Detection of early-stage lung cancer in sputum using automated flow cytometry and machine learning

Respiratory Research

Table 3 CyPath lung performance

	LSRII	LSRII (nodules all < 20 mm)	Navios
Total samples	150	132	32
Cancer	28	13	6
Non-cancer	122	119	26
Sensitivity (95% CI)	0.82 (0.64–0.92)	0.92 (0.67–0.99)	0.83 (0.44–0.97)
Specificity (95% CI)	0.88 (0.81–0.92)	0.87 (0.80–0.92)	0.77 (0.58–0.89)
Accuracy (95% CI)	0.87(0.80–0.91)	0.88 (0.81–0.92)	0.78 (0.61–0.89)
Area under ROC curve (95% CI)	0.89 (0.83–0.96)	0.94 (0.89–0.99)	0.85 (0.71–0.98)
Positive predictive value (95% CI)
Cancer prevalence in data set	0.61 (0.48–0.72)	0.44 (0.33–0.57)	0.45 (0.27–0.65)
Prevalence reported in high-risk^a	0.05 (0.03–0.09)	0.06 (0.04–0.92)	0.03 (0.01–0.06)
Prevalence in LDCT positive^b	0.17 (0.11–0.25)	0.18 (0.12–0.26)	0.10 (0.05–0.19)
Negative predictive value (95% CI)
Cancer prevalence in data set	0.96 (0.91–0.98)	0.99 (0.94–1.00)	0.95 (0.77–0.99)
Prevalence reported in high-risk^a	1.00 (1.00–1.00)	1.00 (1.00–1.00)	1.00 (1.00–1.00)
Prevalence in LDCT positive^b	0.99 (0.99–1.00)	1.00 (0.98–1.00)	0.99 (0.96–1.00)
Positive diagnostic likelihood ratio (PDLR)^c	6.31	7.08	3.61

^a0.83% cancer prevalence in NLST 2013 [1]
^b2.9% cancer prevalence in NLST 2013 LDCT positive cases only
^cSensitivity/(1 − specificity) see Pepe et al. [37]
Wilson confidence intervals (CIs) for sensitivity, specificity and accuracy were calculated using BinomCI (“method = Wilson”) from the R package DescTools [38]
Area under ROC curve CIs were determined by bootstrapping using the R package pROC [39]
CIs of the positive and negative predictive values were calculated using the R package bdpv [40] per Mercaldo et al. [41]

ISSN: 1465-993X