Figure 1
Study design and workflow. Schematic overview of the study design. Retinal traits (10 retinal layers and 25 retinal phenotypes from UK Biobank) and cardiovascular diseases (CVDs, including HF, CHD, HBP, AF, CA, MI, and stroke from the Finland R12 dataset) were analyzed. Linkage disequilibrium score regression (LDSC) was used to evaluate genetic correlations, followed by Mendelian randomization (MR) to infer causal effects, with sensitivity analyses included. Cross-sectional validation was performed in clinical cohorts to test associations between retinal features, circulating biomarkers, and CVDs.
Figure 2
Genetic associations between retinal traits and cardiovascular diseases. Forest plots of MR estimates showing causal associations between different retinal layer thicknesses and risks of major cardiovascular diseases (CHD, HF, HBP, MI, and stroke). Odds ratios (ORs) and 95% confidence intervals (CIs) are presented. Significant associations after false discovery rate (FDR) correction are highlighted, suggesting that retinal structural alterations, particularly in photoreceptor, RPE, and nerve fiber layers, are linked with increased CVD risk.
Figure 3
Causal associations between retinal traits and cardiovascular diseases. Forest plots of MR analyses evaluating the causal effects of retinal structural traits (e.g., ELM–ISOS thickness, GC–IPL thickness, RNFL thickness, INL thickness, RPE thickness, and overall macular thickness) on cardiometabolic outcomes. Red squares represent effect sizes (OR), horizontal bars show 95% CIs. Significant results (FDR < 0.05) are marked in red. The findings indicate specific retinal thinning patterns are causally related to CVD susceptibility.
Figure 4
Causal associations between circulating biomarkers and cardiovascular diseases. (A) Mendelian randomization (MR) results for HBP, AF, CA, and CHD. (B) MR results for HF, MI, and stroke. Forest plots display the causal effects of circulating biomarkers—including lipids, glycemic markers, liver/kidney function indicators, and additional biochemical traits—on each cardiovascular outcome. Red squares represent OR estimates with 95% confidence intervals. Associations surviving FDR correction are highlighted. The results underscore the critical roles of lipid metabolism, glucose regulation, and hepatic/renal biomarkers in cardiovascular disease risk across different phenotypes.
Table 1
The baseline information of seven cardiac vascular disorders, retina layers, and biomarkers. Bold values indicate statistically significant differences (p < 0.05).
| PHENOTYPE | CONTROL | CHD | p | HF | p | MI | p | STROKE | p | CA | p | AF | p | CVD TOTAL | p |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| N | 35 | 25 | 11 | 21 | 10 | 12 | 10 | 89 | |||||||
