Table 1.

Summary of the experimental setup and major improvements for each modality

Experimental setup		Histopathology	Radiology	Dermatology
Diffusion model	Conditioning variable	Diagnosis and hospital ID	Diagnosis	Diagnosis and demographic attribute
	Access to unlabeled data (additional to the diagnostic model’s training data)	Yes	No	Yes
	Using OOD unlabeled data	No	No	Yes
Diagnostic model	Synthetic/real data ratio	50:50	100:0	75:25
	Performance metric	Top-1 accuracy	ROC-AUC	High-risk sensitivity
	Relative performance improvement with regard to baseline without augmentations	48.5%	5.2%	27.3%
	Absolute fairness improvement with regard to baseline	↓ 30.0% in-distribution	↓ 0.031 OOD	↓ 0.044 OOD

Performance improvements are reported with respect to the respective baseline method, while fairness improvements are reported in absolute terms with respect to the baseline for the corresponding performance metric.