筛选对睑板腺功能障碍(MGD)诊断有价值的特征性指标并建立诊断模型。

收集2016年3月至11月由首都医科大学附属北京同仁医院眼科中心、北京市眼科研究所、陕西省西安市第一医院、中国医学科学院北京协和医院及暨南大学附属第一医院等23家多中心医疗单位眼科门诊连续观察的MGD患者923例(923只眼)和健康志愿者114例(114只眼)。其中，男性353例(353只眼)，女性684例(684只眼)；年龄18~88岁，平均年龄(42.5±13.9)岁。通过问卷调查收集人口学信息，自我评估眼表疾病评分指数(OSDI)；检查所有患者右眼的泪膜破裂时间(TBUT)、角结膜荧光素染色评分、泪液分泌测试(SchirmerⅠ)、睑板腺红外照相、泪膜脂质层厚度(LLT)、睑板腺开口及分泌物性状评分。采用数字表法随机将所有患者拆分为70%模型组和30%验证组。按照临床诊断，判定70%模型组中MGD患者和健康者各指标诊断的阴阳性结果，数据以例数和百分比表示，组间的比较采用卡方检验。采用工作特征曲线(ROC)分析获得曲线下面积(AUC)验证MGD各检测指标的诊断效能，并将有效指标通过Logistic回归模型建立判别模型。而后使用30%验证组MGD患者及健康者进行判别，以验证诊断模型的有效性。

在70%模型组中，MGD患者与健康者女性分别为402例和91例，分别占62.2%和79.8%，两组性别比较的差异有统计学意义(χ²＝196.189，P<0.05)。70%模型组中MGD患者OSDI评分、角结膜荧光素染色、SchirmerⅠ值试验、睑板腺缺失率、LLT、睑板腺腺体开口数量、睑板腺睑脂评分及TBUT与临床诊断结果一致者分别为478例、180例、492例、604例、512例、504例、575例及583例，分别占74.0%、27.9%、76.2%、93.5%、79.4%、78.0%、89.0%及90.4%。两组比较的差异均有统计学意义(χ²＝44.46，442.54，25.20，17.54，12.52，6.91，13.90，1.75；P<0.05)。经ROC曲线分析，尽管OSDI评分、角结膜荧光素染色、睑板腺腺体开口数量、睑板腺睑脂评分及TBUT诊断的AUC均小于0.800(P<0.05)，但当将性别(X₁)、OSDI评分(X₂)、角结膜荧光素染色评分(X₃)、TBUT(X₄)、睑板腺腺体开口数量(X₅)及睑板腺睑脂评分(X₆)有诊断价值的6项指标联合采用多元Logistic回归模型建立MGD诊断判别模型时，AUC能达到0.800，说明此模型的诊断效果已达到良好的标准。公式为Y＝1/(1+e^-Z)，Z＝-4.546+1.069X₁-0.020X₂+0.286X₃-0.553X₄-0.220X₅+0.124X₆。当Y<0.196时，判别模型MGD诊断的敏感度为83.4%，特异度为61.4%。将30%验证组的数据在判别模型中进行分类判别。两组与临床诊断一致者分别为229例和44例，分别占82.7%和40.6%，两组比较的差异无统计学意义(χ²＝0.17，P>0.05)。敏感度为82.7%，特异度为61.4%。

性别、OSDI评分、睑板腺腺体开口数量、睑板腺睑脂评分、TBUT及角结膜上皮荧光素染色有助于MGD诊断，基于这些指标建立的Logistic判别模型有较高的灵敏度和特异度。

The aim of this study was to select the characteristic indicators for the diag-nosis of meibomian gland dysfunction (MGD) and establish a model to diagnose MGD.

From March to November 2016, 923 MGD (923 eyes) and 114 cases of healthy volunteers (114 eyes) were collected in 23 multi-center medical institutions including Beijing Tongren Hospital, affiliated to the Capital Medical University, Beijing Institute of Ophthalmology, The First Hospital of Xi′an Shaanxi Province, Chinese Academy of Medical Sciences and the First Hospital affiliated to Jinan University. Among of them, there were 330 male (eyes) and 684 female (684 eyes) with an average age (42.5±13.9) years-old (ranged from 18 to 88 years-old). Demographic information and self-assessment of ocular surface disease index(OSDI) were collected through questionnaires. The tear film break-up time (TBUT), fluoreciin staining of conjunctiva, tear secretion test (SchirmerⅠ), infrared photography of eyelid gland, tear film lipid layer thickness (LLT), eyelid gland opening and secretion trait score were performed. All MGD patients were randomly divided into 70% model group and 30% checking group. According to clinical diagnosis, the results of each index diagnosis were determined positive and negative. Data was expressed in cases and percentages. Chi-square test was used for comparison between groups. The area under the curve (AUC) was used to check the diagnostic efficiency of indices for MGD through receiver operating characteristic curve (ROC) analysis. A discriminant model was established for effective indicators by Logistic regression model. Then, 30% patients with MGD and healthy controls were discriminated to verify the efficiency of the diagnostic model.

In 70% model group, there were 402 (62.2%), 91 (79.8%) females in patients with MGD and healthy controls, respectively with statistically significant differences (χ²=196.189, P<0.05). In 70% model group, the number of positive results consistent with the clinical diagnosis in OSDI scores, fluoreciin staining of conjunctiva, SchirmerⅠ, loss rate of meibomian gland, LLT, eyelid gland opening and secretion trait score, TBUT were 478 (74.0%), 180 (27.9%), 492 (76.2%), 604 (93.5%), 512 (79.4%), 504 (78.0%), 575 (89.0%), and 583 (90.4%) cases, respectively with statistically significant differences (χ²=44.46, 442.54, 25.20, 17.54, 12.52, 6.91, 13.90, 1.75; P<0.05). ROC curve analysis shown that AUC value of OSDI scores, fluoreciin staining of conjunctiva, eyelid gland opening and secretion trait score, TBUT were less than 0.800 (P<0.05), while these effective indices were established by the Logistic discriminant model, such as gender (X₁), OSDI (X₂), fluoreciin staining of conjunctiva (X₃), TBUT (X₄), eyelid gland opening (X₅), and secretion trait score (X₆), and then AUC value was 0.800, which had better diagnostic efficiency. The formula was as follows: Y=1/(1+ e^-Z), Z= -4.546+ 1.069X₁ -0.020X₂+ 0.286X₃ -0.553X₄-0.220X₅+ 0.124X₆. When the diagnostic model predictive value was greater than 0.196, it was defined as a positive diagnosis; the sensitivity of the diagnosis of MGD was 83.4%, and the specificity was 61.4%. In 30% checking group, the number of positive results consistent with the clinical diagnosis were 229 (82.7%) and 44 (40.6%) with non-significant difference (χ²=0.17, P>0.05). The sensitivity and specificity of the discriminant results were 82.7% and 61.4%, respectively.

Abnormal eyelid margin, lack of tear lipid layer, tear film instability, corneal epithelial fluorescein staining are helpful for the diagnosis of MGD. The Logistic discriminant model established based on these indicators has high sensitivity and specificity.