DAG-informed regression modelling, agent-based modelling and microsimulation modelling: a critical comparison of methods for causal inference

Incarceration's lingering health effects on Black men: impacts persist into retirement

Background

The unique challenges Black men face within the criminal justice system underscore structural and systemic factors driving widespread inequalities. The long-term effects of these challenges on economic, health, and social outcomes as individuals transition to retirement remain poorly understood, highlighting a critical gap in our knowledge of life trajectories long after justice system involvement.

Objectives

This study investigated the enduring health impacts of incarceration on Black men, particularly focusing on the transition into retirement. It aimed to explore the influence of race and gender on experiences of incarceration before age 50, and how such experiences affected self-rated health during the retirement transition.

Methods

Utilizing data from the Health and Retirement Study, which followed individuals aged 50-59 for up to thirty years, this research examined the interplay of race, gender, incarceration history, and self-rated health during the retirement transition. Logistic regression and path modeling were employed for data analysis.

Results

Logistic regression results indicated that being Black, male, and having lower educational attainment significantly increased the likelihood of experiencing incarceration before the age of 50 (p < 0.05). This suggests that Black men with lower levels of education are at the greatest risk of incarceration. The path model revealed a correlation between incarceration experiences before age 50 and poorer self-rated health at the time of retirement.

Conclusion

The findings highlighted the disproportionately high risk of incarceration among Black men, especially those with lower educational attainment, and its persistent negative impacts on health decades later, including during the transition into retirement. Addressing structural racism and the mass incarceration of Black men is crucial for achieving racial health equity as individuals retire.

Short-term impacts of Universal Basic Income on population mental health inequalities in the UK: A microsimulation modelling study

BACKGROUND

Population mental health in the United Kingdom (UK) has deteriorated, alongside worsening socioeconomic conditions, over the last decade. Policies such as Universal Basic Income (UBI) have been suggested as an alternative economic approach to improve population mental health and reduce health inequalities. UBI may improve mental health (MH), but to our knowledge, no studies have trialled or modelled UBI in whole populations. We aimed to estimate the short-term effects of introducing UBI on mental health in the UK working-age population.

METHODS AND FINDINGS

Adults aged 25 to 64 years were simulated across a 4-year period from 2022 to 2026 with the SimPaths microsimulation model, which models the effects of UK tax/benefit policies on mental health via income, poverty, and employment transitions. Data from the nationally representative UK Household Longitudinal Study were used to generate the simulated population (n = 25,000) and causal effect estimates. Three counterfactual UBI scenarios were modelled from 2023: "Partial" (value equivalent to existing benefits), "Full" (equivalent to the UK Minimum Income Standard), and "Full+" (retaining means-tested benefits for disability, housing, and childcare). Likely common mental disorder (CMD) was measured using the General Health Questionnaire (GHQ-12, score ≥4). Relative and slope indices of inequality were calculated, and outcomes stratified by gender, age, education, and household structure. Simulations were run 1,000 times to generate 95% uncertainty intervals (UIs). Sensitivity analyses relaxed SimPaths assumptions about reduced employment resulting from Full/Full+ UBI. Partial UBI had little impact on poverty, employment, or mental health. Full UBI scenarios practically eradicated poverty but decreased employment (for Full+ from 78.9% [95% UI 77.9, 79.9] to 74.1% [95% UI 72.6, 75.4]). Full+ UBI increased absolute CMD prevalence by 0.38% (percentage points; 95% UI 0.13, 0.69) in 2023, equivalent to 157,951 additional CMD cases (95% UI 54,036, 286,805); effects were largest for men (0.63% [95% UI 0.31, 1.01]) and those with children (0.64% [95% UI 0.18, 1.14]). In our sensitivity analysis assuming minimal UBI-related employment impacts, CMD prevalence instead fell by 0.27% (95% UI -0.49, -0.05), a reduction of 112,228 cases (95% UI 20,783, 203,673); effects were largest for women (-0.32% [95% UI -0.65, 0.00]), those without children (-0.40% [95% UI -0.68, -0.15]), and those with least education (-0.42% [95% UI -0.97, 0.15]). There was no effect on educational mental health inequalities in any scenario, and effects waned by 2026. The main limitations of our methods are the model's short time horizon and focus on pathways from UBI to mental health solely via income, poverty, and employment, as well as the inability to integrate macroeconomic consequences of UBI; future iterations of the model will address these limitations.

CONCLUSIONS

UBI has potential to improve short-term population mental health by reducing poverty, particularly for women, but impacts are highly dependent on whether individuals choose to remain in employment following its introduction. Future research modelling additional causal pathways between UBI and mental health would be beneficial.

Transportability Without Positivity: A Synthesis of Statistical and Simulation Modeling

Studies designed to estimate the effect of an action in a randomized or observational setting often do not represent a random sample of the desired target population. Instead, estimates from that study can be transported to the target population. However, transportability methods generally rely on a positivity assumption, such that all relevant covariate patterns in the target population are also observed in the study sample. Strict eligibility criteria, particularly in the context of randomized trials, may lead to violations of this assumption. Two common approaches to address positivity violations are restricting the target population and restricting the relevant covariate set. As neither of these restrictions is ideal, we instead propose a synthesis of statistical and simulation models to address positivity violations. We propose corresponding g-computation and inverse probability weighting estimators. The restriction and synthesis approaches to addressing positivity violations are contrasted with a simulation experiment and an illustrative example in the context of sexually transmitted infection testing uptake. In both cases, the proposed synthesis approach accurately addressed the original research question when paired with a thoughtfully selected simulation model. Neither of the restriction approaches was able to accurately address the motivating question. As public health decisions must often be made with imperfect target population information, model synthesis is a viable approach given a combination of empirical data and external information based on the best available knowledge.

