By comprehensively applying data analysis and visualization methods, we completed the exploration of key variable distributions, comparison of variable characteristics, and model evaluation.

Project Background

Chronic Kidney Disease (CKD) refers to a group of disorders characterized by progressive, irreversible damage to renal structure and function. It has evolved into an increasingly severe global public health challenge. According to data from the 2023 Global Burden of Disease Study (GBD), the global prevalence of CKD has surged to 14.2%. Exhibiting a sustained upward trajectory over the past several decades, CKD is now the ninth leading cause of death worldwide, imposing substantial medical and economic burdens on affected individuals. During the clinical course of CKD, Acute Kidney Injury (AKI) represents one of the most life-threatening complications. CKD complicated by AKI is defined as a clinical syndrome characterized by a rapid decline in glomerular filtration rate over a short period due to various predisposing factors, in the context of pre-existing chronic renal impairment. Compared with AKI in the general population, AKI complicating CKD is associated with far poorer clinical outcomes. In the short term, renal function often fails to recover to baseline levels; some patients may progress directly to renal failure, necessitating urgent renal replacement therapy. In the long term, such patients face a significantly elevated risk of progressing to End-Stage Renal Disease (ESRD), accompanied by an increased incidence of cardiovascular complications. This study aims to develop a prognostic model for AKI in CKD patients, which is crucial for the prevention of AKI in this population.

Methods

In the process of data analysis, we not only need to explore the distribution of key variables in the data, but also compare the variable characteristics based on different outcomes, and evaluate the established models simultaneously. This project completed the above work through data visualization methods using tools such as histograms, box plots, confusion matrices, and ROC curves, providing an intuitive basis for data analysis and decision-making.

Analysis Results

Key Variable Distribution Characteristics

Through this histogram, we can intuitively observe the distribution of key variables, which helps to grasp the central tendency and dispersion of the data.

Comparison of Variable Characteristics Based on Different Outcomes

This box plot shows the differences in variable characteristics based on different outcomes, providing a basis for further analysis and decision-making.

Model Evaluation - Confusion Matrix

The confusion matrix indicates that the model has high accuracy in predicting the incidence of AKI, demonstrating good performance in this aspect.

Model Evaluation - ROC Curve

The AUC (Area Under the Curve) is 0.781, which indicates that the model has good predictive efficacy for the incidence of AKI in patients with CKD.

Core Implementation

The following is the implementation of the core code:

# Data Processing Section
# Handle missing values
data = data.dropna()
# Extract features and conduct comparison of baseline data
columns = ['age', 'sex', 'Ethnicity', 'region', "TDI","Employment","education_level","Final_Healthy_diet_score","Smoking","alcohol_intake",
           "BMI","tv","sleep_duration","grip_strength","MVPA_self","urea","urate", "LDL_C","SBP","DBP","creatinine","hba1c"]
categorical = ['sex', 'Ethnicity', 'region', "Employment","education_level","Smoking","alcohol_intake"]
groupby = "Incidental.AKI.HDC.ICD10"
table = TableOne(data,columns=columns,categorical=categorical,groupby=groupby,pval=True)
# Draw histograms and box plots for feature analysis
for idx, var in enumerate(vars_to_plot):
    sns.histplot(x=var,data=data,alpha=0.7,kde=True,color=plt.cm.Set2(idx/len(vars_to_plot)),edgecolor='black',linewidth=0.5,ax=axs_flat[idx])
for idx, var in enumerate(vars_to_plot):
    sns.boxplot(data=data,x=group_var,y=var,ax=axs_flat[idx],palette='Set2',linewidth=1,fliersize=2)
# Model Construction
from sklearn.linear_model import LogisticRegression
model = LogisticRegression()
# Model Evaluation - Confusion Matrix
def confusion_matrix_plot(y_true, y_pred_prob, threshold=0.52, title='Confusion Matrix (Scaled Model)'):
    y_pred = (y_pred_prob > threshold).astype(int)
    cm = confusion_matrix(y_true, y_pred)
    fig, ax = plt.subplots(figsize=(6, 5))
    sns.heatmap(cm,annot=True,fmt='d',cmap='Blues',ax=ax,xticklabels=['Non-AKI(0)', 'AKI(1)'],yticklabels=['Non-AKI(0)', 'AKI(1)'])
# ROC Curve
def plot_roc_curve(y_true, y_pred_prob, title='ROC Curve (Kidney Disease Prediction - Scaled Model)'):
    fpr, tpr, thresholds = roc_curve(y_true, y_pred_prob)
    auc_score = roc_auc_score(y_true, y_pred_prob)
    youden_index = tpr - fpr)
# SHAP Interpretability Analysis
scaler = pipeline.named_steps['scaler']
lr_model = pipeline.named_steps['lr']
explainer = shap.Explainer(lr_model, X_train)
shap_values = explainer(X_train)
shap.summary_plot(shap_values, X_train)
#KMeans Clustering -Elbow Method
for i in range(1, max_clusters + 1):
    kmeans = KMeans(
        n_clusters=i, 
        init='k-means++', 
        random_state=42,
        n_init=10)
    kmeans.fit(data_clustering_scaled)
    wcss.append(kmeans.inertia_)
    print(f"Number of clusters={i}, WCSS={kmeans.inertia_:.2f}")