Log in to post a reply
Oct 30, 2021 05:40PM
In my humble opinion, breast cancer risk models in general are more art than science. They may give you a 3 figure number (such as 47.8%, taking a number out of the air), but they do not tell you how much confidence you should put in that number. They don't tell you how well they know that 48.7% number.
I do not have the expertise to tell you why.
Since I am not a saint, scientist or epidemiologist, the numbers I want are what is MY risk of breast cancer, not how many people in my country will get breast cancer because I am selfish.
I am NOT a statistic guru. I have classic LCIS, and a gene mutation, while NOT BRCA, is in the BRCA pathway. Only after about 4 consults, including 2 to an NCI-certified tertiary institution, that my GP told me we really don't know my lifetime risk.
I have been given numbers as varied as "something between 10% and 40% but probably closer to 10% than 40% - if you want better numbers go to the literature" from one of the NCI-certified consults (note that the 'average' woman in the US has about a 12% lifetime risk of BC, and I had never heard of LCIS reducing breast cancer risk), to, in a non-peer reviewed model, up to 85% lifetime risk. Well, 10% is a bit different than 85%. But these are different models. I was told I was high risk, but in other academic papers I would be grouped in a medium risk group. If my risk was really 10%, that would be lower than the average for the general population of women!
So, how 'high' is 'high risk'? How long is a piece of string? We don't know.
From what I understand, to develop a model, they need to have all sorts of data on people that do and do not have breast cancer. After they develop the model, they should be comparing their results with their population. So, taking numbers out of the air, the model may predict that 148 people in a population will get breast cancer, and the actual number may be 150. That would be a good model _for the population_.
But it does not tell you WHICH individual people will get breast cancer. It can give you no information if Leticia, Peter and Nicole will get breast cancer and Felipe, Anh and A.J. will not. A model can be great for the first (148 instead of 150 people in a particular population will get breast cancer, so they can hire the correct number of oncologists, for example.) It can also be awful in the second: if you place Leticia in the breast cancer group, there might be a 60% chance that you are correct and a 40% chance you are wrong. Remember, 50%-50% is a random coin toss, so a 50%-50% model is no good at all. That is what happened in the Gail model in this paper .https://academic.oup.com/jnci/article/98/23/1673/2...So a model that is correct 60% of the time and wrong 40% of the time is better than pure random chance, but not really very good.
This paper opined the Tyrer-Cusick model was not accurate for atypical hyperplasia https://pubmed.ncbi.nlm.nih.gov/20606088/ or for LCIS https://pubmed.ncbi.nlm.nih.gov/31559544/
From what I understand, you have to have a risk factor at least as potent as a BRCA mutation to tell a breast cancer population from a benign population. And even then there is quite a bit of overlap. They just do not understand what exactly puts someone at risk for breast cancer.
I have seen this happen in other medical situations, where a doctor might tell you that you are at 60% risk for something. But they forget to tell you how WELL they know that number. Especially for YOUR risk, not the risk of people in the whole population.
Unfortunately, there is a lot we don't know about breast cancer.
Classic LCIS.If knowledge can create problems, it is not through ignorance that we can solve them- Isaac Asimov
12/8/2005, LCIS, ER+/PR-
1/24/2006 Lumpectomy: Left
7/15/2006 Tamoxifen pills (Nolvadex, Apo-Tamox, Tamofen, Tamone)