视网膜作为心肾代谢疾病窗口的综述
视网膜眼组学通过眼底图像无创评估心肾代谢(CKM)综合征,视网膜病变和血管指标与卒中、心血管死亡及冠心病相关,AI模型在系统性生物标志物估计和心血管风险分层方面较成熟,但肾脏特异性模型仅限横断面CKD检测,且证据多为观察性、回顾性,晚期CKD、透析和移植人群代表性不足。
要点
- 视网膜病变和定量血管指标与 incident stroke、CV死亡和冠心病相关。
- AI视网膜模型在系统性生物标志物估计和CV风险分层最成熟,肾脏模型限于横断面CKD检测。
- 白蛋白尿、eGFR持续下降、KRT和肾脏相关死亡尚未被视网膜模型靶向。
核心结论
- 视网膜AI有望补充传统肾脏生物标志物,作为低负担的CKM损伤整合窗口。
- 但现有证据多为观察性、回顾性,需填补肾脏终点、晚期CKD人群和管理影响试验等空白。
研究设计与样本
- 综述性文章,未报告系统检索范围。
主要发现
- 视网膜病变和定量血管指标与 incident stroke、CV死亡和冠心病关联最强。
- 视网膜病变与 new-onset hypertension、diabetes 和 incident CKD 也有关联,但CKD关联在调整传统肾脏标志物后减弱。
- AI视网膜模型在系统性生物标志物估计和CV风险分层最成熟,肾脏特异性模型限于横断面CKD检测和CKD发展预测。
- 白蛋白尿、eGFR持续下降、KRT和肾脏相关死亡尚未被靶向。
- 晚期CKD、透析和移植人群代表性不足,模型在族裔不同队列中判别力下降,校准报告不常见。
临床怎么用
- 视网膜AI可能用于筛查、风险分层、进展监测和治疗优先级排序。
- 在早期CKM中用于筛查,在晚期CKD中用于补充表型分析。
局限性
- 证据主要为观察性、回顾性且依赖横断面替代指标。
- 未靶向白蛋白尿、eGFR持续下降、KRT和肾脏相关死亡。
- 晚期CKD、透析和移植人群在开发和验证队列中代表性不足。
- 无视网膜模型报告CKD中的CV风险预测或与肾脏风险方程的正式比较。
- 管理影响试验未开展,成本效益未评估,算法不透明和成像标准化未解决。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
The cardiovascular-kidney-metabolic (CKM) syndrome reframes cardiovascular (CV), kidney, and metabolic disease as an integrated continuum, yet its management relies on reactive laboratory markers with substantial resource burdens. Retinal oculomics offers a noninvasive window into this continuum, grounded in structural and functional parallels between the retina and kidney. Their shared vulnerability to metabolic and hemodynamic stressors allows the retina to reflect subclinical CKM injury. Evidence is strongest for CV end points, where retinopathy and quantitative vessel metrics are associated with incident stroke, CV mortality, and coronary heart disease. Similar associations are reported for new-onset hypertension, diabetes, and incident CKD, although the CKD association attenuates after adjustment for conventional kidney markers. Artificial intelligence (AI) extends retinal analysis beyond categorical grading and predefined vessel metrics by learning latent features from fundus photographs. Systemic biomarker estimation and CV risk stratification are the most mature, whereas kidney-specific models remain confined to cross-sectional CKD detection and prediction of CKD development. However, retinal AI is supported by evidence that remains largely observational, retrospective, and dependent on cross-sectional surrogates. Albuminuria, sustained decline in eGFR, KRT, and kidney-related mortality have not been targeted. Advanced CKD, dialysis, and kidney transplant populations are underrepresented in development and validation cohorts. No retinal model has reported CV risk prediction in CKD or direct comparison with established kidney risk equations. Discrimination in these models falls in ethnically distinct cohorts, and calibration is infrequently reported. Management-impact trials have not been conducted, and cost-effectiveness remains unevaluated. Algorithmic opacity and imaging standardization remain unresolved. Once these gaps are addressed, retinal AI may support screening, risk stratification, progression monitoring, and treatment prioritization, shifting from screening in early CKM to complementary phenotyping in advanced CKD. Retinal AI would then complement conventional kidney biomarkers as an integrative, low-burden window into CKM injury.
引用
- Park JT, Tan YY, Ko YE, et al. The Eye as a Window to Cardio-Kidney-Metabolic Disease. Clin J Am Soc Nephrol. 2026. PMID: 42616587. doi:10.2215/CJN.0000001224
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。