Wednesday, January 10, 2018

Informally Summarizing the FMI CDX Concordance Studies

The FDA homepage for NGS PMA tests is here, product code PQP.  The FMI F1 CDX large panel test is P170019, here.

This is a very brief, freehand, and informal summary of some key S&E FDA review information. 

On page 36, the test has coverage of 702-793X in exomes.  Sequencing error rate was below the requirement of .01, specifically, about .003 (about 30% of the requirement).

Specimens passing all QC ranged from 96-100.   Four tissue types had <90% adequacy rates: pancreatic (83%), appendix (88%), pericardium (79%), prostate (84%).   

Longevity of paraffin block specimens was directly tested to 6 months with further tests underway.  However, two of the concordance studies used clinical samples with much older blocks.

Concordance studies were broken in seven groups:

  1. EGFR Exon 19 DEL & L858R, cobas V2.
    1. These two errors are about 85% of EGFR mutations.
  2. EGFR T790M, cobas V2. 
    1. This is a resistance mutation so only seen after treatment.  It is about 50% of resistance mutations.
  3. ALK rearrangements  Ventana and Vysis
  4. KRAS   Therascreen
  5. ERBB2/HER2   Dako FISH
  6. BRAF V600    Cobas V600 ThXID
  7. BRCA 1-2    FMI Prior Art / FoundationFocus CDx BRCA

click to enlarge

Generally, concordance was based on Noninferiority of Li et al. (2016).   

Controls were CCD1 and CCD2 (duplicates of the control, which provides variance of the prior art PMA assay).  

Only two of the examples were based on clinical trial blocks (EGFR T790M, ALK), and these had more problems with missing samples or poor performance of CCD2 (possibly due to age).  Generally, about 200-300 blocks were used and samples were weighted to be about balanced positive and negative (based on CCD1).   Data was presented also as corrected for expected real world prevalence, which as FDA notes gives data somewhat lower than the lab data on the artificial prevalences.

In those two cases where clinical trial blocks were used, the data assessment is based on analytical validity (analytical concordance).   Also note that while FMI reports full sequence analysis, the FDA PMA evaluations can be based on much simpler point-mutation concordance.


Samples were convenience samples pre screened to give positives and negatives.   406 samples were tested.  84 were duplicates.  267 samples were available for final testing after discarding various process failures.

As in other "pure external concordance" cases, agreement was quite high.  There were 112 positives and 170 negatives.   Of the few mismatches a few were explainable by unusual mutations.  However, when corrected for real world prevalence, % agreement was slightly lower.  

Nearly all the % agreement were >97%.  


Blocks were based on AURA trials for osimertinib.  However, of 354 samples, 227 had full comparison data.   There were 145 positives, 167 negatives.  CCD2 had more common mismatches (e.g. 27 CCD2 negative out of 145 CCD1 positive).  This could be due to block age.

Most of the % agreement were anywhere from 72% to 95%.


317 retrospective convenience samples were available.   There were 125 positive, 192 negative.  CCD2 had some mismatches, e.g. 12 of 125 and 9 of 192.   Of 113 CCD double positives, 12 were F1 negative.

Most of the % agreement were from 80% to 96%. 


175 samples were from Roche ALEX alectinib vs crizotinib study (including screen failures to provide negatives).   Repeat Ventana IHC was not available, Retro VYSIS data was used as a surrogate.   Of 487 samples, 172 had insufficient slides.  Total positives 134, negatives 139.   Final samples were 175 after excluding missing data.  

For example, of 103 CCD double positives, 6 were F1 negative.  Of 134 CCD1 positives, 11 were CCD2 negative.  A smaller sample analysis was also presented (Table 41, 140 samples).  

% agreement ranged from 65% to 98%.   


343 retrospective convenience samples.  As with other convenience samples, correlations were very high.   There were 177 positive and 165 negative.  

% agreement mostly >98%.


305 retrospective convenience samples from advanced melanoma.   167 positive and 138 negative.  Concordance was nearly perfect.

% agreement was mostly > 98%.


The two tests use the same reagents, equipment, and procedures except for lower DNA input for library construction and "enhancements in the analysis pipeline" which "have no impact" on the assay. See PMA P160018.

Summary Figures

For a 27 page deck on the Japanese regulatory approach to NGS CDX, see here.

No comments:

Post a Comment