Skip to main content

Table 2 Comparison of performance for invasive mechanical ventilation prediction models in NICU with a consistent specificity threshold

From: Early prediction of need for invasive mechanical ventilation in the neonatal intensive care unit using artificial intelligence and electronic health records: a clinical study

Models (≥ threshold)

Specificity

Sensitivity

PPV

NPV

LHR + 

LHR-

NEWS ≥ 2

0.8492

0.3248

0.0209

0.9921

2.1545

0.7950

XGBoost(SpO2,FiO2) (≥ 0.3383)

0.8535

0.5150

0.0336

0.9944

3.5165

0.5681

XGBoost (≥ 0.2487)

0.8492

0.6845

0.0430

0.9963

4.5401

0.3714

Random forest (≥ 0.4018)

0.8492

0.6774

0.0426

0.9962

4.4948

0.3797

Proposed (≥ 0.5207)

0.8493

0.7250

0.0455

0.9968

4.8116

0.3237

NEWS ≥ 3

0.9608

0.1812

0.0438

0.9916

4.6334

0.8520

XGBoost(SpO2,FiO2) (≥ 0.5187)

0.9552

0.3121

0.0646

0.9929

6.9723

0.7200

XGBoost (≥ 0.4932)

0.9608

0.5108

0.1145

0.9949

13.0595

0.5090

Random forest (≥ 0.4960)

0.9608

0.4868

0.1097

0.9947

12.4397

0.5340

Proposed (≥ 0.6554)

0.9608

0.5739

0.1268

0.9956

14.6601

0.4434

NEWS ≥ 4

0.9848

0.0838

0.0520

0.9908

5.5429

0.9302

XGBoost(SpO2,FiO2) (≥ 0.6668)

0.9845

0.1440

0.0843

0.9914

9.3021

0.8693

XGBoost (≥ 0.6669)

0.9848

0.3992

0.2067

0.9939

26.3014

0.6100

Random forest (≥ 0.5283)

0.9848

0.3248

0.1749

0.9932

21.3943

0.6855

Proposed (≥ 0.7303)

0.9848

0.4825

0.2403

0.9948

31.9286

0.5253

NEWS ≥ 5

0.9954

0.0225

0.0467

0.9903

4.9483

0.9818

XGBoost(SpO2,FiO2) (≥ 0.7207)

0.9931

0.0880

0.1131

0.9909

12.8826

0.9182

XGBoost (≥ 0.7887)

0.9954

0.2622

0.3638

0.9927

57.7161

0.7411

Random forest (≥ 0.5829)

0.9954

0.1836

0.2869

0.9919

40.6202

0.8200

Proposed (≥ 0.8065)

0.9954

0.3455

0.4289

0.9935

75.8233

0.6574

  1. Abbreviations: LHR Likelihood ratio, NEWS Newborn early warning score system, NICU Neonatal intensive care unit, NPV Negative predictive value, PPV Positive predictive value, XGBoost Extreme gradient boosting