It’s a question every prostate cancer patient wants an answer to: “OK doc. I get it. But how long am I gonna live with this?”
It’s also a question that doctors have a very great deal of difficulty answering, for a whole bunch of reasons. So …
For those of you with a mathematical and statistical background (and I know there are some of you who are regular readers) here’s a link to the full text of a new paper by Mogenen et al. (a group of Danish researchers) who were trying to work out the most accurate way to make such estimates only for men with metastatic, castration-resistant prostate cancer (mCRPC). And what they found, based on the available data sets, was that they really weren’t able to come up with a highly accurate predictive model — even with data from several trials to work from.
To be fair, they think their new and very complex model is better than earlier models, but even they don’t seem overly happy with what they were able to achieve.
What was of particular interest, however, is that many of the most relevant information seemed to have little to do with the prostate cancer itself but rather more to do with the effects of the prostate cancer on certain commonly measured laboratory data, e.g., aspartate aminotransferase (AST), alkaline phosphatase (ALP), albumin, and hemoglobin levels, and the the patient’s ECOG performance status (from 0 to 5). Important prostate cancer-specific prognostic factors included things like whether the patient had soft tissue metastases to places like the liver and the adrenal glands and the prior use of different types of analgesic agents (such as opioids) for management of serious pain.
For those of you without a strong background in mathematics and statistics, I’d skip trying to make any sense of this. Your sitemaster is “competent” in math but less so in statistics and he had a really hard time understanding a good deal of this paper. He only tried at all because he knows how helpful it would be for everyone if we were better able to use the available clinical data to come up with better individualized answers to this important question.
Filed under: Living with Prostate Cancer | Tagged: death, mortality, prognosis, risk |
THE "NEW" PROSTATE CANCER INFOLINK 
This competition looks like it was a lot of fun for very nerdy people. I’m astonished by the quality of the competitors and the tight clustering of the leaders right towards the very end.
For me, I was more thrilled reading the first four paragraphs of the introduction than listening to a play-by-play sportscast. Yes, my nerd flag flies proudly.
If I may be so bold as to give a summary and interpretations:
• Statistical models were compared in a head-to-head competition.
• Each was given a complex and messy data set — 2000 patients with 93 possible variables, where many patients lacked values for many variables.
• The conventional-wisdom reference model (Halabi) had a good accuracy of about 74.29% in the final scoring.
• The best competitor managed beat the standard model with an accuracy just shy of 79.15%.
What would an accuracy of 100% look like?
It would mean that if 10,000 European and American patients with mCRPC were followed for 50 years, then the curve of their overall aggregate survival (i.e., how many died within 1 year, 2 years, etc.) would match the curve predicted by the model quite well. There would of course be variations, but they would fall within measurement error, not modeling error.
The curves would necessarily match less well for a 1000-patient cohort, and even less well for a 100-patient cohort. For an individual patient, they’d be of little use. (Well, except for patients like me, who like to know the probabilistic chances of dying in 2 years vs 20, knowing full well that a “50% chance of dying within 7 years” is only a median — if the curve has a long right tail, the “7-year expected survival” doesn’t rule out survival for 14 or even 40 years.)
Thanks for sharing this!
I knew someone out there would be into this stuff! Thanks Paul. :O)