NEWS.md

In postcard: Estimating Marginal Effects with Prognostic Covariate Adjustment

postcard 1.0.0

Major overhaul of package. Still focuses on analysing data with the use of prognostic scores, but takes a more general approach that allows any distribution of response and covariates within the scope of generalised linear models (GLMs) and does not necessarily run on a number of data sets created by simulation.

The package provides novel methods for:

• rctglm: Finding any marginal effect estimand and estimating the standard error using influence functions to avoid inflation of type 1 error
• rctglm_with_prognosticscore: Do the above, but leveraging historical data to increase precision with prognostic scores.

Additionally, the package includes functionalities for

• fitting a discrete super learner in fit_best_learner, which is leveraged in rctglm_with_prognosticscore
• approximating power using
• standard methods for ANCOVA models (see help topic power_linear)
• a novel method for any model estimating marginal effects (power_marginaleffect)
• generating data from a GLM (glm_data)

postcard 0.2.1

Features

• Added function simulate_collection that takes function arguments for how to simulate covariates and model the outcome in the historical and "current" data to give the user full flexibility (previously a multivariate normal distribution was assumed)

• sim.lm which simulates data from a multivariate normal distribution and models the outcome with a linear model is now a wrapper of the new - more general - simulate_collection.

postcard 0.2.0

Features

• Added option to use sandwich HC estimators for the covariance matrix in sim.lm

• Updated default value of ATE_shift in sim.lm

Code cleanup

• Modularised code. Fx. split lm.hist into lm.procova and lm.psm

• Renamed some functionalities

• Correcting errors in documentation

Best practices package

• Updated DESCRIPTION

• Created README

• Added explicit package imports in form of foo::xx

• Added a few tests

postcard 0.1.0

Initial package created from local files. Package contains functionalities to create simulation study for a specific purpose related to an article. Functionalities include generation of a collection of data sets and a way to analyse these data sets assuming a special case of multivariate normal distribution of covariates with a linear model of the response. In addition, functionalities to estimate the power of certain parameter tests based on the results.

postcard documentation built on April 12, 2025, 1:57 a.m.