Thursday, May 10, 2018

Very Brief Blog: Tissue Type Distribution in FMI F1 CDx FDA Data

Based on its PMA approval last November, the Foundation Medicine F1 CDx test has two types of indications.  The first set is indication(s) for several CDx genes, generally obtaining the CDx authorization by comparison to an FDA-approved legacy single gene test in 150-200 blocks.   The second indication is for pan-solid-cancer tumor "profiling" which is "for use by a physician."   This second indication gets FMI pan solid tumor coverage under the recent CMS NCD for NGS tests in oncology.

On pages 34-35 of the FDA P170019B approval, FMI lists its tissue comparability data (document section 10.)   A large scale retrospective analysis was conducted used 80,715 specimens from 43 tissue types, "to demonstrate that genomic profiling can be performed on DNA derived from FFPE specimens of any tissue type."   The document states that "the data set for analysis consisted of routine clinical samples analyzed using the F1 LDT from March 25, 2015 to March 13, 2017."

I typed the FDA data into Excel to sort it.   FMI provides technical success data for each of the 43 tumor types after DNA extraction, library capture (LC) and hybridization capture (HC).   The vast majority of specimens pass all QC steps, but note that they had presumably been screened or submitted according to some instructions about the expected number of slides or blocks or percent tumor.

Assuming the 80,000-odd blocks were routine clinical workflow, that's about 10,000 cases per quarter over 8 quarters, which more or less foots to FMI statements about clinical volume.   The top ten tumor types, reaching 75% of specimens, are shown below.   Note that I've left the data exactly as in the FDA tables except for merging pancreas together with Whipple resection.

FMI F1 CDx PMA 170019B p 35 table 24

Three cancers - lung, liver, and brain - along with "lymph node" samples, are about half of all specimens.   The top ten together are about 75%, so the next 30 are only 25%.   So that's from the perspective of concentration.

You can also look from the perspective of dispersion - thirty-some types of cancer all log in at less than 2% each.  18% of samples had rather vague denominations like "lymph node" or "effusion."

For comparison, the annual cancer deaths (a proxy for advanced cancers) are lung, 150,000; colorectal, 50,000; pancreatic, 45,000; breast, 40,000; liver, 30,000; prostate, 30,000; leukemias, 25,000.   Brain cancer would be farther down, around 15,000.  In the FDA data, lung and liver FMI tests are about the same, but the incident US death ratio is 5:1 favoring lung.  Colon and brain FMI tests are about the same, but the US death ratio favoring colon is over 3:1.

The original FDA document is here, see pages 34-35.  My typed and sorted table is in the cloud here.

The FDA doesn't seem to require 80,000 blocks in 43 tumors to merit the pan-cancer label; the MSKCC IMPACT test also is labeled as pancancer based on about 10,000 blocks in 17 cancers. 


Update: A numeric error was corrected a few minutes after the original post.
ACS Cancer Facts 2018, here. Odds of dying of 20 cancers (e.g. testicular cancer, odds, 1 in 5094, here.