Fig. 2.

Strategies for identifying diabetes subtypes. (a) Hierarchical (‘hard’) clustering distributes people into discrete subtypes. These clusters are defined using a series of traits, which may include phenotypic and/or genotypic criteria. (b) In a ‘soft’ clustering approach, discrete subtypes are also defined using a series of traits; however, people may have features belonging to more than one cluster. Clusters that represent a distinct pathobiological mechanism may be referred to as endotypes. (c) Alternatively, clinical traits may be integrated into a regression model, yielding a continuous measurement of various outcomes (e.g. response to a certain drug or risk of developing a certain complication). Clinical decisions (e.g. to start a certain medication) are implemented for people who fall above a specified threshold. This figure is available as part of a downloadable slideset