Table 1 Summary Performance for Sepset signature and classifier

From: A machine learning and centrifugal microfluidics platform for bedside prediction of sepsis

RNA-Seq Dataset

AUC

Sensitivity

Specificity

Precision

NPV

Accuracy

Reference

COVID-19 Sepsis (N = 359)

0.85

85%

72%

85%

71%

79%

An39

ICU Validation (N = 176)a

0.9

92%

82%

88%

88%

87%

This study

ER Validation (N = 338)a

0.69

70%

59%

63%

67%

65%

This study

Scicluna (N = 802)

0.99

97%

90%

99%

61%

94%

Scicluna29

Davenport (N = 371)

0.83

86%

62%

59%

88%

74%

Davenport32

Burnham (N = 327)

0.92

87%

83%

74%

92%

85%

Burnham62

Kalantar (N = 152)

0.81

83%

65%

61%

86%

74%

Kalantar63

McClain (N = 201)

0.96

92%

91%

58%

99%

91%

McClain64,65

Tsalik (N = 280)

0.86

75%

88%

70%

91%

82%

Tsalik66

Pankla (N = 138)

0.97

94%

100%

100%

92%

97%

Pankla67

Arunachalam (N = 34)

0.89

83%

95%

91%

91%

89%

Arunachalam68

Aggregated Data (N = 3178)

0.88

87%

78%

80%

83%

83%

 

Negative Control Data Sets

Stage 3/4 Cancer (n = 1755)

0.5

0%

100%

-

63%

50%

TCGA/GDC69

Cardiogenic shock (n = 33)

0.53

0%

100%

-

52%

50%

Yang 202270

Coronary Artery Disease (N = 353)b

0.51

3%

100%

100%

51%

51%

McCaffrey71

Inflammatory Bowel Disease (N = 1030)c

0.53

0%

100%

-

21%

51%

Argmann72

Bacterial infection (N = 170)b

0.39

83%

9%

38%

43%

46%

Smith73

Viral infection (N = 64)c

0.48

0%

93%

0%

83%

46%

Dapat74

  1. Performance of the Sepset signature was evaluated in different validation and testing datasets comprised of transcriptomic analyses of samples collected from various regions and settings. Aggregated results include analysis of patient samples subjected to transcriptomic analysis and available in the literature (see specific references). The internal data sets, samples indicated as “This Study”, were from an independent subset of patients from the Biomarkers of Lung Injury study (ClinicalTrials.gov ID NCT04747782), a prospective, observational, clinical study that collected whole blood from patients who presented to the ICU or ward with acute respiratory distress and suspected sepsis. All datasets are publicly available Supplementary Table 1. Definitions: NPV Negative Predictive Value, ICU intensive care unit, ER emergency room.
  2. a These datasets were distinct subsets of the training set and performance criteria might be affected by overfitting.
  3. bMcCaffrey71; 353 patients with coronary artery disease comparing 189 positives (MID/MID + /CAD) to 164 negatives (LOW CAD).
  4. cArgmann72; comparing 819 inactive/Mild/Moderate/Severe IBD blood-based transcriptomes to 211 healthy controls.
Back to article page