Within the rapidly developing field of HIA no standard method on quantification has emerged yet 
, but three approaches predominate the field: regression based methods, quantitative risk assessment, and population health models. The regression based methods originated in econometrics and usually estimate the long term relationship between exposure (e.g. per-capita consumption) or proxy variables (e.g. tax rate on alcohol) and health outcomes of interest on an aggregate level, adjusting for further variables as suggested by (economic) theory. This approach usually takes only limited notice of underlying epidemiological mechanisms. Quantitative risk assessment, originating from (environmental) exposure assessment of toxic substances, makes explicit use of dose-response relationships derived through epidemiological studies. These approaches are usually static, i.e. do not account for changes over time in real-life populations. Population health models combine epidemiological evidence and insights on causality to dynamically quantify the effect of risk-factors on population health.
DYNAMO-HIA fills a gap among the already existing population health models that are suggested for application in HIA 
. Compared to existing models, DYNAMO-HIA strikes a balance between being sufficiently technically accurate and ensuring wide usability. Technically equal or more complex models – e.g. POHEM, ARMADA, RIVM-CDM – allow for greater flexibility in modeling but are not publicly available, and require highly specialized input data and proficiency in specialized programming languages (except ARMADA). More accessible models – e.g. PREVENT, Proportional Multi-state Life Table (MSLT), GBD – lack dynamic projection capabilities (except PREVENT and multiple cohort versions of the MSLT 
) and do not have explicit risk-factor states. This technical simplification ignores mortality selection and may lead to biased estimates 
DYNAMO-HIA is specially designed to fit within the standard framework of HIA, synthesizing elements of already well established modeling approaches. Our approach allows for a flexible risk-factor configuration (categorical, duration dependent, continuous); generic chronic diseases as specified by the user (with intermediate diseases, partially fatal diseases, and/or diseases with a cured fraction); arbitrary specification of– age and sex-specific – relative risks; and minimal data needs by requiring only population level data (see ). Furthermore, a mouse-driven graphic user interface allows straightforward handling of the software, i.e. no knowledge of a programming language is required. In addition to exporting the existing, partly customizable, graphs into files – e.g. detailed plots of mortality rates or prevalences of risk-factors or diseases, both over time and age-specific – most calculated data can be exported for use in separate software (e.g. Excel). These raw output data allow further analyses, such as grouping diseases into categories (e.g. IHD and stroke or all cancers), including costs, or constructing additional graphs.
Overview of required input data (age- and sex-specific).
DYNAMO-HIA simulates the effect of a single risk-factor on a population without migration. However, the categorical risk-factor can be used to partition the population in up to ten distinctive categories. For example, a population could be partitioned along a risk-factor – say, non-smokers and smokers – and socio-economic status – say, with and without college education – having in total four different groups to assess policies that are more successful for people with certain socio-economic status. The possibility of partitioning a population also allows quantification of the effect of an environmental hazard. In this case, for example, the population is partitioned according to their proximity to the hazard source – say, noise exposure or air pollution due to a new airport – with 5% of the total population living less than 5 km from the hazard source, 5% to 10% living less than 10 km and so on. This requires, of course, sufficient insight into which part of the population is affected and knowledge of the relative risks of the modeled exposure on the included diseases and total mortality.
A category may also represent a combination of known risk-factors: For example, smoking status and BMI – smoking/non-smoking and normal weight/overweight/obese – could be modeled by partitioning the population into six distinctive risk-factor categories. However, this would require knowledge about the relative risk of the combined risk-factor class – say, relative risk of being obese and a smoker on the included diseases and total mortality.
The overall performance of a model crucially depends on the quality of the input data. In particular for dynamic models, the epidemiological data has to be mutually consistent, otherwise projected changes in the prevalences might be caused by mismatching data and not by the changes in the risk-factors. A limitation is that an autonomous trend in the rates, e.g. annual reduction in overall mortality or disease incidence, cannot be specified. Autonomous trends are often observed for past time periods and caused by a number of factors; chief among them improved curative interventions and changed risk-factor behavior. In a risk-factor based model, however, the specification of a future autonomous trend must be net of any underlying risk-factor behavior, as this is already specified explicitly at some other place in the model. Such specific data on future trends is hardly reliably available, if at all, and would, in most cases, only modestly affect the difference between reference and intervention scenarios. Hence, an ordinal ranking of policy alternatives would be rarely affected while still revealing the most effective intervention.