| AGE | 67.2 ± 9.7 | 65.2 ± 8.6 | 0.408 | 70.4 ± 7.3 | 0.331 | 65.2 ± 9.6 | 0.450 | 66.4 ± 11.7 | 0.821 | 66.4 ± 8.9 | 0.807 | 63.0 ± 5.8 | 0.199 | 65.9 ± 8.9 | 0.541 |
| GENDER | 0.825 | 0.703 | 1.000 | 0.872 | 0.943 | 0.872 | 0.806 | ||||||||
| Male | 15 (42.9%) | 10 (40.0%) | 4 (36.4%) | 9 (42.9%) | 4 (40.0%) | 5 (41.7%) | 4 (40.0%) | 36 (40.4%) | |||||||
| Female | 20 (57.1%) | 15 (60.0%) | 7 (63.6%) | 12 (57.1%) | 6 (60.0%) | 7 (58.3%) | 6 (60.0%) | 53 (59.6%) | |||||||
| BP | 0.452 | 0.179 | 0.143 | 0.648 | 0.107 | 0.026 | 0.043 | ||||||||
| No | 16 (45.7%) | 9 (36.0%) | 3 (27.3%) | 6 (28.6%) | 4 (40.0%) | 3 (25.0%) | 2 (20.0%) | 27 (30.3%) | |||||||
| Yes | 19 (54.3%) | 16 (64.0%) | 8 (72.7%) | 15 (71.4%) | 6 (60.0%) | 9 (75.0%) | 8 (80.0%) | 62 (69.7%) | |||||||
| ALCOHOL | 0.042 | 0.048 | 0.338 | 0.186 | 0.035 | 0.776 | 0.676 | ||||||||
| No | 21 (60.0%) | 11 (44.0%) | 5 (45.5%) | 11 (52.4%) | 5 (50.0%) | 5 (41.7%) | 6 (60.0%) | 43 (48.3%) | |||||||
| Yes | 14 (40.0%) | 14(56.0%) | 6 (55.5%) | 10 (47.6%) | 5 (50.0%) | 7(58.3%) | 4(40.0%) | 46(51.7%) | |||||||
| SMOKE | 0.759 | 0.786 | 0.862 | 0.315 | 0.038 | 0.287 | 0.150 | ||||||||
| No | 17 (48.6%) | 11 (44.0%) | 5 (45.5%) | 10 (47.6%) | 6 (60.0%) | 8 (66.7%) | 5 (50.0%) | 45 (50.6%) | |||||||
| Yes | 18 (51.4%) | 14 (56.0%) | 6 (18.2%) | 11(52.4%) | 4(40.0%) | 4(33.3%) | 5 (50.0%) | 44(49.4%) | |||||||
| RNFL | 102.7 ± 7.0 | 90.4 ± 9.6 | <0.001 | 100.6 ± 8.7 | 0.134 | 88.6 ± 12.8 | <0.001 | 89.7 ± 8.0 | <0.001 | 98.5 ± 5.1 | 0.066 | 100.1 ± 13.2 | 0.417 | 93.3 ± 11.1 | <0.001 |
| GCIPL | 85.8 ± 4.7 | 85.3 ± 5.0 | 0.723 | 85.1 ± 8.3 | 0.438 | 85.0 ± 4.4 | 0.445 | 85.3 ± 3.6 | 0.511 | 84.5 ± 4.7 | 0.422 | 85.1 ± 5.2 | 0.698 | 85.1 ± 5.1 | 0.275 |
| INL | 40.4 ± 3.7 | 40.8 ± 3.8 | 0.736 | 40.1 ± 4.9 | 0.786 | 44.5 ± 5.0 | 0.002 | 41.1 ± 5.4 | 0.671 | 40.6 ± 3.8 | 0.902 | 40.7 ± 3.9 | 0.841 | 41.6 ± 4.6 | 0.153 |
| OPONL | 89.7 ± 6.4 | 89.5 ± 5.8 | 0.870 | 90.1 ± 6.3 | 0.908 | 90.1 ± 6.8 | 0.819 | 90.7 ± 6.8 | 0.725 | 89.8 ± 6.1 | 0.932 | 88.6 ± 6.0 | 0.615 | 89.8 ± 6.1 | 0.978 |
| PR-IS/OS | 65.2 ± 2.6 | 69.1 ± 3.2 | <0.001 | 64.4 ± 3.0 | 0.288 | 70.0 ± 3.7 | <0.001 | 65.6 ± 4.1 | 0.773 | 64.7 ± 3.3 | 0.263 | 67.8 ± 3.9 | 0.017 | 67.6 ± 4.1 | 0.002 |
| RPE-BM | 23.2 ± 2.3 | 22.2 ± 2.8 | 0.105 | 22.3 ± 2.4 | 0.292 | 20.5 ± 2.6 | <0.001 | 21.2 ± 3.1 | 0.015 | 23.0 ± 3.1 | 0.883 | 22.5 ± 3.5 | 0.434 | 21.8 ± 2.9 | 0.010 |
| INNER | 228.9 ± 9.3 | 216.4 ± 10.5 | <0.001 | 225.8 ± 17.9 | 0.131 | 218.1 ± 14.4 | 0.006 | 216.1 ± 10.1 | <0.001 | 223.6 ± 7.9 | 0.109 | 225.9 ± 16.1 | 0.462 | 220.0 ± 13.2 | <0.001 |
