A Simple Blood Test Can Tell Early Whether Immunotherapy Is Working for Lung Cancer

Table of Contents

• Key Points
• Why This Research Matters
• How the Study Was Designed
• Who Took Part in the Study
• How Researchers Measured ctDNA Response
• Key Finding: Blood Test Matches Scan Results
• Four Patterns of ctDNA Change Over Time
• ctDNA Response and Survival
• When Scans and Blood Tests Disagree
• Safety and Side Effects
• What This Means for Patients
• Limitations of the Study
• Recommendations and Next Steps
• Frequently Asked Questions
• Source Information

Key Points

• In a 50-patient trial, a blood test for tumor DNA matched CT scan results 82% of the time when ctDNA cleared and 75% when it persisted.
• Patients whose ctDNA disappeared lived longer without cancer growing (5.03 vs 2.6 months) and had longer overall survival (median not reached vs 7.23 months).
• The test helped sort out ambiguous stable disease scans: 80% of those patients had molecular progression, while two had molecular response and did well.
• Side effects were within the expected range; pneumonitis was the most common serious side effect, affecting 5 of 50 patients (10%).
• A second trial stage will randomly assign patients without molecular response to treatment intensification or continuation to test if acting on results improves outcomes.

Why This Research Matters

Liquid biopsies (blood tests that detect fragments of tumor DNA) are gaining momentum in immuno-oncology, the field that uses the immune system to fight cancer. These tests can rapidly and accurately determine whether a treatment is working, especially in advanced cancer that has spread (metastatic disease).

As immune-based therapies and clinical trials expand, doctors face real challenges. Patients respond very differently to the same drug. Standard imaging scans often cannot capture treatment effects quickly or precisely enough.

Today's predictive biomarkers (tests used to guess who will respond) have limits. These include programmed death-ligand 1 (PD-L1) expression, a protein measured on tumor cells, and tumor mutation burden (TMB), the number of genetic mistakes in a tumor. Neither test consistently predicts who will benefit.

These gaps create an urgent need for molecular response-driven approaches (strategies guided by changes in tumor DNA) to interpret outcomes and guide treatment choices. In non-small cell lung cancer (NSCLC), the most common type of lung cancer, the stakes are high. Both pembrolizumab (Keytruda, an immunotherapy drug) alone and pembrolizumab combined with platinum-based chemotherapy are approved first-line options in the United States and Canada.

So how do doctors decide who should get immunotherapy alone versus combination immuno-chemotherapy? Current tests cannot answer that question reliably. That is where ctDNA (circulating tumor DNA, tumor genetic material found in the bloodstream) may help, by tracking real-time tumor burden.

Still, several questions had to be answered before blood-based ctDNA monitoring could enter routine care. What is the best way to track circulating tumor burden? Which ctDNA features best capture survival? Should testing be tumor-informed (tailored to each patient's known tumor mutations) or tumor-agnostic (a broad panel)? What counts as a ctDNA molecular response, and when does it occur? How well does ctDNA response match scan-based response? And which patients benefit most from this molecular approach? The BR.36 trial was built to answer these questions.

How the Study Was Designed

BR.36 is an international, multi-center, open-label, biomarker-directed phase 2 trial. "Open-label" means patients and doctors knew which treatment was given. "Biomarker-directed" means results from a biological test guided the study. The trial tests ctDNA molecular response-adaptive immuno-chemotherapy (treatment adjusted based on blood test results) in patients with previously untreated NSCLC.

The trial has two stages. Stage 1, reported here, was observational. Patients with advanced or metastatic NSCLC who were eligible for standard-of-care single-agent pembrolizumab enrolled in a single-arm study. Researchers used serial liquid biopsies (repeated blood draws) to find the best definition and timing of ctDNA molecular response and to check how well it agreed with radiographic response (what CT scans show).

Stage 2 will be the interventional, randomized portion. It will test whether tailoring treatment to ctDNA molecular response actually improves outcomes.

Key eligibility requirements included:

• Metastatic NSCLC negative for clinically actionable EGFR and ALK mutations (specific gene changes that have their own targeted drugs)
• No prior immune checkpoint inhibitor therapy (drugs that release brakes on the immune system)
• No prior chemotherapy
• PD-L1 expression level of at least 1%

Blood samples were collected before treatment on cycle 1, day 1 (C1D1, the baseline before any drug); cycle 2, day 1 (C2D1, about 3 weeks later); and cycle 3, day 1 (C3D1, about 6 weeks in). Pembrolizumab was given every 3 weeks.

