Comment by cant_kant | Hacker News Reader

Comment by cant_kant | Hacker News Reader

Serum nucleic acid sampling to screen for malignancies has a few problems. One issue is that it runs into the realities of Bayes theorem.

You may want to consider what happened to Pathway Genomics. Their screening test claiming to detect 10 cancers in healthy people drew the ire of the U.S. Food and Drug Administration past year, which sent the company a letter saying:

"We have [...] examined published literature and have not found any published evidence that this test or any similar test has been clinically validated as a screening tool for early detection of cancer in high risk individuals," the letter states.

... We have reviewed the information presented on your website in the white paper, entitled “Liquid Biopsy for the Detection and Monitoring of Cancer: Analysis of 96 Hotspot Mutations via Plasma Derived Circulating Tumor DNA,” dated September 2015. It is unclear how the literature that you cited, addressing the presence of circulating tumor DNA (ctDNA) in already-diagnosed patients, is adequate to support the expansive claims of screening for early cancer detection using ctDNA in undiagnosed patients for up to 10 different cancers with the CancerInterceptTM Detect...

We believe you are offering a high risk test that has not received adequate clinical validation and may harm the public health.."

For the full letter from the FDA, see http://www.fda.gov/downloads/MedicalDevices/ResourcesforYou/... ( it is a pdf )

According to the Warning Letter, Pathway never registered with the FDA, never received clearance, and never performed any (reported) clinical trials. The key sentence from the letter is:

"We have conducted a review of our files and have been unable to identify any Food and Drug Administration (FDA) records reflecting the approval, clearance, or listing of these devices."

They were seeking to fly under the radar as a "lab-derived test" or LDT, using an FDA loophole that the FDA is increasingly seeking to close (also see Theranos). The "we've reviewed the literature" part is just the FDA's attempt to dig into what's been published to see if they can find ANY support for their marketing claims — even if there were volumes of papers on their method, they still would have been sent the warning letter (which is a Very Serious thing to get from the FDA).

The issue wasn't that "this test is too risky for us" it's that they were 100% out of compliance. If you sell a medical device, it must be classified for risk and at the very least registered with the FDA. In the case of a cancer diagnostic, there is zero chance it would be a Class I (lowest) device, so they (at least) would have needed "clearance" requiring a submission of something called a 510(k) documenting tests they've done on their product, audits of suppliers, etc. etc. In addition to the 510(k) submission itself, they would need documented evidence of having put in place a "quality management system" that puts standard operating procedures around 100% of product development from cradle to grave. This is why health tech/biotech companies need to raise so much money, as this is a lot of overhead.

If their test were classified as Class III (highest risk), they would need "approval" rather than "clearance," the bar for which is even higher (although this is primarily limited to ingestibles/implantables). Just the FDA submission itself costs something like a quarter million dollars.

GRAIL, on the other hand, is raising a huge amount of money partially because they know how much it costs to run proper trials and earn a clearance or approval. It's apples and oranges to compare the two.

Some example of Class III devices from Wikipedia :

> Examples of Class III devices that currently require a premarket notification include implantable pacemaker, pulse generators, HIV diagnostic tests, automated external defibrillators, and endosseous implants.[19]

I don't see how an cancer test can not be less than Class III. A misdiagnosis is potentially lethal.

A CBC (complete blood count) test is a class I...

There are also major medical training systems, for technicians and pathologists, who perform and interpret them.

The devices are class 2. Putting a QBC Star in every primary care clinic improves the quality of care delivered across the board and they are designed toward sensitivity: an automated counter has to have a low threshold for abnormal, which then kicks it to a human for review.

And the techs are trained to kick anything that looks like a blast to a pathologist. And it's not a one-time, in the school house tid-bit of education: every time they do a manual diff, they can't get from their cell counts to the signature block without acknowledging they need to show blasts to a pathologist.

> Serum nucleic acid sampling to screen for malignancies has a few problems. One issue is that it runs into the realities of Bayes theorem.

Which is that the occurrence of a genuine positive result is so rare that the false positives vastly outnumber the real positives unless your test is amazingly good.

And this is a bad thing when the interventions are as horrific as most of our current cancer treatments.

This is the reason for the recent recommendations for both breast and prostate cancers to reduce screening unless you have actual risk factors (family history, obesity, smoking, etc.)

Reply on news.ycombinator.com