditional specification (FCS) and multivariate normal imputation (MVNI). This question is brought into focus by the use of categorized versions of semi-continuous exposure variables in analyses (eg, no drinking, drinking below binge level, binge drinking, heavy binge drinking), raising the question of how best to achieve congeniality between imputation and analysis models. We performed a simulation study comparing nine approaches for imputing semi-continuous exposures requiring categorization for analysis. Three methods imputed the categories directly: ordinal logistic regression, and imputation of binary indicator variables representing the categories using MVNI (with two variants). Six methods (predictive mean matching, zero-inflated binomial imputation, and two-part imputation methods with variants in FCS and MVNI) imputed the semi-continuous variable, with categories derived after imputation. The ordinal and zero-inflated binomial methods had good performance across most scenarios, while MVNI methods requiring rounding after imputation did not perform well. There were mixed results for predictive mean matching and the two-part methods, depending on whether the estimands were proportions or regression coefficients. The results highlight the need to consider the parameter of interest when selecting an imputation procedure.

1173The impact of adverse and positive experiences on inflammatory outcomes in Australian and UK children

Background

Inflammation is one of key mechanisms linking childhood experiences to later chronic disease risk. Childhood adversity is associated with inflammation, but little is known about positive experiences. We examine how adverse and positive experiences are associated with inflammatory markers in late childhood, and whether they have an interaction effect.

Methods

Data sources: Longitudinal Study of Australian Children (LSAC; N = 1237) and Avon Longitudinal Study of Parents and Children (ALSPAC; N = 3488). Exposures: Adverse and positive experiences assessed from 0 to 11 (LSAC) and 0-14 years (ALSPAC). Adversity indicators included parent legal problems, family violence, mental illness, substance abuse, harsh parenting, parental divorce, neighbourhood violence, family member death, and bullying victimization. Positive experiences included positive parenting practice, trusting and supportive relationships, supportive neighbourhood and home learning environments, social engagement and enjoyment. Outcomes: Inflammation quantified by high-sensitivity C-reactive protein (hsCRP) and glycoprotein acetyls (GlycA). Analyses: Linear regression was used to estimate relative change in inflammatory markers, adjusted for sociodemographics. Outcomes were log-transformed.

Results

Exposure to adversity was associated with higher levels of inflammation (e.g., CRP: β = 8.8%, 95% CI= -16.5% to 34.2% in LSAC), whereas exposure to positive experiences was associated with lower levels (e.g., CRP: β=-18.9%, 95% CI= -45.8% to 7.9% in LSAC), after adjusting for sociodemographics. There was no interaction effect of adverse and positive experiences on inflammation.

Conclusions

Adverse and positive experiences have independent and small effects on children’s inflammation across two cohorts.

Key messages

Positive experiences are critical to inform interventions to improve inflammatory outcomes for children who face adversity.

923Quantifying multiple causes of death: Observed patterns in Australia, 2006–2017

Background

Four fifths of deaths in Australia involve multiple causes, but statistics typically use a single underlying cause of death (UC). The UC approach alone is insufficient for understanding the impact of non-underlying causes and identifying comorbid disease associations at death. Analysis of multiple causes of death (MC) is needed to measure the impact of all causes. We described MC patterns, considering cause-of-death coding and certification practices in Australia.

Methods

Using deaths registered in Australia from 2006 to 2017 (n = 1773525) coded to the International Classification of Diseases (ICD) and an extended classification (n = 136 causes) based on a World Health Organization short list, we described MCoD data by cause. Age-standardised rates based on UC and MC were compared using the standardised ratio of multiple to underlying causes (SRMU) to estimate the contribution of the cause to mortality compared to using the UC approach. Comorbidity was explored using the cause of death association indicator (CDAI) to compare the observed joint frequency of a contributory-underlying cause combined with expected frequency of the contributory cause (with any UC).

