Methodological Issues in Oral Health Research

Main content

Pre-Conference Workshop

Lecture hall Cavum

How to extract more (and better quality) information from your study?

“Simple questions’ is not the same as simple analyses”

10 May 2016, Full Day Workshop, 09:30-17:00


  • Professor Emmanuel Lesaffre- L-Biostat, KU Leuven, Leuven, Belgium
  • Professor Dominique Declerck- KU Leuven, Leuven, Belgium



  • Make oral health researchers aware of the complexity of (most) oral health research data sets.
  • Show possibilities for extracting more out of oral health data sets.

Who should attend (target audience)?

  • Oral health researchers with some (but not elaborate) knowledge of statistics.
  • Epidemiologists and biostatisticians who wish to learn about the specific problems involved in analyzing oral health data.


  • Workshop
  • Using examples from different disciplines within oral health research as a starting point, participants will be invited to critically reflect on the methodological aspects of the research work. This will serve as a starting point to elaborate on following topics in more detail:
    • Multi-site and split-mouth studies: oral health data often exhibit an hierarchical structure (surfaces of a tooth within teeth, teeth within mouth, etc.) and require that the correlation of the data is taken into account. This involves multi-level analyses.
    • Misclassification issues: Rarely data are recorded without error. For instance, in caries research the caries experience status is often misclassified meaning that there are false positives and negatives in detecting caries. In most studies only a kappa statistic is reported, but no attempt is made to evaluate the effect of misclassification.
    • Time to event studies: longitudinal studies provide a wealth of information and are often considered superior to cross-sectional studies. However, longitudinal studies are also more demanding. Subjects need to be followed up in time to record an event (e.g. failure of an implant), but (luckily) not all subjects experience the event and then one states that the time to event is censored. Several types of censoring occur in practice and need to be taken into account in the statistical analysis.
    • Missing data: longitudinal studies also suffer from subjects that miss a planned examination and/or drop out. Ignoring the missing data mechanism most often implies wrong conclusions. Current statistical techniques are capable of dealing with such a problem.

For each of the above topics, the possible pitfalls of ignoring the problems by making use of a too simple statistical analysis will be exemplified. Examples from the research of the two course instructors are taken to present more appropriate statistical approaches. Emphasis in the course lies in the intuitive ideas and the math will be downplayed.


  • One-day pre-conference course

Number of participants

  • Limited to 25 participants

Course material

  • Case studies from literature
  • Handouts of teaching slides

Recommended literature

 Statistical and Methodological Aspects of Oral Health Research. Editors Lesaffre E, Feine J, Leroux B, Declerck D; Wiley (2009). ISBN: 978-0-470-51792-5


Post-Conference Workshop

Lecture hall Cavum

Analysis of longitudinal data

Mixed Effects Models

14 May 2016, 09:00-13:00


  • Postdoc Roy M Nilsen- University of Bergen, Bergen, Norway
  • Professor Stein Atle Lie- University of Bergen, Bergen, Norway


Simple linear models are the working horse in statistical analyses of continuous data in medicine and in research in general. For simple independent data structures t-tests, analyses of variance (ANOVA), and linear regression are often sufficient tools.

More complex dependent data structures have become more common later years in clinical medicine and dentistry. In this workshop we start on simple data structures and expand to more complex structures for repeated measures. The main focus will be on the interpretation of the results from more complex models with relation to those found in simpler analyses. The workshop will cover methods for clustered observations with the focus on repeated observations using mixed models. The statistical program Stata will be used for examples.


Recommended literature

Rabe-Hesketh, S. and Skrondal, A. (2012). Multilevel and Longitudinal Modeling Using Stata (Third Edition). College Station, TX: Stata Press. Volume I: Continuous Responses.