| OUTER | 178.2 ± 7.1 | 180.8 ± 6.6 | 0.159 | 176.7 ± 7.3 | 0.527 | 180.7 ± 7.6 | 0.189 | 177.5 ± 9.5 | 0.460 | 177.5 ± 7.8 | 0.723 | 178.9 ± 8.6 | 0.787 | 179.2 ± 7.6 | 0.526 |
| ALT | 24.7 ± 17.3 | 21.2 ± 7.8 | 0.341 | 22.4 ± 7.8 | 0.718 | 22.9 ± 12.3 | 0.624 | 21.9 ± 12.2 | 0.967 | 25.2 ± 13.8 | 0.678 | 23.0 ± 12.8 | 0.773 | 22.5 ± 10.7 | 0.872 |
| AST | 24.8 ± 13.7 | 21.4 ± 6.2 | 0.252 | 21.4 ± 4.2 | 0.877 | 22.2 ± 8.0 | 0.793 | 22.4 ± 12.0 | 0.642 | 23.0 ± 9.7 | 0.845 | 23.1 ± 8.1 | 0.711 | 22.1 ± 7.8 | 0.711 |
| GGT | 29.1 ± 10.6 | 30.8 ± 17.8 | 0.656 | 31.7 ± 13.3 | 0.827 | 32.5 ± 20.6 | 0.741 | 31.1 ± 49.0 | 0.005 | 32.1 ± 18.4 | 0.971 | 30.1 ± 15.1 | 0.816 | 31.4 ± 22.7 | 0.377 |
| TBIL | 9.5 ± 4.6 | 10.4 ± 5.0 | 0.474 | 10.4 ± 8.7 | 0.652 | 11.7 ± 6.1 | 0.253 | 9.8 ± 3.5 | 0.672 | 13.2 ± 7.1 | 0.133 | 7.9 ± 4.0 | 0.315 | 10.7 ± 5.9 | 0.505 |
| ALB | 44.7 ± 2.3 | 44.7 ± 2.5 | 0.962 | 44.5 ± 2.4 | 0.767 | 44.8 ± 2.8 | 0.953 | 44.1 ± 2.8 | 0.428 | 44.3 ± 3.3 | 0.971 | 43.6 ± 2.4 | 0.193 | 44.5 ± 2.6 | 0.569 |
| CREA | 75.2 ± 15.4 | 71.6 ± 13.7 | 0.365 | 79.1 ± 22.4 | 0.877 | 78.6 ± 19.7 | 0.531 | 67.2 ± 15.5 | 0.263 | 76.6 ± 20.1 | 0.591 | 89.6 ± 31.6 | 0.050 | 76.4 ± 20.3 | 0.918 |
| UREA | 6.0 ± 1.7 | 6.1 ± 1.6 | 0.799 | 6.1 ± 1.5 | 0.757 | 6.1 ± 2.1 | 0.859 | 5.9 ± 1.5 | 0.806 | 6.6 ± 2.9 | 0.942 | 7.2 ± 2.9 | 0.107 | 6.3 ± 2.1 | 0.665 |
| UA | 303.4 ± 63.7 | 315.2 ± 69.5 | 0.499 | 308.5 ± 101.7 | 0.908 | 321.9 ± 88.3 | 0.526 | 316.0 ± 45.5 | 0.495 | 307.7 ± 52.4 | 1.000 | 330.2 ± 72.8 | 0.261 | 316.7 ± 73.5 | 0.361 |
| TC | 5.1 ± 1.1 | 5.8 ± 1.4 | 0.032 | 5.6 ± 1.6 | 0.149 | 5.8 ± 1.4 | 0.077 | 5.8 ± 1.0 | 0.101 | 5.9 ± 1.5 | 0.127 | 5.4 ± 1.2 | 0.437 | 5.7 ± 1.3 | 0.015 |
| TG | 1.4 ± 0.7 | 1.6 ± 1.0 | 0.365 | 1.5 ± 0.5 | 0.511 | 1.7 ± 0.8 | 0.094 | 1.5 ± 0.5 | 0.287 | 1.7 ± 0.3 | 0.019 | 1.8 ± 0.8 | 0.147 | 1.6 ± 0.7 | 0.047 |
| HDL_C | 1.2 ± 0.4 | 1.0 ± 0.3 | 0.018 | 1.3 ± 0.3 | 0.519 | 1.1 ± 0.2 | 0.243 | 1.0 ± 0.3 | 0.275 | 1.1 ± 0.3 | 0.550 | 1.1 ± 0.4 | 0.613 | 1.1 ± 0.3 | 0.169 |
| LDL_C | 2.0 ± 0.8 | 3.1 ± 0.7 | <0.001 | 2.5 ± 0.5 | 0.039 | 2.4 ± 0.6 | 0.074 | 2.7 ± 0.5 | 0.006 | 2.4 ± 0.5 | 0.097 | 2.5 ± 0.5 | 0.081 | 2.7 ± 0.6 | <0.001 |
| Lp(a) | 150.2 ± 187.6 | 201.1 ± 257.1 | 0.379 | 308.6 ± 274.5 | 0.092 | 179.2 ± 111.2 | 0.050 | 177.3 ± 139.9 | 0.072 | 230.6 ± 247.4 | 0.414 | 166.5 ± 130.8 | 0.799 | 206.6 ± 206.5 | 0.062 |
| APOA | 1.1 ± 0.3 | 1.0 ± 0.3 | 0.068 | 1.0 ± 0.2 | 0.455 | 1.0 ± 0.3 | 0.326 | 0.9 ± 0.3 | 0.020 | 1.1 ± 0.3 | 0.942 | 1.0 ± 0.2 | 0.119 | 1.0 ± 0.3 | 0.055 |
| APOB | 0.8 ± 0.2 | 0.9 ± 0.3 | 0.081 | 0.9 ± 0.3 | 0.381 | 0.9 ± 0.3 | 0.101 | 1.1 ± 0.2 | 0.002 | 0.8 ± 0.1 | 0.200 | 0.8 ± 0.2 | 0.969 | 0.9 ± 0.3 | 0.038 |
