- Diabetes is a serious health problem that affects many Americans. In clinical care, there are many risk factors that directly affects the likelihood of a patient with diabetes to be readmitted within 30 days of discharge.
- A Logistic Regression model with SMOTE was performed to classify patients in this category with an accuracy of 62% (formal report), however it was a good starting point to pinpoint the top 10 important features that contribute to the increase of classification for hospital readmission.
Key Findings: The top 10 important features are listed in the table below, along with their regressor weight.
| Num | Feature | Weight | Note |
|---|---|---|---|
| 1 | discharge_disposition_id_8 | 0.612653 | Transferred to home under care of Home IV provider |
| 2 | admission_source_id_7 | 0.269167 | Admission by Emergency Room |
| 3 | gender_Female | 0.268416 | Distribution of females are slightly less than males |
| 4 | metformin-rosiglitazone_No | 0.175598 | Medicine combination used to treat type 2 diabetes |
| 5 | admission_source_id_8 | 0.169067 | Admission by Court/Law Enforcement |
| 6 | num_procedures | 0.113014 | Number of procedures done |
| 7 | max_glu_serum_>300 | 0.070908 | Simple and direct single test for diabetes |
| 8 | diag_3_Diabetes | 0.060908 | Diabetes as one of patient’s diagnoses |
| 9 | miglitol_No | 0.058262 | Oral anti-diabetic drug that helps patient breaks down complex carbohydrates into glucose |
| 10 | glipizide-metformin_Steady | 0.043848 | Medicine combination used to treat high blood sugar levels caused by type 2 diabetes |
A Tableau Dashboard was created to visualize these 10 important features.
- The dataset, obtained from the Center for Machine Learning and Intelligent Systems at University of California, Irvine, represents 10 years (1999-2008) of clinical care at 130 US hospitals and integrated delivery networks. It includes 100,000+ entries and 50 features, representing patient and hospital attributes such as patient identification, diagnosis codes, admission type, source, and discharge disposition, risk related medications and test results, and numerous other hospitalization indicators.
- UCI Diabetes 130-US hospitals for years 1999-2008
- Impact of HbA1c Measurement on Hospital Readmission Rates: Analysis of 70,000 Clinical Database Patient Records
- Replace '?' values with numpy nan
- Remove 'encounter_id' and 'patient_nbr' since unique IDs
- Remove 'weight' since 97% missing
- Remove 'payer_code' and 'medical_specialty' both have 40% and 49% missing
- Remove 'examide' and 'citoglipton' since both have only 1 value (zero variance)
- Impute 'race' randomly based on its categorical proportions
- Classify ranges of 'diag_1', 'diag_2', and 'diag_3' into categorized diagnoses using 2nd link above
- Classify 'admission_type_id', 'admission_source_id', and 'discharge_disposition_id' according to given IDs_mapping.csv, while joining various categories to prepare for visualization
- Convert response variable 'readmitted' into binary ('NO': 0, '>30': 0, '<30': 1)