app.py

#file_path = "cleaned_bmd_medication_data.xlsx"

import streamlit as st
import pandas as pd
import plotly.graph_objs as go

# Constants from linear regression
REGRESSION_CONSTANTS = {
    'Femoral Neck': {
        'Female': {'mu': 0.916852, 'sigma': 0.120754},
        'Male': {'mu': 0.9687385325352573, 'sigma': 0.121870698023835}
    },
    'Total Hip': {
        'Female': {'mu': 0.955439, 'sigma': 0.125406},
        'Male': {'mu': 0.967924895046735, 'sigma': 0.13081439619361657}
    },
    'Lumbar spine (L1-L4)': {
        'Female': {'mu': 1.131649, 'sigma': 0.139618},
        'Male': {'mu': 1.1309707991669353, 'sigma': 0.1201836924980611}
    }
}

# References content
REFERENCES = [
    "Cosman F, Crittenden DB, Ferrari S, et al. FRAME Study: The Foundation Effect of Building Bone With 1 Year of Romosozumab Leads to Continued Lower Fracture Risk After Transition to Denosumab. J Bone Miner Res. 2018;33(7):1219-1226. doi:10.1002/jbmr.3427",
    "McClung MR, Brown JP, Diez-Perez A, et al. Effects of 24 months of treatment with romosozumab followed by 12 months of denosumab or placebo in postmenopausal women with low bone mineral density: a randomized, double-blind, phase 2, parallel group study. J Bone Miner Res. 2018;33(8):1397–1406.",
    "McClung MR, San Martin J, Miller PD, et al. Opposite bone remodeling effects of teriparatide and alendronate in increasing bone mass [published correction appears in Arch Intern Med. 2005 Oct 10;165(18):2120]. Arch Intern Med. 2005;165(15):1762-1768. doi:10.1001/archinte.165.15.1762",
    "Black DM, Schwartz AV, Ensrud KE, et al. Effects of continuing or stopping alendronate after 5 years of treatment: the Fracture Intervention Trial Long-term Extension (FLEX): a randomized trial. JAMA. 2006;296(24):2927-2938. doi:10.1001/jama.296.24.2927",
    "Black, D.M., et al., Randomised trial of effect of alendronate on risk of fracture in women with existing vertebral fractures. Fracture Intervention Trial Research Group. Lancet, 1996. 348(9041): p. 1535-41.",
    "Harris, S.T., et al., Effects of risedronate treatment on vertebral and nonvertebral fractures in women with postmenopausal osteoporosis: a randomized controlled trial. Vertebral Efficacy With Risedronate Therapy (VERT) Study Group. JAMA, 1999. 282(14): p. 1344-52.",
    "McClung, M.R., et al., Effect of risedronate on the risk of hip fracture in elderly women. Hip Intervention Program Study Group. N Engl J Med, 2001. 344(5): p. 333-40.",
    "Reginster, JY., Minne, H., Sorensen, O. et al. Randomized Trial of the Effects of Risedronate on Vertebral Fractures in Women with Established Postmenopausal Osteoporosis. Osteoporos Int 11, 83–91 (2000).",
    "Chesnut CH 3rd, Skag A, Christiansen C, et al. Effects of oral ibandronate administered daily or intermittently on fracture risk in postmenopausal osteoporosis. J Bone Miner Res. 2004;19(8):1241-1249. doi:10.1359/JBMR.040325",
    "10 years of denosumab treatment in postmenopausal women with osteoporosis: results from the phase 3 randomised FREEDOM trial and open-label extension Bone, Henry G et al. The Lancet Diabetes & Endocrinology, Volume 5, Issue 7, 513 - 523.",
    "Ste-Marie LG, Sod E, Johnson T, Chines A. Five years of treatment with risedronate and its effects on bone safety in women with postmenopausal osteoporosis. Calcif Tissue Int. 2004;75(6):469-476. doi:10.1007/s00223-004-0039-7."
]

# Load medication data
@st.cache_data
def load_medication_data():
    file_path = "cleaned_bmd_medication_data.xlsx"
    return pd.read_excel(file_path)

# Calculate predicted BMD after medication
def calculate_bmd(bmd, percentage_increase):
    return bmd * (1 + percentage_increase)

