Mancl L A, Leroux B G, DeRouen T A
Department of Dental Public Health Sciences, University of Washington, Seattle 98195-7475, USA.
J Dent Res. 2000 Oct;79(10):1778-81. doi: 10.1177/00220345000790100801.
The evaluation of risk factors in dental research frequently uses observations at multiple sites in the same patient. For this reason, statistical methods that accommodate correlated data are generally used to assess the significance of the risk factors (e.g., generalized estimating equations, generalized linear mixed models). In applications of these methods, it is typically assumed (implicitly, if not explicitly) that between-subject and within-subject comparisons will produce the same estimated effect of the risk factor. When between- and within-subject comparisons conflict, the statistical methods can give biased estimates or results that are difficult to interpret. For illustration, we present two examples from periodontal disease studies in which different statistical methods give different estimates and significance levels for a risk factor. Statistical analyses in dental research should assess whether different sources of information give similar conclusions about risk factors or treatments.
牙科研究中对风险因素的评估经常采用对同一患者多个部位的观察。因此，通常使用适用于相关数据的统计方法来评估风险因素的显著性（例如，广义估计方程、广义线性混合模型）。在这些方法的应用中，通常（即使未明确说明也是隐含地）假定受试者间和受试者内比较将产生相同的风险因素估计效应。当受试者间和受试者内比较出现冲突时，统计方法可能会给出有偏差的估计或难以解释的结果。为了说明这一点，我们给出牙周病研究中的两个例子，其中不同的统计方法对一个风险因素给出了不同的估计和显著性水平。牙科研究中的统计分析应评估不同信息来源对于风险因素或治疗方法是否给出相似的结论。
