Table 2.

Association of artificial intelligence-based stromal TILs percentage with clinicopathological parameters in the study cohorts.

Artificial intelligence-based stromal TILs percentage
	Discovery set		Test set		External validation cohort
Characteristics	Mean Rank	p-value	Mean Rank	p-value	Mean Rank	p-value
Age at diagnosis (years)
< 50	845	0.01	342	0.4	156	0.8
≥ 50	772		328		160
Menopausal status
Premenopausal	805	0.3	349	0.2	NA	NA
Postmenopausal	780		325
Tumour size (cm)
< 2	775	0.1	327	0.5	160	0.9
≥ 2	816		337		159
Tumour grade
1	569	<0.001	266	<0.001	147	0.007
2	747		329		160
3	1044		414		207
Mitotic count
1	717	<0.001	302	<0.001	154	0.06
2	912		384		170
3	1054		437		196
Nuclear pleomorphism
1	704	<0.001	302	<0.001	184	0.1
2	696		292		154
3	882		376		171
Tubule formation
1	575	<0.001	257	<0.001	134	0.03
2	789		319		157
3	819		349		172
Nottingham Prognostic Index
Good prognostic group	706	<0.001	305	<0.001	154	0.3
Moderate prognostic group	886		363		169
Poor prognostic group	1228		312		184
Histological types
No special type (NST)	845	<0.001	338	0.2	158	0.4
Lobular	650		330		151
Other special types	645		275		188
Mixed NST and other tumour types	760		324		164
Lymphovascular invasion
Negative	767	<0.001	323	0.008	156	0.005
Positive	915		381		217
Lymph node status
Negative	771	0.007	333	0.4	162	0.3
Positive	848		320		146
Progesterone receptor
Negative	886	<0.001	386	<0.001	146	0.4
Positive	764		317		161
Ki67 index
Low (<20%)	300	<0.001	140	<0.001	NA	NA
High (≥20%)	415		183

NA not applicable

Significant p values are in bold (Mann-Whitney test).