Table 2 Performance of ECG-CLIP on internal testing and external validation

From: Diagnosis of cardiac conditions from 12-lead electrocardiogram through natural language supervision

Cardiac condition

Evaluation metrics

Internal testing

External validation

Atrial fibrillation/flutter

AUROC

0.956 (0.954–0.957)

0.949 (0.947–0.951)

AUPRC

0.787 (0.778–0.795)

0.851 (0.845–0.858)

Accuracy

0.927 (0.925–0.929)

0.881 (0.879–0.884)

Sensitivity

0.800 (0.792–0.808)

0.851 (0.844–0.858)

Specificity

0.945 (0.943–0.947)

0.890 (0.887–0.893)

Precision

0.673 (0.664–0.681)

0.683 (0.675–0.690)

F1 score

0.731 (0.724–0.738)

0.758 (0.752–0.763)

Supraventricular tachycardia

AUROC

0.985 (0.969–0.994)

0.971 (0.969–0.974)

AUPRC

0.229 (0.170–0.290)

0.312 (0.282–0.343)

Accuracy

0.992 (0.991–0.993)

0.979 (0.978–0.981)

Sensitivity

0.667 (0.582–0.748)

0.353 (0.321–0.387)

Specificity

0.992 (0.992–0.993)

0.990 (0.989–0.991)

Precision

0.108 (0.085–0.131)

0.358 (0.323–0.390)

F1 score

0.186 (0.149–0.220)

0.356 (0.326–0.385)

Sinus tachycardia

AUROC

0.982 (0.981–0.983)

0.959 (0.958–0.961)

AUPRC

0.764 (0.752–0.776)

0.742 (0.730–0.754)

Accuracy

0.957 (0.955–0.958)

0.912 (0.909–0.914)

Sensitivity

0.900 (0.893–0.907)

0.847 (0.838–0.855)

Specificity

0.962 (0.961–0.963)

0.924 (0.922–0.927)

Precision

0.690 (0.681–0.700)

0.681 (0.671–0.690)

F1 score

0.781 (0.774–0.788)

0.755 (0.748–0.762)

Sinus bradycardia

AUROC

0.981 (0.980–0.982)

0.985 (0.984–0.986)

AUPRC

0.824 (0.814–0.833)

0.956 (0.952–0.960)

Accuracy

0.962 (0.960–0.963)

0.965 (0.963–0.966)

Sensitivity

0.956 (0.952–0.960)

0.973 (0.971–0.976)

Specificity

0.962 (0.961–0.964)

0.960 (0.958–0.962)

Precision

0.789 (0.781–0.796)

0.934 (0.930–0.937)

F1 score

0.864 (0.860–0.869)

0.953 (0.951–0.955)

Premature ventricular complex

AUROC

0.960 (0.957–0.962)

0.943 (0.931–0.955)

AUPRC

0.707 (0.693–0.721)

0.132 (0.107–0.159)

Accuracy

0.953 (0.952–0.955)

0.953 (0.951–0.955)

Sensitivity

0.787 (0.775–0.798)

0.749 (0.701–0.797)

Specificity

0.964 (0.963–0.966)

0.954 (0.952–0.956)

Precision

0.589 (0.577–0.600)

0.097 (0.084–0.110)

F1 score

0.673 (0.664–0.683)

0.172 (0.151–0.192)

Premature atrial complex

AUROC

0.938 (0.935–0.941)

0.846 (0.837–0.855)

AUPRC

0.422 (0.404–0.440)

0.176 (0.158–0.196)

Accuracy

0.885 (0.883–0.887)

0.875 (0.872–0.878)

Sensitivity

0.813 (0.800–0.826)

0.527 (0.500–0.555)

Specificity

0.889 (0.886–0.891)

0.885 (0.883–0.888)

Precision

0.251 (0.243–0.259)

0.121 (0.112–0.129)

F1 score

0.383 (0.373–0.394)

0.197 (0.183–0.208)

Pacemaker rhythm

AUROC

0.968 (0.963–0.971)

Not applicable

AUPRC

0.318 (0.296–0.341)

Accuracy

0.968 (0.966–0.969)

Sensitivity

0.798 (0.776–0.820)

Specificity

0.970 (0.969–0.971)

Precision

0.305 (0.291–0.321)

F1 score

0.442 (0.424–0.459)

Atrioventricular block

AUROC

0.887 (0.883–0.891)

0.822 (0.811–0.833)

AUPRC

0.419 (0.404–0.435)

0.291 (0.266–0.315)

Accuracy

0.806 (0.803–0.808)

0.701 (0.697–0.705)

Sensitivity

0.784 (0.773–0.795)

0.779 (0.759–0.799)

Specificity

0.808 (0.805–0.810)

0.698 (0.694–0.702)

Precision

0.233 (0.227–0.239)

0.084 (0.079–0.088)

F1 score

0.359 (0.352–0.367)

0.151 (0.144–0.159)

1st degree atrioventricular block

AUROC

0.898 (0.894–0.902)

0.868 (0.856–0.879)

AUPRC

0.441 (0.426–0.457)

0.336 (0.305–0.366)

Accuracy

0.827 (0.824–0.830)

0.754 (0.750–0.758)

Sensitivity

0.786 (0.775–0.797)

0.804 (0.781–0.827)

