 Review
 Open Access
 Open Peer Review
Choosing an epidemiological model structure for the economic evaluation of noncommunicable disease public health interventions
 Adam D. M. Briggs^{1}Email author,
 Jane Wolstenholme^{2},
 Tony Blakely^{3} and
 Peter Scarborough^{1}
https://doi.org/10.1186/s1296301600851
© Briggs et al. 2016
 Received: 1 July 2015
 Accepted: 20 April 2016
 Published: 4 May 2016
Abstract
Noncommunicable diseases are the leading global causes of mortality and morbidity. Growing pressures on health services and on social care have led to increasing calls for a greater emphasis to be placed on prevention. In order for decisionmakers to make informed judgements about how to best spend finite public health resources, they must be able to quantify the anticipated costs, benefits, and opportunity costs of each prevention option available. This review presents a taxonomy of epidemiological model structures and applies it to the economic evaluation of public health interventions for noncommunicable diseases. Through a novel discussion of the pros and cons of model structures and examples of their application to public health interventions, it suggests that individuallevel models may be better than populationlevel models for estimating the effects of population heterogeneity. Furthermore, model structures allowing for interactions between populations, their environment, and time are often better suited to complex multifaceted interventions. Other influences on the choice of model structure include time and available resources, and the availability and relevance of previously developed models. This review will help guide modelers in the emerging field of public health economic modeling of noncommunicable diseases.
Keywords
 Modeling
 Costeffectiveness
 Noncommunicable disease
 Economics
 Public health
Introduction
Noncommunicable diseases (NCDs) are the leading causes of mortality and morbidity globally, with much of this disease burden being preventable [1–3]. Health services around the world are experiencing everincreasing demand that may be alleviated through placing a greater emphasis on NCD prevention. For example, in 2014 the National Health Service (NHS) in England announced the need for a “radical upgrade in prevention and public health” [4] and in 2005, the UK National Institute for Health and Care Excellence (NICE) started a new initiative to produce guidance on public health interventions, including costeffectiveness analyses [5]. As of January 2016, 59 public health guidelines had been published [6].

longterm health impacts (e.g., the effects on cardiovascular disease burden and health expenditure of increasing the price of unhealthy foods may manifest and persist for many years after the intervention is introduced);

wider societal costs and consequences (e.g., transport policies to improve physical activity have social costs and outcomes beyond health; and different social costs and outcomes are relevant to different modeled scenarios, for example, influencing alcohol use has an important effect on crime whereas reducing stroke incidence has a more important effect on social care);

impact on inequalities (e.g., more deprived population groups may respond differently to those who are less deprived following a price increase on unhealthy food);

multicomponent interventions (e.g., a policy aimed at increasing physical activity may include both additional bicycle lanes and subsidized gym membership); and

interactions within complex nonhealth sector systems (e.g., if simulating the effect of upgrading home insulation to reduce winter deaths, it may be important to consider interactions with the housing sector, energy sector, and social care).
Examples of models are used throughout this review to highlight how different model structures cope with one or more of these challenges.
Squires emphasizes that when using this framework, it is important to adopt an iterative approach (represented by the doubleheaded arrows and the arrows feeding back) as in practice, the understanding of the problem to be modeled will develop as the model is built. Decisions regarding model inputs, outputs, scope, and structure are influenced by several processes and considerations. These include understanding exactly what should be modeled, views of stakeholders, data availability, and what other models are currently available (where appropriate, the development of new models should be avoided if existing models can be used or adapted) [22].
A taxonomy of epidemiological modeling structures for the economic evaluation of public health interventions
A  B  C  D  

Cohort/aggregatelevel/counts  Individuallevel  
Expected value, continuous state, deterministic  Markovian, discrete state, stochastic  Markovian, discrete state  NonMarkovian, discrete state  
1  No interaction  Untimed  Decision tree rollback or comparative risk assessment  Simulation decision tree or comparative risk assessment  Individual sampling model: Simulated patientlevel decision tree or comparative risk assessment  
2  Timed  Markov model (deterministic)  Simulation Markov model  Individual sampling model: Simulated patientlevel Markov model  
3  Interaction between entity and environment  Discrete time  System dynamics (finite difference equations)  Discrete time Markov chain model  Discretetime individual event history model  Discretetime discrete event simulation 
4  Continuous time  Systems dynamics (ordinary differential equations)  Continuous time Markov chain model  Continuous time individual event history model  Continuoustime discrete event simulation  
5  Interaction between heterogeneous entities/spatial aspects important  x  x  x  Agentbased simulation 
Summary table of epidemiological modeling structures for the economic evaluation of noncommunicable disease public health interventions
Corresponding section of review and table 1  Modeling method  Advantages  Disadvantages  Public health examples 

Section: Decision trees  Decision tree  Can be easy to construct.  No explicit time component.  Comparing exercise referral schemes with usual care to increase physical activity [29]. 
Relatively easy to interpret.  Exponentially more complex with additional disease states.  
Table 1: A1, B1, C1, D1  Can be adapted for cohorts and individuals.  No looping/recurring.  
Poorly suited to complex scenarios.  
Section: Comparative riskassessment  Comparative risk assessment  Can model multiple diseases and risk factors simultaneously.  More complex to build than decision trees.  Return on investment of workplace interventions to improve physical activity [32]. 
Can be used for individuals or cohorts.  No explicit time component.  
No looping/recurring.  
Table 1: A1, B1, C1, D1  Unable to model interactions between individuals, populations, or their environment.  
Section: Markov models without interaction  Markov models without interaction  Relatively straightforward to construct and to communicate.  The Markovian assumptionindividuals have no memory of (are independent of) previous disease states.  Investigating the cost effectiveness of different smoking cessation strategies using the Benefits of Smoking Cessation on Outcomes (BENESCO) model [33–35]. 
Can model populations or individuals.  
Table 1: A2, B2, C2, D2  Has time component.  Can only exist in one disease state.  
Allows looping/recurring.  Exponential increase in complexity with increasing numbers of disease states.  
Section: System dynamics models  System dynamics models  Allows for interactions between populations and the environment.  Models populations rather than individuals.  Modeling the effects of policies aimed at increasing bicycle commuting rather than travelling by car [63]. 
Table 1: A3, A4  Allows for feedback and recurring.  
Section: Markov chain models and individuallevel Markov models with interaction  Markov chain models and Markov individual event history models  Can model individuals or populations.  Markovian assumption still exists (although its impact can be reducedsee main text).  A CDC model evaluating the costeffectiveness of different diabetes prevention strategies [58, 59]. 
Table 1: B3, B4, C3, C4  Allows for interaction between populations or individuals within the model.  Becomes rapidly more complex with added disease states.  
Section: Discrete event simulation  Discrete event simulation  Allows for interaction between individuals and between individuals, populations, and their environment, governed by system rules.  Model structure can be difficult to communicate and interpret.  Evaluating the costeffectiveness of screening programs [67]. 
