Table 2 Performance metrics for prediction of oxygen requirement in patients with COVID-19 using the CXR xAI model with and without clinical information.

From: Prediction of oxygen requirement in patients with COVID-19 using a pre-trained chest radiograph xAI model: efficient development of auditable risk prediction models via a fine-tuning approach

 

Outcome

RF without clinical data

RF with clinical data

Cutoff value

Sensitivity

Specificity

PPV

NPV

Accuracy

Cutoff value

Sensitivity

Specificity

PPV

NPV

Accuracy

24 h

RA (N = 192)

0.504

0.729

0.786

0.791

0.723

0.756

0.326

0.938

0.767

0.811

0.916

0.852

LFO (N = 141)

0.296

0.574

0.741

0.583

0.735

0.677

0.400

0.518

0.951

0.869

0.758

0.784

HFO (N = 15)

0.100

0.133

0.986

0.236

0.964

0.951

0.032

0.533

0.974

0.471

0.980

0.956

MV (N = 17)

0.014

0.588

0.902

0.227

0.978

0.888

0.030

0.765

0.908

0.289

0.988

0.901

72 h

RA (N = 166)

0.398

0.783

0.754

0.726

0.807

0.767

0.608

0.759

0.884

0.846

0.815

0.827

LFO (N = 147)

0.386

0.320

0.858

0.603

0.652

0.641

0.322

0.497

0.872

0.723

0.720

0.721

HFO (N = 20)

0.070

0.250

0.942

0.200

0.956

0.912

0.016

0.550

0.913

0.268

0.972

0.893

MV (N = 32)

0.002

0.563

0.793

0.207

0.950

0.773

0.010

0.531

0.823

0.224

0.948

0.797

  1. RF random forest, RA room air, LFO low flow oxygen, HFO high flow oxygen and MV mechanical ventilation.
Back to article page