University at Buffalo study has reported on the advantages of using artificial intelligence to better understand Type 2 diabetes across the US. The study describes how machine learning - a subset of AI that involves computers acting intelligently without being explicitly programmed - can help explore the prevalence of the disease, which effects more than 34 million Americans, as well as spot future trends. The outcomes were reported in the paper, ‘Geographically weighted machine learning model for untangling spatial heterogeneity of type 2 diabetes mellitus (T2D) prevalence in the USA’, published Scientific Reports.
Dr Zia Ahmed, a senior scientist and associate research professor at the UB RENEW Institute, who led the study said that the prevalence in the US varies substantially, primarily as a result of wide-ranging socioeconomic and lifestyle risk factors.
The study drew upon data reported in the Centers for Disease Control and Prevention's (CDC) US Diabetes Surveillance System, and the CDC's Behavioral Risk Factor Surveillance System. Additional data such as how six risk factors, namely, access to higher education, poverty, obesity, physical inactivity, access to exercise areas like public parks and access to healthy food - came from the US Census Bureau's Population Estimates Program.
The machine learning programme the research team employed, a geographically weighted random forest (GW-RF), a tree-based non-parametric machine learning model, which they believed may help explore and visualise the relationships between T2D and risk factors at the county-level. GW-RF outputs were compared to global (RF and OLS) and local (GW-OLS) models between the years of 2013–2017 using low education, poverty, obesity, physical inactivity, access to exercise, and food environment as inputs.
The results indicate that a non-parametric GW-RF model shows a high potential for explaining spatial heterogeneity of, and predicting, T2DM prevalence over traditional local and global models when inputting six major risk factors. Some of these predictions, however, are marginal. These findings of spatial heterogeneity using GW-RF demonstrate the need to consider local factors in prevention approaches. Spatial analysis of T2DM and associated risk factor prevalence offers useful information for targeting the geographic area for prevention and disease interventions, the researchers noted.
The added that this model outperforms existing methods and believe that the findings could lead to more tailored and effective prevention strategies from a policy perspective, which is critical given the projected increase of diabetes.
“The results of this study may also present opportunities for focused epidemiologic research at the county level to better understand the mechanisms driving T2D prevalence in various regions. The findings of this study may lead to more tailored and effective prevention strategies from a policy perspective, which is critical, given the projected prevalence increase of diabetes in the coming decades,” they concluded. “Understanding the spatial heterogeneity of the associations between T2D and risk factors may enable more advanced research and policy development to address the underlying, spatially varying contributors to T2D across US counties.
"Zia has comprehensive training, research, and supervisory experience in geospatial data sciences in agricultural, health, and environmental applications,” said Amit Goyal, SUNY Distinguished Professor and founding director of UB's RENEW Institute. “His current research interest is explainable artificial intelligence (XAI) for exploring spatial heterogeneity of local contribution in prediction. His knowledge and skills in advanced data techniques and machine learning are impacting various focus areas of RENEW including environmental exposures, genomes and health.”
Ahmed has more than 20 years of experience in environmental modelling and data analysis. Areas of expertise include data mining; geographic information systems, remote/proximal sensing, and geostatistics; linear/non-linear model, mixed effect model, multivariate statistics and machine learning; and database management.
To access this paper, please click here