Update app.py

app.py

@@ -32,6 +32,8 @@ import torch.nn as nn
 import time
 
 
+# Defining wide mode
+st.set_page_config(layout="wide")
 
 # Criar o menu na barra lateral
 st.sidebar.title("📌 Menu")
@@ -277,8 +279,7 @@ elif page== "📊 Tabular Data":
     fixed_feature_vector = {age_group_mapping.get(k, k): v for k, v in feature_vector.items()}
     feature_df = pd.DataFrame([fixed_feature_vector]).reindex(columns=expected_columns, fill_value=0)
     
-    #
-    st.write(feature_df)
+    #st.write(feature_df)
     # Predict probability of readmission
     prediction_proba = tabular_model.predict_proba(feature_df)[:, 1]
     probability = float(prediction_proba[0])  # Convert NumPy array to scalar
@@ -333,6 +334,8 @@ elif page== "📊 Tabular Data":
         import streamlit as st
         import time
         import random
+        from textwrap import dedent
+
 # Título estilizado
         st.markdown("""
             <h1 style='text-align: center; color: #2c3e50;'>🩺 AI-Powered Patient Readmission Analysis</h1>
@@ -350,36 +353,41 @@ elif page== "📊 Tabular Data":
             time.sleep(2)  # Simular carregamento
             try:
                 ai_output = """
-                To understand why the model made a certain prediction, let's analyze the feature impacts provided. The feature impacts represent the change in the predicted outcome (patient readmission) for a one-unit change in the feature, while holding all other features constant. The magnitude and direction of the impact indicate how much each feature contributes to the prediction.
-
-                Key Characteristics Influencing the Prediction:
-                Previous Stays (1.56 impact): This feature has the highest positive impact, suggesting that patients with a history of previous stays at the hospital are more likely to be readmitted. The large impact indicates that the model places significant weight on this factor, likely because repeated hospital visits can signify chronic conditions or complications that are not fully resolved.
-                
-                Number of Medications (n_meds) (0.17 impact): Patients taking a higher number of medications are at a higher risk of readmission. This could be due to the complexity of their medical conditions, potential side effects, or interactions between medications that may lead to further health issues.
+                # 🩺 AI-Powered Patient Readmission Analysis
                 
-                Time Since Last Stay (0.16 impact): The time elapsed since the patient's last hospital stay also positively influences the prediction of readmission. This might indicate that patients who have been discharged recently are at a higher risk of returning, possibly due to incomplete recovery or the nature of their condition requiring ongoing care.
+                ## 🤖 Understanding the Model's Prediction
+                The feature impacts indicate how much each factor contributes to the model's decision to predict patient readmission. The magnitude and direction of the impact determine its importance.
                 
-                Real Age (0.14 impact): The patient's age is another factor that increases the likelihood of readmission. Older patients may have more complex health issues, diminished physiological reserve, and a higher likelihood of comorbid conditions, all of which can contribute to the need for repeat hospitalizations.
+                ---
                 
-                Length of Stay (los_days) (0.07 impact): Although less influential than the top factors, a longer hospital stay during the current or previous admission(s) slightly increases the risk of readmission. This could be indicative of more severe illness, complications, or the need for prolonged recovery periods.
+                ## 🔍 Key Characteristics Influencing the Prediction:
+                - **Previous Stays** (1.56 impact): This feature has the highest positive impact, suggesting that patients with a history of previous hospital stays are more likely to be readmitted. The large impact indicates that the model places significant weight on this factor, likely because repeated hospital visits can signify chronic conditions or complications that are not fully resolved.
+                - **Number of Medications (n_meds)** (0.17 impact): Patients taking a higher number of medications are at a higher risk of readmission. This could be due to the complexity of their medical conditions, potential side effects, or interactions between medications that may lead to further health issues.
+                - **Time Since Last Stay** (0.16 impact): The time elapsed since the patient's last hospital stay also positively influences the prediction of readmission. This might indicate that patients who have been discharged recently are at a higher risk of returning, possibly due to incomplete recovery or the nature of their condition requiring ongoing care.
+                - **Real Age** (0.14 impact): The patient's age is another factor that increases the likelihood of readmission. Older patients may have more complex health issues, diminished physiological reserve, and a higher likelihood of comorbid conditions, all of which can contribute to the need for repeat hospitalizations.
+                - **Length of Stay (los_days)** (0.07 impact): Although less influential than the top factors, a longer hospital stay during the current or previous admission(s) slightly increases the risk of readmission. This could be indicative of more severe illness, complications, or the need for prolonged recovery periods.
                 
