Scholarlink

<p>Blinding eye diseases pose a substantial global health burden, yet current screening strategies are limited by resource demands and poor accessibility, particularly in underserved settings. Leveraging the broad availability of routine blood testing, we developed a scalable and non-invasive machine learning-based multidisease eye screening (MES) framework to identify individuals at higher risk of major blinding eye diseases and prioritize referral for confirmatory ophthalmic evaluation. Using data from 93,839 participants and internally validated in 33,622 individuals, the MES test integrates a binary classifier to detect eye disease and a multiclass classifier to differentiate seven common blinding conditions. Performance was further evaluated in three independent external cohorts (<i>n</i> = 34,087), a prospective hospital-based cohort (<i>n</i> = 43,556) and a large population-based cohort (<i>n</i> = 498,095). Across validation datasets, the MES test achieved high diagnostic performance for detecting any eye disease, with an area under the curve of 0.9264–0.9561, positive predictive values of 0.9127–0.9260 and negative predictive values of 0.8075–0.8917. Subtype-level classification demonstrated a macroaveraged area under the curve of 0.889–0.900. In real-world clinical and community settings, the MES test yielded positive and negative predictive values of 0.959 and 0.960, and 0.931 and 0.991, respectively. Performance remained robust across age and comorbidity subgroups. These results support the potential of the MES framework as a scalable triage aid to identify individuals at higher risk and prioritize confirmatory ophthalmic assessment for major blinding eye diseases.</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

Large-scale screening of blinding eye diseases from routine blood tests

  • 期刊论文

  • 发表日期：

    2026年4月6日

查看全文
  • Shengjie Li,
  • Jun Ren,
  • Jianing Wu,
  • Yingzhu Li,
  • Xuanxuan Wang,
  • Mengyu Zhang,
  • Henggui Hu,
  • Yunxiao Song,
  • Shunxiang Gao,
  • Wenjun Cao,
  • Meiyan Li,
  • Xingtao Zhou

摘要

Blinding eye diseases pose a substantial global health burden, yet current screening strategies are limited by resource demands and poor accessibility, particularly in underserved settings. Leveraging the broad availability of routine blood testing, we developed a scalable and non-invasive machine learning-based multidisease eye screening (MES) framework to identify individuals at higher risk of major blinding eye diseases and prioritize referral for confirmatory ophthalmic evaluation. Using data from 93,839 participants and internally validated in 33,622 individuals, the MES test integrates a binary classifier to detect eye disease and a multiclass classifier to differentiate seven common blinding conditions. Performance was further evaluated in three independent external cohorts (n = 34,087), a prospective hospital-based cohort (n = 43,556) and a large population-based cohort (n = 498,095). Across validation datasets, the MES test achieved high diagnostic performance for detecting any eye disease, with an area under the curve of 0.9264–0.9561, positive predictive values of 0.9127–0.9260 and negative predictive values of 0.8075–0.8917. Subtype-level classification demonstrated a macroaveraged area under the curve of 0.889–0.900. In real-world clinical and community settings, the MES test yielded positive and negative predictive values of 0.959 and 0.960, and 0.931 and 0.991, respectively. Performance remained robust across age and comorbidity subgroups. These results support the potential of the MES framework as a scalable triage aid to identify individuals at higher risk and prioritize confirmatory ophthalmic assessment for major blinding eye diseases.