GD Slade*, S Petti

Previously we have reported that existing methods accounting for examiner error in longitudinal caries studies can bias estimates of caries risk. Objective: to evaluate the validity of a proposed correction factor for Beck’s probability-based adjustment (ADJ) of examiner errors (Beck et al., 1995) in a sample of children at high risk of caries. Methods: secondary analysis of data from a 1981 US school based, randomized controlled trial of toothpaste efficacy (Zacherl, 1981). Data were used from the study’s 438 children aged 10-11 years at their first exam. D(3)MFT at the 2nd examination was categorised into three levels (0 teeth, 1-3 teeth and 4+ teeth) and used as the risk predictor. The outcome was dichotomised caries increment measured between 2nd and 3rd exams. Three existing methods for computing caries increment were used: CRUDE, NET and Beck’s ADJ. We introduced a fourth method (cADJ) by adding a constant of 0.5 to both numerator and denominator parameters of Beck’s ADJ. Validity of each increment measure was evaluated by comparing examiner error predicted by all four methods with verifiable caries status at baseline (EXAM1) and at the 4th examination (EXAM4). Results: Compared with the reference group with D(3)MFT=0, the risk difference (RD) for subjects with D(3)MFT=1-3 was 0.19 using CRUDE, 0.13 using NET, 0.13 using ADJ and 0.20 using cADJ. Corresponding RDs for D(3)MFT=4+ were 0.24, 0.09, 0.23 and 0.24. When validated using EXAM4 data, 25% of EXAM3 increments were due to examiner error and the rate of error was positively correlated with D(3)MFT. NET and ADJ adjustments failed to correct for observed error for all three D(3)MFT groups, whereas cADJ adjusted correctly for children with D(3)MFT=0. Conclusions: cADJ improves on Beck’s method, producing estimates of effect that were not biased by baseline caries experience and more accurate adjustments for examiner error.

Note: * indicates presenter