Table 1: D3, D4  Computationally challenging both in terms of designing the model and running it.  
Allows for modeling of complex scenarios.  
Section: Agentbased simulation  Agentbased simulation  Allow for interactions within and between individuals, populations, and the environment, governed by rules applied to individuals.  More complex than discrete event simulation.  The Archimedes model for modeling the outcomes of changing health care systems, such as investigating diabetes care [70]. 
Table 1: D5  Requires large computational power.  
Allows for individuals to learn.  Difficult to communicate and interpret model structure.  
Allows modeling of complicated systems.  
Table 1: adjunct to A1, B1, C1, D1, A2, B2, C2, D2  Multistate life tables  Can be used with comparative risk assessment and decision tree models to add a time component.  Assumes diseases are independent of each other.  The Australian Assessing Cost Effectiveness in Prevention (ACE Prevention) project [74, 76]. 
Can be combined with Markov models to increase the numbers of possible disease states without exponentially increasing model complexity.  Model limited by underlying model structure, for example, if combined with a Markov model, the Markovian assumption remains.  
Table 1: adjunct to C1, C2, C3, C4, D1, D2  Microsimulation  Can be combined with decision tree, comparative risk assessment, and Markov models to make it easier to model heterogeneous populations or multiple disease states.  Data requirements and simulations can become computationally challenging with complex models.  The NICE obesity health economic model used by Trueman et al. to estimate the costeffectiveness of primary care weight management programs [83]. 
Model limited by underlying model structure, for example, if combined with a Markov model, the Markovian assumption remains. 
In Table 1, columns A to D divide models into population or individuallevel, and separate model structures by how they deal with random events, expected values, and heterogeneous populations. Broadly speaking, by using populationlevel model structures, population heterogeneity (differences between population subgroups; for example, in terms of age, gender, or risk factors) can be simulated by rerunning the model for different cohorts, whereas individuallevel model structures use multiple samples of different types of people (see section: Use of microsimulation with individuallevel decision tree, comparative risk assessment, and Markov models ). The ability to incorporate randomness allows for Monte Carlo simulations to estimate stochastic uncertainty (describing the uncertainty in individual level models resulting from two individuals being in the same situation but, by chance, having different outcomes), and parametric uncertainty (in either population or individuallevel models and describing the uncertainty in the estimates of model parameters) [26]. Rows 1–4 categorize model structures by whether they allow for interactions to occur between entities within the model and between entities and the environment, and how the model deals with time (untimed means the models do not explicitly include a temporal component). Row 5 describes agentbased simulation modeling, which allows for multiple interactions governed by rules affecting individuals within the model, rather than affecting the system, as is the case in rows 3 and 4. Table 2 summarizes the main advantages and disadvantages of the epidemiological modeling structures discussed in this review.
This review presents an analysis of the pros and cons of epidemiological modeling structures giving examples of their application to the economic evaluation of public health interventions for NCDs to guide modelers in the field of public health economic modeling.
Description of model structures
Rows 1 and 2 – no interaction
Decision trees (row 1, columns A, B, C, and D)
Decision trees simulate possible decisions and outcomes using branches to represent each potential event. Branch points are usually described as nodes and can represent a decision (i.e., whether or not a public health education campaign takes place) or a chance event (i.e., development of disease). The options at each chance node are assumed to be mutually exclusive and the probability of each option occurring needs to sum to one. Each branch eventually ends with a terminal node against which the associated morbidity and costs of that patient journey can be attached [27]. In order to calculate the costeffectiveness of an intervention, the costs and morbidity of each patient journey are multiplied by the probability of that journey occurring, and then summed for each intervention and compared.
Decision trees are one of the simpler model structures available for public health economic modeling, and as such they are transparent as well as relatively straightforward to construct and to analyze. Decision trees can operate at the cohort level or at the individual level but do not have any explicit time component, do not allow for looping (recurring) of events, and do not accommodate interactions between individuals or populations [24]. This makes it particularly challenging to model long term chronic conditions or relapses in disease without becoming overly complicated [27].
Although often used by health economists, decision trees are less commonly used in public health economic models as they are particularly constrained by their lack of a temporal component, which other model structures can handle more efficiently [22, 28]. Trueman and Anokye used a decision tree to model exercise referral schemes (ERSs) for promoting physical activity in the UK [29]. Alongside a costutility analysis, Trueman and Anokye estimated the effects of ERSs on a variety of health outcomes and work absenteeism using a costconsequence analysis. A costconsequence analysis (CCA) is a useful way to illustrate the potential impact of public health interventions on wider society that may be relevant to public health decision makers but are not amenable to being quantified through standard measures of quality of life (such as those used to calculate Quality Adjusted Life Years [QALYs]). However, CCAs do not allow for comparisons across disease areas where some diseases include wider societal costs and benefits, and other diseases do not.
Comparative risk assessment (row 1, columns A, B, C, and D)
CRA models have also been used for public health economic modeling of NCDs [30–32]. These are commonly aggregatelevel models that use populationattributable fractions to estimate how parameters describing the relationship between a risk factor and disease outcome would change following an intervention. Individuals can be simulated when combined with microsimulation (see section: Use of microsimulation with individuallevel decision tree, comparative risk assessment, and Markov models ). CRA models do not allow for interactions but can simulate the age and sexspecific effect of an intervention on multiple risk factors and disease processes simultaneously without becoming as complex as the equivalent decision tree. Other population strata can be simulated in the same way, thereby estimating the impact on health inequalities. Furthermore, CRA models can be adapted to include a time component (see section: Using multistate life tables ). Costs and morbidity associated with each disease process can be used to compare total cost and health outcomes with and without an intervention. Examples including costs are the WHO Comparative Risk Assessment project [30], a French model simulating changes to fruit and vegetable consumption [31], and a US model by Trogdon et al. estimating the potential return on investment of workplace obesity interventions [32]. Trogdon et al.’s model estimates some outcomes and costs outside of the health sector by capturing costs from the employer perspective. This approach of estimating wider societal costs and consequences directly from the health outcome (such as lost days at work as a result of sickness) can be applied to all model structures discussed in this paper where data on the societal outcomes and how they relate to the health outcome are available.
Markov models without interaction (row 2, columns A, B, C, and D)
In comparison to decision trees and CRA models, Markov models are much more commonly used in public health economic modeling of NCDs, and are able to simulate more complex scenarios [28]. For example, a Markov model may be used to model the costeffectiveness of different smoking cessation methods incorporating multiple disease outcomes, the probability of relapse, and outcomes over different time horizons [33–35]. Markov models simulate how a population or individual moves between predefined health or disease states at a specific time interval (for example, annually). This incorporates a time component and allows for modeled populations to remain in a health state from one time interval to the next, and to loop back from a diseased state to a healthy state (recur), all based on a given transition probability. The model can then be run for either a predefined number of cycles or, if using a population cohort, until the entire population have reached a certain age or died. In this way, longterm effects of interventions on disease outcomes can be estimated. When computing costeffectiveness from a Markov model, each health state is assigned a measure of disease morbidity and cost. As the model runs through cycles, the costs and morbidity (or utility) can then be summed for the numbers of individuals in each state for each time cycle.