Agent-based simulation and child protection systems: Rationale, implementation, and verification

BACKGROUND AND OBJECTIVE

Simulation models are an important tool used in health care and other disciplines to support operational research and decision-making. In the child protection literature, simulation models are an under-utilized source of research evidence.

PARTICIPANTS AND SETTING

In this paper, we describe the rationale for and the development of an agent-based simulation of a child protection system in the US. Using the investigation, prevention service, and placement histories of 600,000 children served in an urban child welfare system, we walk the reader through the development of a prototype known as OSPEDALE.

METHODS

The governing equations built into OSPEDALE probabilistically simulate the onset of investigations. Then, drawing from empirical survival distributions, the governing equations trace the probability of subsequent interactions with the system (recurrence of maltreatment, service referrals, and placement) conditional on the characteristics of children, their assessed risk level, and prior child protection system involvement.

RESULTS

As an initial test of OSPEDALE's utility, we compare empirical admission counts with counts generated from OSPEDALE. Though the verification step is admittedly simple, the comparison shows that OSPEDALE replicates the empirical count of new admissions closely enough to justify further investment in OSPEDALE.

CONCLUSIONS

Management of public child protection systems is increasingly research evidence-dependent. The emphasis on research evidence as a decision-support tool has elevated evidence acquired through randomized clinical trials. Though important, the evidence from clinical trials represents only one type of research evidence. Properly specified, simulation models are another source of evidence with real-world relevance.

Applications of agent-based modeling in trauma research

OBJECTIVE

Trauma, violence, and their consequences for population health are shaped by complex, intersecting forces across the life span. We aimed to illustrate the strengths of agent-based modeling (ABM), a computational approach in which population-level patterns emerge from the behaviors and interactions of simulated individuals, for advancing trauma research; Method: We provide an overview of agent-based modeling for trauma research, including a discussion of the model development process, ABM as a complement to other causal inference and complex systems approaches in trauma research, and past ABM applications in the trauma literature; Results: We use existing ABM applications to illustrate the strengths of ABM for trauma research, including incorporating interactions between individuals, simulating processes across multiple scales, examining life-course effects, testing alternate theories, comparing intervention strategies in a virtual laboratory, and guiding decision making. We also discuss the challenges of applying ABM to trauma research and offer specific suggestions for incorporating ABM into future studies of trauma and violence; Conclusion: Agent-based modeling is a useful complement to other methodological advances in trauma research. We recommend a more widespread adoption of ABM, particularly for research into patterns and consequences of individual traumatic experiences across the life course and understanding the effects of interventions that may be influenced by social norms and social network structures. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

The effect of recognition on survival after out-of-hospital cardiac arrest and implications for biosensor technologies

BACKGROUND

Biosensor technologies have been proposed as a solution to provide recognition and facilitate earlier responses to unwitnessed out-of-hospital cardiac arrest (OHCA) cases. We sought to estimate the effect of recognition on survival and modelled the potential incremental impact of increased recognition of unwitnessed cases on survival to hospital discharge, to demonstrate the potential benefit of biosensor technologies.

METHODS

We included cases from the British Columbia Cardiac Arrest Registry (2019-2020), which includes Emergency Medical Services (EMS)-assessed OHCAs. We excluded cases that would not have benefitted from early recognition (EMS-witnessed, terminal illness, or do-not-resuscitate). Using a mediation analysis, we estimated the relative benefits on survival of a witness recognizing vs. intervening in an OHCA; and estimated the expected additional number of survivors resulting from increasing recognition alone using a bootstrap logistic regression framework.

RESULTS

Of 13,655 EMS-assessed cases, 11,412 were included (6314 EMS-treated, 5098 EMS-untreated). Survival to hospital discharge was 191/8879 (2.2%) in unwitnessed cases and 429/2533 (17%) in bystander-witnessed cases. Of the total effect attributable to a bystander witness, recognition accounted for 84% (95% CI: 72, 86) of the benefit. If all previously unwitnessed cases had been bystander witnessed, we would expect 1198 additional survivors. If these cases had been recognized, but no interventions performed, we would expect 912 additional survivors.

CONCLUSION

Unwitnessed OHCA account for the majority of OHCAs, yet survival is dismal. Methods to improve recognition, such as with biosensor technologies, may lead to substantial improvements in overall survival.

Effect of biological, psychological, and social factors on maternal depressive symptoms in late pregnancy: a cross-sectional study

Introduction

Depression commonly occurs during pregnancy and has become a major public health concern. Depression not only affects the individual but also causes adverse consequences for families and children. However, little is known regarding the depression status and its influencing factors in women during late pregnancy in China. This study aimed to assess the prevalence of maternal depressive symptoms in late pregnancy during the coronavirus disease 2019 (COVID-19) pandemic and further explore the effect of biological, psychological, and social factors on depressive symptoms.

Methods

An institution-based cross-sectional survey was conducted among eligible women in the late pregnancy stage and underwent prenatal examination at Lianyungang Maternal and Child Health Hospital in Jiangsu Province, Eastern China from December 2022 to February 2023. Data regarding depressive symptoms and biological, psychological, and social factors of the pregnant women were collected via a structured questionnaire. Chi-square test, Fisher's exact tests, and binary logistics regression were used to analyze the data.