Results

On average 3.4 conditions caused each death and 24.4% of deaths had 5 plus causes. Largest SRMUs were for genitourinary diseases (8.0), blood diseases (7.8) and musculoskeletal conditions (6.7). CDAIs showed high associations between, for example, accidental alcohol and opioid poisoning, septicaemia and skin infections, and traumatic brain injury and falls.

Conclusions

MC indicators enhance measures of mortality and reassess the role of causes of death for descriptive and analytical epidemiology.

Key messages

This research demonstrates the value of MC analysis for Australian mortality data.

314Data-adaptive methods for high-dimensional mediation analysis: Application to a randomised trial of tuberculosis vaccination

Focus of Presentation

Statistical methods for causal mediation analysis are useful for understanding the pathways by which a certain treatment or exposure impacts health outcomes. Existing methods necessitate modelling of the distribution of the mediators, which quickly becomes infeasible when mediators are high-dimensional (e.g., biomarkers). We propose novel data-adaptive methods for estimating the indirect effect of a randomised treatment that acts via a pathway represented by a high-dimensional set of measurements. This work was motivated by the Melbourne Infant Study: BCG for Allergy and Infection Reduction (MIS BAIR), a randomised controlled trial investigating the effect of neonatal tuberculosis vaccination on clinical allergy and infection outcomes, and its mechanisms of action.

Findings

The proposed methods are doubly robust, which allows us to achieve (uniformly) valid statistical inference, even when machine learning algorithms are used for the two required models. We illustrate these in the context of the MIS BAIR study, investigating the mediating role of immune pathways represented by a high-dimensional vector of cytokine responses under various stimulants. We confirm adequate performance of the proposed methods in an extensive simulation study.

Conclusions/Implications

The proposed methods provide a feasible and flexible analytic strategy for examining high-dimensional mediators in randomised controlled trials.

Key messages

Data-adaptive methods for mediation analysis are desirable in the context of high-dimensional mediators, such as biomarkers. We propose novel doubly robust methods, which enable valid statistical inference when using machine learning algorithms for estimation.

533The psychosocial profiles of children aged 11-12 years who have self-harmed: A prospective cohort study

Background

Self-harm in very young people can be a clinically ominous event. While most studies to date have focused on self-harm during the teenage years, fewer studies have examined children aged 12 years or under. We aimed to estimate the incidence and correlates of recent self-harm in a population-based, non-treatment-seeking sample of primary school-aged children.

Methods

1059 children from 43 primary schools in Melbourne, Australia were interviewed at the age of 8 years (Wave 1), and followed up annually until the age of 11 years (Waves 2-4). Interviews covered a range of physical and mental health, social, educational and family domains, including (at Wave 4) self-harm during the previous 12 months.

Results

At Wave 4 (mean age: 11.9 years), a total of 28 children (3%; 18 girls [3%], 10 boys [2%]) reported self-harming during the previous 12 months. When compared with children who reported no self-harm, they were more likely to report depression, anxiety, poor emotional control, frequent bullying victimisation (including online bullying), truancy, recent alcohol consumption, and antisocial behaviour during Waves 1-3. They were also more likely to report having few friends.

Conclusions

Self-harm was reported by a proportion of community-dwelling children aged 11-12 years. As these children were more likely to report a range of other adverse behaviours, experiences and health conditions, clinicians should consider the possibility of prior self-harm when assessing children presenting with such behaviours and issues.

Key messages

The focus of intervention efforts aimed at preventing and reducing adolescent self-harm should extend to primary school-aged children, with a focus on mental health and peer relationships during the pubertal transition.

1023The uses and abuses of regression models: a new approach to teaching regression analysis

Focus of Presentation

Multivariable regression models are widely used in epidemiological data analysis. Traditional teaching often focusses on technical aspects with insufficient attention paid to the purposes for which regression methods are used.

Findings

We have addressed these issues in a new short course that provides an introduction to regression analysis in the context of learning about causal inference, beginning from the standpoint that the majority of research questions in epidemiology are causal in nature. This approach leads naturally to using regression models in two different ways, firstly for direct estimation of a causal effect, under an assumption of constancy of the effect across strata of confounders, and secondly for prediction of outcomes, as a necessary step in the estimation of causal effects via g-computation.

