Forecasting the Local Risk of Type 2 Diabetes and the Effects of Prevention Programs Using Spatial Microsimulation: Development and Use of the TropISM Model
MetadataShow full item record
Simulation modelling has become an important tool in social science research, though such models are less commonly used in population health. Spatial microsimulation models provide a unique way to estimate health outcomes at the small area level and forecast the local effects of potential health interventions. Spatial microsimulation models combine geographically rich census data with information rich survey data to generate synthetic small area populations containing the wealth of information available in both types of data. Model outputs can therefore be used to inform the local delivery of health promotion programs. This research developed and validated the Type 2 dIabetes Spatial Microsimulation ("TropISM") model of chronic disease risk factors and outcomes for the 140 neighbourhoods of metropolitan Toronto. The model was developed using Canadian census data from 2006 and the 2005 Canadian Community Health Survey. The five-year incidence of diabetes was estimated using the Diabetes Population Risk Tool (DPoRT 2.0), a population-level risk algorithm that forecasts disease incidence using risk factor information routinely collected in population health surveys, together with the synthetic TropISM population. By leveraging both models, it was possible to evaluate the effects of hypothetical weight loss interventions on potential reductions in diabetes incidence at the neighbourhood level. Synthetic, neighbourhood specific prevalence estimates of diabetes were also used to estimate potential spatial accessibility to diabetes education programs within metropolitan Toronto. Accessibility was estimated using a two-step floating catchment area model, a type of spatial interaction model used to estimate area specific provider-to-population ratios. Results indicate that although the TropISM model accurately replicated demographic characteristics of Toronto's 140 neighbourhoods, it underestimated the true prevalence of type 2 diabetes, hypertension, and heart disease among men. In addition, TropISM captured broad spatial patterns in disease prevalence, but was unable to capture the spatial variability in known prevalence assessed from administrate health databases maintained by the Institute for Clinical Evaluative Sciences. Irrespective of these limitations, when the DPoRT model used synthetic TropISM population to forecast diabetes incidence, the overall five-year forecast incidence was comparable to the true incidence of disease (4.9% vs. 5.8%, respectively; 95% uncertainty interval: 4.2%-5.9%). At the neighbourhood level, the true incidence of diabetes fell within the range of forecast uncertainty in 65 neighbourhoods, while forecast incidence was underestimated in 52 neighbourhoods, most of which were located in Scarborough and Etobicoke. Again, broad spatial patterns in forecast incidence were captured by TropISM even though forecasts did not capture the spatial variability in true incidence rates. When the synthetic TropISM population was used to assess the ex ante effects of population-level weight loss programs on the future incidence of diabetes in silico, the entire population of high-risk, overweight individuals having a body mass index ≥ 25 kg/m² would have to lose 17% of its baseline body weight to produce a reduction in diabetes incidence of just one percentage point across metropolitan Toronto. Greater reductions in incidence were observed in neighbourhoods comprised of larger proportions of visible minorities and immigrants, even though the baseline prevalence of overweight and obesity tended to be slightly lower in these neighbourhoods compared to the metropolitan average (39% vs. 41.2%, respectively). Finally, the two-step floating catchment area model was used with synthetic, neighbourhood-specific counts of type 2 diabetes to conduct an exploratory analysis of spatial accessibility to diabetes education programs located throughout metropolitan Toronto. Results point to a potential mismatch between population demand for services and potential spatial access. In particular, some neighbourhoods within Scarborough had relatively higher prevalence of type 2 diabetes but lower access to diabetes education programs while neighbourhoods within central Toronto tended to have greater spatial access and lower prevalence rates of type 2 diabetes. Disparities in service provision suggest additional resources could be devoted to diabetes management in high-prevalence, low service neighbourhoods. In light of its short-comings, TropISM model results suggest how spatial microsimulation models can be improved to produce more accurate neighbourhood-specific estimates of diabetes prevalence. Importantly, TropISM was able to capture broad spatial patterns in diabetes prevalence and incidence, providing insight into where weight loss programs may contribute to greater reductions in diabetes incidence and identifying factors that might influence those reductions. This information can be used to customize health promotion interventions to the particular needs of specific communities. In conclusion, the TropISM spatial microsimulation model was able to (a) predict the consequences of different weight loss programs on projected diabetes incidence and (b) identify potential mismatches between existing demand for health promotion programs and the geographic availability of those resources. This locally relevant information enables public health planners to better allocate scarce resources to communities of greatest need. This research therefore illustrates how spatial microsimulation modelling can be used as a spatial decision support tool for local public health planning.
Cite this version of the work
Pete Driezen (2017). Forecasting the Local Risk of Type 2 Diabetes and the Effects of Prevention Programs Using Spatial Microsimulation: Development and Use of the TropISM Model. UWSpace. http://hdl.handle.net/10012/11182