Results

In total, 535 women in the late pregnancy stage were included in this study, 75 (14.0%) of whom exhibited depressive symptoms. A binary logistic regression analysis revealed that pregnant women who were multiparous (OR: 2.420, 95% CI: 1.188-4.932) and had moderate or severe insomnia symptoms (OR: 4.641, 95% CI: 1.787-12.057), anxiety (OR: 8.879, 95% CI: 4.387-17.971), high fear of COVID-19 (OR: 2.555, 95% CI: 1.255-5.199), moderate or severe family dysfunction (OR: 2.256, 95% CI: 1.141-4.461), and poor social support (OR: 2.580, 95% CI: 1.050-6.337) tended to show depressive symptoms. Conversely, pregnant women who received regular prenatal care (OR: 0.481, 95% CI: 0.243-0.951) and had good drinking water quality at home (OR: 0.493, 95% CI: 0.247-0.984) were more likely to avoid developing depressive symptoms.

Conclusion

This study found that the prevalence of maternal depressive symptoms during late pregnancy was high and had multiple influencing factors. Thus, screening for depressive symptoms in women in the late pregnancy stage and providing special intervention programs are necessary, especially for those with risk factors.

Simulation-based estimates and projections of global, regional and country-level maternal mortality by cause, 1990-2050

Maternal mortality is a major global health challenge. Although progress has been made globally in reducing maternal deaths, measurement remains challenging given the many causes and frequent underreporting of maternal deaths. We developed the Global Maternal Health microsimulation model for women in 200 countries and territories, accounting for individual fertility preferences and clinical histories. Demographic, epidemiologic, clinical and health system data were synthesized from multiple sources, including the medical literature, Civil Registration Vital Statistics systems and Demographic and Health Survey data. We calibrated the model to empirical data from 1990 to 2015 and assessed the predictive accuracy of our model using indicators from 2016 to 2020. We projected maternal health indicators from 1990 to 2050 for each country and estimate that between 1990 and 2020 annual global maternal deaths declined by over 40% from 587,500 (95% uncertainty intervals (UI) 520,600-714,000) to 337,600 (95% UI 307,900-364,100), and are projected to decrease to 327,400 (95% UI 287,800-360,700) in 2030 and 320,200 (95% UI 267,100-374,600) in 2050. The global maternal mortality ratio is projected to decline to 167 (95% UI 142-188) in 2030, with 58 countries above 140, suggesting that on current trends, maternal mortality Sustainable Development Goal targets are unlikely to be met. Building on the development of our structural model, future research can identify context-specific policy interventions that could allow countries to accelerate reductions in maternal deaths.

A simulation-based comparative effectiveness analysis of policies to improve global maternal health outcomes

The Sustainable Development Goals include a target to reduce the global maternal mortality ratio (MMR) to less than 70 maternal deaths per 100,000 live births by 2030, with no individual country exceeding 140. However, on current trends the goals are unlikely to be met. We used the empirically calibrated Global Maternal Health microsimulation model, which simulates individual women in 200 countries and territories to evaluate the impact of different interventions and strategies from 2022 to 2030. Although individual interventions yielded fairly small reductions in maternal mortality, integrated strategies were more effective. A strategy to simultaneously increase facility births, improve the availability of clinical services and quality of care at facilities, and improve linkages to care would yield a projected global MMR of 72 (95% uncertainty interval (UI) = 58-87) in 2030. A comprehensive strategy adding family planning and community-based interventions would have an even larger impact, with a projected MMR of 58 (95% UI = 46-70). Although integrated strategies consisting of multiple interventions will probably be needed to achieve substantial reductions in maternal mortality, the relative priority of different interventions varies by setting. Our regional and country-level estimates can help guide priority setting in specific contexts to accelerate improvements in maternal health.

Causal evidence in health decision making: methodological approaches of causal inference and health decision science

Objectives

Public health decision making is a complex process based on thorough and comprehensive health technology assessments involving the comparison of different strategies, values and tradeoffs under uncertainty. This process must be based on best available evidence and plausible assumptions. Causal inference and health decision science are two methodological approaches providing information to help guide decision making in health care. Both approaches are quantitative methods that use statistical and modeling techniques and simplifying assumptions to mimic the complexity of the real world. We intend to review and lay out both disciplines with their aims, strengths and limitations based on a combination of textbook knowledge and expert experience.

Methods

To help understanding and differentiating the methodological approaches of causal inference and health decision science, we reviewed both methods with the focus on aims, research questions, methods, assumptions, limitations and challenges, and software. For each methodological approach, we established a group of four experts from our own working group to carefully review and summarize each method, followed by structured discussion rounds and written reviews, in which the experts from all disciplines including HTA and medicine were involved. The entire expert group discussed objectives, strengths and limitations of both methodological areas, and potential synergies. Finally, we derived recommendations for further research and provide a brief outlook on future trends.