# Convert BMD to T-score
def calculate_tscore(bmd, mu, sigma):
    return (bmd - mu) / sigma

# Display References
def display_references():
    with st.expander("References"):
        for i, ref in enumerate(REFERENCES, start=1):
            st.markdown(f"{i}. {ref}")

# Generate prediction table for all drugs
def generate_predictions(medication_data, site, bmd, mu, sigma):
    site_data = medication_data[medication_data['Site'] == site]
    all_results = []
    
    for _, row in site_data.iterrows():
        drug = row['Medication']
        predictions = {
            'Year': ['0'], 
            'Year Index': [0],  # Numeric x-axis for plotting
            'Predicted BMD': [round(bmd, 3)], 
            'Predicted T-score': [round(calculate_tscore(bmd, mu, sigma), 1)]
        }
        
        year_index = 1
        for year in row.index[1:-1]:  # Skip 'Medication' and 'Site' columns
            if not pd.isna(row[year]):
                percentage_increase = row[year]
                predicted_bmd = bmd * (1 + percentage_increase)
                predicted_tscore = calculate_tscore(predicted_bmd, mu, sigma)
                
                predictions['Year'].append(year.replace(" Year", ""))  # Simplify year label
                predictions['Year Index'].append(year_index)  # Numeric x-axis
                predictions['Predicted BMD'].append(round(predicted_bmd, 3))
                predictions['Predicted T-score'].append(round(predicted_tscore, 1))
                year_index += 1
        
        all_results.append({'Drug': drug, 'Predictions': predictions})
    return all_results


# Display results as table and plots
def display_results(predictions, site):
    st.subheader(f"Predictions for {site}")
    
    for result in predictions:
        drug = result['Drug']
        predictions = result['Predictions']
        
        # Display table
        st.write(f"### {drug}")
        st.dataframe(pd.DataFrame(predictions))
        
        # Plot BMD and T-score using Year Index
        bmd_plot = go.Scatter(
            x=predictions['Year Index'], y=predictions['Predicted BMD'], mode='lines+markers',
            name='Predicted BMD', line=dict(color='blue')
        )
        tscore_plot = go.Scatter(
            x=predictions['Year Index'], y=predictions['Predicted T-score'], mode='lines+markers',
            name='Predicted T-score', line=dict(color='green')
        )
        
        # Combine plots in a single row
        col1, col2 = st.columns(2)
        with col1:
            st.plotly_chart(go.Figure(data=[bmd_plot], layout=go.Layout(
                title=f"{drug} - Predicted BMD", xaxis_title="Year", yaxis_title="BMD (g/cm²)",
                xaxis=dict(tickmode='array', tickvals=predictions['Year Index'], ticktext=predictions['Year'])
            )))
        with col2:
            st.plotly_chart(go.Figure(data=[tscore_plot], layout=go.Layout(
                title=f"{drug} - Predicted T-score", xaxis_title="Year", yaxis_title="T-score",
                xaxis=dict(tickmode='array', tickvals=predictions['Year Index'], ticktext=predictions['Year'])
            )))

# Generate summary of medications reaching the target T-score
def generate_goal_summary(predictions, target_tscore=-2.4):
    def year_to_int(year):
        # Convert "1st", "2nd", "3rd", etc., to numeric values
        try:
            return int(year.rstrip("stndrdth"))  # Remove suffixes like "st", "nd", "rd", "th"
        except ValueError:
            return 0  # Default to 0 if year cannot be converted

    goal_reached = []
    
    for result in predictions:
        drug = result['Drug']
        predictions_data = result['Predictions']
        
        for year, tscore in zip(predictions_data['Year'], predictions_data['Predicted T-score']):
            if tscore >= target_tscore:
                # Convert year to an integer using helper function
                numeric_year = year_to_int(year)
                goal_reached.append({'Medication': drug, 'Year': numeric_year})
                break  # Stop checking further years for this drug

    # Sort by year to prioritize earlier achievement
    goal_reached_sorted = sorted(goal_reached, key=lambda x: x['Year'])
    return goal_reached_sorted

