Abstract
A tool is needed to distinguish type 1 diabetes (T1D) and type 2 diabetes (T2D) in adults with new-onset diabetes because correct classification is needed for correct diagnoses and treatments. Current classification methods are usually applied to biomarkers using binary or quantitative classification with a cut point and may not be adequately nuanced. Combinations of clinical features are not necessarily specific for classifying and may not always indicate a single diagnosis. A probabilistic decision tree classification tool with multiple branches per decision node is needed for adults with new-onset diabetes to avoid misdiagnosis of actual T1D as T2D, misdiagnosis of actual T2D or monogenic diabetes as T1D, and misclassified patients in future population health studies which will lead to incorrect conclusions and suboptimal patient outcomes.