This is a somewhat technical look at what SDNA performance (both programmatic and individual) we may expect based on data available before the val study.
One way of looking at the issue of programmatic sensitivity is to consider tissue performance (sens/spec) and stool performance. For practical purposes, given EXAS’ demonstrated highly sensitive methylation detection capability, the tissue results are essentially deterministic (i.e., non-random, dependent on whether any of the test markers “cover” the particular cancer in question) and the stool results attempt to estimate the tissue results, hindered by errors due to random measurement noise. Repeated testing mitigates the random noise effects, and over time brings stool sensitivity close to tissue sensitivity. Repeated testing does not overcome limitations of the marker panel, except that some cancers that are not covered will be detected accidentally at the false positive rate, and the programmatic false positive rate increases with the number of tests performed.
We only know tissue performance imprecisely. A modest sized study reported in the poster at the Anaheim ASCO conference in July reported 100/100 sens/spec for both pre-cancer and cancer for the 4 marker panel, but in reality it seems likely to have tissue sensitivity in the upper 90s. (That’s an informed guess based on limited available data.)
The tissue data in that poster was very encouraging for several reasons. A quote from the abstract:
<< Median methylation level was 76 (range; 0~1,089,202) in
cancers, 4766 (range; 0~7,329,650) in adenomas, and 0 (range; 0~0.1) in normal epithelia. Vimentin methylation level was significantly higher in cancer and adenoma than in normal epithelia (p<0.01).>>
This referred to the smaller of two studies reported, for vimentin only, in tissue. The very high ratio of median vimentin methylation level for adenomas to the maximum level for normals (4766/0.1 = 47,660) tells me that, while a few adenomas don’t have detectable vimentin methylation in tissue, those that do have it have a level high enough to be quite noise-resistant, so stool performance should still be quite good, despite increased measurement noise compared to tissue. (Here, methylation level was expressed as the ratio of methylated to unmethylated.) This is also true, but to a significantly less degree, for CRC. Using four markers is even better, and Table 2 of the poster reports a larger trial (37 cancer, 25 adenoma, 29 normal) with better sens/spec performance (100/100 and 100/100 sens/spec for CRC and pre-cancer). This is among the several reasons I believe val study results will exceed expectations (other reasons previously discussed). Given the lack of competition, getting anywhere near guidance will be a home run IMO.
What will really matter long term is improving outcomes. It’s seems obvious SDNA should save many lives, but precisely how many remains to be seen. It will be interesting to learn how likely it is that any pre-cancers the current marker panel does not cover lead to cancer and death. Does absence of the usual markers indicate a less dangerous pre-cancer? I'd welcome any insight on that.
"We looked at all 32 known methylation markers that are specific for colon cancer, spent about 1/2 a million dollars narrowing down those with a broad sequencing study we did with the Mayo Clinic, to 9 target markers, and we have narrowed that down to a final 3 or 4." (1)
This statement leads me to believe that a lot of work has gone into identifying what, genetically, is different between cancerous and normal tissue. The set of markers is probably not exhaustive, but lets say it is good enough. Of the 9 markers studied, table 2 of the tissue study poster (2) shows that markers A-D are the winners. Looking at table 1A, samples that showed relatively high values for other markers (E-I) also showed relatively high values for the winners, A-D. I didn't notice any samples that showed high values for E-I, but low values for A-D. This would suggest that, at least for this study, all the cancers and adenomas would be caught by looking only at markers A-D.
Going through the exercise of describing one's reasoning to another person motivates good homework and helps solidify the argument in my own mind. That's the great potential of this message board. Thanks Chas.
1) 09/14/10 Rodman & Renshaw 12th Annual Healthcare Conference, index 19:50
i may not have fully understand your 4th paragraph. did you mean getting anywhere near guidance is a homerun - implying that being a bit below is good enough? while you could be right from a clinical practice perspective, i don't believe the stock will react well.
The market is huge enough that getting anywhere remotely near guidance on market penetration, price, schedule, margins etc. supports a far higher eventual price. (Market opportunity over a billion, current price a fraction of a billion.)
The most important performance metric is pre-cancer sens/spec, followed by stage 1 sens/spec. Later stage detection is necessary but not sufficient.
Non-patient friendly methods are severely limited by compliance, also expense, and have other limitations.
Other patient-friendly methods have very low pre-cancer sens (maybe 20% at best), low stage one (under 40%), and FOBT/FIT requires sampling multiple stools each year (which also has compliance problems).
A one stool sample per 3 years test with pre-cancer sens/spec of, say, 45/90, and stage 1 at 80/90, a little under guidance, would be light-years ahead of any competition in both patient-friendliness and performance, could still virtually eliminate cancer deaths for those regularly tested, and would be highly marketable.
Is better performance better for the EXAS bottom line? Of course. But, solid real-world confirmation of performance even a little under guidance would still make the EXAS test very valuable. To suggest that 51% precancer sens would be a home run but 49% would be a flop doesn't make sense to me.
I don't believe current price has built in confirmed performance at guidance. Expectation and confirmation are two different things. (Ask the Epigenomics stockholders.) EXAS is not fully on the radar yet, val study will help with that.
I don't think we'll ever find out for sure, because I think val study results will exceed guidance (very possibly by a lot).