Specificity

0.830 (0.827–0.832)

0.753 (0.749–0.757)

Precision

0.243 (0.236–0.249)

0.078 (0.073–0.082)

F1 score

0.371 (0.362–0.379)

0.142 (0.134–0.150)

Right bundle branch block

AUROC

0.971 (0.969–0.972)

0.969 (0.964–0.974)

AUPRC

0.832 (0.823–0.841)

0.284 (0.258–0.313)

Accuracy

0.963 (0.962–0.964)

0.921 (0.919–0.924)

Sensitivity

0.800 (0.790–0.810)

0.927 (0.906–0.947)

Specificity

0.977 (0.976–0.978)

0.921 (0.919–0.924)

Precision

0.750 (0.739–0.760)

0.147 (0.136–0.157)

F1 score

0.774 (0.766–0.782)

0.253 (0.238–0.269)

Left bundle branch block

AUROC

0.974 (0.971–0.977)

0.940 (0.925–0.952)

AUPRC

0.695 (0.675–0.713)

0.146 (0.113–0.184)

Accuracy

0.961 (0.959–0.962)

0.982 (0.981–0.983)

Sensitivity

0.864 (0.850–0.877)

0.511 (0.444–0.575)

Specificity

0.964 (0.963–0.965)

0.984 (0.983–0.985)

Precision

0.459 (0.446–0.473)

0.149 (0.125–0.173)

F1 score

0.600 (0.587–0.612)

0.231 (0.197–0.263)

Left ventricular hypertrophy

AUROC

0.704 (0.698–0.710)

0.638 (0.618–0.657)

AUPRC

0.246 (0.237–0.254)

0.024 (0.021–0.027)

Accuracy

0.501 (0.498–0.505)

0.577 (0.573–0.582)

Sensitivity

0.790 (0.782–0.799)

0.632 (0.595–0.668)

Specificity

0.468 (0.464–0.471)

0.577 (0.572–0.581)

Precision

0.148 (0.144–0.151)

0.021 (0.019–0.023)

F1 score

0.249 (0.244–0.253)

0.041 (0.037–0.045)

ST-T change

AUROC

0.474 (0.469–0.479)

0.558 (0.543–0.573)

AUPRC

0.167 (0.164–0.171)

0.029 (0.027–0.033)

Accuracy

0.272 (0.269–0.275)

0.588 (0.583–0.592)

Sensitivity

0.960 (0.957–0.963)

0.486 (0.459–0.515)

Specificity

0.121 (0.119–0.124)

0.590 (0.586–0.595)

Precision

0.194 (0.191–0.197)

0.030 (0.028–0.033)

F1 score

0.322 (0.318–0.326)

0.057 (0.052–0.062)

Prolonged QT interval

AUROC

0.760 (0.753–0.766)

0.797 (0.781–0.813)

AUPRC

0.116 (0.110–0.123)

0.025 (0.022–0.030)

Accuracy

0.610 (0.607–0.613)

0.677 (0.673–0.681)

Sensitivity

0.789 (0.776–0.801)

0.815 (0.778–0.853)

Specificity

0.601 (0.598–0.605)

0.676 (0.672–0.680)

Precision

0.089 (0.086–0.092)

0.022 (0.019–0.024)

F1 score

0.160 (0.155–0.165)

0.042 (0.038–0.047)

Low QRS voltage

AUROC

0.693 (0.687–0.700)

0.740 (0.726–0.754)

AUPRC

0.206 (0.198–0.215)

0.071 (0.061–0.082)

Accuracy

0.445 (0.441–0.448)

0.334 (0.330–0.338)

Sensitivity

0.809 (0.801–0.818)

0.930 (0.914–0.944)

Specificity

0.410 (0.407–0.414)

0.320 (0.316–0.324)

Precision

0.114 (0.112–0.117)

0.031 (0.029–0.033)

F1 score

0.200 (0.196–0.205)

0.061 (0.057–0.064)

Normal ECG

AUROC

0.872 (0.869–0.875)

Not applicable

AUPRC

0.631 (0.623–0.640)

Accuracy

0.800 (0.797–0.803)

Sensitivity

0.807 (0.801–0.813)

Specificity

0.798 (0.794–0.801)

Precision

0.538 (0.532–0.544)

F1 score

0.646 (0.640–0.651)

Borderline ECG

AUROC

0.678 (0.673–0.682)

Not applicable

AUPRC

0.384 (0.378–0.392)

Accuracy

0.551 (0.547–0.555)

Sensitivity

0.809 (0.803–0.815)

Specificity

0.464 (0.459–0.468)

Precision

0.337 (0.332–0.342)

F1 score

0.476 (0.471–0.481)

Abnormal ECG

AUROC

0.798 (0.795–0.801)

Not applicable

AUPRC

0.806 (0.802–0.810)

Accuracy

0.718 (0.715–0.722)

Sensitivity

0.841 (0.837–0.845)

Specificity

0.584 (0.579–0.590)

Precision

0.688 (0.684–0.693)

F1 score

0.757 (0.754–0.760)

  1. Values represent mean values with 95% confidence intervals shown in parentheses. Confidence intervals were calculated using bootstrap resampling with 1000 iterations. AUROC Area Under the Receiver Operating Characteristic curve, AUPRC Area Under the Precision-Recall Curve.