Interested to see how the forthcoming data behind the ‘TMB > 10 in all cancer types’ approval addresses a few key issues, having spent a long time thinking about this general topic - some thoughts to follow:
[1/n]
[1/n]
(Heeding wise words of @tmprowell - this is *not intended as critique*, but rather some open thoughts on the matter that I'm excited to see in the final data once available - feedback welcome! https://twitter.com/tmprowell/status/1273043390716948483)
1) Technical:
- Many non-biological factors affect TMB calculation, like sample quality, tumor purity, sequencing artifacts, and NGS fails.
Difference btwn NGS from TCGA samples vs. clinical trial samples vs. real world clinical samples is large. Also: https://www.nejm.org/doi/full/10.1056/NEJMoa1801946
- Many non-biological factors affect TMB calculation, like sample quality, tumor purity, sequencing artifacts, and NGS fails.
Difference btwn NGS from TCGA samples vs. clinical trial samples vs. real world clinical samples is large. Also: https://www.nejm.org/doi/full/10.1056/NEJMoa1801946
- Inherited variants in tumor-only NGS are always a problem, even more in underrepresented minorities. How was that considered? Were patients of diverse heritages included + was performance evaluated specifically in these groups?
https://bmcmedgenomics.biomedcentral.com/articles/10.1186/s12920-017-0296-8 https://genomemedicine.biomedcentral.com/articles/10.1186/s13073-016-0333-9
https://bmcmedgenomics.biomedcentral.com/articles/10.1186/s12920-017-0296-8 https://genomemedicine.biomedcentral.com/articles/10.1186/s13073-016-0333-9
- How will approval be considered for other NGS tests? TMB > 10 in FMI testing is not identical in other tests - see @NIVokes et al:
https://ascopubs.org/doi/pdfdirect/10.1200/PO.19.00171
https://ascopubs.org/doi/pdfdirect/10.1200/PO.19.00171
2) Biological:
- TMB is intended to capture tumors w/ neoantigens that results in immune activity in setting of immunotherapy.
But many background processes can create high TMB mutations and collapsing this all is hard to understand biologically or detect in NGS panels. (cont.)
- TMB is intended to capture tumors w/ neoantigens that results in immune activity in setting of immunotherapy.
But many background processes can create high TMB mutations and collapsing this all is hard to understand biologically or detect in NGS panels. (cont.)
E.g. MSI and HRD both are DNA repair mutational processes that result in higher TMB in prostate cancer (+ others), but only the former has links to ICB response (w/o requiring high TMB).
Did data underlying approval consider these biological confounders?
Did data underlying approval consider these biological confounders?
- Mutations are not equally represented in patient tumor cells. Does TMB > 10 include all mutations or only clonal mutations? Was this considered or evaluated in the cutoff design? H/t @CharlesSwanton @NickyMcGranahan seminal work:
https://pubmed.ncbi.nlm.nih.gov/26940869/ (cont.)
https://pubmed.ncbi.nlm.nih.gov/26940869/ (cont.)
E.g. Patients with subclonally active mutational processes like APOBEC might have a high TMB subclone despite the rest of the tumor being low TMB.
See also our 2018 study for further breakdown of these complex issues - tl;dr it's complicated!: https://pubmed.ncbi.nlm.nih.gov/30150660/
See also our 2018 study for further breakdown of these complex issues - tl;dr it's complicated!: https://pubmed.ncbi.nlm.nih.gov/30150660/
3) Clinical:
- How were prognostic vs. predictive aspects of TMB considered? How many patients had cancers w/o prior approvals for immunotherapy? https://twitter.com/VanAllenLab/status/1273296492242055168?s=20
How were the widely variable clinical courses for rare cancers considered statistically in the analyses?
- How were prognostic vs. predictive aspects of TMB considered? How many patients had cancers w/o prior approvals for immunotherapy? https://twitter.com/VanAllenLab/status/1273296492242055168?s=20
How were the widely variable clinical courses for rare cancers considered statistically in the analyses?
- In our 2018 multi-histology study ( https://pubmed.ncbi.nlm.nih.gov/30150660/ ), AUC for TMB = 0.66 w/o clean cutoff. How was 10 chosen? What multivariate clinical stats were done here? (cont.)
When we ( @dliu_ccb) specifically revisited melanoma + TMB + immunotherapy with deep clinical data for predictive modeling, TMB was not a strong contributor to prediction and confounded
Will be fascinating to see the stats here https://www.nature.com/articles/s41591-019-0654-5
Will be fascinating to see the stats here https://www.nature.com/articles/s41591-019-0654-5
- Do folks understand how to interpret TMB nuances in NGS reports? For patients w/ cancers that already have immunotherapy approval but low TMB on NGS test (e.g. all kidney cancer), will the community know when to ignore that result? (Probably, but also: https://academic.oup.com/jamia/article/25/5/458/4791826)
Generally, I lean liberal for Rx access + I am excited to see how these issues were considered in forthcoming data
Like many, I have seen extraordinary responses to immunotherapy + of course want that for as many patients as possible, though I've also seen toxicities...
Like many, I have seen extraordinary responses to immunotherapy + of course want that for as many patients as possible, though I've also seen toxicities...
Overall, as with much in the immunotherapy molecular correlates space, going from biology to biomarker with complex tech to assess targeting an extraordinarily complicated biological process in diverse clinical settings is a challenge worth taking on - lets work together! [fin]