Diabetes Prevention Program, for example, lifestyle intervention was effective at reducing diabetes incidence compared to placebo even among those with the highest quartile of T2D rsPS [78]. However, there is limited evidence to date that the communication of genetic risk is sufficient to motivate most individuals to undertake the kind of long-term behavioral modification required for sustained benefit [79-81]. There is also some (at least theoretical) risk of harm if the communication of risk information is mishandled. This could arise through failure to use ethnically appropriate scores, or to incorporate other relevant health information. For example, an overweight person with a low T2D polygenic score may be at far greater risk of disease than the polygenic score alone would suggest. Some individuals may be liable to interpret high genetic risk in a deterministic and fatalistic way, failing to appreciate that remediation of risk through lifestyle modification is no less likely to be effective in their case. Finally, there are questions related to implementation. Several countries (Finland, Estonia, UK, Taiwan, amongst others) are expanding the clinical roll-out of genome-wide genetic data, with plans to deliver genetic profiling to the population scale through a combination of sequence- and array-based strategies. Such universal availability of genomic data would open up much wider use of polygenic scores: the costs of acquiring such data (which only needs to be done once in the life of the individual) could be amortized across multiple applications (rather than needing to be justified based on any single indication) and the marginal costs of any specific use of those data would be minimal. Having said that, any valid assessment of clinical utility needs to consider the full costs of any given application: if the consequence of the unregulated use of genetic information is to identify a large proportion of the population as at high risk, there may be substantial financial and health costs to be incurred in follow-up screening, unnecessary treatment, patient stress, and the unproductive use of medical resources. A rigorous pipeline for the interpretation of these findings and their translation into evidence-based clinical interventions at the point of care will need to be created and deployed for multiple phenotypes across health care systems. 4) Partitioned polygenic risk scores So far, in this review, we have focused on the use of restricted (rsPS) and expanded (gePS) polygenic scores, both of which aim to capture the genetic contribution to predisposition for the major disease phenotypes conventionally used to define morbid states – such as T1D and 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 12 T2D. These scores are designed to enable prediction of an individual’s risk of developing of one of these forms of diabetes, or, as described above, to support differential diagnosis in those who have recently been diagnosed with diabetes. For these indications, it makes sense to combine as many risk variants as possible, irrespective of the mechanisms through which they influence that risk. However, these are not the only clinical questions that polygenic scores are equipped to address. Many of the most difficult problems in the clinical management of T2D, in particular, arise out of the clinical and phenotypic heterogeneity that is an obvious feature of this condition. Clinical management of someone with a diagnosis of T2D would be substantially improved if it were possible to sense how fast their diabetes is likely to progress, their propensity for developing macrovascular and microvascular complications, and their likely response to the range of treatments (therapeutic, surgical, and behavioral) that could be deployed to improve outcomes. Since these are questions that relate to clinical and etiological heterogeneity in those with established T2D, polygenic scores based on overall disease risk are unlikely to offer discriminatory value. As discussed earlier, one promising route to capture elements of this clinical heterogeneity is through the use of “partitioned risk scores” (pPS). These seek to “deconstruct” the overall (restricted or extended) polygenic score along biological axes that represent contributory etiological pathways, and thereby provide a framework upon which to map the variable response to clinical outcomes. One way of conceptualizing these pPS is in terms of the “palette” model of diabetes predisposition, which seeks to focus attention not on T2D itself, but on the various intermediary processes that collectively contribute to T2D-risk [3,38]. These include wellstudied processes such as obesity, fat distribution, islet development and function, and insulin sensitivity, though there are likely to be others that are, as yet, less clearly described. Each of these processes is itself under multifactorial (genetic and non-genetic) control, and a given individual may be positioned at any point on the spectrum from “low-T2D-risk” to “highT2D-risk” for each of these. Whilst the overall load of T2D-risk across the set of processes is likely to be a useful measure of the overall T2D-risk of an individual, the disposition of that risk