PHIR methods

Developing appropriate research methods has been a significant challenge in PHIR.3 Intervention research methods can be classified into two broad categories. The first, intervention studies, examines the accrued benefit of interventions in specific (study) populations, including a number of quasi-experimental (eg, natural experiments, interrupted time-series designs, regression discontinuity approaches and comparative research between similar jurisdictions) and controlled study approaches (eg, randomised control trials). The second category of intervention research examines the potential benefit of introducing interventions into a study population, mainly modelling or simulation studies. Evidence from intervention studies is essential to informing studies that model the effects of interventions. While the merits and challenges of intervention studies have been discussed in the literature, the focus of this manuscript is to review the utility of simulation modelling to advancing PHIR. Simulation modelling is complimentary, and not a replacement, to traditional epidemiological methods that examine the etiological relationships between risk factors and disease.

Simulation modelling

Epidemiological simulation models represent causal relationship between the exposure to etiological or prognostic risk factors, treatments and health outcomes.11 These methods can be applied to simulate the impact of population health interventions within and across populations over time. The use of simulation models is well established both in health (eg, economic evaluation and infectious disease) and in other research fields (eg, climate change and economics). However, simulation models remain underutilised in chronic disease research.12 ,13

What do simulation studies add to PHIR on chronic disease?

Simulation models are flexible, cost-effective way to explore population health interventions. A common approach is to estimate the effectiveness of a defined intervention in a real-life setting for a targeted population that is different from the original study population. For example, would introducing a policy that increases price and reduces availability of cigarettes decrease youth smoking prevalence? Another approach examines the benefit of hypothetical interventions without specifying the nature of the intervention used to achieve the desired health benefit. For example, if physical activity rates increase by 20%, to what extent would heart disease prevalence change? When using this approach, the proposed benefits of hypothetical interventions need to be realistic, rooted in best-practice evidence, with careful attention paid to the different mechanisms of achieving the desired population exposure.14

From a research perspective, intervention studies provide valuable information on the effectiveness of population health interventions and will advance knowledge on how to reduce social inequalities in health.3 However, policy-makers can experience barriers translating this information into the implementation of effective population health interventions.2 Policy-makers are tasked with answering a different question than researchers. Not only whether a population health intervention is effective, but also which intervention(s) will have the greatest impact on achieving the desired health goal in their jurisdiction. In answering this question challenges arise from the limited availability of information on the benefits and cost-effectiveness of all potential interventions, making it difficult to reflect on an array of intervention options, particularly without evidence about their differential impacts on social subgroups.15

Therefore, from a decision-maker’s point of view, it is easy to see how researchers may often have the right answers to the wrong questions,2 and those generating the evidence (ie, social epidemiologists) may not be aware of this problem.16 Simulation models can be an invaluable tool for bridging the evidence gap between researchers and policy-makers. The contribution of both PHIR in general and more specifically simulation models to reducing this disparity is contingent on researchers asking more pragmatic, policy-focused questions to begin with. Simulation models are a platform for integrating the best available evidence from multiple studies as well as expertise from a diversity of scientific disciplines to model the social, behavioural, environmental and genetic effects on health.17 By combining these complex factors into a single model, a better understanding can be achieved of how they interact to influence population health. For example, the complex interactions between the individual, community and macro-policy level interventions available to reduce social inequalities in health.10

Advantages of simulation studies for PHIR on chronic disease

Simulation models have a number of practical advantages over other PHIR methods. They enable the synthesis of evidence from multiple data sources, improving our understanding of the complex pathways that influence population health. The process of conceptualising a simulation model linking exposures and outcomes can be invaluable in identifying gaps in the evidence base, illustrated by the lack of evidence surrounding parameters to be included in the model.13 Other types of PHIR methods may only consider the effect of a single intervention or lack the appropriate data or control to disentangle the complex factors necessary to determine the true effect of an exposure on a given health outcome, whereas by combing data from multiple sources in simulation modelling these challenges can sometimes be overcome. The benefits of multiple types of interventions and intervention strategies (eg, targeting high-risk populations vs the whole population) can be estimated. Combined, this information can help policy-makers evaluate the impact of implementing individual or combinations of population health interventions.17 In addition, simulation studies can be a cost-effective approach over other PHIR designs. Moreover, simulation models can help inform interventions when the direct observation of health outcomes is not feasible, practical or ethical to perform. Therefore, this flexible, cost-effective, evidence-based research method produces relevant evidence for policy-makers to make informed decisions regarding the implementation of population health interventions.

