PyTMLE

PyTMLE is a flexible Python implementation of the Targeted Maximum Likelihood Estimation (TMLE) framework for survival and competing risks outcomes.

The package can be installed from PyPI, for example using pip:

pip install pytmle

It is designed to be easy to use with default models for initial estimates of nuisance functions which are applied in a super learner framework. With a pandas dataframe containing event times, indicators, and (binary) treatment group information in specified columns, it is straight-forward to fit a main PyTMLE class object and get predictions and plots for selected target_times:

from pytmle import PyTMLE

tmle = PyTMLE(df, 
              col_event_times="time", 
              col_event_indicator="status", 
              col_group="group", 
              target_times=target_times)

tmle.plot(type="risks") # get estimated counterfactual CIF, or set to "rr" or "rd" for ATE estimates based on RR or RD
pred = tmle.predict(type="risks") # store estimates in a data frame

However, it also allows for custom models to be used for the initial estimates or even passing initial estimates directly to the second TMLE stage.

Have a look at the package's Read the Docs page for the detailed API reference and tutorial notebooks.