Radiographic assessments used RECIST version 1.1 and iRECIST (immune-specific response criteria, developed because immunotherapy can make tumors look larger on scans even when they are shrinking). Scans were done every 6 weeks until week 12, then at longer intervals.

Researchers assumed more than 80% of patients with metastatic NSCLC would have detectable tumor mutations in plasma. Based on that, they calculated that 50 patients would provide acceptable sensitivity (ability to correctly identify true responses) and specificity (ability to correctly identify true non-responses).

The trial opened on 17 October 2019 and closed to enrollment on 5 April 2022 after reaching its target. The first patient enrolled on 26 May 2020, and the last on 5 April 2022. All 50 patients were followed for at least 12 weeks, and the database was locked on 20 September 2022. Median follow-up was 13.5 months (range 2.5–23.0 months; 95% confidence interval [CI]: 6.9–14.5 months). A confidence interval is a range that likely contains the true value. Two major protocol violations occurred due to mismatched timing of laboratory or imaging assessments.

Who Took Part in the Study

The study enrolled 50 patients with a median age of 65.5 years (range 50–87 years). The group was mostly representative of typical lung cancer populations in some ways, but with important features.

• Sex: 52% female (26 patients), 48% male (24 patients)
• Race: 82% White (41 patients), 8% Asian (4), 4% Black (2), 6% not reported (3)
• Age: 56% were 65 or older (28 patients); 44% were under 65 (22 patients)
• Performance status: 76% had an ECOG PS of 1 (38 patients), meaning mild symptoms but still able to carry out daily activities; 24% had an ECOG PS of 0 (12 patients), meaning fully active
• Smoking: 98% were ever-smokers (49 patients), having smoked more than 100 cigarettes in their lifetime; 24% currently smoked (12 patients)
• Prior treatment: 92% had no prior systemic therapy (46 patients); 46% had prior radiation (23 patients)
• Stage: 98% had stage IV disease (49 patients); 1 patient had stage IIIb
• Histology: 76% had adenocarcinoma (38 patients), 12% squamous cell carcinoma (6), and 12% other types (6)
• PD-L1: 96% had a tumor proportion score (TPS, the percentage of tumor cells showing the marker) of 50% or higher (48 patients); 2 patients (4%) had 1–49%

Two patients were not evaluable for radiographic response because of symptomatic progression, acute deterioration, or death during cycle 1 before any imaging could be done. That left 96% (48 of 50 patients) evaluable for scan-based response, which supports the feasibility of restaging scans in this population.

How Researchers Measured ctDNA Response

Researchers used a tumor-agnostic, white blood cell (WBC)-informed next-generation sequencing (NGS) approach. In plain terms, they sequenced DNA in blood plasma broadly, and they also sequenced DNA from white blood cells to filter out changes that come from normal blood cells rather than cancer.

Of 82 unique plasma variants (genetic changes found in blood), the team confirmed the cellular origin of each:

• 14 variants (17%) came from clonal hematopoiesis, age-related changes in blood-forming cells that can mimic cancer signals
• 4 variants (5%) were germline (inherited from birth)
• 64 variants (78%) were tumor derived

Germline and clonal hematopoiesis variants were excluded from further analysis. This filtering step matters because it prevents false alarms.

The test proved highly sensitive. A binomial model estimated a median sensitivity above 99% for detecting an alteration at 0.30% mutant allele fraction (MAF, the proportion of DNA copies carrying a mutation) and 93% at 0.20% MAF. Those two thresholds covered 95% (53 of 56) of the tumor-specific variants detected at cycle 1. Three well-known cancer hotspots were picked up at very low levels: KRAS G12A at a MAF of 0.19% (patient BR360030), NRAS Q61L at 0.14% (BR360041), and KRAS G12C at 0.14% (BR360048). Together, these findings confirm the assay can detect ctDNA mutations sensitively.

The team tracked the maximal mutant allele fraction (maxMAF) of tumor-derived plasma mutations from baseline to the on-therapy timepoints. Clearance of maxMAF signified molecular response (mR). Persistence of maxMAF indicated molecular disease progression (mPD).

Among the 50 participants, five were not evaluable for ctDNA molecular response because of missed plasma collection or non-evaluable scans. Of the remaining 45 patients, 10 had undetectable ctDNA throughout. That left 35 patients (77.8%) evaluable for ctDNA molecular response. The undetectable ctDNA rate of 22.2% (10 of 45) did not differ significantly from the assumed 20% (P = 0.82). Notably, patients in the evaluable subset (n = 35) had the same progression-free survival and overall survival as the full study population (n = 50), suggesting the subset was representative.

