Table 2. Top variables selected by machine learning procedure that increase risk of readmission within 30 days.
Top supportive evidence | |||
Variable class | Variable description | Log Odds Ratio | Log Odds Ratio Standard Error1 |
Lab Results | Lymphocyte % is low | 0.0128 | 0.0027 |
Patterns of Engagement | Patient was admitted in past 6 months | 0.0112 | 0.0031 |
Lab Results | BUN is high | 0.0038 | 0.0012 |
Lab Results | Glucose level random is elevated | 0.003 | 0.0012 |
Lab Results | Monocyte absolute is low | 0.0028 | 0.0012 |
Other Diagnoses | History of nondependent abuse of drugs (ICD9 305.x) | 0.0018 | 0.001 |
Other Diagnoses | History of chronic airway obstruction, not elsewhere classified (ICD9 496.x) | 0.0017 | 0.0008 |
Other Diagnoses | History of gastrointestinal hemorrhage (ICD9 578.x) | 0.0014 | 0.0007 |
Lab Results | AST is elevated | 0.0013 | 0.0006 |
Other Diagnoses | History of cardiomyopathy (ICD9 425.x) | 0.001 | 0.0006 |
Lab Results | Magnesium is low | 0.001 | 0.0006 |
Lab Results | INR is elevated | 0.0009 | 0.0004 |
Patterns of Engagement | Patient has been in isolated room in hospital | 0.0009 | 0.0006 |
Lab Results | BNP is high | 0.0007 | 0.0005 |
These are variables that receive positive log-odds ratio with the largest magnitude.
1. Obtained from sample standard error for cross-validation odds ratios