Shopping and sightseeing
Vancouver, 2009
I have never claimed to be a hotshot mathematician. I always WANTED to be a hotshot mathematician, but no matter how hard I tried, I just wasn’t one. It always seemed like there was something macho about proving your point with mathematics; sometimes, I suspect, I almost would rather have been wrong – but have used flashy math - than to get it right. (No, not really, but almost.) Mostly I just used trig and calculus, and of course the arcane math of paleomagnetism that only another paleomagnetist would care a fig about. One topic I managed to keep at arm’s length for 50 years was – statistics. So now, here I am: running into statistical arguments in most of the papers I read, and not knowing what in hell they are talking about. So, I am uniquely unqualified to write the remainder of this blog – but here it comes, anyway. The topic: early detection of ovarian cancer.
As you know, my group at the Hutch is focused on the early detection of ovarian cancer, mainly using “markers” that can be measured from a blood sample. One of these markers will be familiar to most of you – CA 125. It turns out that CA 125 is another name for the protein mucin 16. (The name CA 125 came about because MUC16 was the 125th cancer antigen used in a study of a particular ovarian cancer cell line. Back in 1981.)
The title of “our” paper almost says it all: “Longitudinal Screening Algorithm That Incorporates Changes Over Time in CA 125 Levels Identifies Cancer Earlier Than a Single-Threshold Rule” It appeared in the Journal of Clinical Oncology just last month. It reports the results of a “retrospective” study of women who had been involved in the PLCO (Prostate, lung, colorectal and ovarian) Cancer Screening Trial. This was a massive effort involving nearly 80,000 women aged 55 to 79, who were followed for many years. “Retrospective” means that the medical histories of these women were known: the research then consisted of using – you guessed it - statistics to determine what test or tests would have done the best job of detecting their tumors early. By “single-threshold rule” is meant a fixed concentration of CA 125, above which further tests are recommended. The “normal” range for CA125 is 0 to 35, so an example of a single threshold rule would be to recommend ultrasound for any woman who had a concentration >35.
There is a big problem with this, if applied to the general population; there are lots of thing other than ovarian cancer that can raise the CA125 level, and among healthy women the CA125 level can vary significantly. Such a test is so low in “sensitivity” and “specificity” that the NIH, or some equally august body – I forget which – has recommended against single-threshold screening for the general population, although the method is useful in the high-risk population. ( I believe that my group at the Rivkin Center is engaged in just this sort of screening.)
There is a big problem with this, if applied to the general population; there are lots of thing other than ovarian cancer that can raise the CA125 level, and among healthy women the CA125 level can vary significantly. Such a test is so low in “sensitivity” and “specificity” that the NIH, or some equally august body – I forget which – has recommended against single-threshold screening for the general population, although the method is useful in the high-risk population. ( I believe that my group at the Rivkin Center is engaged in just this sort of screening.)
So, naturally, any screen that can be tailored to the individual woman should be a big improvement – and that is precisely what my Hutch group is proposing. The tool of interest is called a “parametric empirical Baysian longitudinal algorithm.” What follows (in the paper) is a mind-glazing bit of mathematical explication, which I fail to follow. Fail miserably, in fact. (I will show the equations to my statistical drinking buddy Jay Teachman when I get home; maybe he can enlighten me.) Anyway, the PEB algorithm worked significantly better than the single-threshold test. It detected more tumors, earlier – ten months earlier on average, and at a lower CA 125 level. Ten months isn’t much, but it could give the oncologist a head start.. What is needed, of course, is a test that spots a cancer case at the moment it first protrudes its ugly little head. And then, a way to kill it.
There seems to be a competing method called ROCA, which codes for Risk of Ovarian Cancer algorithm. I gather that it, too, is more sensitive than the single threshold method, but diligent Googling so far has failed to reveal to me how it works*. Maybe somebody can enlighten me.
And so, yeah, this was a dull blog, and overlong to boot. I needed an excuse to post another picture. The weather here today was mid-70s, sunny, light breeze. How was it where you are?
*However, see my Comment, below
*However, see my Comment, below
Do they have a demographic for which they might want to implement these tests? Those with high risk factors perhaps?