Conclusions/Implications

Approaching the teaching of regression methods within a causal inference framework helps to dispel confusion created by traditional statistical approaches that imply the existence of “true models” and encourage the building of models in a way that is unclear about the purpose for which they will be used, for example seeking to identify “risk factors” in an exploratory manner.

Key messages

The teaching and practice of regression methods in epidemiology can be enhanced by emphasising the key differences between three distinct analytic purposes: description, prediction, causal. Regression models may play a role in all three but the way in which models are developed and interpreted differs between them.

892Quantification of mortality incorporating multiple causes of death: Application of weighting strategies to Australian data

Background

Mortality statistics are typically based on a single underlying cause of death (UCoD). Although UCoD provides a useful construct, the relevance of assuming that a single disease caused the death is diminishing, especially with increased life expectancy and high proportions of deaths in older ages from chronic/degenerative diseases. Focussing on common underlying causes of death in Australia, we quantified mortality incorporating weighting strategies for multiple causes of death (MCoD).

Methods

All deaths registered in Australia from 2015-2017 (478,396 deaths) and coded using International Classification of Diseases Version 10 were classified using an extended cause list (n = 136 causes) based on a World Health Organization short list. Age-standardised rates (ASR) were estimated using three weighting methods: (1) traditional approach using UCoD alone; (2) UCoD and associated causes of death (ACoDs) equally weighted and (3) UCoD weighted 0.5 arbitrarily and remaining 0.5 apportioned to the remaining ACoDs.

Results

Common UCoD were ischaemic heart diseases, cerebrovascular diseases, dementia; 57671, 31515 and 27377 deaths respectively. There were substantial changes in ASR depending on the weighting method used. Variation in mortality patterns estimated using the three weighting methods and challenges to further refinement of the weighting strategy will be discussed.

Conclusions

Mortality indicators incorporating MCoD enhance traditional measures of mortality and provide a means to reassess the role of diseases in causing death. Further disease specific methods are required to refine current weighting strategies.

Key messages

Weighting strategies for are useful for quantifying mortality incorporating MCoD, but methodological challenges exist.

1282Causal inference in multi-cohort studies using the target trial approach

Focus of Presentation

Utilising data from multiple cohorts to address causal questions in health research has become increasingly widespread due to a number of advantages. These include improved precision of estimates, in particular to investigate effect heterogeneity as well as rare events and exposures, and the ability to examine the replicability of findings. However, undertaking causal inference in multi-cohort studies also faces several challenges, which makes clear causal thinking even more important than in single-cohort studies. We propose the use of the “target trial” framework for the conduct of causal inference in multi-cohort studies.

Findings

Using two case studies, the first considering the effect of maternal mental health on emotional reactivity and the second examining the influence of exposure to adversity on inflammatory outcomes in childhood, we describe and demonstrate how the target trial approach enables clear definition of the target estimand and systematic consideration of sources of bias. Considering the target trial as the reference point allows the identification of potential biases within each study, so that analysis can be planned to reduce them. Furthermore, the interpretation of findings is assisted by an understanding of the unavoidable biases that may be compounded when pooling data from multiple cohorts, or that may explain discrepant findings across cohorts.

Conclusions/Implications

Use of the target trial framework in multi-cohort studies helps strengthen causal inferences through improved analysis design and clarity in the interpretation of findings.

Key messages

The target trial framework, already well-established for casual inference in single-cohort studies, is recommended for the conduct of causal inference in multi-cohort studies.

1223Handling missing data for causal effect estimation in cohort studies using Targeted Maximum Likelihood Estimation

Background

Causal inference from cohort studies is central to epidemiological research. Targeted Maximum Likelihood Estimation (TMLE) is an appealing doubly robust method for causal effect estimation, but it is unclear how missing data should be handled when it is used in conjunction with machine learning approaches for the exposure and outcome models. This is problematic because missing data are ubiquitous and can result in biased estimates and loss of precision if handled inappropriately.

