Table 2.
Regression coefficients for the association between probit-transformed prevalence of diabetes based on HbA1c and probit-transformed prevalence based on FPG
| Coefficient (95% CI) | p value* | Univariate R2† | Semipartial R2‡ | ||
|---|---|---|---|---|---|
| Intercept | −1·761 (−2·229 to −1·266) | <0·0001 | NA | NA | |
| Probit-transformed prevalence of diabetes based on FPG | 0·799 (0·763 to 0·835) | <0·0001 | 0·915 | 0·075 | |
| Mean age of age–sex group (per 10 years older) | 0·052 (0·042 to 0·062) | <0·0001 | 0·601 | 0·011 | |
| Study midyear (per one more recent year since 2000) | 0·012 (0·009 to 0·015) | <0·0001 | 0·014 | 0·006 | |
| Natural logarithm of per person gross domestic product | 0·076 (0·035 to 0·114) | 0·0001 | 0·052 | 0·003 | |
| Mean BMI | 0·018 (0·010 to 0·027) | <0·0001 | 0·022 | 0·002 | |
| Study representativeness | .. | .. | 0·013 | 0·004 | |
| National | Reference | .. | .. | .. | |
| Subnational | −0·004 (−0·047 to 0·040) | 0·8758 | .. | .. | |
| Community | 0·090 (0·060 to 0·119) | <0·0001 | .. | .. | |
The appendix shows regional random effects. FPG=fasting plasma glucose.
p values using likelihood ratio test, which compares the likelihood of the models with and without the variable of interest.78
Calculated by regressing against each independent variable alone, without the regional random effect; equals the square of the correlation coefficient.
Is the decrease of R2 if one of the independent variables is removed from the full model; however, traditional R2 is not clearly defined for mixed-effect models, we have used the conditional R2 that describes the proportion of variance explained by both fixed and random factors.79 The overall conditional R2 for the model was 0·949.