-                Less Influential but Still Relevant Factors:
-                CCI Score (0.05 impact): The Charlson Comorbidity Index (CCI) score, which predicts the ten-year mortality for a patient who may have a range of comorbid conditions, such as heart disease, diabetes, or cancer. A higher score indicates a higher risk of mortality and, by extension, potentially a higher risk of readmission due to the complexity of the patient's health conditions.
+                ---
                 
-                DRG Severity (-0.04 impact): The Diagnosis-Related Group (DRG) severity, which categorizes hospital cases into one or more groups that are expected to have similar hospital resource use, has a negative impact. This suggests that higher severity cases, as categorized by DRG, might actually have a slightly lower risk of readmission, possibly due to more intensive treatment and monitoring during their initial stay.
+                ## 📉 Less Influential but Still Relevant Factors:
+                - **CCI Score** (0.05 impact): The Charlson Comorbidity Index (CCI) score predicts the ten-year mortality for a patient with comorbid conditions such as heart disease, diabetes, or cancer. A higher score indicates a higher risk of mortality and, by extension, potentially a higher risk of readmission due to the complexity of the patient's health conditions.
+                - **DRG Severity** (-0.04 impact): The Diagnosis-Related Group (DRG) severity categorizes hospital cases based on expected resource use. Its negative impact suggests that higher severity cases might actually have a slightly lower risk of readmission, possibly due to more intensive treatment and monitoring during their initial stay.
+                - **Race** (race_WHITE: -0.04 impact, race_UNKNOWN: 0.03 impact): The impacts of race are relatively small and might reflect underlying socio-economic or healthcare access disparities rather than direct biological factors. However, interpreting these impacts requires caution due to the potential for confounding variables and the ethical considerations surrounding race in healthcare outcomes.
+                - **Blood Cells** (0.03 impact): This factor, likely referring to some measure of blood cell count or health, has a minor positive impact, suggesting that abnormalities in blood cell counts could slightly increase the risk of readmission, potentially due to underlying conditions affecting the blood or bone marrow.
                 
-                Race (race_WHITE: -0.04 impact, race_UNKNOWN: 0.03 impact): The impacts of race are relatively small and might reflect underlying socio-economic or healthcare access disparities rather than direct biological factors. However, interpreting these impacts requires caution due to the potential for confounding variables and the ethical considerations surrounding race in healthcare outcomes.
+                ---
                 
-                Blood Cells (0.03 impact): This factor, likely referring to some measure of blood cell count or health, has a minor positive impact, suggesting that abnormalities in blood cell counts could slightly increase the risk of readmission, potentially due to underlying conditions affecting the blood or bone marrow.
+                ## ✅ Conclusion:
+                The model's prediction of patient readmission is most strongly influenced by the patient's history of previous hospital stays, indicating a potential for chronic or recurring health issues. The number of medications, time since the last stay, and the patient's age are also significant factors, highlighting the complexity of the patient's health condition and the potential for ongoing care needs. 
                 
-                Conclusion:
-                The model's prediction of patient readmission is most strongly influenced by the patient's history of previous hospital stays, indicating a potential for chronic or recurring health issues. The number of medications, time since the last stay, and the patient's age are also significant factors, highlighting the complexity of the patient's health condition and the potential for ongoing care needs. While other factors such as CCI score, DRG severity, race, and blood cell health play a role, their impacts are less pronounced. Understanding these factors can help healthcare providers identify high-risk patients and implement targeted interventions to reduce the likelihood of readmission.
+                While other factors such as CCI score, DRG severity, race, and blood cell health play a role, their impacts are less pronounced. Understanding these factors can help healthcare providers identify high-risk patients and implement targeted interventions to reduce the likelihood of readmission.
+
                 """
         
                 # 🎭 **Show AI Response in a Stylish Chat Format**
                 with st.chat_message("assistant"):
-                    st.markdown(f"**💡 AI Explanation:**\n\n{ai_output}")
-            
+                    st.markdown(f"**💡 AI Explanation:**\n\n")
+                    st.markdown(dedent(ai_output))
             except Exception as e:
                 st.error(f"⚠️ Error retrieving response: {e}")
                 st.stop()