Collaboration for Academic Primary Care (APEx) Blog
Collaboration for Academic Primary Care (APEx) Blog
Posted by ma403
16 April 2025Lung cancer kills more people in the UK each year than any other cancer, accounting for 1 in 5 cancer deaths. Fewer than half the people diagnosed with lung cancer survive for a year after diagnosis, and only 1 in 5 survive for more than 5 years.
These are pretty dismal statistics.
One of the biggest problems with lung cancer is that by the time it has diagnosed it has frequently spread to other parts of the body (metastasised), by which point treatment options are more limited. Many factors contribute, including the ability for lung cancers to develop without producing symptoms.
We know that smoking is a significant risk factor for lung cancer, so a cancer screening programme targeted towards people with a significant smoking history has the potential to achieve a good balance of benefits, harms and costs.
In 2016, I worked with a team in the Peninsula Technology Assessment Group (PenTAG) at the University of Exeter to produce an interim assessment of targeted lung cancer screening. In 2021–22, I then worked with Chris Hyde, Jaime Peters and Ed Griffin to produce a final assessment, which included developing and calibrating a disease natural history model to form a key part of a policy model.
Our work fed directly to the UK National Screening Committee, which was keen to issue guidance following the publication of the Dutch–Belgian NELSON trial of lung cancer screening which confirmed an earlier result of reduced lung cancer mortality found in the US National Lung Screening Trial (NLST).
Policy models can have a lot of inputs, spanning things like how many people have a disease or risk factor, how the disease develops, what impact it has on quality of life, and how much it costs to treat. Normally, the values for those inputs are estimated directly through experiments, administrative data, and surveys. Model calibration takes a different approach, where instead the outputs are estimated, and then input values are found which cause the policy model to produce outputs in line with those.
What I had hoped would be a 6-month project instead ran closer to 18 months, as calibrating the number of inputs to data from NLST on over 50,000 participants was computationally too taxing even for a high-performance computing cluster. I have to give my sincere thanks to TJ McKinley for his moral and technical support during this time!
When all the work was finally finished, we predicted that targeted lung cancer screening would be cost-effective. The work went through a round of external quality assurance, as it was being used to inform a major public spending decision, though this didn’t materially affect the results.
We are now disseminating our research findings, plus taking the model we developed and applying it to address other research questions. We will put the model into the public domain as open source so that other researchers can make use of it, adapting as they see fit.
This project was the toughest I’ve ever undertaken, but my hope is it can also be one of the most rewarding in terms of impact and enabling many more researchers to evaluate lung cancer policies and interventions.