ReplyDeleteIt rained hard all day yesterday and the wind blew on and off. Yuck. Was calm for my walk to work this morning but it's not looking good for later:
Today: A chance of rain in the morning...then rain likely in the afternoon. Highs in the lower 40s. East wind 5 to 20 mph.
They do have a demographic, but I think it varies slighly from one clinical trial to the next. I will look into the specifics, but in general you fall into the high-risk catagory if any of the following is true.
DeleteYou are of eastern European Jewish ethnicity.
You have either the BRCA1 and/or BRCA2 mutation.
You have a first-ddgree relative that had either ovarian or breast cancer: mother, sister.
You have several more distant relatives (cousin, aunt, etc.) that have had either type of cancer.
Contrarily, you tend to be safer if:
You took birth-control pills for an extended period.
You have been pregnant.
You have breast-fed.
And, of course, if you have had your ovaries and follopian tubes removed surgically.
When I get the chance I will check these things from my memory.
Thanks, I do remember your previous post that included this info, maybe from last summer? It's good they have a preliminary plan for implementation.
DeleteFantastic photo! Oh, and it's in the 30s and foggy. Yuck.
ReplyDeleteNow I know a little bit about ROCA. My earlier problems resulted from asking Google to track down “ROCA”, rather than “Risk of Ovarian Cancer Algorithm”. Ask for ROCA as such and you get various kinds of candy, associations of realtors, etc. Somewhere down the line there must have been what I was looking for, but I lacked the patience to search.
ReplyDeleteSo, anyway, ROCA seems to be very like the method proposed by my Hutch group. It was developed at the University of Texas M.D. Anderson Cancer Center. The Texans seem to have beaten us to the battle field, armed with a very similar weapon. As with our study, women (postmenopausal – age range 50 to 74) have their CA125 levels measured, then re-measured at stated intervals. The trend in CA125 number, rather than its absolute values, is used to establish a baseline for the individual woman. Thankfully, the “algorithm” part (i.e., the statistics) was not explained. Whenever a woman’s CA125 trend triggers whatever threshold the stats have defined, she is referred for transvaginal ultrasound and consultation with an oncologist.
The Texas study involved 3238 women. It shows excellent promise, although Dr. Karen Lu, the boss, cautions that “(results of this study) are not practice-changing at this time.” Well, hell, get to work; we’re not dealing with bad-breath or pimples here.
An aside: An interesting statistic I came across is that 1 in 2500 postmenopausal women will contract ovarian cancer. This is considerably greater than what you find in the general population (22,000 per year divided by maybe 150 million women - or roughly 1 in 7000. Obviously the best game plan is to simply not go into menopause – but, if you’re a woman I guess that’s a bit difficult to arrange.
My hatred for computers knows no bounds, and grows each day. Just now (2/13) I attempted to increase the size of the type in the main tesxt of this blog. Insead, I somehow managed to insert two blank lines between the first and second syllable of the word "rather" in the first paragraph. No trick of mine is proof against the stubborness of this machine: "rather" will be divided, come hell or high water. Fortunately Carolyn will be here in a few days: maybe she can figure it out.
ReplyDeleteCarolyn came, took one look at the situation, said "you've inserted a blank table", and deleted it. Thank God for Carolyn. But I still hate computers; they make me feel so, well, old.
ReplyDeleteSo glad to help, Myrl. A small thing to give you after all the hospitality you gave me. As for feeling old, I forgot my hair appointment yesterday afternoon even though in the morning I had it on my mind. Yes, you are not alone in aging.
ReplyDeleteI still don’t REALLY understand Baysian statistics (as in “parametric empirical Baysian algorithm”, but Dick Ingwall sent me a link that – if not explaining it clearly – at least explains it amusingly. I think I have written about this type of statistics several times, but this blog is the only one I can find. You might kill a few minutes and read the following article. It would help if you like mathematics. Most of you don’t, but read it anyway.
ReplyDeletehttp://www.nytimes.com/2014/09/30/science/the-odds-continually-updated.html?smprod=nytcore-ipad&smid=nytcore-ipad-share&_r=0
Thank, Dick. Most of my time these days is given over to admiring my great grandson. This let’s me slack off but not feel guilty.