Ƶ

AI Blood Test Proves Useful in Lung Cancer Detection

— Such an approach would be useful considering the barriers to LDCT among high-risk people

Ƶ MedicalToday
A computer rendering of a DNA helix inside of a test tube of blood.

A blood test powered by the use of artificial intelligence could be an effective pre-screening tool for the noninvasive detection of lung cancer, according to a prospective study.

When combined with clinical risk factors and a protein biomarker, and then followed by low-dose CT (LDCT), the blood test detected 94% of patients with lung cancer across stages and subtypes, including 91% of those with less invasive stage I/II cancers and 96% of those with more advanced stage III/IV cancers, reported Victor Velculescu, MD, PhD, of the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins University in Baltimore, and colleagues in .

The blood test, called DELFI (DNA evaluation of fragments for early interception), uses machine learning to spot unique patterns in the way DNA is shed from cancer cells circulating in the bloodstream.

"This approach basically looks at genome-wide fragmentation profiles and with a very complex artificial analysis takes that information on the DNA fragments throughout the genome and uses it to identify whether someone has the profile of a healthy individual or an individual with cancer," Velculescu told Ƶ. And not only does this approach identify abnormal fragmentation profiles, he pointed out, it also provides information on the cell of origin.

Velculescu and team evaluated 365 patients over seven months as part of the Danish LUCAS study. Most were at high risk for lung cancer (ages 50-80 and a smoking history of >20 pack-years). More than 90% had pulmonary, non-pulmonary, or constitutional symptoms, as well as typical smoking-related symptoms such as coughing.

Of these patients, 129 were determined to have lung cancer a few days after blood collection, 87 had histologically proven benign nodules, and 149 were not biopsied due to low clinical and radiographic suspicion for cancer.

"The resulting fragmentation profiles were remarkably consistent among non-cancer individuals, including those with non-malignant lung nodules," the authors wrote. "In contrast, cancer patients displayed widespread genome-wide variation. Remarkably, the fragmentation profile differences could be observed in multiple regions throughout the genome for the majority of cancer patients, including across stages and histologies."

Velculescu and team determined that non-cancer cases with or without benign lesions had lower DELFI scores compared with cancer cases, with a stepwise increase in DELFI score by stage. The highest median DELFI score was observed in small cell lung cancer cases.

When they examined the relationship between DELFI scores and cancer stage and histology, they found that receiver operator characteristic (ROC) curves representing the sensitivity and specificity of the DELFI approach to identify cancer revealed an area under the curve (AUC) of 0.90 (95% CI 0.86-0.94). While stage I disease was more difficult to identify (AUC 0.76), stages II, III, and IV disease had similarly high performances (AUC 0.89, 0.92, and 0.92, respectively), they noted.

Analyses of different histologic subtypes of lung cancer showed that small cell and squamous cell lung cancers were more easily detected than lung adenocarcinomas.

In a validation cohort of 46 patients with cancer and 385 individuals without cancer, the sensitivities and specificities of this model "were similar to those observed in the LUCAS cohort at different stages of the disease and among different histologic subtypes," the authors wrote. "Overall, these analyses suggest that the DELFI approach is generalizable across different lung cancer cohorts, including across different stages and histologic subtypes."

The authors also determined that using DELFI in combination with clinical risk factors -- such as age, smoking status, and presence of chronic obstructive pulmonary disease (COPD) -- and carcinoembryonic antigen (CEA) levels could improve DELFI's performance.

Velculescu said that he and his team were particularly interested in how this model would work as a lung cancer pre-screening tool. Such an approach, he noted, would be particularly useful considering the barriers that contribute to the low uptake of LDCT among high-risk individuals, such as access, and concerns about radiation exposure and potential harms associated with false-positive imaging results.

"A simple blood test would be a much better initial test, and if positive, gives additional guidance from the physician to the patient as to whether to undergo a further imaging test," he said.

Data suggest that the majority of patients will proceed to screening if it is offered, Robert Smith, PhD, of the American Cancer Society, told Ƶ. "What we can say with confidence is that it is difficult to identify people who formerly smoked and who are eligible for lung cancer screening because they are no longer smoking and their electronic health record does not identify their pack-year history."

In addition, those currently smoking find it challenging to identify their eligibility based on age and smoking history, and many may be uninsured and have limited encounters with healthcare providers, he noted.

While this is likely to change over time, Smith added, "a blood test that has been demonstrated to have good performance in a rigorous study would be welcome because it may add to the uptake we would achieve from imaging tests, and as the authors noted, it could be extended to adults at lower, but still sufficiently high, risk to justify testing."

In order to evaluate the potential impact of DELFI in a scenario in which it is used as a pre-screening tool, the authors examined the performance of the DELFI score and the multimodal approach, followed by standard CT imaging.

In the LUCAS cohort, LDCT alone demonstrated a high sensitivity (>95%) and a low specificity (58%). When the DELFI score was used to pre-screen patients, the observed sensitivity of the combined DELFI/LDCT approach was 90% (stage I 80%, stage II 86%, stage III 94%, and stage IV 90%), with an increase in specificity to 80%.

If the DELFI multimodal approach followed by LDCT was used, it increased the sensitivity to 94% overall (stage I 87%, stage II 100%, stage III 97%, and stage IV 96%), with the same specificity, and would decrease the number of unnecessary procedures by 52% (from 67 with LDCT alone to 32).

DELFI is undergoing further evaluation in the .

"Too many adults die from lung cancer each year who would not qualify for lung cancer screening under the current recommendations, or would be averse to undergoing an imaging test," Smith said. "The reported results of the DELFI test are promising, and we should look forward to further reports. The more methods we have to identify lung cancer early, the greater our progress will be in averting deaths in the leading cause of cancer mortality."

  • author['full_name']

    Mike Bassett is a staff writer focusing on oncology and hematology. He is based in Massachusetts.

Disclosures

Velculescu, as well as several co-authors, are inventors on patent applications submitted by Johns Hopkins University related to cell-free DNA for cancer detection. In addition, he and several co-authors are founders of Delfi Diagnostics. Velculescu is also a founder of Personal Genome Diagnostics, serves on the Board of Directors and as a consultant for both Delfi Diagnostics and Personal Genome Diagnostics, and owns Delfi Diagnostics and Personal Genome Diagnostics stock, which are subject to certain restrictions under university policy.

In addition, Johns Hopkins University owns equity in Delfi Diagnostics and Personal Genome Diagnostics.

Velculescu is an advisor to Bristol-Myers Squibb, Genentech, and Takeda Pharmaceuticals, and within the last 5 years was an advisor to Merck and Ignyta.

Primary Source

Nature Communications

Mathios D, et al "Detection and characterization of lung cancer using cell-free DNA fragmentomes" Nat Commun 2021; DOI: 10.1038/s41467-021-24994-w.