Table 1 Top 10 features for machine learning prediction in descending order of coefficients or feature importance returned by RF and GBT.
LR with aggregate features | RF with aggregate features | GBT with aggregate features | LR with longitudinal features | RF with longitudinal features | GBT with longitudinal features |
|---|---|---|---|---|---|
EHR length | EHR length | Age | EHR length | EHR length | Age |
Max LDL-C | Age | EHR length | Age | Age | EHR length |
Min Creatinine | Max BMI | SD Creatinine | SD Glucose in 2000 | Aspirin in 2006 | Smoking |
Age | Min BMI | Smoking | SD Creatinine in 2000 | Max SBP in 2006 | Heart valve disorders in 2006 |
Max HDL-C | Median BMI | Min BMI | Max HDL-C 2005 | Min BMI in 2006 | Hypertension in 2006 |
Max BMI | Max SBP | Heart valve disorders (Phecode 395) | SD Glucose in 2006 | Median BMI in 2005 | Aspirin in 2006 |
Max Total Cholesterol | Median SBP | Min Glucose | Median LDL-C in 2006 | Median SBP in 2006 | Disorders of lipoid metabolism in 2006 |
Max DBP | SD BMI | Max SBP | Median BMI in 2006 | Max BMI in 2006 | Clopidogrel in 2006 |
Median Triglycerides | MIN SBP | Max Triglycerides | Median Total Cholesterol in 2006 | Min BMI in 2001 | Max SBP in 2006 |
Min Cholesterol | Max DBP | Aspirin | Heart valve disorders in 2006 | Min BMI in 2002 | SD Glucose in 2006 |
