皮肤毛孔照片诊断标准评价的潜在分类变量模型研究

Evaluating Standardized Photogram of Facial Pores and Diagnostic Accuracy of Tests Using Latent Class Models

本文在无金标准情况下探讨皮肤毛孔标准照片制定的合理性和可行性,对医师诊断正确性进行评价。按照毛孔粗大程度制定分类为5水平的毛孔标准照片。对128名女性志愿者制作鼻翼毛孔照片,5位年资相近的皮肤科医师按照诊断标准和标准照片对128例自愿者照片进行独立的等级评分。诊断结果数据采用潜在分类变量模型(Latent Class Model,LCM)进行分析,分别拟合5位医师诊断条件概率一致的模型和诊断条件概率不一致的模型。计算医师诊断的条件概率和后验概率。潜变量分析结果提示诊断标准过于细化且分类模糊,依据条件概率将原始分类重新划分为3类的模型较好地拟合了诊断数据。运用客观和准确的能够真实反应和区分个体情况的诊断标准是诊断试验评价的基础和前提。潜在分类模型能够有效地处理无金标准的诊断重复性或一致性研究数据。

This paper was designed to investigate the criteria of standardized pore photographs without gold standard and assess the diagnostic accuracy of dermatologists. We formulated the standardized pore photographs into 5 stages according to the magnitude of pores and made photograph of pores on nasal ala from 128 female volunteers. Five dermatologists with similar experience classified the 128 photographs into 5 stages with reference to the standardized photograph. Latent class model （LCM） was used to analyze the data. We established two LCMs, one with consistent conditional probability among the 5 dermatologists, the other with inconsistent conditional probability. Conditional probability and posterior probability were also calculated. The outcomes showed that the 5-stage diagnostic criterion was too detailed and ambiguous. The model fitted the data well after reclassifying the data into 3 stages according to conditional probability. The standardized criterion which is objective and accurate, and can truly reflect and differentiate the status of individual is essential to assess diagnostic test. LCM can effectively deal with diagnostic data of consistency and reproducibility.