| GLU | 5.6 ± 1.0 | 5.2 ± 0.8 | 0.179 | 5.3 ± 0.5 | 0.220 | 5.5 ± 0.9 | 0.846 | 5.7 ± 1.2 | 0.521 | 6.2 ± 2.5 | 1.000 | 5.8 ± 1.4 | 0.580 | 5.6 ± 1.3 | 0.507 |
| ALP | 80.7 ± 21.0 | 78.2 ± 17.4 | 0.636 | 82.1 ± 13.6 | 0.562 | 80.8 ± 29.9 | 0.594 | 80.4 ± 21.4 | 0.935 | 78.3 ± 13.9 | 0.903 | 79.5 ± 25.0 | 0.881 | 79.7 ± 21.0 | 0.879 |
Figure 5
Retinal thickness differences between CVD patients and healthy controls. Box plots Box plots illustrate differences in retinal thickness among healthy controls (HC) and patients with six CVD conditions: CHD, HF, MI, stroke, CA, and AF. (A) RNFL thickness; (B) GC–IPL thickness; (C) INL thickness; (D) OP–ONL thickness; (E) PR–IS/OS thickness; (F) RPE thickness; (G) Outer retina thickness; (H) Inner retina thickness. Across multiple retinal layers, patients with CVDs exhibit altered thickness profiles compared with healthy controls, with notable thinning observed in structural layers such as the RNFL, GC–IPL, and RPE.
Table 2
The association between retina layers thickness and blood pressure subgroups.
| PHENOTYPE | CONTROL | HBP1 | HBP2 | HBP3 | p |
|---|---|---|---|---|---|
| N | 43 | 43 | 25 | 13 | |
| AGE | 67.6 ± 9.7 | 67.2 ± 8.7 | 64.4 ± 8.7 | 62.5 ± 8.5 | |
| GENDER | 0.700 | ||||
| Male | 20 (46.5%) | 18 (41.9%) | 8 (32.0%) | 5 (38.5%) | |
| Female | 23 (53.5%) | 25 (58.1%) | 17 (68.0%) | 8 (61.5%) | |
| ALCOHOL | 0.301 | ||||
| No | 26 (60.5%) | 20 (46.5%) | 10 (40.0%) | 8 (61.5%) | |
| Yes | 17 (39.5%) | 23 (53.5%) | 15 (60.0%) | 5 (38.5%) | |
| SMOKE | 0.282 | ||||
| No | 24 (55.8%) | 24 (55.8%) | 9 (36.0%) | 5 (38.5%) | |
| Yes | 19 (44.2%) | 19 (44.2%) | 16 (64.0%) | 8 (61.5%) | |
| RNFL | 100.1 ± 12.3 | 96.7 ± 9.3 | 90.6 ± 12.0 | 87.3 ± 5.1 | <0.001 |
| GCIPL | 88.2 ± 4.7 | 84.7 ± 4.6 | 81.6 ± 4.0 | 78.2 ± 5.1 | <0.001 |
| INL | 41.9 ± 4.8 | 40.9 ± 4.5 | 39.8 ± 3.2 | 43.2 ± 4.3 | 0.151 |
| OPONL | 89.4 ± 5.8 | 90.7 ± 6.3 | 88.8 ± 6.3 | 90.0 ± 6.7 | 0.552 |
| PR.IS.OS | 67.0 ± 4.0 | 66.5 ± 3.6 | 67.5 ± 4.2 | 66.8 ± 3.8 | 0.837 |
| RPE.BM | 22.2 ± 3.0 | 22.0 ± 2.5 | 22.0 ± 2.9 | 23.5 ± 2.9 | 0.314 |
| INNER | 228.7 ± 12.6 | 222.3 ± 10.3 | 213.4 ± 12.8 | 220.0 ± 10.6 | <0.001 |
| OUTER | 178.6 ± 7.7 | 179.2 ± 7.4 | 178.4 ± 8.0 | 180.3 ± 7.1 | 0.817 |
| TC | 5.4 ± 1.3 | 5.4 ± 1.2 | 5.7 ± 1.0 | 6.7 ± 1.5 | 0.025 |
| TG | 1.5 ± 0.7 | 1.6 ± 0.8 | 1.4 ± 0.6 | 2.0 ± 0.8 | 0.099 |
| HDL_C | 1.1 ± 0.3 | 1.1 ± 0.3 | 1.1 ± 0.3 | 1.2 ± 0.4 | 0.827 |
| LDL_C | 2.4 ± 0.7 | 2.5 ± 0.8 | 2.6 ± 0.8 | 2.4 ± 0.9 | 0.789 |
| Lp(a) | 188.8 ± 213.7 | 179.5 ± 207.0 | 199.2 ± 199.7 | 217.4 ± 167.9 | 0.618 |
| APOA | 1.0 ± 0.2 | 1.1 ± 0.3 | 1.0 ± 0.2 | 0.9 ± 0.4 | 0.261 |
| APOB | 0.9 ± 0.3 | 0.8 ± 0.2 | 0.9 ± 0.3 | 0.8 ± 0.2 | 0.356 |
| GLU | 5.7 ± 1.4 | 5.4 ± 1.1 | 5.7 ± 1.1 | 5.5 ± 1.2 | 0.782 |
[i] Grade 1 HBP: 130–139/80–89 mmHg.
