Dealing with misclassification and missing data when estimating prevalence and incidence of caries experience
Article first published online: 27 FEB 2012
© 2012 John Wiley & Sons A/S
Community Dentistry and Oral Epidemiology
Special Issue: 4th International Meeting: Methodological Issues in Oral Health Research – Intervention Studies
Volume 40, Issue Supplement s1, pages 28–35, February 2012
How to Cite
Mutsvari, T., García-Zattera, M. J., Declerck, D. and Lesaffre, E. (2012), Dealing with misclassification and missing data when estimating prevalence and incidence of caries experience. Community Dentistry and Oral Epidemiology, 40: 28–35. doi: 10.1111/j.1600-0528.2011.00663.x
- Issue published online: 27 FEB 2012
- Article first published online: 27 FEB 2012
- Submitted 5 November 2010; accepted 24 July 2011
- hidden Markov model;
Mutsvari T, García-Zattera MJ, Declerck D, Lesaffre E. Dealing with misclassification and missing data when estimating prevalence and incidence of caries experience. Community Dent Oral Epidemiol 2012; 40 (Suppl. 1): 28–35. © 2012 John Wiley & Sons A/S
Abstract – Objectives: The aim of this research was to estimate the prevalence and incidence of caries experience (CE) in first permanent molars while dealing with misclassification and missing of data.
Methods: CE was modeled as a Hidden Markov Model in which the response variable is subject to misclassification and missingness. The proposed analysis extends that of García-Zattera et al. (Stat Med 2010;29:3103) by allowing for various patterns of missing data. Findings were illustrated using data from the Signal Tandmobiel® study that is a longitudinal oral health intervention study.
Results: Differences in the parameter estimates were noted between models that take into account misclassification and missing data and those that do not. Unbiased parameter estimates of prevalence and incidence were obtained without the use of validation data. Models that include subjects with missing data have smaller standard deviations than models that do not.
Conclusions: It is important to account for misclassification to obtain less biased estimates of prevalence and incidence. For a proper estimation of prevalence and incidence in a longitudinal study subject to misclassification, validation data are not needed but when internal they can increase the efficiency in estimating the model. Also, including subjects with missing data increases the efficiency of estimating the parameters.