what seemed interesting here was that they numerically combined two long-accepted trends : svr is critically affected by time-to-und and a background of cirrhosis. After clearing out the clutter, the message from their final 48w logistic regression equation is:
svr can be well predicted by weighing (1) time to und (2) platelets and alt at base (3) ast at w12.
The first two variables are no surprise, and it's impressive that they predicted outcome with an AUC of 87 among the 76 patients not used to calculate the formula. However ast at w12 seems suspicious - what is this measuring? People report all kinds of crazy fluctuations in their enzyme levels during tx that seem to have no bearing on SVR.
Okay, Robert. Fair enough. I'll wait another week and ask a Computer Engineer. ;-> (Sorry...couldn't resist. :)
Seriously though....thanks for your assessment of the authority behind this, that helps for now. Appreciated.
Aaron...likewise. Thanks also. Just a totally different approach and interesting. You mention other studies of his - I wouldn't mind some links - for those boring days when I have nothing better to do. :)
Incidentally, not quite spring enough yet around here. Still cold, although snow is gone. My bike does come out for spring maintenance this weekend though. The buddy who stored it is going to work with me on my bike and we're going to get her road worthy together. Now there's a brave man. :)
Hope you're hanging in there okay, Aaron.
Trish
"Do you think they are applying their theory correctly? This relatively simple mathematical formula - are they applying it to the correct data in an appropriate way? Or perhaps you already answered that and I'm not getting it."
Honestly, I can't speak to the medical reasons why they feel this is valid. These guys are highly trained, highly respected medical researchers.
In a nutshell, they analyzed common parameters using complex statistical methods to find those with the highest correlation to SVR. I don't know that they had a "theory" so to speak. They used the math to find certain relationships (to SVR) that stood out, then used statistics to validate that relationship.
As Aaron pointed out, these guys are high caliber pros, publishing in an internationally respected medical journal. I basically have to just take their word for it.
That's why they make ten times the money I do .
:-)
Robert
Hi Trish,
The doctor and his research teams developed this equation, specifically because, there was no others in existence using simple data that is easy to acquire, other than viral kinetics, and more importantly, because using viral kinetics is not very accurate in predicting Tx outcome.
They set out to develop a more precise model to use , in conjunction with viral kinetics .
The doctor that conducted it has been researching ... and ... publishing scientific papers involved in different types of viral research since the mid 50's .. and published on HCV research since 1991.
So, I would venture to say, since the research teams developed it .. using the simple data they discovered could be applied, yes, it's pretty accurate for their subjects, 100% accurate .. Nope, but with high CI, PPV's & NPV's ... it is thought provoking .
It is an absorbing read ... as are some of his other studies ... If your bored and have nothing better to do ... like ... going for a motorcycle ride on a nice spring day : )
Cheers
LOL! I'll tell him that one - especially the "1 or 0" comment...pretty funny. :)
Sounds like it will be a fun mother-son exercise to do this with him. I used to be good at math but it's been a very long time - you're right, I looked at that formula and my brain revolted. :)
Perhaps I asked the wrong question. Do you think they are applying their theory correctly? This relatively simple mathematical formula - are they applying it to the correct data in an appropriate way? Or perhaps you already answered that and I'm not getting it. Thanks Robert.
Trish
I'm far from a statistician, so I'm not really qualified to speak to that, but I had the usual courses in stats in college. It seems quite rigorous mathematically speaking. The sample size feels a bit small, but the positive and negative predictive values (PPV and NPV) are quite high. They used very well known methods, and a commercial quality statistics software package to do the crunching.
I'm currently looking for published reviews of this study to see what others think. I was hoping willing or one of the other pros would comment on this, since they've seen many, many studies, and are probably familiar with this one.
Will keep you posted.
P.S. Don't let the formulas scare you. When you break them down to individual terms, they are absurdly simple. The hardest part was carefully reading the paper 3 times to get all the definitions right on what went where, units of measure used, etc. If your son is a Computer Engineering student, this will be child's play for him.
Of course, numbers other than 1 or 0 might scare him, LOL.
Just kidding - my degree is in Computer Science, and Comp Eng and Comp Sci majors are always poking fun at each other :-) We tell them without our software, their big fancy machines are nothing but a "box full of silicon rocks".
Robert
Hi Robert, my son is an about-to-be engineer also - in final exams and will complete all this requirements for his Comp Eng degree this coming Monday. AFTER that...his Mom is going to lay this study in front of him and ask him to help me with this math as I'm interested in this one.
What's your interpretation of their methodology here? Thanks.
Trish
I forgot to thank Aaron57 for showing me this study. Thanks!
Robert