2335

Type 2 diabetes mellitus (DM) is a complex multifactorial disease affecting over 30 million people in the United States (9.4% of the population). The last two decades have seen an increasing volume of research on the effects of air pollution (AP) on numerous health outcomes, including DM, with varied results. To address this, we employed an unsupervised ML algorithm, k-means clustering, to assess multiple AP components, which may show interactions between the constituents on health that traditional regression models don’t capture.K-means was performed on 53,284 observations collected by the US Environmental Protection Agency during 2003-2012 and downloaded from their website. The following are the AP constituents used for partitioning: carbon monoxide (CO), nitrogen oxide (NO), nitrogen dioxide (NO2), ozone (O3), sulfur dioxide (SO2), particulate matter with a diameter of 10µm or less and 2.5µm or less (PM10 and PM2.5, respectively). Change in annual DM incidence, data from the US Center for Disease Control and Prevention, was calculated by subtracting annual DM incidence from the following year for each US county, then matched to AP by year. The k-means analysis resulted in six clusters. The change in annual DM incidence was statistically different in all but two clusters. The cluster with the greatest change in DM incidence (0.19 per 1000) also had the highest concentrations of CO, NO, NO2, PM10, and PM2.5. Additionally, the mean SO2 level was greater than twice the mean SO2 for all observations. The cluster with the largest decrease in DM incidence (-0.19 per 1000) also had the lowest levels of CO, NO, NO2, PM10, PM2.5, and SO2.     Using an unsupervised k-means algorithm, we showed multiple AP components were related to increased incidence of DM even when average concentrations were below the National Ambient Air Quality Standards.