构建并验证屈光不正成年人干眼风险的预测模型。

2024年6月至2025年5月在哈尔滨医科大学附属第二医院就诊的成年合并屈光不正患者402例(402只眼)作为研究对象。其中，男性91例(91只眼)，女性311例(311只眼)；年龄18~79岁，平均年龄(51.8±14.5)岁。通过问卷和眼科检查收集人口学、生活方式及临床资料，使用Excel随机排序功能，将全部数据按7：3的比例随机划为训练集282例(282只眼)和验证集120例(120只眼)。纳入单因素分析中P<0.05的变量，构建基于Logistic回归屈光不正成年人干眼风险的预测模型。通过校准曲线验证模型的拟合优度。年龄等连续型变量在满足正态分布时以±s表示，组间比较采用独立样本t检验；不符合正态分布的变量以M(Q1，Q3)表示，组间比较采用秩和检验。分类变量以频数和百分比表示，组间比较采用卡方检验或Fisher确切概率法。先对所有分类变量进行单因素Logistic回归分析，将P<0.05的变量纳入多因素Logistic回归模型，并在此基础上建立临床预测模型并绘制列线图。采用临床决策曲线分析评估模型的应用价值。

训练集和验证集的干眼患病率分别为42.2%和44.2%。在训练集282例(282只眼)中进行单因素Logistic回归分析，并将具有统计学意义的变量纳入多因素Logistic回归模型进行验证，发现有效屈光矫正是一个强相关因素，屈光矫正者发生干眼的风险显著低于未矫正者(OR＝0.172，95%CI：0.101~0.290，P<0.05)。近视眼和老视均为干眼发生的正相关因素(OR＝1.644，2.00；95%CI：1.021~2.647，1.149~3.483；P<0.05)；日常摄入营养食物种类数少于3种与干眼风险相关(OR＝2.645，95%CI：1.058~6.612，P<0.05)。将有效屈光矫正、近视眼、老视及日常摄入营养食物种类少于3种纳入多因素Logistic回归模型进行再分析，在控制了混杂因素后，有效屈光矫正、近视眼及老视是干眼的独立影响因素(OR_调整后＝0.157，1.715；1.928; 95%CI：0.090~0.273, 1.003~2.931；1.039~3.575；P<0.05)。模型在训练集和验证集的曲线下面积分别为0.766(95%CI：0.711~0.821)和0.781(95%CI：0.699~0.863)，校准良好，决策曲线分析显示在广泛阈值范围内具有临床净获益。

本研究构建的干眼风险预测列线图模型具有良好的预测性能和临床适用性，揭示了未矫正状态的屈光不正、近视眼及老视等屈光因素是屈光不正成年人发生干眼的核心独立风险因素。及时有效地矫正屈光对干眼预防具有关键作用，并有助于推动干眼管理的个体化与早期干预。

The aim of this study is to develop and validate a dry eye risk prediction model for adults with refractive error.

A total of 402 adults (402 eyes) with refractive error who visited the Second Affiliated Hospital of Harbin Medical University from June 2024 to May 2025 were enrolled as study subjects. Among them, there were 91 males (91 eyes) and 311 females (311 eyes) with a mean age of (51.8±14.5) years (ranging from 18 to 79 years). Demographic, lifestyle, and clinical data were collected through questionnaires and ophthalmic examinations. Using the random sorting function in Excel, all data were randomly divided into a training set (282 cases, 282 eyes) and a validation set (120 cases, 120 eyes) at a ratio of 7: 3. Variables with P<0.05 in univariate analysis were included to construct a logistic regression-based dry eye risk prediction model for adults with refractive error. Calibration curves were used to validate the goodness-of-fit of the model. Continuous variables that followed a normal distribution were expressed as ±s, with intergroup comparisons using independent sample t-tests; variables not following a normal distribution were expressed as median (Q1, Q3), with intergroup comparisons using rank sum tests. Categorical variables were expressed as frequency and percentage, with intergroup comparisons using chi-square tests, or Fisher′s exact test. Univariate logistic regression analysis was first performed for all categorical variables, and variables with P<0.05 were entered into the multivariate Logistic regression model. Based on this, a clinical prediction model was established and a nomogram was drawn. Decision curve analysis was used to evaluate the clinical applicability of the model.

The prevalence of dry eye was 42.2% in the training set and 44.2% in the validation set.Univariate logistic regression analysis was performed in the training set (282 cases, 282 eyes), and statistically significant variables were then entered into the multivariate Logistic regression model. Effective refractive correction was a strongly associated factor; the risk of dry eye in those with refractive correction was significantly lower than in those without correction (OR=0.172, 95%CI: 0.101 to 0.290, P<0.05). Myopia and presbyopia were both positively associated with dry eye (OR=1.644 and 2.00; 95%CI: 1.021 to 2.647 and 1.149 to 3.483; P<0.05). Consuming fewer than three types of nutrient-rich foods daily was associated with dry eye risk (OR=2.645, 95%CI: 1.058 to 6.612, P<0.05). The effective refractive correction, myopia, presbyopia, and consuming fewer than three types of nutrient-rich foods were entered into the multivariate logistic regression model. After adjusting for confounders, effective refractive correction, myopia, and presbyopia remained independent factors for dry eye (OR_adjusted=0.157, 1.715, and 1.928; 95%CI: 0.090 to 0.273, 1.003 to 2.931, and 1.039 to 3.575; P<0.05). The area under the curve of the model was 0.766 (95%CI: 0.711 to 0.821) in the training set and 0.781 (95%CI: 0.699 to 0.863) in the validation set. Calibration was good, and decision curve analysis showed a net clinical benefit across a wide range of threshold probabilities.

The nomogram-based dry eye risk prediction model constructed has a good predictive performance and clinical applicability. It reveals that refractive factors such as uncorrected refractive error, myopia, and presbyopia are core independent risk factors for dry eye in adults with refractive error. Timely and effective refractive correction plays a key role in dry eye prevention and helps promote individualized and early intervention in dry eye management.