Get the app
Respiratory medicine2021Impact: 1 out of 3, proof of concept

Lung cancer: a blood test analyzed by machine learning to catch the disease early

In a study published in Nature Communications in 2021, a machine learning method that reads DNA fragments in the blood, followed by a CT scan, found 94% of lung cancers at 80% specificity. It was developed in 365 people at risk, then validated in a second group. It is not yet a routine test: the authors call for a large validation before any clinical use.

Nature Communications, 2021Lire en français

Why is lung cancer often found too late?

Lung cancer is often found at a late stage, when treatment works less well. Screening with a low-dose CT scan exists for heavy smokers, and large trials have shown it reduces deaths.

Yet this scan is rarely used: according to the authors, fewer than 6% of people at risk are screened. Reasons include worries about false alarms, radiation and the invasive tests that may follow. A simple blood test could help decide who needs a scan most.

What did the researchers do?

When cells die, they leave small pieces of DNA in the blood, called cell-free DNA. Pieces that come from a tumor are cut slightly differently from those of healthy cells. The researchers' method, called DELFI, measures the size and spread of millions of these fragments across the whole genome.

The team analyzed blood from 365 people at risk, seen at a hospital in Copenhagen, Denmark. Most were aged 50 to 80, were heavy smokers and had symptoms. A machine learning model combined the fragment patterns, risk factors and a blood marker called CEA. It was then tested, unchanged, on 46 people with lung cancer and 385 people without cancer.

94% of cancers found, including early ones

Combining the blood test with a follow-up CT scan, the method found 94% of people with cancer, across all stages and types, at 80% specificity. Specificity is the share of people without cancer who are correctly recognized. Early cancers (stages I and II) were found 91% of the time, and advanced ones 96%.

The fragment patterns could also tell small cell lung cancer apart from other forms with high accuracy. Finally, a higher fragmentation score was linked to shorter survival.

Key figures

  • 94%of lung cancers found (91% at stages I and II), at 80% specificity
  • 365people at risk studied prospectively to develop the method
  • fewer than 6%of people at risk have CT screening

Talking about your lung risk with Mima

This blood test is not available routinely yet. If you smoke or have smoked a lot, the real question today is still CT screening, which few people at risk have.

Mima helps you prepare that conversation: note a lasting cough or breathlessness over the days, and keep the question for your doctor, such as: "Should I be screened for lung cancer with a CT scan?"

Try Mima for free

What this could change for patients

If these results hold up, a blood test could one day help pick out the people who most need a CT scan, and encourage more people at risk to join screening. The authors stress that their approach is low cost and could eventually reach beyond high-risk groups.

But today, this is not a test your doctor can order routinely based on this study. CT screening remains the recommended test for people at risk.

What are the limits of this study?

The authors say so themselves: most people studied already had symptoms, which is not what a screening population looks like. A large prospective validation in a real screening population will be needed before any clinical use.

The groups are small, especially the 46 people with cancer in the validation group. A few advanced cancers were missed. And the link with survival was seen in patients who had not necessarily received the newest treatments, such as immunotherapy.

Frequently asked questions

Is there a blood test for lung cancer screening?

Not routinely today. This study shows that machine learning analysis of DNA fragments in the blood is promising, but the authors call for a large validation before any clinical use.

What is cell-free DNA?

Small pieces of DNA released into the blood when cells die. Pieces from a tumor are cut differently, which is what the DELFI method looks for.

Does this test replace the CT scan?

No. In the study, the blood test is followed by a CT scan. The idea is to target scans better, not to skip them.

Who can be screened for lung cancer?

CT screening is mainly offered to people at high risk, such as heavy current or former smokers. Rules vary by country, so ask your doctor.

The original study

Mathios D, Johansen JS, Cristiano S, et al. “Detection and characterization of lung cancer using cell-free DNA fragmentomes” Nature Communications, August 20, 2021. DOI : 10.1038/s41467-021-24994-w

Read the full study in Nature Communications

This article summarizes a scientific study and is not medical advice.