It was a stat that I came upon probably a dozen times when I did a deep dive on Google Scholar and read every study, within the past ten years, on estrogen-positive breast cancer. The non-compliance for post-surgery adjuvant hormone theory, within the first five years, was repeatedly cited as approximately 50% so I assume that it's an accepted statistic in the breast cancer field. I'm sure you can find it by deep-searching on Google Scholar. That's the venue for filtering out general-public articles and so much of the fluff poor-content cut-and-paste articles clog up the search results. I don't remember though if the 50% non- or fall-off compliance separated stats for tamoxifen and AIs. And an oncologist I saw referred to it, so I just assumed it's current thinking.

There was one very-flawed study of Kaiser-Permanente breast cancer patients, looking at how many, prescribed, free, post-surgery anti-hormone drugs did not, or stopped, taking them within 12 months and the non-compliance rate was over 50%. Kaiser was unable to question all the several thousand (if I recall correctly) women who non-complied but, which is interesting, 80% of the ones they could locate, five years later, had not had a cancer recurrence.

There was one article, I think in JAMA, where an oncologist questioned the ethics of making 49 women take aromatase inhibitors, with their very serious side effects to, only maybe possibly result in 1 of the 49 maybe not having a recurrence of breast cancer, and the remaining 48 suffer any side effects for no defensible reason. His argument was whether that was, in fact, good medical practice and suggested prescribing physicians make extra effort to educate the patients about the side effects that were (so far) known.

If I had the URLs at the ready, I'd be happy to post them but I lost bookmarks on my tablet when upgrading.

As to OncotypeDX numbers, I don't know the difference between the Recurrence Score and the (I assume) deruved Risk of Recurrence result. I 'think' but that's just a guess, the RS might be the industry wide accepted stat for, say, a Stage X invasive DCIS, with Y lymph node involvement of Z size. And the Risk score is where the genetic analysis of the 21 genes puts that patient vis a vis the wider data pool.

But that's just guess. If someone reading this had a higher Risk than Recurrence Score maybe that would be suggestive. And if I ever talk to Oncotype again! I'll ask and post as I'm curious too. But it might be a complicated derivative of their proprietary algorithm and of no value to us here.