One limitation to this approach is that the quality of simulation models is inevitably linked to that of the data, evidence, assumptions specified in the model, and that the combination of these factors may have greater than additive effects on model estimates.18 Understanding the limitations associated with the selected parameters and being transparent about potential biases is important when communicating the results from simulation models.17 Another limitation is the number of simplifying assumptions that need to be made when attempting to model reality. The degree to which these assumptions impact the overall conclusions should be tested using sensitivity analyses.19 Finally, the development, maintenance and updating of simulation models is continually required to ensure their relevancy for answering policy-relevant research questions. While this requires a significant dedication of resources, simulation modelling remains a cost-effective approach to evaluating the effect of population health interventions.

What variables should be included in the simulation model?

The first steps in developing a simulation model are to define the scope and choose the appropriate model for the research question. The inclusion and proper parameterisation of social measures in simulation models is essential for assessing the impact of population health interventions on social inequalities in CHD. To evaluate the effect of changing a given exposure (eg, education) or mediator (eg, obesity) on CHD (eg, incidence or mortality), they all must be included in the model. Therefore, to truly investigate social inequalities in CHD, upstream determinants of health as well as factors outside of the health field need to be included in simulation models (figure 1A). This has been accomplished in simulation models examining policy levers in tobacco control,21 obesity17 and population-based prevention.22

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Building a simulation model to investigate the impact of population health interventions on social inequalities in coronary heart disease (CHD). (A) The selection of measures of inequalities, mediators and outcomes and the need to determine the causal relationships between variables, steps (1) and (2). Step (3) in (B) illustrates three different potential targets for population health interventions to reduce CHD, the effects of which can be evaluated and compared by running different ‘what-if’ counterfactual scenarios in the model. Step (4) in (C) represents the possibility of evaluating three different interventions for the same risk factor to evaluate which intervention will result in the greatest reduction in CHD. Interventions represented in step (4) should be parameterised to allow for different effectiveness by socioeconomic group.

The goal is to make simulation models flexible enough to evaluate interventions that target several pathways between SEP and CHD. For example, what is the impact of intervening on different social factors (eg, education vs income), different risk factors (eg, smoking vs diet), the same risk factor using different interventions (eg, mass media campaigns vs regulation), at different points in the life course (eg, childhood vs adulthood) or on targeted versus population-level interventions (figure 1B,C).

What mechanisms link SEP and CHD?

The next step is to map the relationships linking SEP and CHD. The process can be invaluable in identifying gaps in the evidence base, illustrated by the uncertainty surrounding model parameters.13 Mathematical equations are used to model how these factors interact to predict CHD. To parameterise the equations, one strategy is to use information from a number of data sources, prioritising the highest level of evidence available. Alternatively, established predictive risk algorithms have been developed to predict CHD.23 While these algorithms have traditionally been developed for clinical populations, their relevance for assessing baseline risk in population settings has been demonstrated.24

Once parameters are estimated, models need to be calibrated (refinement of model parameters to reproduce expected or observed results) and validated using an external data source.19 Sensitivity analyses are also used to test the robustness of the different model parameters.

Does the simulation model predict risk consistently across SEP groups?

An important assumption when investigating inequalities is whether simulation models accurately predict CHD within different subpopulations. Evidence from CHD risk prediction algorithms has demonstrated that including SEP measures can improve model calibration in low-SEP groups25; an effect that will otherwise underestimate inequalities between SEP groups. For example, the Framingham Risk Score23 has been demonstrated to underestimate CHD risk in low-SEP individuals; a bias reduced by incorporating SEP.26 Moreover, CHD risk algorithms that include area-based deprivation measures have shown improved CHD risk prediction in low-SEP individuals.27

Further research is needed on how best to include SEP into CHD risk prediction. The selection of SEP indicator(s) has largely been influenced by what information is available in a particular data source, a restriction that does not apply when simulating population-level data. Given the abundant evidence that multiple measures of SEP are likely necessary to capture the full range of socioeconomic differences, considering more than a single measure in simulation models seems prudent.28

Is the effectiveness of population health interventions equal across social groups?

