Repeatedly measured predictors: a comparison of methods for prediction modeling

Table 3 Characteristics of the methods for developing a prediction model with a longitudinal predictor

Method	Flexible with missing values	Flexible with timing of measurements	Encompasses all information on the development of the predictor	Capable of dealing with a great number of repeated measurements	Capable of dealing with a small number of repeated measurements	Straightforward predictor computation (no additional steps that need to be performed before prediction model can be made)
1. All original measurements			+		+	+
2. Single “best” measurement				+	+	+
3. Summary (mean or maximum etc.)	+	+		+	+	*
4. Change between measurements		*	+		+	*
5. Conditional measurements			+		+	*
6. Growth curve parameters	+	+	+	+

⁺advantage that is present; *advantage that is partially present; an empty cell indicates an advantage that is not present. See discussion section "Methods to develop prediction models with a longitudinal predictor" for more information

ISSN: 2397-7523