Model development
A prediction model was developed using multivariable binary logistic
regression with backward stepwise selection, excluding variables with
p-values > 0.10. We excluded predictors with less than
0.010 contribution to total model performance, expressed in area under
the ROC curve (AUC). Categories were only replaced by calcium richer
subcategories when substitution led to an increase in AUC. We used the
final regression output to calculate individual probabilities of having
an adequate calcium intake.
Model discrimination was assessed using the AUC. We used the Hosmer and
Lemeshow test and Brier score as measures for goodness of fit. Internal
validity was assessed through a bootstrapping procedure using 5
bootstrap samples, resulting in optimism corrected estimates of model
performance, which are presented in the results section. Model
calibration was assessed by a calibration plot, comparing the predicted
probabilities of having an adequate calcium intake with the actual
proportion of participants with an adequate intake.