Skip to main content

Table 3 Lung cancer genomic sequencing classifier validation performance (Down, up classification). *Cancer prevalence calculation includes local benign subjects as Prevalence = \( \frac{\# Malignant}{\# Malignant+\# Benign+\# Local\ Benign} \). The local benign subjects had local label as benign but did not have an adjudicated label. NPV, PPV and % Reclassified are all functions of prevalence (estimated including local benign subjects), sensitivity and specificity (both estimated excluding local benign subjects)

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

AUC

Pre-test Cancer Risk

*Cancer prevalence

Cancer risk re-stratification

Specificity

Sensitivity

Post-test NPV/PPV

§% Re-stratified

73.4% [68.3–78.4]

Low

5%

Low to Very Low

57.4% [44.8–69.3]

100% [39.8–100]

100% NPV [91.0–100]

54.5%

Intermediate

28.2%

Intermediate to Low

37.3% [27.9–47.4]

90.6% [79.3–96.9]

91.0% NPV [80.8–96.0]

29.4%

Intermediate to High

94.1% [87.6–97.8]

28.3% [16.8–42.3]

65.4% PPV [43.8–82.1]

12.2%

High

73.6%

High to Very High

91.2% [76.3–98.1]

34.0% [25.0–43.8]

91.5% PPV [77.9–97.0]

27.3%

  1. §% Reclassified (Low to Very Low, Intermediate to Low) = (1- Prevalence) specificity + Prevalence (1-sensitivity)
  2. §% Reclassified (Intermediate to High, High to Very High) = Prevalence sensitivity + (1-Prevalence) (1- specificity)
  3. * There are 8, 33 and 4 local benign subjects in low, intermediate and high-risk group