Skip to main content

Table 1 Ten most important variables for model to predict outcomes

From: Machine learning to predict hemorrhage and thrombosis during extracorporeal membrane oxygenation

Random forest model*

Chi-square model

Hemorrhage

 Heparin drip rate—maximum dosage

Heparin drip rate—maximum dosage

 Heparin drip rate—mean dosage

Heparin drip rate—mean dosage

 PTT—lowest value

Heparin drip rate—minimum dosage

 Activated clotting time—highest value

PTT—highest value

 Platelet count—highest value

PTT—mean value

 Race

PTT—lowest value

 ECMO configuration

INR—highest value

 ECMO—double-lumen cannulation

INR—mean value

 Drainage cannula size

INR—lowest value

 Drainage cannula site

Activated clotting time—highest value

Thrombosis

 ECMO—double-lumen cannulation

Heparin drip rate—maximum dosage

 Platelet—lowest value

Heparin drip rate—mean dosage

 Transfusion of cryoglobulin

Heparin drip rate—minimum dosage

 Transfusion of platelets

PTT—highest value

 Body mass index

PTT—mean value

 Renal replacement therapy

PTT—lowest value

 ECMO—duration

INR—highest value

 ECMO indication—status asthmaticus

INR—mean value

 ECMO indication—PH/right ventricular failure

INR—lowest value

 Platelet count—mean value

Activated clotting time—highest value

  1. ECMO extracorporeal membrane oxygenation, PH pulmonary hypertension, PTT partial thromboplastin time, INR international normalized ratio
  2. *p > 0.05, none of the individual features significantly contributed to the model’s performance