Of the 35 evaluable patients, 15 fell into the mR category, for an evaluable mR rate of 43% (90% CI: 0.29–0.58).

Key Finding: Blood Test Matches Scan Results

The trial met its primary endpoint. The sensitivity of molecular response (mR) for RECIST best overall response was 82% (90% CI: 52–97%), and specificity was 75% (90% CI: 56.5–88.5%). In plain terms, out of 100 patients whose scans showed shrinkage, the blood test correctly identified about 82. And out of 100 patients whose scans showed no shrinkage, the blood test correctly identified about 75.

Both sensitivity and specificity beat the prespecified goal of 70%. The lower bounds of the 95% confidence intervals for both measures also exceeded 50%, meaning the result was not likely a fluke. Median time to ctDNA molecular response was 2.1 months (90% CI: 1.5–2.6 months).

When researchers used iRECIST (immune-specific response criteria) instead of RECIST, the sensitivity of molecular response was 83% (90% CI: 56–97%) and specificity was 78% (90% CI: 60–91%). Again, the lower bounds of both measures were better than 50%.

For context on how scans performed overall: the best overall RECIST response rate was 32% (90% CI: 21–44%), which was significantly lower than the assumed 45% (P = 0.04). The best overall iRECIST response rate was 36% (90% CI: 25–49%), not significantly different from the assumed 45% (P = 0.13). Median duration of radiographic response was 10.1 months (90% CI: 5.6–19.2 months) for RECIST and 9.7 months (90% CI: 5.6–10.7 months) for iRECIST.

Four Patterns of ctDNA Change Over Time

Researchers identified four distinct patterns of ctDNA kinetics (how ctDNA levels changed over time):

1. ctDNA maxMAF clearance at cycle 2, day 1 (about 3 weeks)
2. maxMAF clearance at cycle 3, day 1 (about 6 weeks)
3. maxMAF reduction greater than 85% at cycle 2 or 3
4. ctDNA persistence (no meaningful clearance)

Of the 15 patients with ctDNA clearance, two still had detectable ctDNA at cycle 2 but cleared it by cycle 3. These two were classified in the mR group. Two other patients showed marked maxMAF reduction (more than 85% but not complete) and were classified in the mPD group.

ctDNA Response and Survival

Patients with molecular response lived longer without their cancer worsening. Median progression-free survival (PFS, time before the disease progresses) was 5.03 months for patients with mR versus 2.6 months for patients with mPD.

The overall survival (OS) difference was even more striking. Median OS was not reached for patients with mR (meaning more than half were still alive at the time of analysis) versus 7.23 months for patients with mPD.

For comparison, RECIST response was less effective at separating patients. Patients with a RECIST complete or partial response (CR/PR) had a median PFS of 8.31 months versus 4.27 months for patients with stable or progressive disease (SD/PD), and median OS was not reached versus 16.89 months.

In post hoc analyses (additional analyses done after the main results), researchers examined the mixed group of patients with RECIST stable disease. Patients with stable disease who also had a molecular response (n = 2) had PFS and OS that were not reached, compared with 4.27 months (PFS) and 8.08 months (OS) for patients with stable disease and molecular progression (n = 8). This suggests ctDNA testing can sort out a group that scans alone cannot.

The depth of ctDNA response also predicted radiographic response. The area under the receiver operating characteristic curve (AUC, a measure of predictive accuracy where 0.5 is no better than chance and 1.0 is perfect) was 0.77 (CI: 0.59–0.94) for cycle 2, day 1 assessments and 0.81 (CI: 0.67–0.95) for cycle 3, day 1 assessments. Similar results appeared for other ctDNA measures. Change in mean MAF and median MAF predicted radiographic RECIST response with AUCs of 0.78 (CI: 0.62–0.95) at cycle 2 for both measures, and 0.82 (CI: 0.68–0.95) for mean MAF and 0.81 (CI: 0.67–0.95) for median MAF at cycle 3.

Researchers also explored baseline maxMAF (ctDNA level before treatment) and outcomes. Patients who achieved a radiographic response did not have significantly different baseline maxMAF than those who did not (Wilcoxon P = 0.43). However, the 10 patients with undetectable ctDNA at baseline had numerically longer PFS and OS than the 35 with detectable ctDNA: median PFS 8.31 months versus 2.96 months, and median OS 16.89 months versus 10.94 months.

