Liu Zhenqiu, Yuan Huangbo, Suo Chen, Jin Li, Zhang Tiejun, Chen Xingdong
Human Phenome Institute, Research and Innovation Center, Shanghai Pudong Hospital, Fudan University, Shanghai, 201203, China; Fudan University Taizhou Institute of Health Sciences, Taizhou, 225316, China.
Human Phenome Institute, Research and Innovation Center, Shanghai Pudong Hospital, Fudan University, Shanghai, 201203, China; Fudan University Taizhou Institute of Health Sciences, Taizhou, 225316, China.
Dig Liver Dis. 2025 Sep;57(9):1819-1825. doi: 10.1016/j.dld.2025.06.016. Epub 2025 Jul 12.
Existing polygenic risk scores (PRS) for severe liver disease (SLD) have limited predictive ability, highlighting a possible reorientation for PRS application in clinical practice.
Using non-overlapping subsets of the UK Biobank cohort, we first conducted a genome-wide association study of magnetic resonance imaging-derived hepatic fat content (HFC; n = 12,838), and then constructed a polygenic risk score to capture genetically predicted HFC (gHFC), which was applied in an independent sample (n = 426,529) to stratify individuals and evaluate the performance of clinical fibrosis scores.
Among 426,529 participants, 4417 developed SLD during follow-up. gHFC alone showed limited predictive power for SLD, and adding it to fibrosis scores did not improve AUROC. However, population stratification by gHFC substantially improved the performance of Fibrosis-4 (FIB-4), Forns, and Aspartate aminotransferase-to-Platelet Ratio Index (APRI), particularly for hepatocellular carcinoma (HCC). In the highest gHFC quintile, the areas under the receiver operating characteristic curve for HCC were 0.819 (FIB-4), 0.877 (Forns), and 0.851 (APRI), significantly higher than in the lowest quintile. Similar trends were observed using two alternative HFC PRSs.
Stratifying the population by PRS before using clinical fibrosis scores to predict SLD is a more effective approach than considering PRS as an alternative or an addition to clinical risk models.
现有的严重肝病（SLD）多基因风险评分（PRS）预测能力有限，这凸显了PRS在临床实践中应用可能需要重新定位。
我们使用英国生物银行队列的非重叠子集，首先对磁共振成像衍生的肝脏脂肪含量（HFC；n = 12,838）进行全基因组关联研究，然后构建一个多基因风险评分以获取基因预测的HFC（gHFC），并将其应用于一个独立样本（n = 426,529）中对个体进行分层，并评估临床纤维化评分的性能。
在426,529名参与者中，4417人在随访期间发生了SLD。单独的gHFC对SLD的预测能力有限，将其添加到纤维化评分中并未改善曲线下面积（AUROC）。然而，通过gHFC进行人群分层显著改善了Fibrosis-4（FIB-4）、Forns评分和天冬氨酸转氨酶与血小板比值指数（APRI）的性能，尤其是对于肝细胞癌（HCC）。在gHFC最高的五分位数中，HCC的受试者操作特征曲线下面积分别为0.819（FIB-4）、0.877（Forns）和0.851（APRI），显著高于最低五分位数。使用另外两种HFC PRS也观察到了类似趋势。
在使用临床纤维化评分预测SLD之前，通过PRS对人群进行分层是一种比将PRS视为临床风险模型的替代或补充更有效的方法。