Markov models can be relatively straightforward to develop and to represent graphically, thereby making them transparent to peerreviewers and decisionmakers. Furthermore, transitional tunnel states can be added to increase complexity and make the model more realistic. For example, in order to more accurately capture the increased costs and morbidity associated with the first year of having a heart attack compared to subsequent years, all individuals who have a heart attack may spend one cycle in a tunnel state with associated high morbidity and cost, before having a 100 % probability of leaving that state. Markov models have been commonly used in a range of different public health economic analyses of NCDs in different settings and countries, for example the Dutch Rijksinstituut voor Volksgezondheid en Milieu (RIVM) Chronic Disease Model (CDM) [36–39], the Benefits of Smoking Cessation on Outcomes (BENESCO) model [33–35], the Australian Quit Benefits Model [40, 41], the US Centers for Disease Control and Prevention (CDC) Measurement of the Value of Exercise (MOVE) model [42], the Australian Coronary Heart Disease Prevention Model [43], the US Coronary Heart Disease (CHD) Policy Model [44, 45], and bespoke models for analyzing public health interventions internationally [46], in the Netherlands [47], Switzerland [48], Finland [49, 50], Germany [51], US [52, 53], UK [54, 55], and in Australia [56, 57].
Markov models have some important assumptions. Firstly, populationlevel models have no memory (the Markovian assumption) meaning that the morbidity, cost, and transition probabilities associated with a given health state are the same irrespective of previous health states or how long an individual has been in a health state. This can, in part, be dealt with by using tunnel states, as used by the Australian Quit Benefits Model [40], or by using microsimulation (see section: Use of microsimulation with individuallevel decision tree, comparative risk assessment, and Markov models ). Secondly, individuals can also only exist in one state at a time. This means that to add a new disease to a model that can coexist with the previously modeled diseases, the number of health states included in the model must each time be exponentially increased (as new health states are required for each disease combination). Further complexity is introduced if trying to model a heterogeneous population where population subgroups have different transition probabilities. This can be addressed by using weighted average costs, disabilities, and transition probabilities at the aggregate level. Alternatively, multiple cohorts, each with different characteristics, can be run through the model with cohortspecific transition parameters to give results by population subgroup, which can also be useful when estimating effects on inequalities; however, this adds model complexity and run time. An example of using multiple Markov models to model a heterogeneous population with multiple disease states is the US CDC diabetes prevention model, discussed in more detail in section: Markov chain models and individuallevel Markov models with interaction [58, 59].
Rows 3 and 4  interaction allowed
System dynamics models (rows 3 and 4, column A)
System dynamics models allow for populations to interact both with each other and with their environment. The probabilities of events occurring in the model (the system) change through feedback as the model runs, governed by algebraic or differential equations [60]. Such a model can be made increasingly complex as increasing numbers of factors influencing the system are added (requiring increasing amounts of data). This makes system dynamics models better able to simulate interactions within complex nonhealth sector systems, and to estimate effects of multicomponent interventions than previously discussed model structures. Costs can be applied to either the disease state, or to all factors within the model, and then cost and health outcomes with and without the intervention can be compared. System dynamics models can usually be represented graphically, facilitating communication of the model with stakeholders. Such models are wellestablished for communicable diseases [61, 62] and are increasingly being applied to NCD risk factors, such as Macmillan et al. who used a system dynamics model to explore the potential effect of different transport policies on bicycle commuting in Auckland, New Zealand [63]. The authors not only estimated health outcomes, but also the effect on air pollution, carbon emissions, and fuel costs over a period of 40 years. In this way, longterm health and economic impacts were estimated, and some nonhealth outcomes were quantified. The authors monetized the model’s outcomes and a costbenefit analysis was used to compare different policies. Through monetizing nonhealth outcomes and assigning utilities to health outcomes it would be possible to perform a costeffectiveness analysis using the same approach. A potential limitation of system dynamics models is that the dynamic element of the model (the rate of change of parameters with time) is deterministic, although parametric uncertainty can be modeled.
Markov chain models and individuallevel Markov models with interaction (rows 3 and 4, columns B and C)
In discrete or continuous time Markov chain models, state transition probabilities can depend on (interact with) the proportion of different populations in different disease states, and on the time that has elapsed in the model. These interactions are the key difference between Markov chain models and those discussed in section: Markov models without interaction , and provide the model with some degree of memory, in part overcoming the Markovian assumption. However, when simulating cohorts using timedependent transition factors, it is not possible to completely overcome this assumption because the model cannot know how long different proportions of the population in any given health states have been in that state. By contrast, individuallevel Markov models simulate individuals separately (also called microsimulation, see relevant section on microsimulation later in article), making it possible to ”know” how long an individual has been in each state and to alter transition probabilities as a function of time in a given state. The simulation may or may not retain “memory” of which states an individual has been in previously. Markov chain and individuallevel Markov model structures with interaction are better able to cope with system complexity than CRA and decision tree model structures, however they are unable to explicitly model nonhealth sector system interactions.
Examples of discrete time aggregatelevel Markov models allowing for interaction are a version of the RIVM CDM (also cited in Markov models without interaction (row 2, columns A, B, C, and D) section) which includes disease incidence parameters that depend on time from smoking cessation [64], and the US CDC diabetes prevention model, described in detail in a technical report by Hoerger et al. available from Herman et al. as an online supplement [59]. In this cohort model, transition probabilities are dependent on time since diagnosis of diabetes, as well as on levels of glycemia and hypertension. Furthermore, the model simulates multiple disease processes simultaneously by allowing the cohort to coexist in five different disease pathways which are linked to the overall Markov model by bridge models (see online supplement from Herman et al. for full description of the model) [59, 65]. Bridge models allow the overall Markov model to collect accumulated data on the number of events that have occurred, and keep track of the proportion of the cohort remaining in each disease state in any given year and the proportion who have left either through death or remission. Finally, Herman et al. account for a heterogeneous population by simulating 560 different cohorts, each with individual state transition probabilities dependent on age, sex, race, hypertension, cholesterol, and smoking status.