Methods

Based on a motivating example from the Victorian Adolescent Health Cohort Study, we conducted a simulation study to evaluate the performance of available approaches for handling missing data when using TMLE with machine learning. These included complete-case analysis; an extended TMLE approach incorporating an outcome missingness probability model; the missing indicator approach for missing covariate data (MCMI); and multiple imputation (MI) using standard parametric approaches or machine learning algorithms. We considered 11 missingness mechanisms typical in cohort studies, and a simple and a complex setting, in which exposure and outcome generation models included two-way and higher-order interactions.

Results

MI using regression with no interactions and MI with random forest yielded estimates with the highest bias. MI with regression including two-way interactions was the best performing method overall. Of the non-MI approaches, MCMI performed the worst

Conclusions

When using TMLE with machine learning to estimate the average causal effect, avoiding standard MI with no interactions and MCMI is recommended.

Key messages

We provide novel guidance for handling missing data for causal effect estimation using TMLE.

1230Should multiple imputation be stratified by exposure group when estimating causal effects via outcome regression?

Background

Outcome regression remains widely applied for estimating causal effects in observational studies, in which causal inference is conceptualised as emulating a randomized controlled trial (RCT). Multiple imputation (MI) is a commonly used method for handling missing data, but while in RCTs it has been shown that MI should be conducted by treatment group to reduce bias, whether imputation should be conducted by exposure group in observational studies has not been studied.

Methods

We conducted a simulation study to evaluate the performance of seven methods for handling missing data: Complete-case analysis (CCA), MI of main effect, MI with interactions (between exposure and: outcome, a strong confounder, outcome and a strong confounder, all incomplete), and MI conducted by exposure group. We simulated data based on an example from the Victorian Adolescent Health Cohort Study. Three exposure prevalences and seven outcome generation models were considered, the latter ranging from no interaction to strong-positive or negative exposure-confounder interaction. Various missingness scenarios were examined: with incomplete outcome only or also incomplete confounders, and three levels of complexity regarding the missingness mechanism.

Results

For all scenarios, MI by exposure led to the least bias, followed by MI approaches that included exposure-confounder interactions.

Conclusions

If MI is adopted in outcome regression, we recommend conducting MI by exposure group and, when not feasible, including exposure-confounder interactions in the imputation model.

Key messages

Similar to RCTs, MI should be conducted by exposure group when estimating average causal effects using outcome regression in observational studies.

1000Quantifying multiple causes of death: A systematic review and audit of methods and practice

Background

Mortality reporting and research are typically focused on a single underlying cause of death (UCoD) selected from multiple reported causes. The need to incorporate multiple causes of death (MCoD) in mortality statistics is now recognised internationally, but there is scant methodological work to guide analytical approaches. This review aims to identify and appraise current methods and practices used to analyse MCoD data.

Methods

The Web of Science, Medline, Pubmed and Scopus (from inception to December 2019) were queried. Studies reporting MCoD alone or in comparison with single UCoD were included. The review is supplemented by qualitative interview with international experts.

Results

3491 studies were identified; 141 full texts were included in the review. The measures usually estimated when analysing MCoD can be broadly categorised into descriptive measures (n = 93 studies), measures of associations between diseases (n = 46 studies) and advanced statistical methods (n = 11 studies). Descriptive statistics commonly used include standardized ratio of multiple to underlying cause (SRMU) and mortality rates based on any mention of a disease. Approaches used to assess measures of associations between diseases include the Cause-of-Death Association Indicator (CDAI) and social network analysis. The advanced statistical methods include weighting MCoD and lethal defect-wear model of mortality. Audit results will be discussed.

Conclusions

This review provides a comprehensive and updated summary of methodological approaches used to analyse MCoD data. The merit of each analytical framework is discussed.

Key messages

More work is needed to develop methodological frameworks that could be used to support routine consideration of MCoD in practice.

