Читать книгу Clinical Dilemmas in Diabetes - Группа авторов - Страница 62

Ahlqvist et al. undertook a data‐driven cluster analysis in patients with newly diagnosed diabetes from a Swedish cohort. They utilized six variables measured at the time of diagnosis – age at diagnosis, BMI, HbA_1c, GAD antibodies, and homoeostatic model assessment (HOMA) estimates of β‐cell function and insulin resistance. The investigators identified five clusters based on these variables – one cluster was antibody positive, and relatively lean with poor metabolic control. The second cluster differed from the first in that antibodies were absent, while the third cluster had an elevated BMI and insulin resistance. Cluster 4 was also obese but defects in insulin action as measured by HOMA were not as severe as in the 3rd cluster. Finally, patients in the 5th cluster were older but, like cluster 4, had less severe metabolic abnormalities. These clusters were subsequently validated in 3 independent cohorts. Over subsequent follow‐up, the incidence of diabetic ketoacidosis and of non‐alcoholic fatty liver disease were highest in the first and third cluster, respectively. This latter cluster also had the highest risk of chronic kidney disease [54]. HOMA measures have significant limitations and poor predictive value for progression to diabetes [55] – although in fairness the investigators used C‐peptide rather than insulin to estimate β‐cell function. Whether this will help guide therapeutic interventions and improve outcomes in individual patients remains to be ascertained.

Another approach to characterizing different sub‐types of diabetes is based on common genetic variation and its influence on quantitative traits such as glucose tolerance in response to a standardized challenge. For example, Dimas et al. identified 5 clusters of genetic risk loci that altered glycemic traits [56]. In this study examining data from ~58 000 non‐diabetic subjects, one cluster altered insulin sensitivity. A second cluster altered fasting glucose while a third cluster altered the ratio of proinsulin to insulin in the fasting state. Another cluster was primarily characterized by defects in post‐challenge insulin secretion. The final cluster of risk loci did not alter glycemic traits. While this exercise has certainly helped understand the effect of genotype on phenotype, the effect size of each risk allele on a given phenotypic trait is so small as be unhelpful in terms of predicting individual clinical behavior.

Pharmacogenomics has also held promise as a way of classifying diabetes. This hope was reinforced by the discovery that the target for thiazolidinediones and sulfonylureas are risk loci for type 2 diabetes. Unfortunately, their effect on response to therapy is difficult to discern at an individual level and genotype at PPARG and KCNJ11 should not alter therapeutic choices in type 2 diabetes [57, 58].