Discrete event simulation (rows 3 and 4, column D)
Discrete event simulation (DES) is an extremely flexible modeling structure that simulates a system changing over time with a sequence of discrete individual events [66]. Rather than simulating populations or individuals through states for a fixed length of time, multiple future events are in competition and the model jumps to whichever event occurs next based on predefined probabilities. The occurrence of an event can directly lead to a series of contemporaneous events, as well as affect the probability of future events. The probability of a given event occurring can also vary with time and be affected by interactions between individuals, populations, and their environment at each event. A set of system rules and probabilities govern the behavior of the population or individuals in a DES model, and these can be changed based on the intervention being modeled. As each event occurs, costs and utilities based on the event, consequence, and time to event are estimated. However, due to the large number of variables possible in DES and the need to simulate many individuals, models can be computationally slow to run (particularly when estimating uncertainty) and require large quantities of data for each disease outcome. Within public health economic modeling of NCDs, this approach has been commonly adopted for evaluating screening programs. [67] Along with other model structures that can simulate interactions between the population being modeled and the environment, DES is wellequipped for addressing interactions within complex systems.
Row 5 – interaction between heterogeneous entities/special aspects important
Agentbased simulation (row 5, column D)
Agentbased simulation (ABS) has many similarities to DES in that it allows for the probability of events occurring within the system being modeled to change as a consequence of interactions between individuals (agents), between agents and the environment, and with time. ABS is therefore also able to deal with the challenges of multicomponent interventions and interactions within complex nonhealth sector systems. The difference, however, is that ABS models apply rules to agents or groups of agents, and responses depend on individual agent characteristics which can change either over time or following interactions with other agents or the environment. This is compared to systembased rules found in DES [22, 68]. A heterogeneous population of agents is therefore able to ”learn” over time and affect the system, which, as Squires discusses, allows for more accurate representation of spatial effects, such as social networks [22]. However, ABS is again more complex than DES and can require considerably more data to represent heterogeneous population groups. Costs and morbidity can be applied to each event and disease state in order to derive estimates of costeffectiveness of interventions that affect the behavior of agents (other societal costs and outcomes could also be estimated in a similar fashion). Populationlevel results emerge from the aggregate of all agentlevel results.
An example of ABS being used in public health economic modeling of NCDs is the Archimedes Model [69, 70]. This model was designed to simulate a wide range of interventions modeling a whole suite of clinical outcomes through changes to physiological risk factors. It is therefore a good example of simulating multicomponent interventions. It allows interactions between variables both within and between individuals, and with the system as a whole. As such, it relies on a large amount of processing power and data.
Using multistate life tables and microsimulation to increase model flexibility
Use of multistate life tables with decision tree, comparative risk assessment, and Markov models with no interaction (rows 1 and 2, columns A, B, C and D)
For the purposes of this review, multistate life tables are defined as life tables that model an individual’s, or proportion of a population’s, probability of developing a given disease at different ages and subsequent case fatality rates once the disease is acquired. These can simulate multiple diseases simultaneously and can be used to add a temporal component to decision tree or CRA models. For each intervention scenario being analyzed, the decision tree or CRA model can be used to generate population impact fractions to alter the multistate life table disease incidence, case fatality, and overall mortality rates. Rerunning the multistate lifetable model then allows scenarios to be compared over the number of years in the life table allowing for longterm health effects and economic impacts to be estimated.
In a similar way, multistate life tables can be used with population or individual Markov models to simulate multiple diseases without the need for large increases in the number of disease states. To do this, proportions of the population can coexist in more than one disease state in the multistate life table and for each disease a Markov model can simulate the probability of moving between diseased and not diseased states (thereby performing a similar function to bridge models, as discussed in section: in Markov chain models and individuallevel Markov models with interaction ). These properties usually assume that diseases are independent of one another (for example, the probability of developing ischemic heart disease does not change with a concurrent diagnosis of cancer). Published examples include the Australian Assessing Cost Effectiveness in Prevention (ACE Prevention) program of research (which estimates the effects of various interventions over the lifetime of the Australian population) [71–76], the WHO PopMod model (which estimates effects over a 100year time period) [77, 78], and a New Zealand model estimating the effects of increasing tobacco taxation (which estimates effects over the lifetime of the New Zealand population) [79].
Use of microsimulation with individuallevel decision tree, comparative risk assessment, and Markov models (rows 1 and 2, columns C and D; and rows 3 and 4, column C)
In order to overcome the complexity of modeling multiple diseases and heterogeneous populations in decision tree, CRA, and Markov model structures, an alternative approach is to use individual patient simulation models (microsimulation). These allow for a population of heterogeneous individuals to move through the model based on probabilities appropriate to their characteristics (such as demographic factors or physiological characteristics). The model is run at the individual level with all members, or randomly selected members of a predefined population, being simulated until either a prespecified outcome occurs or a certain length of time has elapsed (e.g., death or reaching age 100). Once completed, the individual results can be aggregated to calculate a single populationlevel result or for percentile (or other) variations in results across individuals to be reported (thereby also allowing reporting of inequalities). Microsimulation is also particularly useful when estimating population means of skewed effects (such as the growth rates of different cancers when a few may be very quickgrowing and have different subsequent events compared to the majority being slowgrowing), and is very flexible at modeling interactions.
Model parameters can be changed for different scenario analyses in the same way as with multistate life tables see section: Use of multistate life tables with decision tree, comparative risk assessment, and Markov models with no interaction by using decision tree, CRA, or Markov models. However, in microsimulation this is at the individual level and parameters are specific to the individual’s characteristics (such as their age and gender). Drawbacks of individuallevel simulation are that they are computationally intense although modern computers and software are increasingly able to cope with the many thousands of iterations often computed. Examples of patientlevel simulation Markov models within public health economic modeling of NCDs include the UK Health Forum model [80], the World Health Organisation and Organisation for Economic Cooperation and Development (OECD) Chronic Disease Prevention (CDP) model [81, 82], and the NICE Obesity Health Economic Model [83], as well as examples from Australia [84], Korea [85], and Sweden [86, 87].
Conclusions
In the context of an everincreasing global burden of preventable NCDs, decisions need to be made as to how to approach their prevention and management. Modeling the costeffectiveness of NCDrelated public health interventions is an expanding academic field that is starting to embrace more sophisticated modeling structures (see Table 2 for a summary of model structures). Public health economic modeling of NCDs has thus far primarily used Markov models, with or without the use of life tables and microsimulation, to investigate the health impacts of a given intervention and its costeffectiveness. However, there is scope for more complex systems to be modeled with a wider range of possible outcomes by using model structures such as DES and ABS. Balancing transparency and parsimony with complexity when developing such models is crucial for model results to be readily interpreted and used by decisionmakers [24, 88–90].
We identify many NCD public health economic models that, to a greater or lesser extent, address the specific challenges of epidemiological, public health, and economic modeling when compared to health technology appraisals; namely assessing longterm health and economic impacts, quantifying societal costs and consequences, identifying the potential impact on inequalities, simulating multicomponent interventions, and simulating interactions within complex nonhealth sector systems.
The solutions to these challenges presented in this review are either specific to a given model structure, or can be applied across all structures. Longterm health and economic impacts are impossible to estimate using crosssectional model structures such as CRA and decision tree models structures. However, combining these with a multistate life table can add a longitudinal component to the model thereby enabling longterm outcomes to be simulated [74, 78].