896Quantifying cause-related mortality incorporating multiple causes: challenges and opportunities

Key contact person

Dr Grace Joshy, Fellow, Research School of Population Health, Australian National University.

Focus and outcomes for participants

Mortality statistics are typically based on a single underlying cause of death (UCoD), derived from multiple conditions on the death certificate, and have provided critical evidence for policy and practice for over a century. There have been radical shifts in patterns of death in the past couple of decades; deaths in older ages are increasingly from chronic and degenerative diseases. The relevance of assuming that a single disease is causing the death is diminishing, especially with an aging population structure and increasing life expectancy. This symposium will enable participants to understand the complexities associated with mortality reporting/coding, strengths and limitations of available statistical methods for using multiple causes of death (MCoD) and the importance of quantifying mortality incorporating MCoD.

Rationale for the symposium, including for its inclusion in the Congress

The use of a single UCoD rather than MCoD means that vast amounts of potentially useful data are largely ignored, which is likely to bias mortality estimates (including under- and over-reporting of the importance of certain causes of death). Despite global recognition of the urgent need to better integrate data on MCoD into mortality statistics, use of these data are challenging and limited. Complexities arise from the way mortality information is reported on death certificates and coded to form mortality collections; limited understanding of available statistical methods also adds to the complexity.

International Classification of Diseases 10th Revision (ICD-10) has been translated into 43 languages and it is being used by over 100 countries to report mortality data, a primary indicator of health status. The 2018 release of the 11th revision of the International Classification of Diseases, enriching data on multiple parameters including comorbidity, confers further urgency and a unique opportunity to optimise the use of MCoD in mortality reporting.

The World Congress of Epidemiology 2020 will provide a unique platform for wider discussions on the challenges and opportunities for using MCoD data. The symposium will provide a deeper understanding and enhanced the use of MCoD data. The speakers are engaged in cutting-edge NHMRC-funded research on mortality incorporating MCoD and development of novel statistical methods.

Presentation program

The symposium will feature presentations from six speakers.

Names of presenters

James Eynstone-Hinkins, Lauren Moran, Saliu Balogun, Karen Bishop, Margarita Moreno-Betancur, Grace Joshy

855Multiple Imputation in the Context of Case-Cohort Studies: Simulation and Case Study

Background

Multiple imputation (MI) is commonly used to address missing data in epidemiological studies, but valid use requires compatibility between the imputation and analysis models. Case-cohort studies use unequal sampling probabilities for cases and controls which are often accounted for during analyses through inverse probability weighting (IPW). It is unclear how to apply MI for missing covariates while achieving compatibility in this setting.

Methods

A simulation study was conducted with missingness in two covariates, motivated by a case-cohort investigation within the Barwon Infant Study. MI methods considered involved including interactions between the outcome (as a proxy for weights) and analysis variables, stratification by weights, and ignoring weights, within the context of an IPW analysis. Factors such as the target estimand, proportion of incomplete observations, missing data mechanism and subcohort selection probabilities were varied to assess performance of MI methods.

Results

There was similar performance in terms of bias and efficiency across the MI methods, with expected improvements compared to IPW applied to the complete cases. Precision tended to decrease as the subcohort selection probability decreased. Similar results were observed irrespective of the proportion of incomplete cases.

Conclusions

Our results suggest that it makes little difference how weights are incorporated in the MI model in the analysis of case-cohort studies, potentially due to only two weight classes in this setting.

Key messages

If and how the weights are incorporated in the imputation model may have little impact in the analysis of case-cohort studies with incomplete covariates

Cannabis and tobacco use prior to pregnancy and subsequent offspring birth outcomes: a 20-year intergenerational prospective cohort study