# Display summary of goal-reaching medications
def display_goal_summary(goal_summary, target_tscore):
    st.subheader(f"Goal Treatment Summary (T-score ≥ {target_tscore:.1f})")
    
    if not goal_summary:
        st.info("No medications reach the target T-score.")
    else:
        summary_table = pd.DataFrame(goal_summary)
        st.table(summary_table)

# Medication Selection with Collapsible Categories
def select_medications():
    st.subheader("Select Medications to Display")
    show_all = st.checkbox("Show All Medications", key="show_all")
    
    selected_medications = []
    if not show_all:
        # Define categories and medications
        categories = {
            "Bisphosphonates": [
                "Alendronate", "Risedronate", "Ibandronate oral", 
                "Zoledronate", "Ibandronate IV (3mg)"
            ],
            "RANK Ligand Inhibitors": [
                "Denosumab", "Denosumab + Teriparatide"
            ],
            "Anabolic Agents": [
                "Teriparatide", "Teriparatide + Denosumab"
            ],
            "Sclerostin Inhibitors": [
                "Romosozumab", "Romosozumab + Denosumab", 
                "Romosozumab + Alendronate", "Romosozumab + Ibandronate", 
                "Romosozumab + Zoledronate"
            ]
        }
        
        # Create collapsible sections
        for category, medications in categories.items():
            with st.expander(category):
                for med in medications:
                    # Use a unique key for each checkbox
                    if st.checkbox(med, key=f"{category}_{med}"):
                        selected_medications.append(med)
    else:
        # Include all medications if "Show All" is selected
        selected_medications = [
            "Alendronate", "Risedronate", "Ibandronate oral", 
            "Zoledronate", "Ibandronate IV (3mg)", "Denosumab", 
            "Denosumab + Teriparatide", "Teriparatide", 
            "Teriparatide + Denosumab", "Romosozumab", 
            "Romosozumab + Denosumab", "Romosozumab + Alendronate", 
            "Romosozumab + Ibandronate", "Romosozumab + Zoledronate"
        ]
    
    return selected_medications

# Streamlit UI
# Main function
def main():
    st.title("BMD and T-score Prediction Tool")
    
    # Add a descriptive line for the tool
    st.markdown(
        "### A tool for initiating and evaluating osteoporosis treatment to achieve individualized T-score goals."
    )
    
    # DEXA Machine Selection
    dexa_machine = st.selectbox("DEXA Machine", ["LUNAR"])
    
    # Gender Selection
    gender = st.selectbox("Gender", ["Female", "Male"])
    
    # Location (Site) Selection with Mapping
    site_mapping = {
        'Lumbar spine (L1-L4)': 'LS',
        'Femoral Neck': 'FN',
        'Total Hip': 'TH'
    }
    site_options = list(site_mapping.keys())
    selected_site = st.selectbox("Select Region (Site)", site_options)
    site = site_mapping[selected_site]  # Map to the actual value in the dataset
    
    # Input patient data
    bmd_patient = st.number_input(
        "Initial BMD",
        min_value=0.000, max_value=2.000,
        value=0.800, step=0.001,
        format="%.3f"
    )
    
    # Input for personalized T-score goal
    treatment_goal_tscore = st.number_input(
        "Targeted T-score Goal (default: -2.4):",
        min_value=-2.4, max_value=-1.0, step=0.1, value=-2.4,
        format="%.1f"
    )
    
    # Medication Selection
    selected_medications = select_medications()
    
    # Load constants and medication data
    medication_data = load_medication_data()
    constants = REGRESSION_CONSTANTS[selected_site][gender]

    # Generate and display predictions for selected medications
    if st.button("Predict"):
        all_predictions = generate_predictions(medication_data, site, bmd_patient, constants['mu'], constants['sigma'])
        filtered_predictions = [pred for pred in all_predictions if pred['Drug'] in selected_medications]
        
        if not filtered_predictions:
            st.warning("No medications selected. Please select at least one medication or use the 'Show All' option.")
        else:
            # Generate and display goal treatment summary using the specified T-score goal
            goal_summary = generate_goal_summary(filtered_predictions, target_tscore=treatment_goal_tscore)
            display_goal_summary(goal_summary, target_tscore=treatment_goal_tscore)
            
            # Display individual medication results
            display_results(filtered_predictions, selected_site)

    # References always visible at the bottom
    display_references()
    
if __name__ == "__main__":
    main()