Simulation models need to incorporate different levels of community effectiveness for interventions by social group. The ‘equity effectiveness loop’ describes the hypothesised differential equity effectiveness as relating to (1) lower awareness, access or coverage; (2) screening, diagnosis or targeting; and (3) compliance of providers and adherence of consumers.29 Simulation models should be user-friendly, allowing easy manipulation of these factors across SEP groups on the estimated intervention benefits.

Currently, there is a poor understanding of intervention effectiveness across social groups.30 The assessment of equity in individual studies and systematic reviews will help contribute to reducing this knowledge gap.31 Collecting data on the benefits of interventions both in terms of population and social inequalities health is important. For example, a recent systematic review that demonstrated the effectiveness of public health interventions on both population health, but also on reducing social inequalities in smoking.32 Further, systematic reviews that aggregate the effectiveness of interventions that target the wider social determinants of health will be useful in demonstrating how intervening on these factors can contribute to reducing health inequities.30 These studies provide information regarding which interventions should be considered in simulation studies that investigate solutions for reducing social inequalities in CHD.

Additionally, from a modelling perspective, assessing the effectiveness of interventions across different social groups is invaluable. For example, a recent systematic review provided evidence on the types of interventions that may, despite their best intentions, have the unintended consequence of generated inequalities between groups.33 These studies provide an enriched understanding used to parameterise models around the uptake and effectiveness of interventions.

Modelling past trends in social inequalities in CHD

Simulation studies have been used to better understand the substantial decline in recent decades of CHD mortality across Westernised countries.20 For example, the IMPACT CHD model seeks to evaluate the contribution of the two broad categories attributed for the reduction in CHD mortality: changes in population health CHD risk factors and improvement of medical and surgical techniques.34 The contribution of these factors is now being evaluated both in terms of population health and by deprivation level.34 The findings from this model provide useful information for making decisions about future resource allocation and which targets should be prioritised to reduce the population burden and social inequalities in CHD.

Evaluating the impact of interventions on future trends in CHD

Simulation models enable the testing of multiple ‘what-if’ intervention scenarios with respect to future changes in CHD risk factors, the distribution of SEP in the population (eg, income distribution) and the structural determinants of health inequalities (eg, macro-level policies to improve housing). This methodology enables the testing of different strategies for achieving health equity. For example, Hoffmann et al35 investigated the impact of the social redistributing of CHD risk factors (smoking, alcohol, physical inactivity and body mass index) on mortality. This approach levels down the population prevalence of risk factors to those observed in the highest educational attainment group to understand the proportion of mortality that would be reduced if social inequalities across all risk factors were eliminated. Alvarado et al36 examined different scenarios for how the US Healthy People 2010 targets could be achieved and what the corresponding effect would be on population and social inequalities in CHD. In addition to levelling down, three other intervention scenarios were investigated: (1) ideal (all social groups achieve risk factors target), (2) more realistic (risk factor target is achieved, but the social gradient across risk factors remains) and (3) poverty focused (only the most disadvantaged group is targeted and benefits from a risk factor reduction). Benach et al37 present an example of using simulation modelling to investigate a new typology of policies to reduce social inequalities in health, defined as targeted, universal with additional targeting, redistributive and proportionate universalism.

Simulation models can also be used to estimate how introducing a set of interventions into a population can impact social inequalities in CHD. For example, in the Whitehall Study, Kivimaki et al38 used a risk equation approach, accounting for major CHD risk factors (systolic blood pressure, total cholesterol, blood glucose in pre-diabetic people, non-insulin-dependent diabetes and cigarette smoking) to determine expected CHD mortality both before and after applying best-practice risk factor interventions.

To date, simulation studies have focused on downstream determinants of health to understand the potential benefits of population health interventions on reducing social inequalities in CHD. These interventions may be less likely to reduce social inequalities in health and may actually increase inequalities,39 depending on whether absolute or relative inequalities are considered.40 However, these studies have demonstrated the types of questions that can be investigated using simulation models to reduce social inequalities in health. The extension of simulation models to improve modelling of more complex systems linking social exposures and disease will enhance our ability to answer both research and policy questions regarding how best to reduce social inequalities in CHD.