Results

Causal inference methods aim for drawing causal conclusions from empirical data on the relationship of pre-specified interventions on a specific target outcome and apply a counterfactual framework and statistical techniques to derive causal effects of exposures or interventions from these data. Causal inference is based on a causal diagram, more specifically, a directed acyclic graph (DAG), which encodes the assumptions regarding the causal relations between variables. Depending on the type of confounding and selection bias, traditional statistical methods or more complex g-methods are needed to derive valid causal effects. Besides the correct specification of the DAG and the statistical model, assumptions such as consistency, positivity, and exchangeability must be checked when aiming at causal inference. Health decision science aims for guiding policy decision making regarding health interventions considering and balancing multiple competing objectives of a decision based on data from multiple sources and studies, for example prevalence studies, clinical trials and long-term observational routine effectiveness studies, and studies on preferences and costs. It involves decision analysis, a systematic, explicit and quantitative framework to guide decisions under uncertainty. Decision analyses are based on decision-analytic models to mimic the course of disease as well as aspects and consequences of the intervention in order to quantitatively optimize the decision. Depending on the type of decision problem, decision trees, state-transition models, discrete event simulation models, dynamic transmission models, or other model types are applied. Models must be validated against observed data, and comprehensive sensitivity analyses must be performed to assess uncertainty. Besides the appropriate choice of the model type and the valid specification of the model structure, it must be checked if input parameters of effects can be interpreted as causal parameters in the model. Otherwise results will be biased.

Conclusions

Both causal inference and health decision science aim for providing best causal evidence for informed health decision making. The strengths and limitations of both methods differ and a good understanding of both methods is essential for correct application but also for correct interpretation of findings from the described methods. Importantly, decision-analytic modeling should be combined with causal inference when developing guidance and recommendations regarding decisions on health care interventions.

Effects of poverty on mental health in the UK working-age population: causal analyses of the UK Household Longitudinal Study

BACKGROUND

Addressing poverty through taxation or welfare policies is likely important for public mental health; however, few studies assess poverty's effects using causal epidemiology. We estimated the effect of poverty on mental health.

METHODS

We used data on working-age adults (25-64 years) from nine waves of the UK Household Longitudinal Survey (2009-19; n = 45 497/observations = 202 207 following multiple imputation). We defined poverty as a household equivalized income <60% median, and the outcome likely common mental disorder (CMD) as a General Health Questionnaire-12 score ≥4. We used double-robust marginal structural modelling with inverse probability of treatment weights to generate absolute and relative effects. Supplementary analyses separated transitions into/out of poverty, and stratified by gender, education, and age. We quantified potential impact through population attributable fractions (PAFs) with bootstrapped standard errors.

RESULTS

Good balance of confounders was achieved between exposure groups, with 45 830 observations (22.65%) reporting poverty. The absolute effect of poverty on CMD prevalence was 2.15% [%-point change; 95% confidence interval (CI) 1.45, 2.84]; prevalence in those unexposed was 20.59% (95% CI 20.29%, 20.88%), and the odds ratio was 1.17 (95% CI 1.12, 1.24). There was a larger absolute effect for transitions into poverty [2.46% (95% CI 1.56, 3.36)] than transitions out of poverty [-1.49% (95% CI -2.46, -0.53)]. Effects were also slightly larger in women than men [2.34% (95% CI 1.41, 3.26) versus 1.73% (95% CI 0.72, 2.74)]. The PAF for moving into poverty was 6.34% (95% CI 4.23, 8.45).

CONCLUSIONS

PAFs derived from our causal estimates suggest moves into poverty account for just over 6% of the burden of CMD in the UK working-age population, with larger effects in women.

Estimating the total incidence of type 1 diabetes in children and adolescents aged 0-19 years from 1990 to 2050: a global simulation-based analysis

BACKGROUND

Previous studies of type 1 diabetes in childhood and adolescence have found large variations in reported incidence around the world. However, it is unclear whether these reported incidence levels are impacted by differences in country health systems and possible underdiagnosis and if so, to what degree. The aim of this study was to estimate both the total and diagnosed incidence of type 1 diabetes globally and to project childhood type 1 diabetes incidence indicators from 1990 to 2050 for each country.

METHODS

We developed the type 1 diabetes global microsimulation model to simulate the natural history and diagnosis of type 1 diabetes for children and adolescents (aged 0-19 years) in 200 countries and territories, accounting for variability in underlying incidence and health system performance. The model follows an open population of children and adolescents in monthly intervals and simulates type 1 diabetes incidence and progression, as well as health system factors which influence diagnosis. We calibrated the model to published data on type 1 diabetes incidence, autoantibody profiles, and proportion of cases diagnosed with diabetic ketoacidosis from 1990 to 2020 and assessed the predictive accuracy using a randomly sampled test set of data withheld from calibration.

FINDINGS

We estimate that in 2021 there were 355 900 (95% UI 334 200-377 300) total new cases of type 1 diabetes globally among children and adolescents, of which 56% (200 400 cases, 95% UI 180 600-219 500) were diagnosed. Estimated underdiagnosis varies substantially by region, with over 95% of new cases diagnosed in Australia and New Zealand, western and northern Europe, and North America, but less than 35% of new cases diagnosed in west Africa, south and southeastern Asia, and Melanesia. The total number of incident childhood cases of type 1 diabetes is projected to increase to 476 700 (95% UI 449 500-504 300) in 2050.

INTERPRETATION

Our research indicates that the total global incidence of childhood and adolescent type 1 diabetes is larger than previously estimated, with nearly one-in-two children currently undiagnosed. Policymakers should plan for adequate diagnostic and medical capacity to improve timely type 1 diabetes detection and treatment, particularly as incidence is projected to increase worldwide, with highest numbers of new cases in Africa.

FUNDING

Novo Nordisk.

