Longitudinal Models

Longitudinal Models

Coordinator: Prof. J.W.R. Twisk

The research line longitudinal models is based on three themes: applied longitudinal data analysis, clinical prediction models and missing data.

Applied longitudinal data analysis
Regarding applied longitudinal data analysis, much research is performed to compare different methods with each other in order to find the best way to analyse longitudinal data. Furthermore, research is performed to find the best way to analyse longitudinal data with specific distributions of the outcome variable, such as distributions with an excess of zeroes and longitudinal IRT methods. This resulted over the last years in many applied methodological papers and several textbooks (published by Cambridge University Press).

Clinical prediction models
Clinical prediction models are an important tool for applied researchers and clinicians. Prediction models involve combining information from patients to estimate the probability of a particular illness or outcome as accurately as possible . The problems which can occur when developing prediction models include the difficulty of selecting the most important predictors from a large number of variables. General applicability - i.e. the accuracy of the prediction model when applied to new patients in the future - is an important aspect. Also challenging is to develop and validate prediction model in longitudinal data designs with varying predictors and outcomes. How to develop and validate these models is a goal within our research line.

Missing data
Although researchers do their best to avoid missing data, it is a common problem in medical and epidemiological studies. There are several simple and advanced methods to deal with missing data. Simple solutions are to ignore the missing values and delete all cases with missing values from the analysis or to use a single regression model to estimate the missing values. An advanced method is Multiple Imputation with the Multivariate Imputation with Chained Equations (MICE). With Multiple Imputation several complete datasets are generated and results are pooled using special calculation rules (called Rubin's rules). These steps are challenging especially in multilevel and longitudinal study designs with missing values in covariates. Furthermore, not for all statistical test results pooling rules are available and it is often not clear which rules can be applied for which statistical tests. This makes missing data an important research topic within our research line.

Projects and Education

Mediation analysis

Judith Rijnhart

The topic of my PhD-project is the application of mediation analysis within epidemiological research. Mediation analysis is a statistical technique that can be used to unravel treatment effects and mechanisms of disease development. Traditionally mediation analysis was mostly used in psychology and sociology, but recently also epidemiologists started to show their interest in this analysis technique. In the past years, several methods for mediation analysis have been developed. In my PhD-project I am studying the differences and similarities between these methods  in different types of data situations (e.g. cross-sectional and longitudinal data). To do this, I apply the different methods for mediation analysis to both empirical and simulated data. The overall goal of my PhD-project is to aid epidemiologists in their method choice.


Marieke Welten

My PhD research is part of the ProCOR (Prediction Of Child CardiOmetabolic Risk) project. This project is a collaboration between the ErasmusMC, RIVM and VUmc and is funded by  ZonMW. The ProCOR project aims to develop dynamical prediction models to identify  0 to 6 year-old children with an increased risk of future (age 10 years) overweight, hypertension or prehypertension, low HDL-C levels, and/or high total cholesterol to HDL-C ratio. These models will be converted into a digitalized and user-friendly screening tool to be used by child health care, which can update the children's risk using newly obtained data from visitations. This way the ProCOR project would like to make a contribution to the primary prevention of overweight and cardiovascular disease in the future
The research of my PhD project is mostly focused on the development of a prediction model to identify 0 to 6 year-old children at high risk of becoming overweight in the future. Data from the PIAMA birth cohort will be used to develop this model.  Moreover, during my research I will also take a look at how longitudinal data of repeatedly measured predictors can best be analyzed to develop a dynamic prediction model.

Literature on the ProCOR project:
De Kroon ML, Wijga A, Vergouwe Y, et al. Prediction of Preadolescent Overweight and Poor Cardiometabolic Outcome in Children up to 6 Years of Age: Research Protocol. Eysenbach G, ed. JMIR Research Protocols. 2016;5(2):e85. doi:10.2196/resprot.5158. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4937175/

Sick leave due to LBP

Lisa Bosman

The research within my PhD-project is aimed at sick leave due to low back pain and is embedded at ArboNed. At first we developed a prediction model for the risk on sick leave due to low back pain, based on data registered at ArboNed. Secondly, we will develop a prediction model for the prognosis of sick leave due to low back pain with data from our cohort study Predict2Work. In this cohort study we are following employees that are sick-listed due to low back pain.
Furthermore, I would like to evaluate the effectiveness of treatments for employees who sick-listed due to low back pain, based on observational data by using new statistical methods.

Other PhD-projects
Rosalie Gorter
Alette Spriensma
Guus Panken
Dagmar Nieboer
Marieke van Hoffen
Maartje Klaver

Trynke Hoekstra
Iris Eekhout
Marieke ter Wee
Corne Roelen

Education is a very important activity for the research line. Within the post initial master education programme EpidM several courses are coordinated by researchers from the research line longitudinal models (i.e. longitudinal data analysis, multilevel analysis, mixed models, clinical prediction models and missing data). Besides that courses are given within the Bachelor Medical Science.