iptw-survival 0.1.4

A Python package for IPTW-based survival analysis with bootstrapped metrics and covariate balance diagnostics.

iptw-survival

iptw-survival is a Python package for calculating inverse probability of treatment weights (IPTW) and performing survival analysis using real-world data. It includes functionality for estimating stabilized weights, checking covariate balance, visualizing propensity scores, and calculating bootstrapped survival metrics.

Features

• Calculate stabilized or unstabilized IPTW using logistic regression
• Visualize propensity score distributions for treatment groups
• Assess covariate balance using standardized mean differences (SMD), including a Love plot
• Generate a bootstrapped Kaplan-Meier survival summary DataFrame with point estimates and bootstrapped 95% confidence intervals (useful for plotting with your preferred visualization library)
• Compute survival metrics -- all with bootstrapped 95% condience intervals -- including:
  • Probability of survival at fixed timepoints
  • Restricted mean survival time (RMST)
  • Median survival

Standard variance estimates for survival curves can be biased when using IPTW, especially when weights are non-integer or extreme. To address this, iptw-survival uses bootstrapping to provide robust confidence intervals for survival metrics.

Installation

pip install iptw-survival

Quick Start

from iptw_survival import IPTWSurvivalEstimator

# Instantiate and fit model
estimator = IPTWSurvivalEstimator()
iptw_df = estimator.fit_transform(df,
                                  treatment_col = 'treatment',
                                  cat_var = ['stage', 'ecog'],
                                  cont_var = ['age', 'creatinine'],
                                  binary_var = ['surgery', 'medicaid'],
                                  lr_kwargs = {
                                    'class_weight': 'balanced',
                                    'random_state': 42},
                                  clip_bounds = (0.01, 0.99),
                                  stabilized = True)

# Plot propensity score distribution
estimator.propensity_score_plot()

# Assess covariate balance
smd_df, fig = estimator.standardized_mean_differences(return_fig = True)

# Kaplan-Meier plot with bootstrapped CIs
km_df = estimator.km_confidence_interval(df = iptw_df
                                         event_col='event', 
                                         duration_col = 'duration',
                                         n_bootstrap = 500
                                         random_state = 42)

# Use km_df to plot survival curves 
import matplotlib.pyplot as plt

plt.plot(km_df['time'], km_df['treatment_estimate'], label = 'Treatment')
plt.fill_between(km_df['time'], km_df['treatment_lower_ci'], km_df['treatment_upper_ci'], alpha = 0.1)
plt.xlabel('Time (months)')
plt.ylabel('Survival probability')
plt.legend()
plt.title('IPTW-adjusted Kaplan-Meier Curve')
plt.show()

# Calculate survival metrics
results = estimator.survival_metrics(df = iptw_df
                                     event_col='event', 
                                     duration_col = 'duration',
                                     psurv_time_points = [24, 36],
                                     rmst_time_points = [12],
                                     median_time = True
                                     n_bootstrap = 500
                                     random_state = 42)

Example Tutorial

A full walkthrough using the Flatiron Health advanced urothelial cancer dataset is available in tutorial/tutorial.ipynb.

Requirements

Built and tested in python 3.13

Core dependencies:

• pandas
• numpy
• scikit-learn
• lifelines
• matplotlib

Contact

Contributions and feedback are welcome. Contact: [e-mail]