HOSPITAL READMISSION PREDICTION USING PYTHON
Using EHR data to build machine learning models to predict the likelihood that a patient could be readmitted to hospital within 30 days after discharging.
This post aimed at showing how to process EHR data before building machine learning models.
The readmission rate is an important KPI to measure the percentage of patients who are admitted to hospitals again after a specific time duration such as 30 days. This metrics helps to assess the quality of care, patient outcomes and the effectiveness of post-discharge follow-up.
I use the simulated hospital data tables from Synthea to do the analysis.
LOADING DATA
In [ ]:
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, confusion_matrix, accuracy_score
In [ ]:
import pandas as pd
df = pd.read_csv("/Users/nnthieu/Downloads/Data/Readmission_models.csv")
df.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 176 entries, 0 to 175 Data columns (total 43 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 id 176 non-null object 1 start 176 non-null object 2 stop 176 non-null object 3 patient 176 non-null object 4 organization 176 non-null object 5 provider 176 non-null object 6 payer 176 non-null object 7 encounterclass 176 non-null object 8 code 176 non-null int64 9 description 176 non-null object 10 base_encounter_cost 176 non-null float64 11 total_claim_cost 176 non-null float64 12 payer_coverage 176 non-null float64 13 reasoncode 176 non-null int64 14 reasondescription 176 non-null object 15 id-2 176 non-null object 16 birthdate 176 non-null object 17 deathdate 130 non-null object 18 ssn 176 non-null object 19 drivers 176 non-null object 20 passport 176 non-null object 21 prefix 176 non-null object 22 first 176 non-null object 23 middle 59 non-null object 24 last 176 non-null object 25 suffix 1 non-null object 26 maiden 30 non-null object 27 marital 176 non-null object 28 race 176 non-null object 29 ethnicity 176 non-null object 30 gender 176 non-null object 31 birthplace 176 non-null object 32 address 176 non-null object 33 city 176 non-null object 34 state 176 non-null object 35 county 176 non-null object 36 fips 32 non-null float64 37 zip 176 non-null int64 38 lat 176 non-null float64 39 lon 176 non-null float64 40 healthcare_expenses 176 non-null float64 41 healthcare_coverage 176 non-null float64 42 income 176 non-null int64 dtypes: float64(8), int64(4), object(31) memory usage: 59.3+ KB
Create variables such as 'age', 'length of stay' based on the date of birth, hospitalization and discharge.
PROCCESSING DATA
In [ ]:
df['admitted_date'] = pd.to_datetime(df['start'])
df['discharged_date'] = pd.to_datetime(df['stop'])
df['birthdate'] = pd.to_datetime(df['birthdate'])
# Calculate the length of stay
df['length_of_stay'] = (df['discharged_date'] - df['admitted_date']).dt.days
# Calculate the age
df['age'] = (df['admitted_date'] - df['birthdate']).apply(lambda x: x.days // 365)
Create a variable called 'readmitted' to mark encounters at that patients admitted again within 30 days from previous admission. at first, it must sort data by patients and admitted days.
In [ ]:
# Sort data by patient_id and admitted_date
df.sort_values(by=['patient', 'admitted_date'], inplace=True)
In [ ]:
# Create a column to indicate if a patient was readmitted within 30 days
df['readmitted'] = 0
for i in range(len(df) - 1):
if df.iloc[i]['patient'] == df.iloc[i + 1]['patient']:
if (df.iloc[i + 1]['admitted_date'] - df.iloc[i]['discharged_date']).days <= 30:
df.at[i, 'readmitted'] = 1
Check for missing data and count the encounters with readmission.
In [ ]:
df['readmitted'].value_counts()
Out[ ]:
readmitted 0 122 1 54 Name: count, dtype: int64
In [ ]:
df.isnull().sum()
Out[ ]:
id 0 start 0 stop 0 patient 0 organization 0 provider 0 payer 0 encounterclass 0 code 0 description 0 base_encounter_cost 0 total_claim_cost 0 payer_coverage 0 reasoncode 0 reasondescription 0 id-2 0 birthdate 0 deathdate 46 ssn 0 drivers 0 passport 0 prefix 0 first 0 middle 117 last 0 suffix 175 maiden 146 marital 0 race 0 ethnicity 0 gender 0 birthplace 0 address 0 city 0 state 0 county 0 fips 144 zip 0 lat 0 lon 0 healthcare_expenses 0 healthcare_coverage 0 income 0 admitted_date 0 discharged_date 0 length_of_stay 0 age 0 readmitted 0 dtype: int64
In [ ]:
df['gender'] = df['gender'].replace({'F': 0, 'M': 1})
/var/folders/cx/3wbhcqyd3cld6gvk_xjkvr_40000gn/T/ipykernel_59497/3294156851.py:1: FutureWarning: Downcasting behavior in `replace` is deprecated and will be removed in a future version. To retain the old behavior, explicitly call `result.infer_objects(copy=False)`. To opt-in to the future behavior, set `pd.set_option('future.no_silent_downcasting', True)`
df['gender'] = df['gender'].replace({'F': 0, 'M': 1})
Select nessessory variables for building models.
In [ ]:
df_f = df[['total_claim_cost','gender','income','length_of_stay','age','readmitted']]
In [ ]:
df_f.head()
Out[ ]:
| total_claim_cost | gender | income | length_of_stay | age | readmitted | |
|---|---|---|---|---|---|---|
| 161 | 1012.63 | 0 | 77504 | 1 | 34 | 0 |
| 162 | 3480.38 | 0 | 77504 | 4 | 49 | 0 |
| 151 | 199646.60 | 1 | 29497 | 22 | 63 | 1 |
| 24 | 5750.55 | 0 | 47200 | 1 | 25 | 0 |
| 25 | 83463.56 | 0 | 47200 | 7 | 47 | 0 |
BUILDING MODELS
Building a Random Forest Classifier model.
In [ ]:
# Define features and target variable
X = df_f.drop(columns=['readmitted'])
y = df_f['readmitted']
In [ ]:
# Split data into training and testing sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)
In [ ]:
# Train a Random Forest Classifier
model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_train, y_train)
Out[ ]:
RandomForestClassifier(random_state=42)In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
RandomForestClassifier(random_state=42)
In [ ]:
# Make predictions
y_pred = model.predict(X_test)
Evaluate its performance
In [ ]:
# Evaluate the model
print('Confusion Matrix:')
print(confusion_matrix(y_test, y_pred))
print('Classification Report:')
print(classification_report(y_test, y_pred))
print('Accuracy:', accuracy_score(y_test, y_pred))
Confusion Matrix:
[[33 5]
[ 7 8]]
Classification Report:
precision recall f1-score support
0 0.82 0.87 0.85 38
1 0.62 0.53 0.57 15
accuracy 0.77 53
macro avg 0.72 0.70 0.71 53
weighted avg 0.77 0.77 0.77 53
Accuracy: 0.7735849056603774