Table 2.

The predictive quality of prediction models developed using different methods to include longitudinal predictor BMI-SDS

Method	Model includes	Outcome at 10y
		Overweight		BMI-SDS
		Nk R2	AUC	R 2	AUC
1. All original measurements	BMI-SDS at age 0 days, 3 months, 6 months, 14 months, 2 years, 3 years, 5.5 years	0.244a	0.807a	0.339b	0.801b
2. Single ‘best’ measurement	BMI-SDS at age 5.5 years	0.230	0.799	0.329	0.799
3. Summary measurement	Mean (BMI-SDS at age 0 days, 3 months, 6 months, 14 months, 2 years, 3 years, 5.5 years)	0.168	0.767	0.238	0.767
3. Summary measurement	Maximum (BMI-SDS at age 0 days, 3 months, 6 months, 14 months, 2 years, 3 years, 5.5 years)	0.130	0.737	0.177	0.737
4. Change between measurements	BMI-SDS at age 0 days and BMI-SDS changes between ages 3m-0d, 6m-3m, 14m-6m, 2y-14m, 3y-2y, 5.5y-3y	0.244c	0.807c	0.339d	0.801d
5. Conditional measurements	BMI-SDS at age 0 days and conditional BMI-SDS at age 3 months, 6 months, 14 months, 2 years, 3 years, 5.5 years	0.244e	0.807e	0.348	0.806
6. Growth curve parameters	Mean and regression coefficients of the cubic growth curve ($\mathit{\text{mean}},b_{\mathit{age}},b_{{\mathit{age}}^2},b_{{\mathit{age}}^3}$)	0.241	0.803	0.337	0.803

Values are the explained variance of each prediction model developed in the broken stick dataset expressed in adjusted Nagelkerke R² (Nk R²) or adjusted R² (R²) and the area under the curve (AUC). The models predicting the dichotomous outcome overweight no/yes were analyzed using logistic regression. The prediction models predicting the continuous outcome BMI-SDS at age 10 were analyzed using linear regression

Due to collinearity: a. The model did not contain BMI-SDS at 5.5 years; b. The model did not contain BMI-SDS at 3 years; c. The model did not contain ΔBMI-SDS between 5.5_y-3_y; d. The model did not contain BMI-SDS at 0 days; e. The model did not contain conditional BMI-SDS at 5.5 years