Summary statistics to assess binary liver enzymes in predicting incident T2D
| Variable | Best cut-offs | Multivariable model | |||
|---|---|---|---|---|---|
| Discrimination (AUC (95% CI)) | Calibration (AIC) | NRI | IDI | ||
| Base model | 0.74 (0.71 to 0.77) | 1394 | |||
| ALT | 21 | 0.75 (0.73 to 0.78)* | 1370 | 0.53 | 0.02 |
| Base model | 0.76 (0.72 to 0.80) | 613 | |||
| GGT | 23 | 0.77 (0.73 to 0.81)† | 609 | 0.41 | 0.01 |
| Base model | 0.76 (0.72 to 0.80) | 613 | |||
| ALT+GGT | – | 0.77 (0.73 to 0.81)† | 609 | – | – |
Base model included education level (no, primary school, secondary and above), weekly moderate-to-vigorous activity (<0.5, 0.5–3.9, and ≥4 hours/week), history of hypertension (yes, no), plasma concentrations of triglycerides (mmol/L) (tertiles) and high-density lipoprotein cholesterol (mmol/L) (tertiles), and BMI (continuous).
Multivariable model adjusted for all the variables included in the base model.
*ALT concentrations were measured in 1142 participants (571 case–control pairs). Compared with the base model, the increment in AUC value was statistically significant (p<0.05).
†GGT concentrations were measured in 510 participants (255 case–control pairs with cases having baseline HbA1c <48 mmol/mol (6.5%)). Compared with the base model (AUC, 0.76 (0.72–0.80)), increment in AUC value by adding GGT to the base model was not significant (p=0.36). Additionally, increment in AUC value by adding GGT together with ALT to the base model was not significant either (p=0.20).
AIC, Akaike information criterion; ALT, alanine aminotransferase; AUC, area under ROC curve; BMI, body mass index; GGT, γ-glutamyl transferase; IDI, integrated discrimination improvement; NRI, net reclassification improvement; T2D, type 2 diabetes.