A Retrospective Cohort Study Using Causal Inference Tools to Evaluate the Efficacy of Treatments for Pancreatic Cancer

Pancreatic cancer affects nearly half a million people worldwide each year and causes almost as many deaths. It is one of the deadliest cancers because it is usually diagnosed late, when treatment options are limited. Only a small number of patients are diagnosed early enough for surgery, which is the only potential cure. For most people, treatment such as chemotherapy (medicine that kills cancer cells or slows their growth) or radiotherapy (high-energy rays used to destroy cancer cells) is mainly used to control symptoms and slow the cancer’s progression. As the number of cases continues to rise, pancreatic cancer is expected to become one of the leading causes of cancer-related deaths in Europe.

Because pancreatic cancer progresses quickly, researchers often measure treatment success by looking at how long patients live after diagnosis. The best way to compare treatments is through randomized clinical trials, where patients are randomly assigned to one treatment or another. However, these trials are expensive, take many years to complete, and may not reflect how treatments work in real-world patients.

We will study a new way to estimate how well treatments work without relying solely on traditional clinical trials. We have developed a method called Personalised Synthetic Controls, in which we use statistical models to predict how each patient would have responded to a standard treatment. We can then compare this predicted outcome with the patient’s actual outcome after receiving a different treatment. This allows us to estimate treatment benefit for each individual.

We have already created two prediction models for pancreatic cancer: one for patients who receive treatment after surgery (the adjuvant setting) and one for patients whose cancer has already advanced. In this project, we will use patient information available through the Vivli data-sharing platform to test, refine, and apply these models.

When we obtain data on patients who received the same standard treatment used in our models, we will check how well our predictions match their real outcomes. This will help us improve the accuracy of the models. When we obtain data on patients who received different treatments, we will use our method to estimate how effective those treatments may be and explore whether certain groups of patients benefit more than others. Our goal is to support future research and help improve treatment decisions for people living with pancreatic cancer.