G-Computation and Agent-Based Modeling for Social Epidemiology: Can Population Interventions Prevent Posttraumatic Stress Disorder?

Agent-based modeling and g-computation can both be used to estimate impacts of intervening on complex systems. We explored each modeling approach within an applied example: interventions to reduce posttraumatic stress disorder (PTSD). We used data from a cohort of 2,282 adults representative of the adult population of the New York City metropolitan area from 2002-2006, of whom 16.3% developed PTSD over their lifetimes. We built 4 models: g-computation, an agent-based model (ABM) with no between-agent interactions, an ABM with violent-interaction dynamics, and an ABM with neighborhood dynamics. Three interventions were tested: 1) reducing violent victimization by 37.2% (real-world reduction); 2) reducing violent victimization by100%; and 3) supplementing the income of 20% of lower-income participants. The g-computation model estimated population-level PTSD risk reductions of 0.12% (95% confidence interval (CI): -0.16, 0.29), 0.28% (95% CI: -0.30, 0.70), and 1.55% (95% CI: 0.40, 2.12), respectively. The ABM with no interactions replicated the findings from g-computation. Introduction of interaction dynamics modestly decreased estimated intervention effects (income-supplement risk reduction dropped to 1.47%), whereas introduction of neighborhood dynamics modestly increased effectiveness (income-supplement risk reduction increased to 1.58%). Compared with g-computation, agent-based modeling permitted deeper exploration of complex systems dynamics at the cost of further assumptions.

Applications of Complex Systems Models to Improve Retail Food Environments for Population Health: A Scoping Review

Retail food environments (RFEs) are complex systems with important implications for population health. Studying the complexity within RFEs comes with challenges. Complex systems models are computational tools that can help. We performed a systematic scoping review of studies that used complex systems models to study RFEs for population health. We examined the purpose for using the model, RFE features represented, extent to which the complex systems approach was maximized, and quality and transparency of methods employed. The PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) guidelines were followed. Studies using agent-based modeling, system dynamics, discrete event simulations, networks, hybrid, or microsimulation models were identified from 7 multidisciplinary databases. Fifty-six studies met the inclusion criteria, including 23 microsimulation, 13 agent-based, 10 hybrid, 4 system dynamics, 4 network, and 2 discrete event simulation models. Most studies (n = 45) used models for experimental purposes and evaluated effects of simulated RFE policies and interventions. RFE characteristics simulated in models were diverse, and included the features (e.g., prices) customers encounter when shopping (n = 55), the settings (e.g., restaurants, supermarkets) where customers purchase food and beverages (n = 30), and the actors (e.g., store managers, suppliers) who make decisions that influence RFEs (n = 25). All models incorporated characteristics of complexity (e.g., feedbacks, conceptual representation of multiple levels), but these were captured to varying degrees across model types. The quality of methods was adequate overall; however, few studies engaged stakeholders (n = 10) or provided sufficient transparency to verify the model (n = 12). Complex systems models are increasingly utilized to study RFEs and their contributions to public health. Opportunities to advance the use of these approaches remain, and areas to improve future research are discussed. This comprehensive review provides the first marker of the utility of leveraging these approaches to address RFEs for population health.

Assessing causality in epidemiology: revisiting Bradford Hill to incorporate developments in causal thinking

The nine Bradford Hill (BH) viewpoints (sometimes referred to as criteria) are commonly used to assess causality within epidemiology. However, causal thinking has since developed, with three of the most prominent approaches implicitly or explicitly building on the potential outcomes framework: directed acyclic graphs (DAGs), sufficient-component cause models (SCC models, also referred to as 'causal pies') and the grading of recommendations, assessment, development and evaluation (GRADE) methodology. This paper explores how these approaches relate to BH's viewpoints and considers implications for improving causal assessment. We mapped the three approaches above against each BH viewpoint. We found overlap across the approaches and BH viewpoints, underscoring BH viewpoints' enduring importance. Mapping the approaches helped elucidate the theoretical underpinning of each viewpoint and articulate the conditions when the viewpoint would be relevant. Our comparisons identified commonality on four viewpoints: strength of association (including analysis of plausible confounding); temporality; plausibility (encoded by DAGs or SCC models to articulate mediation and interaction, respectively); and experiments (including implications of study design on exchangeability). Consistency may be more usefully operationalised by considering an effect size's transportability to a different population or unexplained inconsistency in effect sizes (statistical heterogeneity). Because specificity rarely occurs, falsification exposures or outcomes (i.e., negative controls) may be more useful. The presence of a dose-response relationship may be less than widely perceived as it can easily arise from confounding. We found limited utility for coherence and analogy. This study highlights a need for greater clarity on BH viewpoints to improve causal assessment.

Assessing causality in epidemiology: revisiting Bradford Hill to incorporate developments in causal thinking

The nine Bradford Hill (BH) viewpoints (sometimes referred to as criteria) are commonly used to assess causality within epidemiology. However, causal thinking has since developed, with three of the most prominent approaches implicitly or explicitly building on the potential outcomes framework: directed acyclic graphs (DAGs), sufficient-component cause models (SCC models, also referred to as 'causal pies') and the grading of recommendations, assessment, development and evaluation (GRADE) methodology. This paper explores how these approaches relate to BH's viewpoints and considers implications for improving causal assessment. We mapped the three approaches above against each BH viewpoint. We found overlap across the approaches and BH viewpoints, underscoring BH viewpoints' enduring importance. Mapping the approaches helped elucidate the theoretical underpinning of each viewpoint and articulate the conditions when the viewpoint would be relevant. Our comparisons identified commonality on four viewpoints: strength of association (including analysis of plausible confounding); temporality; plausibility (encoded by DAGs or SCC models to articulate mediation and interaction, respectively); and experiments (including implications of study design on exchangeability). Consistency may be more usefully operationalised by considering an effect size's transportability to a different population or unexplained inconsistency in effect sizes (statistical heterogeneity). Because specificity rarely occurs, falsification exposures or outcomes (i.e., negative controls) may be more useful. The presence of a dose-response relationship may be less than widely perceived as it can easily arise from confounding. We found limited utility for coherence and analogy. This study highlights a need for greater clarity on BH viewpoints to improve causal assessment.

