TABLE 4.

Evaluating the impact of key components through ablation on PMC-15M and NIH ChestX-ray14.

Model	PMC-15M dataset				NIH ChestX-ray14 dataset
	Accuracy	Recall	F1 score	AUC	Accuracy	Recall	F1 score	AUC
w./o. Disentangled Temporal Representation	87.83 $\pm$ 0.02	84.74 $\pm$ 0.02	85.20 $\pm$ 0.02	90.45 $\pm$ 0.03	89.55 $\pm$ 0.02	86.92 $\pm$ 0.02	87.04 $\pm$ 0.03	90.41 $\pm$ 0.02
w./o. Ontology-Aware Alignment	88.96 $\pm$ 0.03	85.33 $\pm$ 0.02	86.45 $\pm$ 0.02	91.14 $\pm$ 0.02	89.11 $\pm$ 0.03	87.24 $\pm$ 0.02	87.68 $\pm$ 0.02	91.27 $\pm$ 0.03
w./o. Causal-Aware Refinement	88.42 $\pm$ 0.02	86.27 $\pm$ 0.02	86.38 $\pm$ 0.03	91.72 $\pm$ 0.02	90.14 $\pm$ 0.02	88.20 $\pm$ 0.02	88.51 $\pm$ 0.03	92.02 $\pm$ 0.02
Ours	89.74 $\pm$ 0.02	86.95 $\pm$ 0.02	87.84 $\pm$ 0.03	92.61 $\pm$ 0.02	91.86 $\pm$ 0.02	89.33 $\pm$ 0.02	90.07 $\pm$ 0.03	93.15 $\pm$ 0.02

The values in bold are the best values.