Polygenic scores for longitudinal prediction of incident type 2 diabetes in an ancestrally and medically diverse primary care physician network: a patient cohort study

Table 2 Association of polygenic scores as a categorical variable (top 5% vs IQR) with incident T2D

	Scenario 1	Scenario 2	Scenario 3	Scenario 4
n of T2D cases (n in top 5%, n in IQR)	1073 (199, 874)	1002 (184,818)	827 (152, 675)	712 (132, 580)
Total n (n in top 5%, n in IQR)	8092 (736, 7356)	7518 (684, 6834)	5427 (494, 4933)	4032 (367, 3665)
Clinical variables only model
C-index	0.675	0.752	0.816	0.801
CRS HR (CI) (p-val)				1.71 (1.56–1.88) (1.6e−29)
PGS only model
C-index	0.608	0.607	0.603	0.608
PGS HR (CI) (p-val)	2.43 (2.08–2.85) (1.0e−28)	2.40 (2.04–2.83) (7.1e−26)	2.43 (2.03–2.91) (3.5e−22)	2.58 (2.13–3.14) (7.3e−22)
Combined clinical and PGS model
C-index	0.702	0.769	0.822	0.803
CRS HR (CI) (p-val)				1.71 (1.56–1.88) (6.6e−29)
PGS HR (CI) (p-val)	2.80 (2.39–3.28) (1.3e−37)	2.65 (2.25–3.12) (3.3e−31)	2.40 (2.0–2.88) (4.9e−21)	2.09 (1.72–2.55) (1.7e−13)
C-index improvement	0.027	0.017	0.006	0.002
LRT p-value	8.77E−31	6.12E−26	5.44E−18	7.09E−12

Longitudinal models were constructed with either clinical variables included in each scenario, polygenic scores (PGS) only, or both the clinical variables and PGS in a combined model. PGS were converted into a categorical value differentiating participants in the top 5% of the PGS compared to those in the interquartile range (IQR) of the PGS. Clinical risk factors in each scenario are as follows: scenario 1 age, sex; scenario 2 age, sex, BMI, family history of T2D, SBP; scenario 3 age, sex, BMI, family history of T2D, SBP, random glucose; scenario 4 age, sex, BMI, family history of T2D, SBP, triglycerides, total cholesterol, and HDL combined into a clinical risk score (CRS) and random glucose. We reported the concordance index (C-index), hazard ratio (HR) for the PGS for being in the top 5% compared to the IQR of the PGS or CRS per standard deviation depending on if they are included in the model, and the log-likelihood ratio test (LRT) p-value from comparing the difference in performance between the combined clinical and PGS model with the clinical variables only model

ISSN: 1756-994X