There is increasing evidence that the life-course origins of health and development begin before conception. We examined associations between timing and frequency of preconception cannabis and tobacco use and next generation preterm birth (PTB), low birth weight (LBW) and small for gestational age. 665 participants in a general population cohort were repeatedly assessed on tobacco and cannabis use between ages 14-29 years, before pregnancy. Associations were estimated using logistic regression. Preconception parent (either maternal or paternal) daily cannabis use age 15-17 was associated with sixfold increases in the odds of offspring PTB (aOR 6.65, 95% CI 1.92, 23.09), and offspring LBW (aOR 5.84, 95% CI 1.70-20.08), after adjusting for baseline sociodemographic factors, parent sex, offspring sex, family socioeconomic status, parent mental health at baseline, and concurrent tobacco use. There was little evidence of associations with preconception parental cannabis use at other ages or preconception parental tobacco use. Findings support the hypothesis that the early life origins of growth begin before conception and provide a compelling rationale for prevention of frequent use during adolescence. This is pertinent given liberalisation of cannabis policy.

Multiple imputation for handling missing outcome data in randomized trials involving a mixture of independent and paired data

Randomized trials involving independent and paired observations occur in many areas of health research, for example in paediatrics, where studies can include infants from both single and twin births. Multiple imputation (MI) is often used to address missing outcome data in randomized trials, yet its performance in trials with independent and paired observations, where design effects can be less than or greater than one, remains to be explored. Using simulated data and through application to a trial dataset, we investigated the performance of different methods of MI for a continuous or binary outcome when followed by analysis using generalized estimating equations to account for clustering due to the pairs. We found that imputing data separately for independent and paired data, with paired data imputed in wide format, was the best performing MI method, producing unbiased point and standard error estimates for the treatment effect throughout. Ignoring clustering in the imputation model performed well in settings where the design effect due to the inclusion of paired data was close to one, but otherwise led to moderately biased variance estimates. Including a random cluster effect in the imputation model led to slightly biased point estimates for binary outcome data and variance estimates that were too small in some settings. Based on these results, we recommend researchers impute independent and paired data separately where feasible to do so. The exception is if the design effect due to the inclusion of paired data is close to one, where ignoring clustering may be appropriate.

Preventing postnatal depression: a causal mediation analysis of a 20-year preconception cohort

Postnatal depression (PND) is common and predicts a range of adverse maternal and offspring outcomes. PND rates are highest among women with persistent mental health problems before pregnancy, and antenatal healthcare provides ideal opportunity to intervene. We examined antenatal perceived social support as a potential intervention target in preventing PND symptoms among women with prior mental health problems. A total of 398 Australian women (600 pregnancies) were assessed repeatedly for mental health problems before pregnancy (ages 14-29 years, 1992-2006), and again during pregnancy, two months postpartum and one year postpartum (2006-2014). Causal mediation analysis found that intervention on perceived antenatal social support has the potential to reduce rates of PND symptoms by up to 3% (from 15 to 12%) in women with persistent preconception symptoms. Supplementary analyses found that the role of low antenatal social support was independent of concurrent antenatal depressive symptoms. Combined, these two factors mediated up to more than half of the association between preconception mental health problems and PND symptoms. Trialling dual interventions on antenatal depressive symptoms and perceived social support represents one promising strategy to prevent PND in women with persistent preconception symptoms. Interventions promoting mental health before pregnancy may yield an even greater reduction in PND symptoms by disrupting a developmental cascade of risks via these and other pathways. This article is part of the theme issue 'Multidisciplinary perspectives on social support and maternal-child health'.

Maternal prenatal gut microbiota composition predicts child behaviour

BACKGROUND

Murine studies demonstrate that maternal prenatal gut microbiota influences brain development and behaviour of offspring. No human study has related maternal gut microbiota to behavioural outcomes during early life. This study aimed to evaluate relationships between the prenatal faecal microbiota, prenatal diet and childhood behaviour.

METHODS

A sub-cohort of 213 mothers and 215 children were selected from a longitudinal pre-birth cohort. Maternal prenatal exposure measures collected during the third trimester included the faecal microbiota (generated using 16S rRNA amplicon sequencing), and dietary intake. The behavioural outcome used the Childhood Behaviour Checklist at age two. Models were adjusted for prenatal diet, smoking, perceived stress, maternal age and sample batch.

