Skip to main content

Table 2 Overall performance of queries in the 361-patient manually abstracted cohort

From: Performance of a rule-based semi-automated method to optimize chart abstraction for surveillance imaging among patients treated for non-small cell lung cancer

Image annotation

    

Image type: any relevant image

  

Gold standard manual abstraction

  

Relevant study

Not relevant study

Automated Lucene tool result

Relevant study

2548

133

Not relevant study

463

47

Sensitivityβ

(95% CI)

Specificityγ

(95% CI)

PPVδ

(95% CI)

F1 scoreε

85%

(83–86%)

26%

(20–33%)

95%

(94–96%)

0.90

Indication: surveillance

  

Gold standard manual abstraction

  

Surveillance study

Not surveillance study

Automated Lucene tool result

Surveillance study

690

292

Not surveillance study

264

1763

Sensitivity

(95% CI)

Specificity

(95% CI)

PPV

(95% CI)

F1 score

72%

(69–75%)

86%

(84–87%)

70%

(67–73%)

0.71

Finding: suspicious

  

Gold standard manual abstraction

  

Suspicious finding on study

No suspicious finding on study

Automated Lucene tool result

Suspicious finding on study

247

755

No suspicious finding on study

84

1922

Sensitivity

(95% CI)

Specificity

(95% CI)

PPV

(95% CI)

F1 score

75%

(70–79%)

72%

(70–73%)

25%

(22–27%)

0.37

Finding: recurrence only

  

Gold standard manual abstraction

  

Recurrence on study

No recurrence on study

Automated Lucene tool result

Recurrence on study

105

353

No recurrence on study

19

2531

Sensitivity

(95% CI)

Specificity

(95% CI)

PPV

(95% CI)

F1 score

85%

(77–91%)

88%

(87–89%)

23%

(19–27%)

0.36

  1. βSensitivity = true positive/ (true positive + false negative)
  2. γSpecificity = true negative/ (true negative + false positive)
  3. δPositive predictive value = true positive/ (true positive + false positive)
  4. εF1 score = 2 ((sensitivity*PPV)/ (sensitivity + PPV))