Reducing the bias of probing depth and attachment level estimates using random partial-mouth recording

Abstract

Objectives: To evaluate the bias and precision of probing depth (PD) and clinical attachment level (CAL) estimates of random and fixed partial examination methods compared with full-mouth examinations.

Methods: PD and CAL were calculated on six sites for up to 28 teeth (considered to be the gold standard with no bias) and three fixed-site selection methods (FSSMs) that resulted in a partial examination of sites: the Ramfjord method, and the NIDCR methods used in NHANES I, and NHANES 2000. Finally, seven random-site selection methods (RSSMs) were created by sampling the following number of sites: 84, 42, 36, 28, 20, 15, 10 and 6. To compare bias and precision of the methods we calculated percent relative bias and relative error.

Results: Estimates of means, standard deviations (SD), relative bias and relative error for RSSMs were almost identical to the full-mouth examination, but SDs increase slightly when fewer than 28 sites were sampled and relative bias and error increase for methods sampling fewer than 20 sites. The FSSMs had very low relative error, but much higher relative bias indicating underestimation. The FSSM with the smallest bias and error was the Ramfjord method, but the Random 36 method had less bias and less relative error. The NHANES 2000 method was the FSSM with the lowest bias and relative error for estimates of Extent Scores (percent of sites > or =3, 4, 5, or 5 mm PD or CAL) but random methods sampling fewer sites performed just as well. Both FSSMs and RSSMs underestimated prevalence, especially prevalence of less frequently occurring conditions, but most RSSMs were less likely to underestimate prevalence than the FSSMs.

Conclusion: The promise of reducing bias and increasing precision of the estimates support the continued development and examination of RSSMs.