Abstract Diabetes incidence and prevalence remain at record highs in the United States. Understanding diabetes disease progression and how it varies among America’s heterogeneous population is critical, given unequal risks and outcomes for individuals of different racial/ethnic groups. Diabetes outcome prediction and simulation models allow prediction of a person’s risk for diabetes complications and death. A recent review of 19 such models found that the majority—16 models—relied at least partly on transition functions developed by the United Kingdom Prospective Diabetes Study (UKPDS). The UKPDS draws on data from a trial that began in 1977 and involved 5100 patients who were followed for a total of 89,760 person years. The sample consisted of mostly white British citizens. Only 8% and 10% of the UKPDS sample were Indian Asian and Afro- Caribbean patients, respectively. The major racial/ethnic groups that make up the US population were not included, and the variables studied in the UKPDS did not include any behavioral data. Long term, longitudinal patient data on diabetes outcomes is costly to collect and all information on the UKPDS Outcomes Models has been transparently reported and made publicly available. This has left the UKPDS risk models as the best option for many risk engines, despite the small, dated and nondiverse sample that it is based on. Capitalizing on Kaiser Permanente Southern California (KPSC) Electronic Health Records (EHR) data and legacy data systems, we identified over 527,000 patients with incident diabetes that were diagnosed and treated at KPSC from 1993 to 2020. Our sample provides more than 4.4 million person-years of follow up. More than 34,000 patients could be followed up for 21 or more years. The incident diabetes cohort from KPSC is 34.4% Hispanic, 10.6% Asian and 12.7% African American or Black allowing us to update the risk equations for all UKPDS outcome models by major race-ethnicity groups directly relevant for the U.S population. These updated models will allow us to identify disparities in diabetes, assure statistical fairness, and improve prediction of diabetes outcomes for diverse population groups. Because diabetes outcomes are largely influenced by health behaviors, we will also analyze behavioral data captured in the EHR including data on exercise and referrals to diabetes and weight management education classes. We will use cutting edge parametric, semi-parametric and non-parametric models to re- estimate risk equations using standard split sample cross-validation. We will report our methodology and results transparently in the same format as the UKPDS. Our study will help to update existing simulation models and support more timely and equitable clinical decision support and patient education.