fedeca.competitors

class PooledIPTW(treated_col='treated', event_col='E', duration_col='T', ps_col='propensity_scores', effect='ATE', variance_method='naive', n_bootstrap=200, seed=None, cox_fit_kwargs=None)

Bases: BaseSurvivalEstimator, BootstrapMixin

Class for the Pooled IPTW.

Parameters:

• variance_method (Literal['naive', 'robust', 'bootstrap']) –

• n_bootstrap (int) –

• seed (int | Sequence[int] | BitGenerator | SeedSequence | Generator | None) –

fit(data, targets=None)

Estimate the treatment effect via Inverse Probability Treatment Weighting.

Option to estimate the variance of estimation by bootstrapping.

targets: pd.DataFrame, optional

pre-computed propensity scores. It is possible to pass pre-computed propensity scores to the fit function to use in the IPTW estimator instead of estimating the scores using logistic regression.

Parameters:

• data (DataFrame) –

• targets (DataFrame | None) –

point_estimate(data)

Return a point estimate of the treatment effect.

Return type:

ndarray

Parameters:

data (DataFrame) –

class MatchingAjudsted(treated_col='treated', event_col='E', duration_col='T', ps_col='propensity_scores', variance_method='naive', n_bootstrap=200, seed=None, cox_fit_kwargs=None)

Bases: BaseSurvivalEstimator, BootstrapMixin

Implement Matching-Adjusted Indirect Comparisons class.

We consider that we have access to individual patients data for one of the centers and that for the other centers we only have access to aggregated data. This method proposes a way to balance the distribution of the indivual patients data to match the mean (and std) of a list of covariates available in both studies.

Parameters:

• variance_method (Literal['naive', 'robust', 'bootstrap']) –

• n_bootstrap (int) –

• seed (int | Sequence[int] | BitGenerator | SeedSequence | Generator | None) –

bootstrap_sample(data, seed=None)

Resampling only the individual patient data (IPD) with replacement.

In the setting of an estimation using MAIC, the caller is suppposed to have access only to the individual patient data, assumed here to be marked by non- zero treatment allocations in the data. Therefore during the resampling, only accessible data should be resampled.

Return type:

DataFrame

Parameters:

• data (DataFrame) –

• seed (int | Sequence[int] | BitGenerator | SeedSequence | Generator | None) –

fit(data, targets=None)

Estimate the treatment effect via Inverse Probability Treatment Weighting.

Option to estimate the variance of estimation by bootstrapping.

Return type:

None

Parameters:

• data (DataFrame) –

• targets (DataFrame | None) –

targets: pd.DataFrame, optional

pre-computed propensity scores. It is possible to pass pre-computed propensity scores to the fit function to use in the IPTW estimator instead of estimating the scores using logistic regression.

point_estimate(data)

Return a point estimate of the treatment effect.

Return type:

ndarray

Parameters:

data (DataFrame) –

class NaiveComparison(treated_col='treated', event_col='E', duration_col='T', ps_col='propensity_scores', variance_method='naive', n_bootstrap=200, seed=None, cox_fit_kwargs=None)

Bases: BaseSurvivalEstimator, BootstrapMixin

Naive comparison as if in a randomized setting.

Parameters:

• variance_method (Literal['naive', 'robust', 'bootstrap']) –

• n_bootstrap (int) –

• seed (int | Sequence[int] | BitGenerator | SeedSequence | Generator | None) –

fit(data, targets=None)

Estimate the treatment effect via Inverse Probability Treatment Weighting.

Option to estimate the variance of estimation by bootstrapping.

Return type:

None

Parameters:

• data (DataFrame) –

• targets (DataFrame | None) –

targets: pd.DataFrame, optional

pre-computed propensity scores. It is possible to pass pre-computed propensity scores to the fit function to use in the IPTW estimator instead of estimating the scores using logistic regression.

point_estimate(data)

Return a point estimate of the treatment effect.

Return type:

ndarray

Parameters:

data (DataFrame) –