Comparing Four Caries Detection Methods in a School Setting

Methods: Eight dentists assessed dental disease levels in thirty-five school children, using one of four different caries detection methods: three in current use within the Scottish National Dental Inspection Programme (NDIP) and the other in use internationally [International Caries Detection System (ICDAS)]. Two dentists were allocated to each method. The detection systems were labelled NDIP Basic, AMR (Active caries/Missing/Restored), NDIP Detailed, and ICDAS. Twenty-five children were examined by all dentists. There were four outcome variables: caries experience (dentinal threshold) (methods: Basic, AMR, Detailed and ICDAS; eight comparisons), numbers of decayed, missing, and filled teeth (D₃MFT/d₃mft) (methods: AMR, NDIP Detailed, and ICDAS; six comparisons) and numbers of decayed or filled surfaces (D₃FS/d₃fs) (NDIP Detailed, and ICDAS; four comparisons) and the time taken to complete the assessment (eight comparisons). Statistical analysis was carried out using a generalised estimating equation (GEE) approach (applicable to both discrete and continuous outcomes). Overall caries experience was modelled using a binomial with probit link function and a negative binomial to model mean D₃MFT/d₃mft and D₃FS/d₃fs counts.

Results: Marginal plots and summary statistics for each outcome variable (except time taken) were found to be similar regardless of which dentist was performing the assessment. When comparing the different assessment methods, however, statistically significant differences were found between estimations of overall caries experience (Wald χ²=29.09, df=6, p<0.000), D₃MFT/d₃mft count (Wald χ²=23.51, df=5, p<0.000), and time taken (Wald χ²=368.60, df=7, p<0.000).

Conclusions: It is concluded that there were differences between assessments depending on the criteria employed in the different methods. The GEE approach was found to be a useful tool for the quantisation of similarities and differences between groups of dependent data, in particular those cases where a standard normal distribution assumption is not appropriate.