Network Meta-Analysis of survival outcomes with non-proportional hazards using flexible M-splines
This paper proposes a flexible Network Meta-Analysis model using M-splines and a novel weighted random walk prior to effectively handle non-proportional hazards in time-to-event outcomes, implemented in the R package `multinma` and demonstrated on non-small cell lung cancer data.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the world of healthcare decisions, doctors and policymakers often face a difficult question: which treatment helps patients live longer or stay disease-free the longest? To answer this, they rely on a statistical method called network meta-analysis. Imagine trying to compare three different medicines, but no single study has tested all three against each other. Instead, you have one study comparing medicine A to B, another comparing B to C, and a third comparing A to C. Network meta-analysis acts like a bridge, connecting these separate pieces of evidence to create a complete picture of how all the treatments stack up.
For outcomes that happen over time, such as how long a patient survives or how long it takes for a cancer to return, researchers traditionally used a method that assumes the risk of an event happening stays in a constant ratio between treatments. This means if one drug is twice as good as another at the start, it is assumed to remain twice as good forever. However, modern treatments, particularly immunotherapies that train the body's immune system to fight disease, do not behave this way. Their effects can be delayed, or they might work differently at different stages of the disease. When the risk ratio changes over time, the old methods fail to capture the true story, potentially leading to incorrect conclusions about which treatment is best.
A team of researchers at the University of Bristol has developed a new way to handle these complex, changing risks. They created a flexible mathematical model that does not force the data into a rigid shape. Instead of assuming the risk stays in a constant ratio, their model allows the risk to bend and shift over time, following the actual patterns seen in the patient data. They achieved this by using a technique called M-splines, which acts like a flexible ruler that can curve to fit the specific shape of the survival data without breaking. To ensure this flexibility does not lead to the model making up patterns that aren't really there, they added a special statistical rule that gently pulls the curves toward a smooth, sensible shape, preventing the model from overreacting to small, random fluctuations in the data.
The researchers tested this new approach using data from four major clinical trials involving treatments for non-small cell lung cancer. These trials compared different combinations of chemotherapy and immunotherapy. When they looked at the raw data, they saw that the survival curves for different treatments crossed each other and moved in ways that the old, rigid models could not explain. The new flexible model, however, fit the data beautifully. It successfully captured the complex timing of when the treatments started working and how their effects evolved over weeks and months.
Crucially, the new method did more than just fit the existing data; it allowed the researchers to make predictions for treatments that were not directly compared in every single study. By connecting the evidence across the network, the model could estimate how a specific treatment would perform in a specific group of patients, even if that exact comparison had never been made in a single trial. When the researchers compared their new model to older methods, they found that the new approach provided a better fit to the data and was less likely to be misled by the complex, changing nature of the survival curves.
The team also showed that this method is robust. They tested whether the results changed if they adjusted the number of points used to shape the curves, and found that the model remained stable and reliable. This suggests that the new method is not overly sensitive to the specific choices made by the researcher. The work was implemented in a software tool that other scientists can use, making it possible for them to apply this flexible thinking to their own studies of survival and disease progression.
By moving away from the assumption that treatment effects are constant, this research offers a more accurate way to understand how modern medicines work over time. It provides a clearer view of the benefits and risks of different treatments, which is essential for making informed decisions about patient care. The researchers demonstrated that when treatments behave in complex ways, the tools used to evaluate them must be equally sophisticated, allowing the data to tell the true story rather than forcing it into a shape that does not fit.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.