data expands. However, it is already possible to identify a series of obstacles that need to be overcome before the full potential of this approach can be realized. The most critical is the need to ensure that the benefits of accurate, robust polygenic score determination are equally available to all. As others have pointed out, most GWAS and sequence data have been derived from the European-descent individuals who live in the developed nations of Europe and North America, and the polygenic scores generated from these data perform best when applied to the same populations [74,75]. There is a critical need to generate equivalent data and polygenic scores in other populations, to explore and characterize the extent to which transethnic portability of polygenic scores can be tolerated, and to define strategies for their deployment in special situations such as recently-admixed and isolate populations. Concerns about the impact of population stratification and the limits of transethnic portability provide arguments for the use of rsPS over gePS [74-77]. This may be particularly true for T1D and T2D given the limited increment in performance available with more extended scores. Wider recognition needs to be given that, for multifactorial traits with an appreciable nongenetic component, a wholly genetic explanation of disease prediction and state will never provide a perfect clinical instrument. In some settings, the information from genetics may simply recapitulate measures already available from other risk factors. The clinical use of cholesterol measures as a biomarker for CAD risk provides a counterexample, reflecting the benefits it offers as an integrator of both genetic and environmental risk. At the same time, some of those who are less enthusiastic about the clinical value of polygenic scores often fail to acknowledge that many established clinical tools (for example the use of BMI to predict T2D risk, or the use of islet cell antibodies for the differential diagnosis of T1D in late-onset diabetes) are likely to have performance metrics that limit their discriminative power. As the costs associated with the generation and interpretation of individual genomic information decline, there will be a growing roster of clinical applications where polygenic scores can add value. There is clearly a need to develop novel approaches to establish the clinical validity and utility of polygenic scores in medical practice that take account not just of the marginal cost of acquiring the data, but the full costs of implementation. Randomized clinical trials are unlikely to be the answer here, not least because the dynamic nature of the underpinning genetic databases means that polygenic scores are likely to evolve, rapidly rendering 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 18 redundant any precise quantification of cost and benefit based around a historical set of scores [106]. There will need to be concomitant efforts to document the provenance, content and performance of polygenic scores using standardized metrics and conventions which do not currently exist. There will need to be education of citizens and professionals to appreciate the benefits and limitations of polygenic scores [105]. It should be clear that genetics represents only one contributor to individual disease risk and profile, that genetically-defined risk should not, for multifactorial traits at least, be considered deterministic, and that most of the evidence indicates that behavioral modifications are just as likely to succeed (and in fact to be even more beneficial) in those at highest genetic risk [78]. The ease with which polygenic score information can be integrated with conventional approaches to risk profiling that are already widely used in clinical practice (e.g. to estimate future risk of CAD) should facilitate widespread introduction, and minimize the need for the health care professionals involved to develop an intimate knowledge of human genetics. It goes without saying that any clinical application of genetic data will need to fully address issues related to privacy and informed consent [107]. At the heart of precision medicine is the notion that an improved specification of disease risk or subtype will allow better targeted interventions to prevent or treat disease. Such efforts must compete for resources with population-based interventions that seek to achieve the same ends through non-targeted means [108]. In many existing clinical settings (e.g. related to reducing rates of cardiovascular disease, melanoma or breast cancer), these two strategies are seen to be complementary and are pursued in parallel. The development of polygenic scorebased approaches to support targeting of high-risk individuals will not alter these assessments. As now, the balance of effort between targeted and non-targeted approaches to the reduction of disease and disability will, for any clinical indication, continue to be dependent on the relative impact, cost, acceptability and sustainability of these complementary strategies. Box: Polygenic Score Terminology used in this article BOX: Polygenic score terminology 1. Restricted-to-significant Polygenic Scores (rsPS):