FINDINGS

We found evidence that the alpha diversity of the maternal faecal microbiota during the third trimester of pregnancy predicts child internalising behaviour at two years of age (-2·74, (-4·71, -0·78), p = 0·01 (Wald test), R2=0·07). Taxa from butyrate-producing families, Lachnospiraceae and Ruminococcaceae, were more abundant in mothers of children with normative behaviour. A healthy prenatal diet indirectly related to decreased child internalising behaviours via higher alpha diversity of maternal faecal microbiota.

INTERPRETATION

These findings support animal studies showing that the composition of maternal prenatal gut microbiota is related to offspring brain development and behaviour. Our findings highlight the need to evaluate potential impacts of the prenatal gut microbiota on early life brain development.

FUNDING

This study was funded by the National Health and Medical Research Council of Australia (1082307, 1147980), Murdoch Children's Research Institute, Barwon Health and Deakin University.

Preconception depression and anxiety symptoms and maternal-infant bonding: a 20-year intergenerational cohort study

Early maternal-infant bonding problems are often forerunners of later emotional and behavioural difficulties. Interventions typically target the perinatal period but many risks may be established well before pregnancy. Here we examine the extent to which adolescent and young adult depression and anxiety symptoms predict perinatal maternal-infant bonding difficulties. The Victorian Intergenerational Health Cohort Study (VIHCS, est. 2006) is following offspring born to the Victorian Adolescent Health Cohort Study (VAHCS; est. 1992). VAHCS participants were assessed for depression and anxiety symptoms nine times during adolescence and young adulthood (age 14-29 years), and then contacted bi-annually (from age 29-35 years) to identify pregnancies. The Postpartum Bonding Questionnaire (PBQ) was administered to mothers at 2 and 12 months postpartum. A total of 395 women (606 infants) completed the 2-month and/or 12-month postpartum interviews. For most infants (64%), mothers had experienced depression and/or anxiety before pregnancy. Preconception depression and anxiety symptoms that persisted from adolescence into young adulthood predicted maternal-infant bonding problems at 2 months (β = 0.30, 95% CI 0.04, 0.55) and 12 months postpartum (β = 0.40, 95% CI 0.16, 0.63). Depression and anxiety symptoms occurring in young adulthood only, also predicted bonding problems at 12 months postpartum (β = 0.37, 95% CI 0.02, 0.71). Associations between preconception depression and anxiety symptoms and anxiety-related maternal-infant bonding problems at 12 months postpartum remained after adjustment for antenatal and concurrent postpartum depressive symptoms. This study puts forward a case for extending preconception health care beyond contraception and nutrition to a broader engagement in supporting the mental health of young women from adolescence.

Social networking and symptoms of depression and anxiety in early adolescence

BACKGROUND

Use of social networking in later childhood and adolescence has risen quickly. The consequences of these changes for mental health are debated but require further empirical evaluation.

METHODS

Using data from the Childhood to Adolescence Transition Study (n = 1,156), duration of social networking use was measured annually at four time points from 11.9 to 14.8 years of age (≥1 h/day indicating high use). Cross-sectional and prospective relationships between social networking use and depressive and anxiety symptoms were examined.

RESULTS

In adjusted (age, socioeconomic status, prior mental health history) cross-sectional analyses, females with high social networking use had greater odds of depressive (odds ratio [OR]: 2.15; 95% confidence interval [CI]: 1.58-2.91) and anxiety symptoms (OR: 1.99; 95% CI: 1.32-3.00) than those that used a few minutes at most, while males with high social networking use had 1.60 greater odds of reporting depressive symptoms (95% CI: 1.09-2.35). For females, an increased odds of depressive symptoms at age 14.8 was observed for high social networking use at one previous wave and at two or three previous waves, even after adjustment (OR: 1.76; 95% CI: 1.11-2.78; OR: 2.06, 95% CI: 1.27-3.37, respectively) compared to no wave of high use.

