When Are Changes in Genomic Medicine "Too Small?"

Genomic medicine improves precision, but when are changes "too small?"

I've seen two examples in recent reading:

• Andre' et al, reported by Caroline Hopkins, look directly at the "long tail" of genes in multi hundred gene CGP testing.   They assert that little if any benefit resides in the "long tail."
• Regarding lung nodule prediction, Mazzone et al. note that expert opinion about matches the best nomograms (and molecular testing is incrementally better than those).   Then, regarding prostate cancer prediction, Li et al. find that AI-supported MRI is about as predictive as molecular profiling of prostate cancer.  

CGP TESTING

There are many arguments for more comprehensive genomic testing in advanced cancer.  CMS has an NCD covering all PMA/companion diagnostic gene panel tests, like the Foundation Medicine test.   And the Personalized Medicine Coalition just posted a new report on defining clinical utility for genomic testing, and defining it more broadly.  See Pritchard et al. (2022), open access here, "CLinical Utility of Genomic Testing in Cancer Care."  See also the older Joseph et al. AMP consensus report, 2016, here.

That said, there are other lines of evidence, too.  See a deep-dive article at Genomeweb/Precision Oncology News, by Caroline Hopkins, in December 2021, here.  In a study by French authors https://pubmed.ncbi.nlm.nih.gov/33321450/they found that only the higher-evidence genes were actionable with improved clinical outcomes, and genes with little evidence, but present on large panels, did not seem to improve outcomes.   The study uses European Society Medical Oncology system ESCAT, actionability of targets, as a rating system.

MOLECULAR PROGNOSIS VS NOMOGRAMS AND CLINICAL DATA:

LUNG NODULES, PROSTATE CANCER

Since the introduction of prognostic gene expression tests like Oncotype Dx in 2006, clinical and technology evaluators have worked to access their incremental utility over other standards of care.   I was told at the time that MolDx discontinued coverage of the CardioDx test in part because of new insights at MolDx that much of the clinical power was driven by clinical variables like age.  

On this theme, two things caught my eye this past week.  

Lung Nodules. One is a review article on pulmonary nodule management in JAMA.   See Mazzone 2022, JAMA here.   (See also Maldonado 2020, JAMA, a review of prediction models for small nodules found from LDCT, here.)   Mazzone notes that clinical judgement (older patients with larger lesions!) already has an AUC of .70-.85, whereas calculators may have AUC of .72.-77.   At least some molecular tests for nodules so far provide "incrementally better" AUC values against these benchmarkers.  Medicare covers the XL2 test (L37054 here).  to Mazzone's point, see e.g. Fig 3 in Silvestri 2018 (here). 

Prostate Prediction.  In another interesting study, see Li et al. 2021 in eBiomedicine here.  The team studies an AI-supported MRI-based classifier for prostate cancer outcomes.  Argues that the AI-assisted MRI scores with clinical information may at least equal performance of molecular gene panel expression testing.   

See an article on "Defining clinical utility" in translational research by Smith et al, 2021.

https://www.cell.com/ajhg/fulltext/S0002-9297(21)00334-7

Conceptualization of utility in translational clinical genomics research.  AJHG 108:2027.

And Pritchard et al. 2022

https://ascopubs.org/doi/10.1200/PO.21.00349

Clinical utility of genomic testing in cancer care.  JCO Precision Oncol (epub).