Man’s Best Friend and Other Lung Cancer Biomarkers

Low dose CT scans provide an imaging-based biomarker useful for screening patients at high risk for developing lung cancer. Well established algorithms guide the workup and subsequent interventions for patients with suspicious nodules, but only a small percentage of those who qualify undergo screening. Furthermore, the test sometimes yields indeterminate findings that lead to more questions than answers.

One of the more innovative ideas for an inexpensive, scalable lung cancer screening test is to analyze volatile organic compounds in exhaled breath.¹ Trained sniffer dogs might play a future role in this domain.² The technology is not quite ready for widespread implementation, though, since the data are still a bit, um, ruff.

At the moment, the most thoroughly vetted biomarkers involve immunohistochemical or molecular profiling that primarily drive decisions about systemic therapy selection for patients with advanced or metastatic disease. Testing for PDL-1 expression is routinely accessible in nearly every pathology lab, as are analyses for common oncogenes such as EGFR or Ros1 mutations, ALK gene rearrangements, BRAF mutations, and a few others.

Clinical outcomes for patients with targetable mutations identified by these biomarkers are often remarkable. Responses like the one shown in Figure 1 are now commonplace, but it should be appreciated that such outcomes were simply unheard of even just 20 years ago, before the advent of new classes of systemic agents.

Figure 1: A 64-year-old female never smoker was found to have widespread EGFR mutation+ non-small cell lung cancer. Images at left show the initial PET scan and brain MRI, which had several dozen small lesions, some marked by arrows in this slice. The images at right show a sustained cranial and extracranial complete response after 18 months of osimertinib.

Perhaps most challenging to traditional dogma³ for patients like the one shown in the figure has been a recalculation of how to manage asymptomatic brain metastases in patients with tumors expected to be sensitive to agents that easily cross the blood-brain barrier. Recent publications demonstrate that durable responses, or at least clinically meaningful downstaging, can be achieved with these agents, allowing for a more selective approach to treatment with radiosurgery,^4,5 a rubric endorsed in NCCN guidelines. Patients in this category also very often eventually have oligoprogression in extracranial locations for which locally ablative or palliative-intent therapy in the form of radiation treatment might play an important role.⁶ The patient in Figure 1, for example, has had three courses of radiotherapy to various extracranial sites in her over four years (and counting) of survivorship since diagnosis.

There have been efforts to identify a molecular signature that would identify patients whose oligometastatic state might be confirmed as a predictor of future clinical outcome, but working against these efforts is the bedeviling Catch-22 of many cancers whereby different foci of local or metastatic deposits can have different combinations of genetic mutations. Annoying, to say the least. Efforts to identify a genetic profile that predicts lung cancer radiosensitivity are also ongoing, and one of the most intriguing is a single nucleotide polymorphisms signature in DNA repair pathway genes ERCC1/ERCC2, shown to have possible implications for identifying patients who might benefit from dose escalation.⁷ Other such markers are in various stages of refinement.⁸

Last but not least, the requisite query about the role of artificial intelligence in this space: where are we with AI and lung cancer at this time? The overarching field of AI offers a number of promising opportunities to analyze data and aid in the diagnosis, treatment and follow-up of lung cancer patients.⁹ Of course, an important caveat for AI is that its value is predicted on the accuracy and completeness of the data available for crunching. A collective aspirational goal might be to work with the electronic record systems commonly used in radiation oncology departments (i.e., Aria, Mosaiq and others) and electronic medical records (e.g., Epic) to autopopulate mega databases that can then be analyzed through AI to create treatment algorithms, decision trees, and predictive models. Opportunities are endless but it will require a large, community-wide commitment to move the needle meaningfully.

Putting it all together, I personally think we have only just begun to see how the burgeoning field of biomarkers of all types will ultimately help lung cancer patients, and I look forward to days ahead when the powers will be truly unleashed.¹⁰