PP36 Analysis of population-based health forecasting mathematical simulation models for obesity: results of a systematic literature review

Background Population-based health mathematical forecasting models provide useful information about how changes in different variables may impact the future health of individuals based on the population’s characteristics. They can be used to simulate hypothetical scenarios to evaluate potential effects and costs of different interventions. The objective of this study was to review the different mathematical simulation models and their methods used to forecast obesity prevalence in the general population. Methods A systematic review was carried out searching PUBMed and Embase/Ovid online databases for papers published between 1980 and January 2015. Articles were included if they contained a simulation modelling technique or mathematical projections, evaluated an intervention or scenario (lifestyle modification intervention, or reduction of risk factors applicable at a population level) and with projections made within a country, state level or subgroups within the population. The papers had to include as primary outcomes: weight, BMI classification, and/or waist circumference. Secondary outcomes were measures of NCDs, health outcomes or economic outcomes. The articles were categorised in two groups according to the mathematical methods used for their projections: mathematical projections (MP) and simulation models (SM). Results A total of 45 articles were selected for review. MPs methods of choice were analysis of trend, regressions or time series. MPs calculated only a limited number of possible outcomes and had reduced flexibility for evaluating possible scenarios. The most frequent type of SM used was microsimulation, which was used as a modelling method in 10 of the reviewed articles, and was the preferred model for forecasting, and evaluation of possible policies or interventions. The most common modelling technique found in the SMs were Markov process and Monte Carlo simulations . However, eight of these studies were “combined models”. These forecasting models used MP as a subsidiary method to calculate future projections of BMI and then used the results as input for a SM to calculate projections of obesity complications. Conclusion Forecasting obesity simulation models varied greatly in the methods used to project their results and in the number of outcomes. It is difficult to decide which type of model is better as the models are commonly built based on the specific policy question to answer. Researchers should choose their obesity model taking into account the information they have available and the outcomes of interest.