Fig. 1
Development and assessment of the LRS in the SAFHS cohort. a Manhattan plot showing the association of the entire lipidome with incident T2D. All results are from mixed effects Cox proportional hazards models that account for degree of relationship among participants. All models are adjusted for age, age2, sex, age x sex interaction, age2 × sex interaction, systolic and diastolic blood pressures, waist circumference, body mass index, total serum cholesterol, serum high-density lipoprotein cholesterol, serum triglycerides and use of anti-lipid and anti-hypertensive drugs. The dots are color-coded for different lipid classes. The blue horizontal line represents a nominal type I error rate of 0.05 while the red horizontal line represents a global, FDR-corrected type I error rate of 0.2. All lipid species above the red line were simultaneously included in stepwise regression models and the species that were retained in the final multivariate model are indicated at the top of the plot. b Association of the LRS with incident T2D. Based on the regression coefficients from this final model, we derived a LRS as 0 · 4176*i(Cer(d18:0/18:0))-0 · 3443*i(LPC(O-22:1)) + 0 · 5361*i(TG(16:0_18:0_18:1)) where i(L) represents the inverse-normalized plasma concentration of lipid species L. The individuals were classified into three groups based on tertiles of the LRS. Shown is a Kaplan-Meier plot of progression to T2D based on the LRS. The results of mixed effects Cox proportional hazards models are after adjusting for the same covariates as those listed in Panel A. RH, relative hazard. c Association of the LRS with incident T2D independent of prediabetes. The plot shows relative hazard (colored bars), 95 % confidence intervals (error bars) and p-values (rotated numbers above bars). Model 1 – stratified mixed effects Cox PH model with presence of prediabetes as the stratifying variable; Model 2 – mixed effects Cox PH model restricted to NGT participants only; Model 3 – Interactive model that included interaction terms between presence of prediabetes and the LRS tertiles. All models use participants in the lower tertile of LRS as the reference category. NGT, normal glucose tolerance; Prediab, presence of prediabetes. d Clinical value of the LRS as a biomarker of incident T2D. Plots show bars representing improvement in the indicated characteristic achieved by adding LRS to the color coded models. Three models were used: clinical score for Mexican Americans (pink bars), oral glucose tolerance test (yellow bars) and clinical score combined with oral glucose tolerance test (orange bars). Statistical significance of the improvement is shown by the rotated numbers above the bars. *, the plot shows decrease in this parameter as a measure of improvement; LRχ2, likelihood ratio χ 2 statistic; AIC, Akaike information criterion; Uno’s C, Uno’s C statistic for survival data; IDI, integrated discrimination improvement; NRI, continuous net reclassification index