Wider societal costs and consequences (e.g., air pollution or effects on employment) can be estimated using any model structure. The challenge is less a problem of choosing the appropriate epidemiological modeling structure, but more an issue of identifying robust cost and nonhealth data with which to parameterize the model; although, some model structures are more adaptable than others for estimating multiple nonhealth outcomes. For example, if data are available it may be possible within any model structure to estimate some of the effects on productivity costs of an intervention directly from a model’s health outcomes by inferring what may happen to work absenteeism as a consequence of ill health [29]. It is less easy to do this for nonhealth related outcomes such as effects on air pollution as a result of changes to traffic policy and numbers of cyclists. Macmillan et al. do this using a system dynamics model and relate air pollution (and its cost) to part of the causal pathway (number of vehicles on the road) between the intervention and health outcome [63]. For outcomes that cannot be directly estimated from a step on a model’s causal pathway or result from interactions with other model outcomes (for example, if increased air pollution affected weather patterns which in turn affected cycling habits) then model structures allowing for interactions are more appropriate.
The potential impact of a simulated intervention on inequalities can also be estimated using any epidemiological modeling structure by simulating population groups of interest separately if using a cohort model (for example, Blakely et al. estimated the effects of a tobacco tax by ethnic group in New Zealand using a multistate lifetable model [79]) or aggregating results from the two or more population groups of interest when using microsimulation (such as that used by Feldman et al. when estimating the effects of lifestyle interventions on different risk groups for metabolic syndrome [87]). Estimating wider societal costs and consequences is dependent on available data.
The challenge of multicomponent interventions can be addressed both using epidemiological model structures that do and do not allow for interaction. Without interaction, a model assumes that each component of a multicomponent intervention acts independently on a disease outcome or risk factor. If interaction between multiple interventions is necessary to simulate, then model structures that allow interactions are required. Both for multicomponent interventions and for simulating interactions within complex nonhealth sector systems, microsimulation models (such as ABS and DES) offer more flexibility than populationlevel models.
However, it is important to note that when dealing with any of these challenges, a model is only as good as the data used to parameterize it and adding more complexity may only serve to make the model more uncertain and more difficult to communicate. As Whitty discusses, a model that is simple, timely, and lays bare its problems is far more useful to a policymaker than one that is more detailed, more complicated, possibly more accurate, but less interpretable and arrives after the policy decision is made [90].
No one model identified in this review addresses all the challenges of modeling economic evaluations of public health interventions for NCDs, and the choice of which epidemiological model structure to adopt will depend on what is being modeled: the interventions being evaluated, the outputs required, and the needs of the decisionmaker. We therefore provide a revised version of Brennan et al.’s taxonomy of model structures for the economic evaluation of health technologies to act as a guide to modelers in the field of public health economic modeling [24].
Declarations
Acknowledgments
We would like to thank and acknowledge Frederieke van der Deen and Linda Cobiac for their insights and critiques of early drafts of this paper.
ADMB is funded by the Wellcome Trust (grant number 102730/Z/13/Z); PS is funded by the British Heart Foundation (grant number 021/P&C/Core/2010/HPRG); TB is funded by the Health Research Council of New Zealand (grant number 10/248). The funding bodies had no role in the writing of the manuscript or the decision to submit the manuscript for publication.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Authors’ Affiliations
References
 GBD 2013 Mortality and Causes of Death Collaborators. Global, regional, and national age–sex specific allcause and causespecific mortality for 240 causes of death, 1990–2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet. 2015;385:117–71.View ArticleGoogle Scholar
 Lim SS, Vos T, Flaxman AD, et al. A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990–2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet. 2012;380:2224–60.View ArticlePubMedPubMed CentralGoogle Scholar
 Murray CJL, Richards MA, Newton JN, et al. UK health performance: findings of the Global Burden of Disease Study 2010. Lancet. 2013;381:997–1020.View ArticlePubMedGoogle Scholar
 National Health ServiceEngland. Five year forward view. London: National Health ServiceEngland; 2014. https://www.england.nhs.uk/wpcontent/uploads/2014/10/5yfvweb.pdf. Accessed 25 Apr 2016.Google Scholar
 Kelly M, Morgan A, Ellis S, Younger T, Huntley J, Swann C. Evidence based public health: a review of the experience of the National Institute of Health and Clinical Excellence (NICE) of developing public health guidance in England. Soc Sci Med. 2010;71:1056–62.View ArticlePubMedGoogle Scholar
 National Institute for Health and Care Excellence (2015) Public health guidance list. http://www.nice.org.uk/guidance/published?type=ph. Accessed 25 Apr 2016.
 National Institute for Health and Care Excellence. Methods for the development of NICE public health guidance. Third edition. London: National Institute for Health and Care Excellence; 2012. http://www.nice.org.uk/article/pmg4/resources/nonguidancemethodsforthedevelopmentofnicepublichealthguidancethirdeditionpdf. Accessed 25 Apr 2016.Google Scholar
 National Institute for Health and Care Excellence. Guide to the methods of technology appraisal 2013. London: National Institute for Health and Care Excellence; 2013. http://www.nice.org.uk/article/pmg9/resources/nonguidanceguidetothemethodsoftechnologyappraisal2013pdf. Accessed 25 Apr 2016.Google Scholar
