be identified whose pathophysiology is predominantly driven by one process, and in whom the monogenic paradigm of a drug targeting that very process could be applied effectively. Whether, and in whom, such approaches may prove successful will require the conduct of appropriately designed precision clinical trials. These pPS approaches to analysing phenotypic heterogeneity, which build out from genetic risk, offer a complementary perspective to the results emerging from the analysis of real world data [100, 101]. These real-world methods have focused on efforts to classify T2D into distinct subtypes, analogous to the categorization of monogenic forms of disease. Such an objective, if successful, would offer clinical expediency. However, these efforts to sift individuals into discrete subtypes of disease would appear to run counter to the evidence that points to a complex, graded, architecture of risk, one that is consistent with a multifactorial etiology, composed of genetic predisposition dominated by multiple common variants of modest effect, and pervasive exposures contributing to risk. In one recent study, Ahlqvist et al. used basic clinical information from patients with newlydiagnosed adult-onset diabetes, to define five subtypes of T2D: an autoimmune form (covering T1D and other related clinical entities), two severe forms (one dominated by insulin deficiency, the other by insulin resistance), and two milder forms (termed “obesityrelated“ and “age-related” diabetes) [101]. Whereas the genetic clusters that form the basis of pPS are defined at the level of the variants, these clinical subtypes are defined at the level of the individual, and based on biomarkers and clinical data gathered at a specific point in the progression of an individual from health disease. The latter is likely to limit their relevance to those who have not yet developed disease, and/or those who are on treatment. It is worth emphasizing the different, but complementary, nature of these two approaches: the partitioned risk approach involves first clustering genetic signals by mechanism to derive pPS, and then exploring how the quantitative pPS scores perform across individuals. In contrast, the phenotypic clustering approach attempts to hard cluster individuals on the basis of their physiology. Further work is required to understand how these two approaches to capturing clinical heterogeneity relate to each other, and to objective measures of clinical utility. One of the fundamental issues – which pervades diverse aspects of precision medicine – relates to the relative merits of retaining as much quantitative information on an individual as possible until the point when a substantive (typically binary) clinical decision needs to be made, as opposed to early diagnostic categorization of the individual in a way that bases subsequent clinical decision-making on the optimized outcomes of the group to which they have been assigned. While further investigation is needed, a recent analysis by Dennis et al. in the ADOPT and RECORD clinical trials indicated that the former approach – considering phenotypic traits as continuous measures – provided better predictive value of treatment response, than an approach that binned individuals using the phenotypic clustering approach of Ahlqvist et al [101, 102]. 5) Summary and further discussion After many years of frustration at the slow progress that had been made in the translation of recent discoveries in human genetics – notably the many risk variants for common, multifactorial forms of diabetes identified through GWAS and sequencing – there is now ADVANCE ARTICLE: Endocrine Reviews Downloaded from https://academic.oup.com/edrv/advance-article-abstract/doi/10.1210/er.2019-00088/5535575 by 81225740 user on 24 July 2019 ADVANCE ARTICLE Endocrine Reviews; Copyright 2019 DOI: 10.1210/er.2019-00088 17 growing optimism that the use of polygenic scores will offer substantial clinical benefit and contribute to efforts to forestall the growing morbidity and mortality associated with these conditions. Some early clinical applications have emerged – mostly related to positive identification of those who have developed, or at highest imminent risk of developing, T1D [57,65-68]. It is inevitable that clinical applications of the polygenic score approach will roll out at a different pace across disease conditions, with a focus on different clinical questions, dictated by the additional clinical benefit that they provide, and the extent of the unmet clinical need. One size certainly does not fit all, and the relative merits of the different types of polygenic score described in this review (gePS, rsPS, pPS) will differ according to the specific clinical situation. It also remains to be determined how or whether pPS and phenotypic trait clustering will impact clinical care and be deployed in practice. Recent developments in relation to the potential clinical use of polygenic scores have led to heated debate between those who are enthusiastic about the potential, and those who are of the view that the clinical value of human genetics discovery has been consistently hyped, and who feel that polygenic scores represent just the latest chapter in that story of scientific overselling [103,104]. As in other similar situations, the outcome of this debate will become clearer as theoretical and basic knowledge develops and the collection of real-world