Improving lung cancer risk stratification leveraging whole transcriptome RNA sequencing and machine learning across multiple cohorts

Table 2 Demographic and clinical characteristics of training and test sets focusing on within-indication subjects

	Training		Test		P-value
Characteristic	AEGIS (N = 189)	Registry (N = 122)	AEGIS (N = 246)	Registry (N = 166)
Sex					0.36
Female	72	65	83	84
Male	117	57	163	82
Median age (IQR)	62 (54–70)	64 (57–71)	62 (54–70)	65 (58–71)	0.45
Race					0.59
White	141	106	192	132
Black	34	14	42	29
Other	11	2	12	4
Unknown	3	0	0	1
Smoking status					0.45
Current	79	48	107	73
Former	110	74	139	93
Median cumulative tobacco use (IQR) – pack-year	40 (18–57)	35 (20–50)	35 (20–56)	35 (20–56)	0.82
Lesion size					< 0.001
Infiltrate	0	0	25	0
< 2 cm	42	61	88	80
2 to 3 cm	30	29	48	29
> 3 cm	41	26	75	44
Unknown	60	6	10	13
Lesion location					0.47
Central	50	9	72	10
Peripheral	78	107	108	144
Central and peripheral	46	0	53	0
Unknown	15	6	13	12
Lung-cancer histologic type					0.025
Small-cell	8	3	8	1
Non-small-cell	69	48	100	43	0.18
Adenocarcinoma	30	25	58	25
Squamous	28	12	26	10
Large-cell	6	1	4	0
Non-small-cell not otherwise specified	5	10	12	8
Other	0	2	0	2
Unknown	21	3	3	6
Diagnosis of a benign condition					< 0.001
Fibrosis	1	0	1	0
Granuloma	15	6	26	10
Infection	30	15	36	15
Inflammation	4	2	1	2
Multiple	6	0	8	0
Other	17	4	25	2
Resolution of Stability	18	39	38	40
Clinically benign	0	0	0	45

ISSN: 1755-8794