 Kaltenthaler E, Tappenden P, Paisley S, Squires H. NICE DSU technical support document 13, Identifying and reviewing evidence to inform the conceptualisation and population of costeffectiveness models. 2011. http://www.nicedsu.org.uk/TSD%2013%20model%20parameters.pdf. Accessed 25 Apr 2016.Google Scholar
 Fischer AJ, Threlfall A, Meah S, Cookson R, Rutter H, Kelly MP. The appraisal of public health interventions: an overview. J Public Health. 2013;35:488–94.View ArticleGoogle Scholar
 Threlfall AG, Meah S, Fischer AJ, Cookson R, Rutter H, Kelly MP. The appraisal of public health interventions: the use of theory. J Public Health. 2015;37:166–71.View ArticleGoogle Scholar
 Marsh K, Phillips CJ, Fordham R, Bertranou E, Hale J. Estimating costeffectiveness in public health: a summary of modelling and valuation methods. Health Econ Rev. 2012;2.Google Scholar
 Weatherly H, Drummond M, Claxton K, Cookson R, Ferguson B, Godfrey C, Rice N, Sculpher M, Sowden A. Methods for assessing the costeffectiveness of public health interventions: key challenges and recommendations. Health Policy. 2009;93:85–92.View ArticlePubMedGoogle Scholar
 Drummond M, Weatherly H, Claxton K, Cookson R, Ferguson B, Godfrey C, Rice N, Sculpher M, Sowde A. Assessing the challenges of applying standard methods of economic evaluation to public health interventions. York: Public Health Research Consortium; 2007. http://phrc.lshtm.ac.uk/papers/PHRC_D105_Final_Report.pdf. Accessed 25 Apr 2016.Google Scholar
 Edwards RT, Charles JM, LloydWilliams H. Public health economics: a systematic review of guidance for the economic evaluation of public health interventions and discussion of key methodological issues. BMC Public Health. 2013;13:1001.View ArticlePubMedPubMed CentralGoogle Scholar
 Payne K, McAllister M, Davies LM. Valuing the economic benefits of complex interventions: when maximising health is not sufficient. Health Econ. 2013;22:258–71.View ArticlePubMedGoogle Scholar
 Kelly MP, McDaid D, Ludbrook A, Powell J. Economic appraisal of public health interventions. London: Health Development Agency; 2005. http://www.cawt.com/Site/11/Documents/Publications/Population%20Health/Economics%20of%20Health%20Improvement/Economic_appraisal_of_public_health_interventions.pdf. Accessed 25 Apr 2016.Google Scholar
 Shiell A, Hawe P, Gold L. Complex interventions or complex systems? Implications for health economic evaluation. BMJ. 2008;336:1281–3.View ArticlePubMedPubMed CentralGoogle Scholar
 Cooper K, Brailsford SC, Davies R. Choice of modelling technique for evaluating health care interventions. J Oper Res Soc. 2007;58:168–76.View ArticleGoogle Scholar
 Caro JJ, Briggs AH, Siebert U, Kuntz KM. Modeling good research practicesoverview: a report of the ISPORSMDM Modeling Good Research Practices Task Force1. Value Health. 2012;15:796–803.View ArticlePubMedGoogle Scholar
 Eddy DM, Hollingworth W, Caro JJ, Tsevat J, McDonald KM, Wong JB. Model transparency and validation: a report of the ISPORSMDM Modeling Good Research Practices Task Force7. Value Health. 2012;15:843–50.View ArticlePubMedGoogle Scholar
 Squires H. A methodological framework for developing the structure of Public Health economic models. Sheffield: University of Sheffield; 2014. http://etheses.whiterose.ac.uk/5316/1/Final%20doctoral%20thesis%20with%20appendices.pdf. Accessed 25 Apr 2016.Google Scholar
 Barton P, Bryan S, Robinson S. Modelling in the economic evaluation of health care: selecting the appropriate approach. J Health Serv Res Policy. 2004;9:110–8.View ArticlePubMedGoogle Scholar
 Brennan A, Chick SE, Davies R. A taxonomy of model structures for economic evaluation of health technologies. Health Econ. 2006;15:1295–310.View ArticlePubMedGoogle Scholar
 National Institute for Health and Care Excellence. Section 3, The manufacturer’s submission. In: Ticagrelor for the treatment of acute coronary syndromes. London: NICE technology appraisal guidance TA 236; 2011. https://www.nice.org.uk/guidance/ta236/chapter/3Themanufacturerssubmission. Accessed 25 Apr 2016.Google Scholar
 Groot Koerkamp B, Stijnen T, Weinstein MC, Hunink MGM. The combined analysis of uncertainty and patient heterogeneity in medical decision models. Med Decis Making. 2011;31:650–61.View ArticlePubMedGoogle Scholar
 Petrou S, Gray A. Economic evaluation using decision analytical modelling: design, conduct, analysis, and reporting. BMJ. 2011;342:d1766.View ArticlePubMedGoogle Scholar
 The EConDA project. Costeffectiveness of interventions to prevent, screen and treat chronic diseases: A review, The EConDA Project. 2014. http://econdaproject.eu/publications_18_361135336.pdf. Accessed 25 Apr 2016.Google Scholar
 Trueman P, Anokye NK. Applying economic evaluation to public health interventions: the case of interventions to promote physical activity. J Public Health. 2013;35:32–9.View ArticleGoogle Scholar
 Asaria P, Chisholm D, Mathers C, Ezzati M, Beaglehole R. Chronic disease prevention: health effects and financial costs of strategies to reduce salt intake and control tobacco use. Lancet. 2007;370:2044–53.View ArticlePubMedGoogle Scholar
 Dallongeville J, Dauchet L, de Mouzon O, Réquillart V, Soler LG. Increasing fruit and vegetable consumption: a costeffectiveness analysis of public policies. Eur J Public Health. 2011;21:69–73.View ArticlePubMedGoogle Scholar
 Trogdon J, Finkelstein EA, Reyes M, Dietz WH. A returnoninvestment simulation model of workplace obesity interventions. J Occup Environ Med. 2009;51:751–8.View ArticlePubMedGoogle Scholar
 Howard P, Knight C, Boler A, Baker C. Costutility analysis of varenicline versus existing smoking cessation strategies using the BENESCO Simulation model: application to a population of US adult smokers. Pharmacoeconomics. 2008;26:497–511.View ArticlePubMedGoogle Scholar
 Bolin K, Wilson K, Benhaddi H, de Nigris E, Marbaix S, Mork AC, Aubin HJ. Costeffectiveness of varenicline compared with nicotine patches for smoking cessationresults from four European countries. Eur J Public Health. 2009;19:650–4.View ArticlePubMedGoogle Scholar
 Hoogendoorn M, Welsing P, Ruttenvan Mölken MPMH. Costeffectiveness of varenicline compared with bupropion, NRT, and nortriptyline for smoking cessation in the Netherlands. Curr Med Res Opin. 2008;24:51–61.View ArticlePubMedGoogle Scholar
 Jacobsvan der Bruggen MAM, Bos G, Bemelmans WJ, Hoogenveen RT, Vijgen SM, Baan CA. Lifestyle interventions are costeffective in people with different levels of diabetes risk: results from a modeling study. Diabetes Care. 2007;30:128–34.View ArticlePubMedGoogle Scholar