Emulating Target Trials to Improve Causal Inference From Agent-Based Models

Agent-based models are a key tool for investigating the emergent properties of population health settings, such as infectious disease transmission, where the exposure often violates the key "no interference" assumption of traditional causal inference under the potential outcomes framework. Agent-based models and other simulation-based modeling approaches have generally been viewed as a separate knowledge-generating paradigm from the potential outcomes framework, but this can lead to confusion about how to interpret the results of these models in real-world settings. By explicitly incorporating the target trial framework into the development of an agent-based or other simulation model, we can clarify the causal parameters of interest, as well as make explicit the assumptions required for valid causal effect estimation within or between populations. In this paper, we describe the use of the target trial framework for designing agent-based models when the goal is estimation of causal effects in the presence of interference, or spillover.

Comparison of smoking reduction with improvement of social conditions in early life: simulation in a British cohort

BACKGROUND

Health care evaluation models can be useful to assign different levels of priority to interventions or policies targeting different age groups or different determinants of health. We aimed to assess early mortality in counterfactual scenarios implying reduced adverse childhood experience (ACE) and/or improved educational attainment (childhood and early life characteristics), compared with a counterfactual scenario implying reduced smoking in adulthood.

METHODS

We used data from the 1958 National Child Development Study British birth cohort, which initially included 18 558 subjects. Applying a potential outcome approach, scenarios were simulated to estimate the expected mortality between ages 16 and 55 under a counterfactual decrease by half of the observed level of exposure to (i) ACE, (ii) low educational attainment (at age 22), (iii) ACE and low educational attainment (a combined exposure) and (iv) smoking at age 33. Estimations were obtained using g-computation, separately for men and women. Analyses were further stratified according to the parental level of education, to assess social inequalities.

RESULTS

The study population included 12 164 members. The estimated decrease in mortality in the counterfactual scenarios with reduced ACE and improved educational attainment was close to the decreased mortality in the counterfactual scenario with reduced smoking, showing a relative difference in mortality of respectively -7.2% [95% CI (confidence interval) = (-12.2% to 1.2%)] versus -7.0% (-13.1% to +1.2%) for women, and -9.9% (-15.6% to -6.2%) versus -12.3% (-17.0% to -5.9%) for men.

CONCLUSIONS

Our results highlight the potential value of targeting early social characteristics such as ACE and education, compared with well-recognized interventions on smoking.

A Mediterranean diet microsimulation modeling in relation to cardiovascular disease burden: the ATTICA and GREECS epidemiological studies

BACKGROUND/OBJECTIVES

To quantify the changes in 10-year cardiovascular disease (CVD) onset, recurrence, and mortality, in relation to transitioning from low to a higher level of adherence to the Mediterranean diet.

SUBJECTS/METHODS

An individual-level microsimulation was created based on ATTICA (2002-2012, n = 3042 subjects free-of-CVD) and GREECS (2004-2014, n = 2172 patients with acute coronary syndrome (ACS)) studies (in total n = 5214). Eight scenarios regarding the proportion of participants and the size of improvement of the level of adherence to the Mediterranean diet (corresponding to one to ten point increases in MedDietScore) were compared in terms of relative change in CVD incidence and mortality, as well as, the number of preventable CVD events and deaths.

RESULTS

Improving adherence to the Mediterranean diet in at least 10% of the population, a significant relative percentage reduction could be observed in 10-year CVD onset, recurrence, and mortality. At least 851 first CVD events, 374 recurrent CVD events, and 205 CVD deaths per 100,000 of the population could be averted or delayed. In addition, Mediterranean diet clustering revealed that scoring higher in fruits, vegetables, whole wheat products, and legumes was more important than achieving higher scores in low consumption of meat and full-fat dairy products against CVD (all HRs in the former cluster were lower than the latter, indicating a stronger protective effect).

CONCLUSIONS

This microsimulation process confirms the added value of the Mediterranean diet in primary and secondary CVD prevention having great achievements even with modifications in a small part of the population (10%), while challenges the orientation of Mediterranean-diet interventions giving higher weights to plant-based part.

Complex systems models for causal inference in social epidemiology

Systems models, which by design aim to capture multi-level complexity, are a natural choice of tool for bridging the divide between social epidemiology and causal inference. In this commentary, we discuss the potential uses of complex systems models for improving our understanding of quantitative causal effects in social epidemiology. To put systems models in context, we will describe how this approach could be used to optimise the distribution of COVID-19 response resources to minimise social inequalities during and after the pandemic.

