Table 2.

The potential list of candidates for the multivariable model in Table 3 was generated from a random forest model. Candidate variables for inclusion in the multivariable model were generated from a random forest that included indicator variables shown above for every ICD-9 diagnosis code, procedure code, lab, image, and pharmaceutical. 1000 conditional inference trees were fit to bootstrap samples, and aggregated by averaging the observation weights. Conditional variable importance was used to select candidate variables. This was performed using the ‘Party’ package for R v.3.2.

Factors Considered for model
Antibiotics	120000 Anti-infective agents
Fluids	146218 Sodium Chloride
Blood products	99.04 transfusion of pRBC 99.05 transfusion of platelets 99.06 transfusion of coagulation factors 99.08 transfusion of blood expander
Vasoactive Agents	131311 dopamine 131305 dobutamine 131321 epinephrine 131111 milrinone 131351 norepinephrine
Infectious testing	361100 Bacterial cultures, unspecified (includes aerobic and anaerobic from any source) 361200 Bacterial test, unspecified (antibody and PCR based tests 362000 Yeast and fungi, unspecified 363000 Parasites, unspecified 364000 Viruses, unspecified
Other lab testing (Lactate, Blood Gas)	313300 Lactate 311100 Blood gas
Radiology	87.44 Chest X-ray
Procedures
	38.91 arterial catheterization
	38.97 central venous catheter placement with guidance
	39.65 ECMO
	89.61 systemic arterial pressure monitoring
	89.62 central venous pressure monitoring
	93.90 non-invasive mechanical ventilation
	96.70–96.72 continuous invasive mechanical ventilation
	96.04 insertion of endotracheal tube
Diagnoses
	995.92 Severe sepsis
	785.52 Septic shock
	038.9 Unspecified septicemia
	287.5 Thrombocytopenia, unspecified
	518.81 Acute respiratory failure
	276.2 Acidosis
	276.69 Other fluid overload
	255.41 Glucocorticoid deficiency