 Bemelmans W, van Baal P, WendelVos W, Schuit J, Feskens E, Ament A, Hoogenveen R. The costs, effects and costeffectiveness of counteracting overweight on a population level. A scientific base for policy targets for the Dutch national plan for action. Prev Med. 2008;46:127–32.View ArticlePubMedGoogle Scholar
 van Gils PF, Over EAB, Hambergvan Reenen HH, de Wit GA, van den Berg M, Schuit AJ, Engelfriet PM. The polypill in the primary prevention of cardiovascular disease: costeffectiveness in the Dutch population. BMJ Open. 2011;1, e000363.PubMedPubMed CentralGoogle Scholar
 Hoogenveen RT, van Baal PHM, Boshuizen HC. Chronic disease projections in heterogeneous ageing populations: approximating multistate models of joint distributions by modelling marginal distributions. Math Med Biol. 2010;27:1–19.View ArticlePubMedGoogle Scholar
 Hurley SF, Matthews JP. The Quit Benefits Model: a Markov model for assessing the health benefits and health care cost savings of quitting smoking. Cost Eff Resour Alloc. 2007;5:2.View ArticlePubMedPubMed CentralGoogle Scholar
 Hurley SF, Matthews JP. Costeffectiveness of the Australian National Tobacco Campaign. Tob Control. 2008;17:379–84.View ArticlePubMedGoogle Scholar
 Roux L, Pratt M, Tengs TO, et al. Cost effectiveness of communitybased physical activity interventions. Am J Prev Med. 2008;35:578–88.View ArticlePubMedGoogle Scholar
 Liew D, Lim SS, Bertram M, McNeil JJ, Vos T. A model for undertaking effectiveness and costeffectiveness analyses of primary preventive strategies in cardiovascular disease. Eur J Cardiovasc Prev Rehabil. 2006;13:515–22.View ArticlePubMedGoogle Scholar
 Lazar LD, Pletcher MJ, Coxson PG, BibbinsDomingo K, Goldman L. Costeffectiveness of statin therapy for primary prevention in a lowcost statin era. Circulation. 2011;124:146–53.View ArticlePubMedGoogle Scholar
 Pletcher MJ, Lazar L, BibbinsDomingo K, Moran A, Rodondi N, Coxson P, Lightwood J, Williams L, Goldman L. Comparing impact and costeffectiveness of primary prevention strategies for lipidlowering. Ann Intern Med. 2009;150:243–54.View ArticlePubMedGoogle Scholar
 Palmer AJ, Roze S, Valentine WJ, Spinas GA, Shaw JE, Zimmet PZ. Intensive lifestyle changes or metformin in patients with impaired glucose tolerance: modeling the longterm health economic implications of the diabetes prevention program in Australia, France, Germany, Switzerland, and the United Kingdom. Clin Ther. 2004;26:304–21.View ArticlePubMedGoogle Scholar
 Adarkwah CC, Gandjour A, Akkerman M, Evers SM. Costeffectiveness of angiotensinconverting enzyme inhibitors for the prevention of diabetic nephropathy in The Netherlandsa Markov model. PLoS One. 2011;6, e26139.View ArticlePubMedPubMed CentralGoogle Scholar
 Galani C, Schneider H, Rutten FFH. Modelling the lifetime costs and health effects of lifestyle intervention in the prevention and treatment of obesity in Switzerland. Int J Public Health. 2007;52:372–82.View ArticlePubMedGoogle Scholar
 Lindgren P, Lindström J, Tuomilehto J, Uusitupa M, Peltonen M, Jönsson B, de Faire U, Hellénius ML. Lifestyle intervention to prevent diabetes in men and women with impaired glucose tolerance is costeffective. Int J Technol Assess Health Care. 2007;23:177–83.View ArticlePubMedGoogle Scholar
 Peura P, Martikainen J, Soini E, Hallinen T, Niskanen L. Costeffectiveness of statins in the prevention of coronary heart disease events in middleaged Finnish men. Curr Med Res Opin. 2008;24:1823–32.View ArticlePubMedGoogle Scholar
 Menn P, Leidl R, Holle R. A lifetime Markov model for the economic evaluation of chronic obstructive pulmonary disease. Pharmacoeconomics. 2012;30:825–40.View ArticlePubMedGoogle Scholar
 Pignone M, Earnshaw S, Tice JA, Pletcher MJ. Aspirin, statins, or both drugs for the primary prevention of coronary heart disease events in men: a costutility analysis. Ann Intern Med. 2006;144:326–36.View ArticlePubMedGoogle Scholar
 Smith KJ, Hsu HE, Roberts MS, Kramer MK, Orchard TJ, Piatt GA, Seidel MC, Zgibor JC, Bryce CL. Costeffectiveness analysis of efforts to reduce risk of type 2 diabetes and cardiovascular disease in southwestern Pennsylvania, 2005–2007. Prev Chronic Dis. 2010;7:A109.PubMedPubMed CentralGoogle Scholar
 Gulliford MC, Charlton J, Bhattarai N, Charlton C, Rudisill C. Impact and costeffectiveness of a universal strategy to promote physical activity in primary care: populationbased cohort study and Markov model. Eur J Heal Econ. 2014;15:341–51.View ArticleGoogle Scholar
 Ward S, Lloyd Jones M, Pandor A, Holmes M, Ara R, Ryan A, Yeo W, Payne N. A systematic review and economic evaluation of statins for the prevention of coronary events. Health Technol Assess. 2007;11(1–160):iii–iv.PubMedGoogle Scholar
 Dalziel K, Segal L. Time to give nutrition interventions a higher profile: costeffectiveness of 10 nutrition interventions. Health Promot Int. 2007;22:271–83.View ArticlePubMedGoogle Scholar
 Palmer AJ, Tucker DMD. Cost and clinical implications of diabetes prevention in an Australian setting: a longterm modeling analysis. Prim Care Diabetes. 2012;6:109–21.View ArticlePubMedGoogle Scholar
 The CDC Diabetes Costeffectiveness Group. Costeffectiveness of Intensive glycemic control, intensified hypertension control, and serum cholesterol level reduction for type 2 diabetes. JAMA. 2002;287:2542.View ArticleGoogle Scholar
 Herman WH, Hoerger TJ, Brandle M, Hicks K, Sorensen S, Zhang P, Hamman RF, Ackermann RT, Engelgau MM, Ratner RE. The costeffectiveness of lifestyle modification or metformin in preventing type 2 diabetes in adults with impaired glucose tolerance. Ann Intern Med. 2005;142:323–32.View ArticlePubMedPubMed CentralGoogle Scholar
 Homer JB, Hirsch GB. System dynamics modeling for public health: background and opportunities. Am J Public Health. 2006;96:452–8.View ArticlePubMedPubMed CentralGoogle Scholar
 Kaplan EH, Craft DL, Wein LM. Emergency response to a smallpox attack: the case for mass vaccination. Proc Natl Acad Sci U S A. 2002;99:10935–40.View ArticlePubMedPubMed CentralGoogle Scholar
 Dangerfield BC. System dynamics applications to European health care issues. J Oper Res Soc. 1999;50:345–53.View ArticleGoogle Scholar
 Macmillan A, Connor J, Witten K, Kearns R, Rees D, Woodward A. The societal costs and benefits of commuter bicycling: simulating the effects of specific policies using system dynamics modeling. Environ Health Perspect. 2014;122:335–44.PubMedPubMed CentralGoogle Scholar
