Scholarlink

Purpose <p>To identify which features best differentiate Sjögren’s-related from non-Sjögren’s related dry eye disease (DED) using machine-learning models.</p> Methods <p>A retrospective study was conducted on 324 individuals with DED based on Dry Eye Workshop-II criteria, including 162 with and 162 without Sjögren’s disease (SjD). One eye per subject was randomly selected. Demographics, Ocular Surface Disease Index (OSDI), and clinical signs were collected. A symptom–sign discordance score was calculated (range − 1 to + 1), positive values indicated greater symptom burden relative to signs, and negative values indicated greater clinical signs relative to symptoms. Multiple machine learning models were developed, and the one with the highest accuracy was selected as optimal. Feature importance, Weight of Evidence (WoE), and Information Value (IV) were computed for the top five predictors in the optimal model.</p> Results <p>The mean age was 47.4 years, and 82% of individuals were female. Compared to the non-SjD group, those with SjD had lower OSDI scores (31.8 vs. 44.6, <i>p</i> &lt; 0.001) despite higher ocular surface staining (OSS) (Oxford IV 5.5% vs. 4.9, <i>p</i> = 0.015), shorter non-invasive tear break-up time (NIBUT) (8.6 vs. 10.6&#xa0;s), lower Schirmer values (7.5 vs. 14.4&#xa0;mm), and a higher rate of positive InflammaDry tests (73.4% vs. 25.3%, <i>p</i> &lt; 0.001). Discordance scores for OSS (− 0.34 vs. +0.09, <i>p</i> &lt; 0.001), NIBUT (− 0.60 vs. −0.39, <i>p</i> &lt; 0.001), and Schirmer (− 0.46 vs. −0.39, <i>p</i> = 0.003) were more negative in the SjD compared to the non-SjD group, indicating that objective signs exceeded symptoms across all three metrics in the SjD group. The optimal model (accuracy = 0.96) identified the most predictive features of SjD-related DED as a positive InflammaDry test (WoE: 1.034, IV: 0.45), OSS discordance (score &lt; − 0.5) (WoE: 1.376, IV: 0.42), NIBUT discordance (score &lt; − 0.5) (WoE: 2.079, IV: 0.22), Schirmer test &lt; 5&#xa0;mm/5 min (WoE: 1.007, IV: 0.18), and Schirmer discordance (score &lt; − 0.5) (WoE: 0.746, IV: 0.09).</p> Conclusion <p>SjD-related DED is characterized by ocular surface inflammation, symptom-sign discordance with signs outweighing symptoms, and a Schirmer test &lt; 5&#xa0;mm/5 min.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Which features best differentiate Sjögren’s-related from non-Sjögren’s-related dry eye disease?

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月8日

查看全文
  • Germán Mejía-Salgado,
  • William Rojas-Carabali,
  • Carlos Cifuentes-González,
  • Juanita Téllez-Zambrano,
  • Blanca Aguilar-Barrera,
  • Laura Daniela Rodriguez-Camelo,
  • Santiago Espinosa-Lugo,
  • Guillermo Marroquín-Gómez,
  • Martha Lucía Moreno-Pardo,
  • Juliana Tirado-Ángel,
  • Anat Galor,
  • Alejandra de-la-Torre

摘要

Purpose

To identify which features best differentiate Sjögren’s-related from non-Sjögren’s related dry eye disease (DED) using machine-learning models.

Methods

A retrospective study was conducted on 324 individuals with DED based on Dry Eye Workshop-II criteria, including 162 with and 162 without Sjögren’s disease (SjD). One eye per subject was randomly selected. Demographics, Ocular Surface Disease Index (OSDI), and clinical signs were collected. A symptom–sign discordance score was calculated (range − 1 to + 1), positive values indicated greater symptom burden relative to signs, and negative values indicated greater clinical signs relative to symptoms. Multiple machine learning models were developed, and the one with the highest accuracy was selected as optimal. Feature importance, Weight of Evidence (WoE), and Information Value (IV) were computed for the top five predictors in the optimal model.

Results

The mean age was 47.4 years, and 82% of individuals were female. Compared to the non-SjD group, those with SjD had lower OSDI scores (31.8 vs. 44.6, p < 0.001) despite higher ocular surface staining (OSS) (Oxford IV 5.5% vs. 4.9, p = 0.015), shorter non-invasive tear break-up time (NIBUT) (8.6 vs. 10.6 s), lower Schirmer values (7.5 vs. 14.4 mm), and a higher rate of positive InflammaDry tests (73.4% vs. 25.3%, p < 0.001). Discordance scores for OSS (− 0.34 vs. +0.09, p < 0.001), NIBUT (− 0.60 vs. −0.39, p < 0.001), and Schirmer (− 0.46 vs. −0.39, p = 0.003) were more negative in the SjD compared to the non-SjD group, indicating that objective signs exceeded symptoms across all three metrics in the SjD group. The optimal model (accuracy = 0.96) identified the most predictive features of SjD-related DED as a positive InflammaDry test (WoE: 1.034, IV: 0.45), OSS discordance (score < − 0.5) (WoE: 1.376, IV: 0.42), NIBUT discordance (score < − 0.5) (WoE: 2.079, IV: 0.22), Schirmer test < 5 mm/5 min (WoE: 1.007, IV: 0.18), and Schirmer discordance (score < − 0.5) (WoE: 0.746, IV: 0.09).

Conclusion

SjD-related DED is characterized by ocular surface inflammation, symptom-sign discordance with signs outweighing symptoms, and a Schirmer test < 5 mm/5 min.