Reliability of single and double probing attachment level measurements.

Clarification of the reliability associated with probing attachment level measurements and with diagnostic rules derived from them is necessary to examine the potential of these measurements as reliable markers of the disease process and as an outcome measure in clinical treatment studies. The purpose of the present study was to describe and compare the longitudinal reliability of single and double probing attachment level measurable by estimating false positive and false negative rates. In 20 systemically healthy adults with untreated advanced adult periodontitis, probing attachment levels and probing depth were measured at 6 sites in all teeth at the start of the study, and every 30 days thereafter for 11 months. Attachment levels were double measured, using acrylic onlays providing reference points and an electronic pressure sensitive probe. Attachment loss from baseline of 2.5 mm or more was used to accept dynamic attachment loss. The statistical methods for the cross-sectional analysis included mean absolute differences between double measurements and cross-correlations. Longitudinal analyses, expressing estimates of diagnostic error rates were made using maximum likelihood methods. 5 measurement protocols were compared. The results showed that the mean absolute difference between replicate measurements was 0.095 mm in bicuspids and 0.107 mm in molars. The mean absolute differences decreased over time from 0. 128 mm in visit 3 to 0.08 mm in visit 7. Correlation coefficients for replicate measurements were higher than 0.98. In general, false positive rates were markedly lower (< or = 0.02) than false negative rates (< or = 0.31). The relative sample size required to obtain comparable statistical power, was minimized in premolars when using the first of 2 measurements or the mean of 2 measurements, and in molars when using the first of 2 measurements. The methodology used in the present study provides guidelines for designing clinical studies which maintain statistical power by balancing off examiner reliability and sample size.