Skip to main content

Table 3 Predictive performance of the full model and the simplified model

From: Development of a data-driven COVID-19 prognostication tool to inform triage and step-down care for hospitalised patients in Hong Kong: a population-based cohort study

 

Full Model* (based on 30 features)

On Day 1 of admission

On Day 5 of admission#

Predicted class

Predicted class

Critical/serious

Stable

Satisfactory

Critical/serious

Stable

Satisfactory

Actual class

 Critical/serious

6

0

4

10

0

0

 Stable

6

86

5

0

96

1

 Satisfactory

1

0

100

0

0

101

Sensitivity

 By class

60.0%

88.7%

99.0%

100.0%

99.0%

100.0%

 Macro averaged

82.6%

99.7%

 Micro averaged

92.3%

99.5%

Specificity

 By class

96.5%

100.0%

91.6%

100.0%

100.0%

99.1%

 Macro averaged

96.0%

99.5%

 Micro averaged

96.1%

99.5%

Accuracy

92.3%

99.5%

 

Simplified model* (based on 7 features)

On Day 1 of admission

On Day 5 of admission#

Predicted Class

Predicted Class

Critical/serious

Stable

Satisfactory

Critical/serious

Stable

Satisfactory

Actual class

 Critical/serious

7

1

2

10

0

0

 Stable

7

86

4

9

87

1

 Satisfactory

4

0

97

2

0

99

Sensitivity

 By class

70.0%

88.7%

96.0%

100.0%

89.7%

98.0%

 Macro averaged

84.9%

95.9%

 Micro averaged

91.3%

94.2%

Specificity

 By class

94.4%

99.1%

94.4%

94.4%

100.0%

99.1%

 Macro averaged

96.0%

97.8%

 Micro averaged

95.7%

97.1%

Accuracy

91.3%

94.2%

  1. *Model performance based on testing dataset (n = 208)
  2. #upon discharge if hospital discharged before Day 5