Table 2.
Incidence (n/N) | AROC (95 % CI) | HL Chi-SQ statistic | P for HL | AIC | BIC | |
---|---|---|---|---|---|---|
Risk prediction models: variables | ||||||
Model 1: Variables from AUSDRISK a | 147/1655 | 0.76 (0.72,0.80) | 5.29 | 0.726 | 895 | 960 |
Model 2: Model 1 with variables from other risk models b | 147/1655 | 0.82 (0.79,0.86) | 4.84 | 0.775 | 847 | 987 |
Model 3: Model 2 with total testosterone (continuous variable) | 147/1655 | 0.82 (0.79,0.86) | 4.45 | 0.815 | 844 | 990 |
Model 4: Model 2 with total testosterone (<16 vs ≥16 nmol/L) | 147/1655 | 0.83 (0.79,0.86) | 3.97 | 0.860 | 846 | 992 |
Model 5: Backwards selection modelc | 147/1655 | 0.82 (0.78,0.85) | 5.43 | 0.711 | 825 | 885 |
Sensitivity analyses | ||||||
Model 6: Model 4 without imputation (15.5 % missing) | 126/1399 | 0.82 (0.78,0.86) | NA | NA | NA | NA |
Model 7: Model 4 for NWAHS cohort | 62/820 | 0.79 (0.74,0.85) | NA | NA | NA | NA |
Model 8: Model 4 for FAMAS cohort | 85/835 | 0.84 (0.79,0.88) | NA | NA | NA | NA |
Notes: AROC area under receiver operating characteristic curve (imputation with median AROC reported), HL Hosmer-Lemeshow, AIC Akaike information criterion; Bayesian information criterion, NA not applicable because we are not fitting a new model - just testing the existing model; a Age, ethnicity/country of birth, family history of diabetes, IFG instead of self reported high blood glucose, currently taking blood pressure medications, current smoking status, physical inactivity, waist circumference category; b Pre-diabetes, BMI category, diagnosed cardiovascular disease, high blood pressure, high triglycerides, low HDL-C, income category; c Backwards selection model started with all available variables included in Model 4, dropped if p > 0.25, re-entered if p < 0.20 (retained variables were family history of diabetes, currently taking blood pressure medications, current smoking, waist circumference category, pre-diabetes, high blood pressure, low HDL-C low serum testosterone <16 nmol/L); Net changes in AROC were: Model 5 vs Model 2, 0.0093 (95 % CI: −0.0032,0.0218, P = 0.1455)