Analyzing Activity and Injury: Lessons Learned from the Acute:Chronic Workload Ratio

Injuries occur when an athlete performs a greater amount of activity than what their body can withstand. To maximize the positive effects of training while avoiding injuries, athletes and coaches need to determine safe activity levels. The International Olympic Committee has recommended using the acute:chronic workload ratio (ACWR) to monitor injury risk and has provided thresholds to minimize risk when designing training programs. However, there are several limitations to the ACWR and how it has been analyzed which impact the validity of current recommendations and should discourage its use. This review aims to discuss previously published and novel challenges with the ACWR, and strategies to improve current analytical methods. In the first part of this review, we discuss challenges inherent to the ACWR. We explain why using a ratio to represent changes in activity may not always be appropriate. We also show that using exponentially weighted moving averages to calculate the ACWR results in an initial load problem, and discuss their inapplicability to sports where athletes taper their activity. In the second part, we discuss challenges with how the ACWR has been implemented. We cover problems with discretization, sparse data, bias in injured athletes, unmeasured and time-varying confounding, and application to subsequent injuries. In the third part, conditional on well-conceived study design, we discuss alternative causal-inference based analytical strategies that may avoid major flaws in studies on changes in activity and injury occurrence.

Evidence-Based Medicine in Otolaryngology, Part XI: Modeling and Analysis to Support Decisions

OBJECTIVE

To provide a resource to educate clinical decision makers about the analyses and models that can be employed to support data-driven choices.

DATA SOURCES

Published studies and literature regarding decision analysis, decision trees, and models used to support clinical decisions.

REVIEW METHODS

Decision models provide insights into the evidence and its implications for those who make choices about clinical care and resource allocation. Decision models are designed to further our understanding and allow exploration of the common problems that we face, with parameters derived from the best available evidence. Analysis of these models demonstrates critical insights and uncertainties surrounding key problems via a readily interpretable yet quantitative format. This 11th installment of the Evidence-Based Medicine in Otolaryngology series thus provides a step-by-step introduction to decision models, their typical framework, and favored approaches to inform data-driven practice for patient-level decisions, as well as comparative assessments of proposed health interventions for larger populations.

CONCLUSIONS

Information to support decisions may arise from tools such as decision trees, Markov models, microsimulation models, and dynamic transmission models. These data can help guide choices about competing or alternative approaches to health care.

IMPLICATIONS FOR PRACTICE

Methods have been developed to support decisions based on data. Understanding the related techniques may help promote an evidence-based approach to clinical management and policy.

Optimizing the impact of medications for opioid use disorder at release from prison and jail settings: A microsimulation modeling study

BACKGROUND

We examined the impact of expanded access to medications for opioid use disorder (MOUD) in a unified prison and jail system on post-release, opioid-related overdose mortality.

METHODS

We developed a microsimulation model to simulate a population of 55,000 persons at risk of opioid-related overdose mortality in Rhode Island. The effect of an extended-release (XR) naltrexone only intervention and the effect of providing access to all three MOUD (i.e., methadone, buprenorphine, and XR-naltrexone) at release from incarceration on cumulative overdose death over eight years (2017-2024) were compared to the standard of care (i.e., limited access to MOUD).

RESULTS

In the standard of care scenario, the model predicted 2385 opioid-related overdose deaths between 2017 and 2024. An XR-naltrexone intervention averted 103 deaths (4.3% reduction), and access to all three MOUD averted 139 deaths (5.8% reduction). Among those with prior year incarceration, an XR-naltrexone only intervention and access to all three MOUD reduced overdose deaths by 22.8% and 31.6%, respectively.

CONCLUSIONS

Expanded access to MOUD in prison and jail settings can reduce overdose mortality in a general, at-risk population. However, the real-world impact of this approach will vary by levels of incarceration, treatment enrollment, and post-release retention.

Causally speaking: Challenges in measuring gentrification for population health research in the United States and Canada

The planet is rapidly urbanizing, the need for actionable evidence to guide the design of cities that help (not hinder) our health has never felt more urgent. One essential component of healthy city design is improving neighborhood conditions in previously disinvested areas. To ensure equitable city design, policy makers, city planners, health practitioners, and researchers are interested in understanding the complex relationship between urban change, gentrification, and population health. Yet, the causal link between gentrification and health outcomes remain unclear. Without clear and consistent gentrification measures, researchers struggle to identify populations who are exposed to gentrification, and to compare health outcomes between exposed and unexposed populations. To move the science forward, this paper summarizes the challenges related to gentrification measurement in the United States and Canada when aspiring to conduct studies to analyze causal relationships between gentrification and health. The paper concludes with a series of recommendations for studies aimed at examining both causes and consequences of gentrification and health.

Invited Commentary: Making Causal Inference More Social and (Social) Epidemiology More Causal

A society's social structure and the interactions of its members determine when key drivers of health occur, for how long they last, and how they operate. Yet, it has been unclear whether causal inference methods can help us find meaningful interventions on these fundamental social drivers of health. Galea and Hernán propose we place hypothetical interventions on a spectrum and estimate their effects by emulating trials, either through individual-level data analysis or systems science modeling (Am J Epidemiol. 2020;189(3):167-170). In this commentary, by way of example in health disparities research, we probe this "closer engagement of social epidemiology with formal causal inference approaches." The formidable, but not insurmountable, tensions call for causal reasoning and effect estimation in social epidemiology that should always be enveloped by a thorough understanding of how systems and the social exposome shape risk factor and health distributions. We argue that one way toward progress is a true partnership of social epidemiology and causal inference with bilateral feedback aimed at integrating social epidemiologic theory, causal identification and modeling methods, systems thinking, and improved study design and data. To produce consequential work, we must make social epidemiology more causal and causal inference more social.

