Table 6.

Kolmogorov–Smirnov tests for marginal distributions on CheXpert.

	Single task disease detection model
	No find./Pleur. eff.	White/Asian	Asian/Black	Black/White	Male/Female
Marginal			p-values
PCA mode 1	<0.0001∗∗	1.00	0.019∗	0.42	1.00
PCA mode 2	0.010∗	1.00	<0.0001∗∗	<0.0001∗∗	0.25
PCA mode 3	<0.0001∗∗	0.024∗	0.29	1.00	1.00
PCA mode 4	0.12	1.00	0.42	0.17	0.15
Logit ‘no finding’	<0.0001∗∗	0.031∗	0.00098∗∗	1.00	0.18
Logit ‘pleural effusion’	<0.0001∗∗	1.00	<0.0001∗∗	0.00025∗∗	0.24

	Multitask disease detection model
	No find./Pleur. eff.	White/Asian	Asian/Black	Black/White	Male/Female
Marginal			p-values
PCA mode 1	<0.0001∗∗	<0.0001∗∗	0.00025∗∗	<0.0001∗∗	<0.0001∗∗
PCA mode 2	<0.0001∗∗	0.18	0.20	1.00	0.38
PCA mode 3	0.092	<0.0001∗∗	<0.0001∗∗	<0.0001∗∗	0.029∗
PCA mode 4	0.034∗	<0.0001∗∗	0.077	<0.0001∗∗	0.018∗
Logit ‘no finding’	<0.0001∗∗	<0.0001∗∗	<0.0001∗∗	0.70	0.00085∗∗
Logit ‘pleural effusion’	<0.0001∗∗	0.75	0.00017∗∗	0.018∗	0.034∗

Two-sample Kolmogorov–Smirnov tests are performed between the pairs of subgroups indicated in each column. The p-values are adjusted for multiple testing using the Benjamini–Yekutieli procedure and significance is determined at a 95% confidence level. Statistically significant results are marked with ∗p < 0.05 and ∗∗p < 0.001.