Machine learning model for early prediction of acute kidney injury (AKI) in pediatric critical care

Table 2 Final predictors categorized by predictor type and ranked by the order in which the predictor was selected by the model

Type	Predictor	Statistic	Unit	p-value
Vitals	Shock index*	Max	bpm/mmHg	< 0.001
	SpO₂	Mean	%	< 0.001
Laboratory values	Blood urea nitrogen	Last	mg/dL	< 0.001
	Serum creatinine rate of change	Last	mg/dL/hr	< 0.001
	Bilirubin	Last	mg/dL	< 0.001
	PaCO₂	Max	%	< 0.001
	Anion gap	Last	mmol/L	0.005
	White blood cell count (WBC)	Last	10⁹/L	< 0.001
	Serum albumin	Last	g/dL	< 0.001
	Serum chloride	Last	mmol/L	< 0.001
	Gentamicin trough	Last	mg/L	< 0.001
Medications	Number of vasoactive drugs administered	–	–	< 0.001
	Number of high nephrotoxic potential drugs administered	–	–	< 0.001
Ventilation†	Mean airway pressure	Median	cmH₂O	0.277
Others	Time since admission	–	hours	< 0.001

^*Shock index = heart rate/blood pressure. †Ventilation-related predictors are treated as missing data for patients not on ventilation. When not available, predictor values are entered as not a number (NAN), which the model is capable of handling

ISSN: 1364-8535