When researchers tested continuous baseline maxMAF against OS, it was not significantly associated with survival after adjusting for ctDNA molecular response status (hazard ratio [HR] = 0.986, 95% CI: 0.946–1.027, P = 0.56). A hazard ratio compares risk between groups. By contrast, ctDNA kinetics and molecular response were the strongest predictor of outcome (HR = 0.158, 95% CI: 0.052–0.48). This suggests what matters most is not how much tumor DNA a patient starts with, but how that amount changes during treatment.

As part of post hoc analyses, researchers tested different baseline maxMAF cutoffs. Using 3% as the threshold, sensitivity and specificity were 80% and 69% for maxMAF below 3%, and 83% and 88% for maxMAF of 3% or higher. Using 4% as the threshold, sensitivity and specificity were 88% and 69% for maxMAF below 4%, and 67% and 88% for maxMAF of 4% or higher. These analyses had limited statistical power but support the stability of ctDNA response assessment across a wide range of mutation levels.

When Scans and Blood Tests Disagree

ctDNA and radiographic responses were largely concordant, but important exceptions stood out.

• All patients with a RECIST complete or partial response at week 6 (cycle 3, day 1) had a ctDNA molecular response.
• Patient BR360019 had radiographic disease progression at both 6 and 12 weeks but showed a ctDNA molecular response. That patient's overall survival was 14.1 months and still ongoing at the time of data lock. The blood test captured the benefit that scans missed.
• Patient BR360041 had a radiographic partial response, which disagreed with ctDNA molecular progression. The ctDNA result better captured that patient's short radiographic response of only 1.47 months.
• Among patients with stable disease on scans (n = 10), most (80%) were classified as molecular progression, while two showed molecular response. Patient BR360017, who had both molecular response and stable disease on scans, had ongoing PFS and OS beyond 13 months.

Radiographic response assessments are challenging during immune checkpoint blockade and may not precisely reflect tumor burden dynamics. These cases show that molecular tracking of circulating tumor burden can rapidly and more accurately capture long-term outcomes, especially in the heterogeneous group of patients whose scans show stable disease.

Safety and Side Effects

Adverse events fell within the expected spectrum of immune-related side effects historically reported with this class of drugs. The study recorded one grade 4 event (severe) and 20 grade 3 events (serious but not life-threatening).

The most frequent serious adverse event was pneumonitis (inflammation of the lung tissue), reported in five patients (10%).

Four grade 5 events (deaths) occurred. Two were due to disease progression, and two were possibly related to drug toxicity.

Of the 50 patients enrolled, two were not evaluable for radiographic response because of symptomatic progression, acute deterioration, or death during cycle 1 without imaging. This is a reminder that some patients with advanced lung cancer decline quickly, before treatment can show measurable benefit.

What This Means for Patients

This trial shows that a simple blood test can identify early whether pembrolizumab is helping. The test matched scan results well, and it identified patients who would live longer, sometimes better than scans alone.

The clinical relevance is significant in several ways:

• Earlier answers. The median time to ctDNA molecular response was 2.1 months. Scans often take longer to reveal a clear picture, and immunotherapy can even make tumors appear larger on scans before they shrink.
• Better sorting of stable disease. Scans showing "stable disease" are ambiguous. A ctDNA test can separate patients who are doing well from those whose cancer is quietly progressing.
• Predicting survival. Patients with a molecular response had about twice the progression-free survival (5.03 months versus 2.6 months) and much longer overall survival (median not reached versus 7.23 months).
• Avoiding unnecessary toxicity. If a blood test can flag likely non-responders early, those patients could be moved to other, potentially more effective treatments sooner.

Importantly, patients with undetectable ctDNA at baseline did well regardless. Their median PFS was 8.31 months and median OS was 16.89 months. This suggests that a low tumor DNA signal at the start is itself a favorable sign.

Limitations of the Study

Several limitations should be recognized. This was a single-arm observational study, so there was no comparison group to test whether acting on ctDNA results actually changes outcomes. That is the purpose of stage 2.

Ten of 45 patients (22.2%) had undetectable ctDNA, meaning they could not be evaluated for molecular response at all. That limits how broadly this approach can be applied.

The trial cohort was 82% White, so results may not fully generalize to more diverse populations. Most participants (96%) had PD-L1 expression of 50% or higher, and most had adenocarcinoma. The findings may not apply to patients with lower PD-L1 levels or other tumor types.

The overall RECIST response rate (32%) was lower than the 45% the researchers had assumed, which may reflect the real-world nature of the study population. The post hoc analyses on baseline maxMAF thresholds (3% and 4%) had limited statistical power, so those specific cutoffs need confirmation in larger studies.

