Variable selection for a prediction model of COVID-19-related death among individuals with laboratory confirmed SARS-CoV-2 infection
| Clinical variable | Base multivariable model (p value)* | Diagnoses promoted for evaluation by LASSO regression (FDR adjusted p value)† | LASSO-derived coefficients for variables retained in the final prediction model‡ | OR for the retained variable§ |
| Age | 1.76×10–144 | -- | 0.06290 | 1.065 |
| Sex-male | 2.68×10–5 | -- | 0.00718 | 1.007 |
| Race-ethnicity—African American | 5.94×10–5 | -- | -- | -- |
| Race-ethnicity—Asian | 0.070 | -- | -- | -- |
| Race-ethnicity—Latino | 0.282 | -- | -- | -- |
| Race-ethnicity—other | 0.721 | -- | -- | -- |
| Smoking status | 3.28×10–6 | -- | -- | -- |
| BMI | 0.675 | -- | -- | -- |
| Creatinine, serum | 7.28×10–16 | -- | 0.12933 | 1.138 |
| Pulmonary: respiratory failure | -- | 1.22×10–18 | 0.58317 | 1.792 |
| Cardiovascular: congestive heart failure | -- | 1.27×10–17 | 0.53928 | 1.715 |
| Pulmonary: chronic obstructive pulmonary disease | -- | 3.69×10–13 | 0.29649 | 1.345 |
| General: supplemental O2 or ventilation | -- | 3.82×10–10 | -- | -- |
| Cardiovascular: coronary artery disease | -- | 3.07×10–7 | 0.07760 | 1.081 |
| Cardiovascular: atrial fibrillation or flutter | -- | 6.27×10–7 | 0.07681 | 1.080 |
| Cardiovascular: cerebrovascular disease | -- | 7.20×10–6 | 0.12774 | 1.136 |
| Gastrointestinal: liver disease, not otherwise specified | -- | 7.20×10–6 | -- | -- |
| General: electrolyte disorder | -- | 7.74×10–6 | -- | -- |
| Infectious disease: sepsis | -- | 1.95×10–5 | -- | -- |
| Cardiovascular: pulmonary embolism or deep vein thrombosis | -- | 2.74×10–5 | -- | -- |
| Musculoskeletal: skeletal disorder affecting thorax | -- | 2.74×10–5 | 0.09123 | 1.096 |
| Psychiatric and substance use: alcohol use | -- | 2.74×10–5 | -- | -- |
| Infectious disease: pneumonia | -- | 2.98×10–5 | -- | -- |
| Cardiovascular: peripheral vascular occlusive disease | -- | 3.01×10–5 | 0.14121 | 1.152 |
| Haematologic: leucopenia | -- | 3.87×10–5 | -- | -- |
| Dermatology: pressure ulcer | -- | 7.11×10–5 | 0.21731 | 1.243 |
| Pulmonary: pulmonary vascular hypertension | -- | 8.32×10–5 | -- | -- |
| Gastrointestinal: cirrhosis | -- | 1.15×10–4 | -- | -- |
| General: nutritional deficiency | -- | 7.88×10–4 | -- | -- |
| Neurologic: traumatic brain injury | -- | 8.84×10–4 | 0.04844 | 1.050 |
| Cardiovascular: structural heart disease | -- | 1.23×10–3 | -- | -- |
| Neurologic: epilepsy | -- | 1.88×10–3 | -- | -- |
| Cardiovascular: cardiomyopathy | -- | 2.74×10–3 | -- | -- |
| Neurologic: dementia | -- | 4.12×10–3 | 0.13412 | 1.144 |
| Endocrine and metabolism: diabetes any | -- | 5.70×10–3 | -- | -- |
| Rheumatologic: sarcoidosis | -- | 7.48×10–3 | -- | -- |
| Haematologic: coagulopathy | -- | 7.49×10–3 | -- | -- |
| Psychiatric and substance use: opioid use | -- | 9.14×10–3 | -- | -- |
| Gastrointestinal: gastro-oesophageal disease | -- | 0.011 | -- | -- |
| Oncologic: solid malignancy | -- | 0.021 | -- | -- |
| Infectious disease: infectious enteritis | -- | 0.021 | -- | -- |
| Gastrointestinal: constipation | -- | 0.030 | -- | -- |
| General: muscle contraction or wasting | -- | 0.038 | -- | -- |
| Neurologic: encephalopathy | -- | 0.041 | -- | -- |
| Rheumatologic: systemic lupus erythematosus | -- | 0.049 | -- | -- |
Risk score=2340.58 + 62.90×age (in years)+7.18×sex (male=1, female=0)+129.33×serum creatinine (in mg/dL)+583.17×history of respiratory failure (yes=1, no=0)+539.28×congestive heart failure (yes=1, no=0)+296.49×chronic obstructive pulmonary disease (yes=1, no=0)+77.60×coronary artery disease (yes=1, no=0)+76.81×atrial fibrillation/atrial flutter (yes=1, no=0)+127.74×cerebrovascular disease (yes=1, no=0)+91.23×musculoskeletal disease affecting thorax (yes=1, no=0)+141.21×peripheral vascular disease (yes=1, no=0)+217.31×pressure ulcer (yes=1, no=0)+48.44×traumatic brain injury (yes=1, no=0)+134.12×dementia (yes=1, no=0).
*The base model included all of the variables shown; p values were derived from the adjusted model.
†P values are adjusted for the false discovery rate. Each clinical variable is separately adjusted by the following model: COVID-19 death ~age+sex+race-ethnicity+smoking status +BMI+serum creatinine +clinical variable. Online supplemental table 1 shows the additional clinical variables which were screened but which did not meet the criteria for evaluation by LASSO regression for prediction model building.
‡Variable weights using in the actual prediction model were obtained by multiplying each coefficient by 1000; this ensured that all model weights were >1. The final 14-variable prediction model had the following form.
§OR for COVID-19-related death according to a one-unit increase in the listed variable (coding and directionality for each variable retained is described in the preceding footnote).
BMI, body mass index; FDR, false discovery rate; LASSO, least absolute shrinkage and selection operator.