 Hoogenveen RT, van Baal PH, Boshuizen HC, Feenstra TL. Dynamic effects of smoking cessation on disease incidence, mortality and quality of life: The role of time since cessation. Cost Eff Resour Alloc. 2008;6:1.View ArticlePubMedPubMed CentralGoogle Scholar
 Weinstein MC, Coxson PG, Williams LW, Pass TM, Stason WB, Goldman L. Forecasting coronary heart disease incidence, mortality, and cost: the Coronary Heart Disease Policy Model. Am J Public Health. 1987;77:1417–26.View ArticlePubMedPubMed CentralGoogle Scholar
 Karnon J, Stahl J, Brennan A, Caro JJ, Mar J, Möller J. Modeling using discrete event simulation: a report of the ISPORSMDM Modeling Good Research Practices Task Force4. Med Decis Making. 2012;32:701–11.View ArticlePubMedGoogle Scholar
 Fone D, Hollinghurst S, Temple M, Round A, Lester N, Weightman A, Roberts K, Coyle E, Bevan G, Palmer S. Systematic review of the use and value of computer simulation modelling in population health and health care delivery. J Public Health Med. 2003;25:325–35.View ArticlePubMedGoogle Scholar
 Siebers PO, Macal CM, Garnett J, Buxton D, Pidd M. Discreteevent simulation is dead, long live agentbased simulation! J Simul. 2010;4:204–10.View ArticleGoogle Scholar
 Schlessinger L, Eddy DM. Archimedes: a new model for simulating health care systemsthe mathematical formulation. J Biomed Inform. 2002;35:37–50.View ArticlePubMedGoogle Scholar
 Eddy DM, Schlessinger L, Kahn R. Clinical outcomes and costeffectiveness of strategies for managing people at high risk for diabetes. Ann Intern Med. 2005;143:251–64.View ArticlePubMedGoogle Scholar
 Haby MM, Vos T, Carter R, Moodie M, Markwick A, Magnus A, TayTeo KS, Swinburn B. A new approach to assessing the health benefit from obesity interventions in children and adolescents: the assessing costeffectiveness in obesity project. Int J Obes (Lond). 2006;30:1463–75.View ArticleGoogle Scholar
 Forster M, Veerman JL, Barendregt JJ, Vos T. Costeffectiveness of diet and exercise interventions to reduce overweight and obesity. Int J Obes (Lond). 2011;35:1071–8.View ArticleGoogle Scholar
 Cobiac LJ, Vos T, Veerman JL. Costeffectiveness of interventions to promote fruit and vegetable consumption. PLoS One. 2010;5, e14148.View ArticlePubMedPubMed CentralGoogle Scholar
 Cobiac LJ, Vos T, Barendregt JJ. Costeffectiveness of interventions to promote physical activity: a modelling study. PLoS Med. 2009;6, e1000110.View ArticlePubMedPubMed CentralGoogle Scholar
 Cobiac L, Vos T, Veerman L. Costeffectiveness of weight watchers and the lighten up to a healthy lifestyle program. Aust N Z J Public Health. 2010;34:240–7.View ArticlePubMedGoogle Scholar
 Cobiac LJ, Vos T, Veerman JL. Costeffectiveness of interventions to reduce dietary salt intake. Heart. 2010;96:1920–5.View ArticlePubMedGoogle Scholar
 Murray CJL, Lauer JA, Hutubessy RCW, Niessen L, Tomijima N, Rodgers A, Lawes CMM, Evans DB. Effectiveness and costs of interventions to lower systolic blood pressure and cholesterol: a global and regional analysis on reduction of cardiovasculardisease risk. Lancet. 2003;361:717–25.View ArticlePubMedGoogle Scholar
 Rubinstein A, García Martí S, Souto A, Ferrante D, Augustovski F. Generalized costeffectiveness analysis of a package of interventions to reduce cardiovascular disease in Buenos Aires, Argentina. Cost Eff Resour Alloc. 2009;7:10.View ArticlePubMedPubMed CentralGoogle Scholar
 Blakely T, Cobiac LJ, Cleghorn CL, Pearson AL, van der Deen FS, Kvizhinadze G, Nghiem N, McLeod M, Wilson N. Health, health inequality, and cost impacts of annual increases in tobacco tax: multistate life table modeling in New Zealand. PLoS Med. 2015;12, e1001856.View ArticlePubMedPubMed CentralGoogle Scholar
 Divajeva D, Retat L, Shaw A, Brown M, Webber L. Risk factor based modelling for Public Health England. London: UK Health Forum; 2014. http://nhfshare.heartforum.org.uk/RMAssets/UKHFreports/PHE%20Report_October%202014%20final.pdf. Accessed 25 Apr 2016.Google Scholar
 Sassi F, Cecchini M, Lauer JA, Chisholm D. Improving lifestyles, tackling obesity: the health and economic impact of prevention strategies, OECD Health Working Papers, No. 48. 2009. http://www.oecdilibrary.org/socialissuesmigrationhealth/improvinglifestylestacklingobesitythehealthandeconomicimpactofpreventionstrategies_220087432153. Accessed 25 Apr 2016.View ArticleGoogle Scholar
 Cecchini M, Sassi F, Lauer JA, Lee YY, GuajardoBarron V, Chisholm D. Tackling of unhealthy diets, physical inactivity, and obesity: health effects and costeffectiveness. Lancet. 2010;376:1775–84.View ArticlePubMedGoogle Scholar
 Trueman P, Haynes SM, Felicity Lyons G, et al. Longterm costeffectiveness of weight management in primary care. Int J Clin Pract. 2010;64:775–83.View ArticlePubMedGoogle Scholar
 Bertram MY, Lim SS, Barendregt JJ, Vos T. Assessing the costeffectiveness of drug and lifestyle intervention following opportunistic screening for prediabetes in primary care. Diabetologia. 2010;53:875–81.View ArticlePubMedGoogle Scholar
 Kang HY, Ko SK, Liew D. Results of a Markov model analysis to assess the costeffectiveness of statin therapy for the primary prevention of cardiovascular disease in Korea: the Korean IndividualMicrosimulation Model for Cardiovascular Health Interventions. Clin Ther. 2009;31:2919–30.View ArticlePubMedGoogle Scholar
 Johansson P, Ostenson CG, Hilding AM, Andersson C, Rehnberg C, Tillgren P. A costeffectiveness analysis of a communitybased diabetes prevention program in Sweden. Int J Technol Assess Health Care. 2009;25:350–8.View ArticlePubMedGoogle Scholar
 Feldman I, Hellström L, Johansson P. Heterogeneity in costeffectiveness of lifestyle counseling for metabolic syndrome risk groups primary care patients in Sweden. Cost Eff Resour Alloc. 2013;11:19.View ArticlePubMedPubMed CentralGoogle Scholar
 Basu S, Andrews J. Complexity in mathematical models of public health policies: a guide for consumers of models. PLoS Med. 2013;10, e1001540.View ArticlePubMedPubMed CentralGoogle Scholar
 Sculpher M, Fenwick E, Claxton K. Assessing quality in decision analytic costeffectiveness models. A suggested framework and example of application. Pharmacoeconomics. 2000;17:461–77.View ArticlePubMedGoogle Scholar
 Whitty CJM. What makes an academic paper useful for health policy? BMC Med. 2015;13:301.View ArticlePubMedPubMed CentralGoogle Scholar