Reducing the bias of probing depth and attachment level estimates using random partial-mouth recording
Reducing the bias of probing depth and attachment level estimates using random partial-mouth recording
复制标题
DOI：
10.1111/j.1600-0528.2006.00252.x
复制
发表时间：
2006-02-01
影响因子：
2.3
通讯作者：
Moss, K
中科院分区：
文献类型：
--
作者：
Beck, JD;Caplan, DJ;Moss, K
目的：评估随机和固定部分检查方法与全口检查方法相比，探测深度（PD）和临床附着水平（CAL）估计值的偏倚和精度。研究方法：PD和CAL计算在6个网站上多达28颗牙齿（被认为是没有偏见的金标准）和三个固定的网站选择方法（FSSM），导致部分检查的网站：Ramfjord方法，NHANES I和NHANES 2000中使用的NIDCR方法。最后，通过对以下数量的站点进行采样创建了七种随机站点选择方法（RSSM）：84、42、36、28、20、15、10和6。为了比较方法的偏倚和精密度，我们计算了相对偏倚百分比和相对误差。结果如下：RSSM的平均值，标准差（SD），相对偏差和相对误差的估计值几乎相同的全口检查，但SD略有增加时，少于28个网站的抽样和相对偏差和误差增加的方法少于20个网站。FSSM的相对误差很低，但相对偏差高得多，表明低估。具有最小偏差和误差的FSSM是Ramfjord方法，但Random 36方法具有较小偏差和较小相对误差。NHANES 2000方法是程度评分估计值（>= 3、4、5或5 mm PD或CAL的位点百分比）的偏倚和相对误差最低的FSSM，但随机方法抽样较少的位点也同样有效。FSSM和RSSM都低估了患病率，特别是不太频繁发生的疾病的患病率，但大多数RSSM比FSSM更不可能低估患病率。结论：减少偏倚和提高估计精度的承诺支持RSSM的持续开发和检查。
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 >= 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.