An interesting article just came out in the Journal of Clinical Oncology from researchers at Duke, led by Dr. Ed Patz of the Radiology Department there (abstract here). Recognizing the problems with detection of lung cancer (LC) based on symptoms (which detects LC far too late) or screening CTs (which detects early LC but also many nodules that aren’t cancer), the authors worked to develop a panel of serum proteins that can distinguish between people who have LC and those who don’t. I’ve covered the idea that one reflection of activity of a subset of LC tumors is elevation of one or more proteins, or tumor markers, in the blood. But there is no individual marker that is reliably elevated enough in LC or normal outside of LC to serve as a useful discriminator. So the Duke group looked backwards, starting with sets of patients with and without LC (50 each) to identify a panel of proteins that were the most useful discriminators, to see if this biomarker panel that could be obtained from a blood draw could replace or, far more likely, enhance the workup of lung nodules detected based on symptoms, incidentally, or in a screening study.
Starting with a “training set” of serum from 50 patients with LC and 50 controls who didn’t have LC, they studied differences in proteins in the serum (“proteomics”) to identify four proteins that appeared to highlight differences between the groups (the individual proteins don’t matter as much as the principle), and the investigators then added two more serum markers that they believed would be relevant, carcinoembryonic antigen (CEA) and squamous cell carcinoma antigen (SCC). Measuring protein levels of these six biomarkers among the 100 samples that comprised the test set, they could come up with a model that placed every case in one of seven “bins” or patterns of activity, and with that could identify 88% of cancer patients and 82% of control patients correctly. While that’s good, that’s clearly not perfect.
They then provided an additional “test set” of another 97 serum samples, evenly split between those patients with LC and controls. Looking for the same protein patterns, or bins, the system correctly identified 71% of the patients with LC correctly, while identifying 67% of control patients. For both the training set and the validation/test set, more than half of the patients were placed into bins that allowed you to be as confident as 90-92% that LC was or was not present. So while this approach wasn’t perfect, for many patients it allowed you to be quite conflident about whether LC is or is not present.
This strategy isn’t commercially available, and it’s not reliable enough to be considered seriously as a screening test on its own. But the authors raised the point that this approach would allow for a serum-based battery of tests could very strinkingly modify the likelihood that a questionable nodule on a CT scan represents cancer not. In this way, people with ambiguous early CT findings could potentially be separated into those who you’d now feel more comfortable following radiographically and others who you’d now have a much lower threshold of obtaining tissue to clarify a diagnosis.
It’s too early for this type of approach to be recommended for general use, but this work is being studied quite actively, and I think it’s likely that in the next 3-5 years a straightforward blood test like this one will be available to help modify our decisions to move ahead with biopsy and treatment, and potentially even alter treatment decisions at some later point.