Read on Twitter
Intriguing @theNCI @NCIDrDougLowy @NEJM study showing lung cancer mortality rates falling since 2013 faster than incidence rates. Study title claims they observed the impact of improved treatments. I think that may be an overly strong conclusion. 1/n https://www.nejm.org/doi/full/10.1056/NEJMoa1916623?query=featured_home
To my read, the 'new Rx's reduced mortality" conclusion is built on on this (and only this) logic: 1) survival improved; 2) new Rx's were intro’d; 3) ergo new Rx's caused improved survival. They could have done so much more. (Warning, fractions appear in subsequent tweets) 2/n
Ask - do either observed or presumed treatment variation correlate with survival changes? Regional uptake variations could have been used, so could presumed ones (like traditional race and age gaps). They didn't and in fact gains were homogenous across race/ethnicity 3/n
Or ask - does this make sense on the back of an envelope? Specifically, could EGFR and ALK agents (erlotinib and crizotinib) plausibly improve 2 year survival in 2013 by 3%? This calc involves only the max survival improvement and the pct of patients possibly treated 4/n
So here goes. For EGFR/erlotinib mutation prevalence is betw 15% & 20% for NSCLC. The absolute improvement in 2 yr overall survival (the authors' metric) is around 2% (my model back calculated from an 8% delta at 5 years on the drug label). 5/n https://www.accessdata.fda.gov/drugsatfda_docs/label/2010/021743s14s16lbl.pdf
Those are best case for prevalence, testing, survival benefit matching clinical trials. These conditions are never met. E.g. testing was 43%/24% for EGFR/ALK respectively in 2013. That dilutes down effects to 0.1%/0.02% and combined 0.12% 6/n https://www.mdedge.com/hematology-oncology/article/141544/lung-cancer/physician-attitudes-and-prevalence-molecular-testing
Or they could have asked if the treatment sales equated to widespread use Roche reported $650 MM in Tarceva US sales in 2013. That tells us that even assuming all for lung (pancreas another indication) treatment rates were low: 7/n https://www.roche.com/dam/jcr:517e5929-1591-4652-bf31-b63debeba974/en/med-cor-2014-01-30-e.pdf.
My math, using price and treatment duration data, comes out with a ceiling treatment number of 10K lung cancer patients in 2013. That is abt 15% of all stage IV NSCLC, lower than the 20% I used in the earlier calculation so even a smaller aggregate effect. 8/n
Frustrating. New lung ca treatments are great. But reconciling them relative to other investments is foundational to good public health planning. FWIW, the 2013 US Tarceva sales figures? They equal typical 50 state annual investment in tobacco control. 9/9 https://www.cdc.gov/mmwr/preview/mmwrhtml/mm6424a5.htm
