Smarter Clinical Trials for Personalised Medicine

Personalised medicine has a simple goal: to ensure that patients receive the treatment best suited to them. People can respond differently to the same therapy, and factors such as age, diagnosis, biomarkers and other individual characteristics can influence both the course of a disease and the effectiveness of a treatment. A key challenge in clinical trials is therefore to determine as efficiently as possible which treatments work best for which groups of patients. 

In a joint study, Andres Alban from Frankfurt School and Stephen E. Chick and Spyros I. Zoumpoulis from INSEAD examine how adaptive clinical trials can improve this learning process. The researchers develop a mathematical approach that combines existing knowledge about patient characteristics with insights gained during a clinical trial. 

Unlike purely random assignment, the approach takes into account what has already been learned during the trial when assigning new participants to different treatment options. As the trial progresses, new information is used to inform subsequent decisions.

Over time, this provides a clearer picture of which treatments may be particularly suitable for different patient groups. 

A key aspect is distinguishing between different types of patient characteristics. Some characteristics can indicate how a person might respond to a particular treatment. Others provide information about the general course of the disease, regardless of the treatment received. Taking such information into account when designing a clinical trial can speed up the learning process. 

The researchers test their approach in simulations using two scenarios based on insights from sepsis treatment. They compare the proposed approach with several other methods for assigning treatments within clinical trials. 

The simulations show that the new approach can identify personalised treatment strategies more efficiently. The approach achieves a comparable level of learning to random assignment with significantly fewer participants. The results, therefore, suggest that existing knowledge about different patient groups can help researchers make better use of data from clinical trials. 

The researchers also address an important practical challenge: before a trial begins, it is not always known which patient characteristics are particularly relevant to treatment outcomes. They investigate whether such relationships can be identified from the data collected during the trial. 

For the design of clinical trials, this means that existing knowledge about patient groups can be incorporated more systematically into the learning process. Rather than focusing exclusively on finding the treatment that works best on average, adaptive methods can help account for differences between patient groups. 

Overall, the study demonstrates how mathematical decision models and existing medical knowledge can be combined to better align clinical trials with the needs of personalised medicine. The goal is to make the best possible use of the available data to identify suitable personalised treatment strategies for future patients.