CONCLUSIONS

Our results suggest weak to moderate increased odds of depression and anxiety in girls and boys with high social networking use versus low/normal use. These findings indicate that prevention programs for early mental health problems might benefit from targeting social networking use in early adolescence.

Mediation effects that emulate a target randomised trial: Simulation-based evaluation of ill-defined interventions on multiple mediators

Many epidemiological questions concern potential interventions to alter the pathways presumed to mediate an association. For example, we consider a study that investigates the benefit of interventions in young adulthood for ameliorating the poorer mid-life psychosocial outcomes of adolescent self-harmers relative to their healthy peers. Two methodological challenges arise. First, mediation methods have hitherto mostly focused on the elusive task of discovering pathways, rather than on the evaluation of mediator interventions. Second, the complexity of such questions is invariably such that there are no well-defined mediator interventions (i.e. actual treatments, programs, etc.) for which data exist on the relevant populations, outcomes and time-spans of interest. Instead, researchers must rely on exposure (non-intervention) data, that is, on mediator measures such as depression symptoms for which the actual interventions that one might implement to alter them are not well defined. We propose a novel framework that addresses these challenges by defining mediation effects that map to a target trial of hypothetical interventions targeting multiple mediators for which we simulate the effects. Specifically, we specify a target trial addressing three policy-relevant questions, regarding the impacts of hypothetical interventions that would shift the mediators' distributions (separately under various interdependence assumptions, jointly or sequentially) to user-specified distributions that can be emulated with the observed data. We then define novel interventional effects that map to this trial, simulating shifts by setting mediators to random draws from those distributions. We show that estimation using a g-computation method is possible under an expanded set of causal assumptions relative to inference with well-defined interventions, which reflects the lower level of evidence that is expected with ill-defined interventions. Application to the self-harm example in the Victorian Adolescent Health Cohort Study illustrates the value of our proposal for informing the design and evaluation of actual interventions in the future.

Adiposity and Endometrial Cancer Risk in Postmenopausal Women: A Sequential Causal Mediation Analysis

BACKGROUND

Adiposity increases endometrial cancer risk, possibly through inflammation, hyperinsulinemia, and increasing estrogens. We aimed to quantify the mediating effects of adiponectin (anti-inflammatory adipocytokine); IL6, IL1-receptor antagonist, TNF receptor 1 and 2, and C-reactive protein (inflammatory status biomarkers); C-peptide (hyperinsulinemia biomarker); and free estradiol and estrone (estrogen biomarkers) in the adiposity-endometrial cancer link in postmenopausal women.

METHODS

We used data from a case-control study within the European Prospective Investigation into Cancer and Nutrition (EPIC). Eligible women did not have cancer, hysterectomy, and diabetes; did not use oral contraceptives or hormone therapy; and were postmenopausal at recruitment. Mediating pathways from adiposity to endometrial cancer were investigated by estimating natural indirect (NIE) and direct (NDE) effects using sequential mediation analysis.

RESULTS

The study included 163 cases and 306 controls. The adjusted OR for endometrial cancer for body mass index (BMI) ≥30 versus ≥18.5-<25 kg/m2 was 2.51 (95% confidence interval, 1.26-5.02). The ORsNIE were 1.95 (1.01-3.74) through all biomarkers [72% proportion mediated (PM)] decomposed as: 1.35 (1.06-1.73) through pathways originating with adiponectin (33% PM); 1.13 (0.71-1.80) through inflammation beyond (the potential influence of) adiponectin (13% PM); 1.05 (0.88-1.24) through C-peptide beyond adiponectin and inflammation (5% PM); and 1.22 (0.89-1.67) through estrogens beyond preceding biomarkers (21% PM). The ORNDE not through biomarkers was 1.29 (0.54-3.09). Waist circumference gave similar results.

CONCLUSIONS

Reduced adiponectin and increased inflammatory biomarkers, C-peptide, and estrogens mediated approximately 70% of increased odds of endometrial cancer in women with obesity versus normal weight.

IMPACT

If replicated, these results could have implications for identifying targets for intervention to reduce endometrial cancer risk in women with obesity.