Mortality-Air Pollution Associations in Low-Exposure Environments (MAPLE): Phase 1

INTRODUCTION

Fine particulate matter (particulate matter ≤2.5 μm in aerodynamic diameter, or PM2.5) is associated with mortality, but the lower range of relevant concentrations is unknown. Novel satellite-derived estimates of outdoor PM2.5 concentrations were applied to several large population-based cohorts, and the shape of the relationship with nonaccidental mortality was characterized, with emphasis on the low concentrations (<12 μg/m3) observed throughout Canada.

METHODS

Annual satellite-derived estimates of outdoor PM2.5 concentrations were developed at 1-km2 spatial resolution across Canada for 2000-2016 and backcasted to 1981 using remote sensing, chemical transport models, and ground monitoring data. Targeted ground-based measurements were conducted to measure the relationship between columnar aerosol optical depth (AOD) and ground-level PM2.5. Both existing and targeted ground-based measurements were analyzed to develop improved exposure data sets for subsequent epidemiological analyses.

Residential histories derived from annual tax records were used to estimate PM2.5 exposures for subjects whose ages ranged from 25 to 90 years. About 8.5 million were from three Canadian Census Health and Environment Cohort (CanCHEC) analytic files and another 540,900 were Canadian Community Health Survey (CCHS) participants. Mortality was linked through the year 2016. Hazard ratios (HR) were estimated with Cox Proportional Hazard models using a 3-year moving average exposure with a 1-year lag, with the year of follow-up as the time axis. All models were stratified by 5-year age groups, sex, and immigrant status. Covariates were based on directed acyclical graphs (DAG), and included contextual variables (airshed, community size, neighborhood dependence, neighborhood deprivation, ethnic concentration, neighborhood instability, and urban form). A second model was examined including the DAG-based covariates as well as all subject-level risk factors (income, education, marital status, indigenous identity, employment status, occupational class, and visible minority status) available in each cohort. Additional subject-level behavioral covariates (fruit and vegetable consumption, leisure exercise frequency, alcohol consumption, smoking, and body mass index [BMI]) were included in the CCHS analysis.

Sensitivity analyses evaluated adjustment for covariates and gaseous copollutants (nitrogen dioxide [NO2] and ozone [O3]), as well as exposure time windows and spatial scales. Estimates were evaluated across strata of age, sex, and immigrant status. The shape of the PM2.5-mortality association was examined by first fitting restricted cubic splines (RCS) with a large number of knots and then fitting the shape-constrained health impact function (SCHIF) to the RCS predictions and their standard errors (SE). This method provides graphical results indicating the RCS predictions, as a nonparametric means of characterizing the concentration-response relationship in detail and the resulting mean SCHIF and accompanying uncertainty as a parametric summary.

Sensitivity analyses were conducted in the CCHS cohort to evaluate the potential influence of unmeasured covariates on air pollution risk estimates. Specifically, survival models with all available risk factors were fit and compared with models that omitted covariates not available in the CanCHEC cohorts. In addition, the PM2.5 risk estimate in the CanCHEC cohort was indirectly adjusted for multiple individual-level risk factors by estimating the association between PM2.5 and these covariates within the CCHS.

RESULTS

Satellite-derived PM2.5 estimates were low and highly correlated with ground monitors. HR estimates (per 10-μg/m3 increase in PM2.5) were similar for the 1991 (1.041, 95% confidence interval [CI]: 1.016-1.066) and 1996 (1.041, 1.024-1.059) CanCHEC cohorts with a larger estimate observed for the 2001 cohort (1.084, 1.060-1.108). The pooled cohort HR estimate was 1.053 (1.041-1.065). In the CCHS an analogous model indicated a HR of 1.13 (95% CI: 1.06-1.21), which was reduced slightly with the addition of behavioral covariates (1.11, 1.04-1.18). In each of the CanCHEC cohorts, the RCS increased rapidly over lower concentrations, slightly declining between the 25th and 75th percentiles and then increasing beyond the 75th percentile. The steepness of the increase in the RCS over lower concentrations diminished as the cohort start date increased. The SCHIFs displayed a supralinear association in each of the three CanCHEC cohorts and in the CCHS cohort.

In sensitivity analyses conducted with the 2001 CanCHEC, longer moving averages (1, 3, and 8 years) and smaller spatial scales (1 km2 vs. 10 km2) of exposure assignment resulted in larger associations between PM2.5 and mortality. In both the CCHS and CanCHEC analyses, the relationship between nonaccidental mortality and PM2.5 was attenuated when O3 or a weighted measure of oxidant gases was included in models. In the CCHS analysis, but not in CanCHEC, PM2.5 HRs were also attenuated by the inclusion of NO2. Application of the indirect adjustment and comparisons within the CCHS analysis suggests that missing data on behavioral risk factors for mortality had little impact on the magnitude of PM2.5-mortality associations. While immigrants displayed improved overall survival compared with those born in Canada, their sensitivity to PM2.5 was similar to or larger than that for nonimmigrants, with differences between immigrants and nonimmigrants decreasing in the more recent cohorts.

CONCLUSIONS

In several large population-based cohorts exposed to low levels of air pollution, consistent associations were observed between PM2.5 and nonaccidental mortality for concentrations as low as 5 μg/m3. This relationship was supralinear with no apparent threshold or sublinear association.