Model | Terms in fitted model | Likelihood ratio statistic (degrees of freedom); P value*† | |
---|---|---|---|
Compared with age-only model | Compared with model indicated | ||
Overall variation between blood glucose groups | |||
A‡ | Age + BG (groups) | 47.0 (20); P < 0.001 | |
Linear and quadratic effects using blood glucose | |||
B | Age + blood glucose (linear) | 13.3 (1); P < 0.001 | |
C | Age + blood glucose (Linear) + blood glucose (quadratic) | 22.5 (2); P < 0.001 | 9.25 (1); P = 0.002 vs. model B |
Linear effect above threshold value (no effect below threshold) | |||
Age + blood glucose ≥3.8 (linear) | 16.7 (1); P < 0.001 | ||
Age + blood glucose ≥4.0 (linear) | 18.5 (1); P < 0.001 | ||
Age + blood glucose ≥4.2 (linear) | 21.9 (1); P < 0.001 | ||
Age + blood glucose ≥4.4 (linear) | 25.9 (1); P < 0.001 | ||
D | Age + blood glucose ≥4.6 (linear) | 28.4 (1); P < 0.001 | 5.46§ vs. model B |
Age + blood glucose ≥4.8 (linear) | 27.6 (1); P < 0.001 | ||
Age + blood glucose ≥5.0 (linear) | 25.5 (1); P < 0.001 | ||
Age + blood glucose ≥5.2 (linear) | 24.6 (1); P < 0.001 | ||
Age + blood glucose ≥5.4 (linear) | 22.3 (1); P < 0.001 | ||
Age + blood glucose ≥5.6 (linear) | 19.9 (1); P < 0.001 | ||
Linear effects above and below threshold value | |||
E | Age + blood glucose ≥4.6 (linear) + blood glucose <4.6 (linear) | 29.4 (2); P < 0.001 | 1.04 (1); P = 0.31 vs. model D |
Men with blood glucose ≥11.1 are excluded. Models compared using likelihood ratio statistics.
↵* The likelihood ratio statistic is −2 (difference in log likelihood of two models). This estimates the increase in variation explained compared with the model specified.
↵† Degrees of freedom for test is equal to the difference in the number of parameters between two models being compared and is given in parentheses.
↵‡ In model A, blood glucose is grouped: <3.2; 0.2-mmol/l intervals from 3.2 to 6.3, 6.4 to 6.7, 6.8 to 7.5, 7.6 to 8.9, and 9.0 to 11.0.
↵§ Models D and C are not nested models so that the difference in the variation explained by these two models cannot be tested formally.