Grade 2 HBP: 140–159/90–99 mmHg.
Grade 3 HBP: ≥160/100 mmHg.
Bold values indicate statistically significant differences (p < 0.05).
Table 3
The association between retina layers thickness and total cholesterol.
| TC | Q1 | Q2 | Q3 | Q4 | p-VALUE |
|---|---|---|---|---|---|
| N | 30 | 32 | 29 | 33 | |
| AGE | 68.9 ± 9.5 | 65.8 ± 8.8 | 66.7 ± 9.5 | 63.9 ± 8.4 | 0.178 |
| GENDER | 0.788 | ||||
| Male | 11 (36.7%) | 12 (37.5%) | 14 (48.3%) | 14 (42.4%) | |
| Female | 19 (63.3%) | 20 (62.5%) | 15 (51.7%) | 19 (57.6%) | |
| RNFL | 101.0 ± 14.9 | 94.9 ± 9.1 | 90.1 ± 8.9 | 86.9 ± 7.0 | <0.001 |
| GCIPL | 84.1 ± 5.5 | 83.9 ± 4.2 | 85.6 ± 5.9 | 84.8 ± 6.5 | 0.653 |
| INL | 40.0 ± 3.0 | 41.7 ± 4.5 | 41.9 ± 4.8 | 41.3 ± 5.0 | 0.328 |
| OPONL | 90.0 ± 6.2 | 89.8 ± 5.4 | 89.2 ± 7.0 | 90.1 ± 6.1 | 0.955 |
| PR-IS/OS | 67.4 ± 3.8 | 67.1 ± 3.8 | 66.0 ± 3.9 | 67.4 ± 4.0 | 0.484 |
| RPE-BM | 23.1 ± 2.0 | 22.1 ± 2.7 | 22.2 ± 2.5 | 20.4 ± 3.2 | 0.001 |
| INNER | 225.1 ± 17.6 | 220.5 ± 9.3 | 217.6 ± 10.0 | 213.1 ± 10.7 | 0.002 |
| OUTER | 180.5 ± 7.6 | 178.9 ± 7.2 | 177.5 ± 7.5 | 177.9 ± 8.4 | 0.448 |
Figure 6
Nonlinear relationships between serum cholesterol and retinal layer thickness. Restricted cubic spline (RCS) models were used to explore the dose–response relationship between TC levels and the thickness of four retinal layers. (A) RNFL thickness; (B) RPE thickness; (C) GC–IPL thickness; (D) Inner retina thickness. Red curves represent the fitted nonlinear associations, and blue dashed lines denote 95% confidence intervals. The results reveal inverse or nonlinear patterns linking higher cholesterol levels with retinal neurodegeneration across multiple retinal layers.
Figure 7
Prediction model for cardiovascular disease risk using retinal traits and biomarkers. (A) LASSO coefficient profiles of all candidate features. (B) Ten-fold cross-validation for optimal parameter (lambda) selection in the LASSO model. (C) Nomogram developed from the selected predictors, incorporating retinal layer thickness and circulating biomarkers. (D) Receiver operating characteristic (ROC) curve for the nomogram, showing discrimination performance (AUC = 0.878). (E) Decision curve analysis (DCA) evaluating the net clinical benefit across threshold probabilities. (F) Calibration plot comparing predicted vs. observed risk, demonstrating good model calibration. Overall, the LASSO-based nomogram shows strong discrimination, clinical utility, and calibration, supporting the predictive value of combining retinal imaging features with systemic biomarkers in assessing cardiovascular disease risk.