In health impact assessment, three criteria are used to assess validity: formal validity, plausibility, and predictive validity 
Formal validity assesses the degree to which correct methods are applied correctly. The model structure of DYNAMO-HIA is well-founded in epidemiological evidence – incidence, prevalence, and excess mortality – and demographic modeling practice, i.e. a multistate Markov-type model of chronic disease with explicit risk-factor states and inclusion of intermediate diseases.
assesses the degree to which an observer deems the theoretical framework understandable, applicable, and plausible. Hence, DYNAMO-HIA deliberately restricts itself to the well-established causal chain “risk-factor exposure -> incidence ->prevalence -> disease-related mortality -> overall population health” and requires only data that is available in sufficient quality for the most common diseases (e.g. cancer, CVD, diabetes, COPD) and risk-factors (e.g. smoking, BMI, alcohol) in developed countries. In the Swedish application example, our results for the number of excess deaths is slightly lower than estimates based on a regression approach utilizing historical relationships and aggregate-data pooled from several Nordic countries 
. One reason for this difference lies in the relative risks on all-cause mortality used in our illustration. These are taken from epidemiological studies and capture only the effect of individual exposure, i.e. drinking behavior. Consequently, our results do not account for broader effects that a change in alcohol consumption may have on population health, i.e. abstainers or moderate drinkers becoming victims of increased alcohol-induced violence or accidents caused by the increased number of intoxicated drinkers.
Plausibility and well-established formal methods should not be mistaken for constantly delivering expected results. Dynamic projections may reveal counterintuitive yet correct results and, hence, lead to important insights. In the smoking application, for example, the number of breast cancer cases in the never-smoker scenario is larger than in the reference scenario, although smoking has no causal link to breast cancer incidence. This seemingly unexpected result is caused by an increase in overall longevity of a healthier living population and, hence, an increased number of females susceptible to breast cancer. This phenomenon is well known among modelers of health care costs; dynamic analysis has shown repeatedly that a population-level reduction in obesity or smoking may lead to higher health care costs in the long run 
is the degree to which predictions are confirmed by facts; according Veerman et al 
, however, this criterion usually cannot be established in the context of HIA. The sometimes decades-long time lag between a change in policy and a change in the corresponding health effects makes it difficult to conduct a full evaluation of the HIA prediction. Moreover, a HIA might influence policy in such a way as to (successfully) invalidate its own predictions.
We emphasize that a software model like DYNAMO-HIA is only a decision-support tool. It helps to quantify the expected differences in population health given two (or more) different scenarios: one of them a baseline scenario (without the intervention) and one (or more) scenario(s) with intervention(s). It does not predict the development of future population health as such. Decision makers must be constantly aware that real-world phenomena are necessarily more complex and that no model can predict future events with 100% accuracy. In HIA, it may be useful to avoid calling the results of mathematical models ‘predictions’, but rather projections
of “what if
” scenarios in a clearly defined and simplifying framework. The term ‘prediction’ should be reserved for the entire process, in which a software model is only one element of the utilized evidence 
Internal validity was extensively tested. To allow future thorough checking of cross validity by outside experts as well, the software and the source code are publicly available (www.DYNAMO-HIA.eu
). In its current form, DYNAMO-HIA also facilitates unproblematic one- and multi-way sensitivity analysis, by allowing easy manipulation of all input parameters. Like most other population health models, however, the current version of DYNAMO-HIA does not include a probabilistic sensitivity analysis (PSA). Implementing a PSA in population health models is time and cost intensive; the extra data needed to conduct a PSA are difficult to obtain and preparing them requires expert knowledge. However, DYNAMO-HIA can be used in batch mode, allowing users with sufficient computing skills to build a PSA shell around the software, if desired.
DYNAMO-HIA is available for free download and includes a data set covering a large number of EU countries (www.dynamo-hia.eu
). This internally consistent data set has prevalence data for three risk-factors (smoking, BMI, alcohol), nine diseases (incidence, prevalence, excess mortality), and population data (e.g. total mortality, projected number of newborns). This data set allows instant use of DYNAMO-HIA for the covered countries. However, DYNAMO-HIA is also usable with external data on other countries, (sub-)populations, diseases, or risk-factors. Furthermore, the already included data set can be easily updated when more recent data become available. DYNAMO-HIA can be used for a range of applications, in particular if additional data are available.
Recent application of DYNAMO-HIA include comparison of tobacco control scenarios 
; the effect of an increase in obesity levels for the Dutch population 
; the EU-wide gains in population health when increasing prices on alcohol 
; and the potential health gains and losses in the EU achievable through feasible prevalences of life-style related risk factors 
. The current focus of DYNAM-HIA is on policies at the national level, but the software can, in principle, also be used for applications at the regional or local level.
DYNAMO-HIA differs from other population health models for HIA 
in several important aspects. From the outset, it has been designed for public use within HIA-applications by featuring a user-friendly graphic interface, and employing a model structure that ensures accurate simulation using epidemiological evidence while having modest data needs.