# Estimating treatment effects using parametric models as counter-factual evidence

## TL;DR

This paper introduces a new method using parametric models to estimate treatment effects when randomized trials are not available, helping evaluate medical treatments more efficiently.

## Contribution

The novel contribution is a Bayesian method for estimating treatment effects using parametric models as counter-factual controls.

## Key findings

- The proposed method can estimate treatment effects using data from a single experimental arm.
- It was demonstrated to be useful for comparing treatments in pancreatic cancer trials.
- The approach can support both observational studies and RCT design under reasonable assumptions.

## Abstract

Randomisation controlled trial are the gold standard for causal inference, however the rapidly increasing development of new treatments and the movement towards personalised medicine mean there is a need to measure efficacy outside of the costly and time-consuming RCT.

Here we propose a method of estimating treatment effects using parametric models to act as control against which to compare data from an experimental arm. This allows for treatment effects to be estimated where data are only available from an experimental arm and can be a tool useful in the analysis of observational cohorts or for the design and analysis of RCTs.

We develop this approach using Bayesian estimation procedures and compare this approach against other casual inference tools. We then demonstrate how this may be used to estimate the efficacy between two treatment in different RCTs for the analysis of Pancreatic Cancer.

It is proposed that with reasonable assumptions, this approach can produce a reliable estimate of efficacy and can have applications in both evaluating currently available data and in the efficient design of future trials.

The online version contains supplementary material available at 10.1186/s12874-025-02540-2.