Title
Longitudinal individual predictions from irregular repeated measurements data
Author
Eekhout, I.
van Buuren, S.
Visser, B.
Bink, M.C.A.M.
Huisman, A.
Publication year
2023
Abstract
Intensive longitudinal data can be used to explore important associations and patterns between various types of inputs and outcomes. Nonlinear relations and irregular measurement occasions can pose problems to develop an accurate model for these kinds of data. This paper focuses on the development, fitting and evaluation of a prediction model with irregular intensive longitudinal data. A three-step process for developing a prediction tool for (daily) monitoring and prediction is outlined and illustrated for intensive weight measurements in piglets. Step 1 addresses a nonlinear relation in the data by developing and applying a normalizing transformation. Step 2 addresses the intermittent nature of the time points by aligning the measurement times to a common time grid with a broken-stick model. Step 3 addresses the prediction problem by selecting and evaluating inputs and covariates in the model to obtain accurate predictions. The final model predicts future outcomes accurately, while allowing for nonlinearities between input and output and for different measurement histories of individuals. The methodology described can be used to develop a tool to deal with intensive irregular longitudinal data that uses the available information in an optimal way. The resulting tool demonstrated to perform well for piglet weight prediction and can be adapted to many different applications.
Subject
Measurements
Data
Prediction
Tools
Individuals
Patterns
Computational biology and bioinformatics
Computational models
Data processing
Epidemiology
Predictive medicine
Software
Statistical methods
To reference this document use:
http://resolver.tudelft.nl/uuid:a513cb45-0da5-4a09-9704-c707961d8c54
DOI
https://doi.org/10.1038/s41598-022-26933-1
TNO identifier
981544
Source
Scientific Reports, 13 (13)
Document type
article