Table 1.

Features extracted for three training datasets: features on first calendar day of admission, last available value and selected 1 day at random from patient’s stay

Dataset sample	Feature engineering	Variable
Data from admission	Quintile binning on training set for continuous variables	Demographic and hospital characteristics: previous positive COVID-19 PCR test during an outpatient or inpatient visit within 60 days, race, age, sex, body mass index (BMI) and days in hospital (current day minus admission date).
Data from first calendar day at admission, last available value, and 1 day selected at random from patient’s stay	Quintile binning on training set variables: current value, first value, minimum value, maximum value, mean value, median value, difference in current value from mean, difference in current value from median, difference in first value from mean, difference in first value from median, difference in max value from mean, difference in max value from median, difference in minimum value from mean and difference in minimum value from median	Laboratory values: albumin, alkaline phosphatase (ALKPHOS), alanine aminotransferase (ALT), anion gap (ANIONGAP), activated partial thromboplastin time (APTT), aspartate aminotransferase (AST), atypical lymphocytes per cent (ATYLYMREL), bands per cent (BANDSPCT), conjugated bilirubin (BILIDB), bilirubin direct (BILIDIRECT), bilirubin total, natriuretic peptide B (BNPEPTIDE), blood urea nitrogen (BUN), calcium, CKTOTAL, chloride, carbon dioxide (CO2), creatinine, C reactive protein (CRP), d-dimer, glomerular filtration rate – African American (EGGRAA), glomerular filtration rate – non-African American (EGFRNONAA), erythrocyte sedimentation rate (ESR), ferritin, fibrinogen, fraction of inspired oxygen arterial blood gas (FIO2ABG), glucose, HCT, haemoglobin, haemoglobin (HA1C), immunoglobulin A (IGA), immunoglobulin G (IGG), glomerular basement membrane (IGBM), absolute immature granulocytes (IMMGRANABS), per cent immature granulocytes (IMMGRANPCT), interleukin-1 beta (INTERL1B), interleukin 6 INTRLKN6, potassium (K), potassium plasma (KPLA), lactate arterial blood gas (LACTATEABG), lactate venous blood gas (LACTATEVBG), lactate dehydrogenase (LDH), lipase, lymphocyte absolute calculated (LYMPABSCAL), lymphocyte per cent (LYMPHPCT), lymphocyte absolute (LYMPHSABS), magnesium (MG), sodium (NA), NEUTABSCAL, neutrophil absolute (NEUTSABS), neutrophils per cent (NEUTSPCT), carbon dioxide in arterial blood (PCO2ART), carbon dioxide in venous blood (PCO2VEN), pH of arterial blood (PHART), phosphorous, pH of venous blood (PHVBG), platelet, P02ABG, P02VB, procalcitonin (PROCAL), total protein (PROTTOTAL), prothrombin time (PT), platelet poor plasma (PTT), red blood cell (RBC), troponin (TROPONINI), troponin point of care (TRPNONPOC) and white blood cell count (WBC).
Data from first calendar day at admission, last available value, and 1 day selected at random from patient’s stay	Quintile binning on training set: current value, first value, minimum value, maximum value, mean value, median value, difference in current value from mean, difference in current value from median, difference in first value from mean, difference in first value from median, difference in max value from mean, difference in max value from median, difference in minimum value from mean and difference in minimum value from median	Vitals: systolic blood pressure, diastolic blood pressure, pulse pressure, oximetry, respiratory rate, pulse and temperature.