Development of statistical method to modeling tooth healthspan curves in Korean

Abstract

The duration of teeth remaining in mouth is being increased by oral health promotion and aging society. Thus it is necessary to develop oral health measure focus on health rather than disease experience such as DMFT index. Tooth healthspan curves visualize the decrease of sound teeth index (ST index), functioning teeth index (FST index), and present teeth index (PT index) using percentile curve. Previous versions of tooth healthspan curves were developed without applying weighting and appropriate smoothing methods. Most of all, the curves had limitation which could not provide accurate values because parametric models were not employed. This study aims to construct statistical models for tooth healthspan curves using representative national survey data, advanced statistical smoothing procedures, and LMS technique. Moreover, the dental health indices were compared with DMFT index to investigate their usefulness.Data collected by Korean National Health and Nutrition Examination Survey (KNHANES) IV and the first and second year of KNHANES V (2010-2011) was used. Subjects with interview of health and dental examination data aged 16 to 89 years were included. The empirical percentiles of ST index, FST index and PT index specified for sex and age groups were calculated. The empirical percentile curves were smoothed with locally weighted regression followed by 7-order polynomial regression procedures. The modified LMS procedure was performed to approximate smoothed curves and estimate transformation parameters. The multiple linear regression whose dependent variable were indices and explanatory variables were oral health related factors was performed. The results are listed as follows.1. The sex-specific smoothed percentile curves for ST index-for-age, FST index-for-age, and PT index-for-age were constructed. The smoothed percentile curves were closely matched to empirical percentile curves.2. The parameters lambda, sigma, and mu were estimated to enable computation of exact z-score and percentiles for every single values for ST index, FST index, and PT index.3. The explained proportion of variance was highest as of 10% in male and female in their 20''s and was decreased to only 5% in 80’s male and female by age when the dependent variable was DMFT and explanatory variables were age, family income, level of education, area of residence, perception of dental health and dental examination within 1 year in multiple linear regression. Otherwise, the explained proportion of variance were 9-26% and greater than any other indices when the dependent variable was the percentile of FST index for male. The explanatory variables were family income, level of education, area of residence, perception of dental health and dental examination within 1 year in multiple linear regression. For female, the explained proportion of variance were 8-10%, greater than any other indices in 20’s and 30’s when the dependent variable were ST index and the explained proportion of variance were 14-17%, greater than any other indices in 40''s to 70''s when the dependent variables were FST index.The study resulted in construction of tooth healthspan curves with sex and age group of specific estimated parameter L, M, and S to calculate exact z-score and percent in continuous manner. Moreover the percentiles of FST index were explained better than DMFT index especially in elderly population. The tooth healthspan curves provide comprehensive information in intuitive way. The curves can be used to provide highly accurate information for individual and expected to be a practical tool in clinical and public health setting providing diagnosis reference and to reach appropriate goal.