Recommendations and Next Steps

The BR.36 trial met its primary endpoint, and its findings are now being incorporated into a second stage. The upcoming stage is ctDNA-driven, interventional, and molecular response-adaptive. In other words, treatment decisions will change based on blood test results.

In stage 2, patients identified as being at risk of progression (those without a molecular response) will be randomly assigned to one of two paths:

1. Treatment intensification (adding or switching therapy)
2. Continuation of the current therapy

This design will directly test whether acting on ctDNA results improves patient outcomes, which stage 1 could not answer.

For patients and caregivers today, some practical points follow from this research:

1. Ask your oncology team whether ctDNA testing (also called liquid biopsy) is available in your situation. It may be part of a clinical trial rather than routine care.
2. Understand that scans and blood tests can disagree. Your care team should weigh both pieces of information.
3. Discuss the meaning of a "stable disease" scan result. Molecular testing may clarify whether you are truly responding.
4. Report new or worsening symptoms such as shortness of breath or cough promptly, because pneumonitis was the most common serious side effect, affecting 10% of patients in this study.
5. Ask about clinical trials if you have advanced NSCLC and are starting immunotherapy, since the field is actively refining how to personalize treatment.

Frequently Asked Questions

What is ctDNA and how is it measured?

ctDNA is tumor DNA floating in the blood. In a 50-patient trial, researchers used a broad blood test that also sequenced white blood cells to filter out normal changes. They tracked the highest level of tumor mutations before and during treatment. Clearance of that signal meant molecular response; persistence meant molecular progression.

Who was eligible for this trial?

Eligible patients had previously untreated advanced non-small cell lung cancer without EGFR or ALK mutations, had not received prior immunotherapy or chemotherapy, and had PD-L1 expression of at least 1%. In the 50 enrolled patients, 96% had PD-L1 of 50% or higher, and 98% had stage IV disease.

How well did the blood test match CT scans?

In the 50-patient trial, when ctDNA cleared, it matched scan shrinkage 82% of the time. When ctDNA stayed detectable, it matched scan non-response 75% of the time. Both results exceeded the prespecified goal of 70%, meaning the blood test agreed with scans in most patients.

What did the blood test mean for survival?

Patients whose ctDNA disappeared lived longer without cancer growing: 5.03 months versus 2.6 months for those whose ctDNA persisted. Overall survival was also longer; median was not reached for the clearance group versus 7.23 months for the persistence group. These findings are from a 50-patient trial.

What are the side effects of pembrolizumab in this trial?

Side effects were within the expected range for this drug class. The trial recorded one severe (grade 4) event and 20 serious but not life-threatening (grade 3) events. The most common serious side effect was pneumonitis, lung inflammation, in 5 of 50 patients (10%). Four deaths occurred, two from disease and two possibly from drug toxicity.

What happens next in this research?

The trial has a second stage that is now underway. Patients without a ctDNA molecular response, meaning they are at risk of progression, will be randomly assigned to either intensify treatment or continue current therapy. This stage will test whether acting on blood test results actually improves outcomes, which the first stage could not answer.

What does it mean if my scan shows stable disease?

In the trial, scans showing stable disease were ambiguous. Among 10 such patients, 80% had molecular progression, while two had molecular response. Those two had ongoing progression-free and overall survival beyond 13 months. A ctDNA test may help clarify whether stable disease truly reflects benefit or quiet progression.

If I have advanced non-small cell lung cancer and my scans show stable disease on pembrolizumab, when should I seek a second opinion about ctDNA testing?

Stable disease on scans is ambiguous, and ctDNA testing can separate patients who are doing well from those whose cancer is quietly progressing. Among patients with stable disease, 80% were classified as molecular progression, while two showed molecular response and had progression-free and overall survival beyond 13 months. A second opinion can help interpret whether a stable scan reflects true response, and whether ctDNA testing is available in your situation, possibly through a clinical trial. Diagnostic Detectives Network provides independent expert second opinions.

Source Information

Original article title: ctDNA response after pembrolizumab in non-small cell lung cancer: phase 2 adaptive trial results

Authors: Valsamo Anagnostou, Cheryl Ho, Garth Nicholas, Rosalyn A. Juergens, Adrian G. Sacher, Andrea S. Fung, et al.

Publication: Nature Medicine, Volume 29, October 2023, pages 2559–2569

DOI: https://doi.org/10.1038/s41591-023-02598-9

ClinicalTrials.gov ID: NCT04093167

Received: 14 March 2023; Accepted: 19 September 2023; Published online: 9 October 2023

This patient-friendly article is based on peer-reviewed research.