AAAS Session Coverage: Evolutionary Personalized Medicine

This session was organized to talk about research in genomics and how it is allowing for personalized medicine with respect to early diagnosis, efficacy of medication, etc.  To now we have used medical experience and clinical trials for knowledge basis to make decisions, going forward we might be able to change to more early diagnosis and optimal treatment based on genomics.

Program Description

I’m going to try and turn my scrawled notes into something coherent and then my thoughts are in italics at the bottom. Any errors in the notes should be assumed to be my poor transcription rather than the speakers.

Speaker:

Sholom Wacholder, National Institutes of Health

(I was a smidge late–needed coffee, so I missed his intro)

The research being discussed was on breast cancer in white women of European descent around age 55.  For about 30 years, breast cancer has used the Gail Model: 7 Questions to calcuate the risk that a woman has of getting breast cancer.  Last year the study was published, identifying genomic regions associated with breast cancer. It was, as one editorial reviewed it, a tiny step closer.  We’re still not good at predicting who will end up getting breast cancer.

The benefits that we are seeing right now are only for those who are getting early treatment for breast cancer and this has to be balanced against the expense and hassle of early screening.  We want to have an intervention for those who most need it, those who are the highest risk and hope genomic research  can shape future recommendation guidelines to take risk into account.  If we are able to have better risk stratification it gives us more potential for benefit.

Our current results have to take into effect

  • Benefits minus costs
  • Sum of costs for everyone in the program
  • There will be winners and losers (those who needed treatment and didn’t get it; those who didn’t and did)

Public health evaluation of costs and benefits may differ from individual and even societal.

We’re moving towards better genetic information that we will combine with epidemiological and clinical data and eventually expand the research to non-European so we can have more accurate models to confirm or clarify the accuracy of our risk estimates and determine if we can improve allocation of income.  Doctors need to get away from “You are not average” to “we have recommendations tailored for your risk” so high risk and low risk can get appropriate treatments.

Question from the audience: What do we do when many docs don’t believe in evolution? How does it affect medicine and research?

Answer focused on research looking at outcomes.  Flippantly commented that if you could actually empirically prove that we could tie zodiac and date of birth to be tied to risk then we would include that as part of our intervention and diagnosis.

Question from the audience: The predictive models aren’t working so well right now. How do we keep going and getting buy in when we’re failing?

Answer: We’re still dealing with extremely new research. We’ve not yet had time to determine how all we’ll use it or how many lives it will change.

Speaker:

Knut M. Wittkowski

µGWAS on a Grid Enabling Small Sample Screening for Common Complex Conditions

Dr. Wittkowski spoke about Personalized Diagnostic Challenges.  The goal is to create signatures and discriminate (not in a bad way) who migh tbest need treatment.  We don’t know how many parameters on our data to discriminate appropriately but we do know that one signature does not fit all

We’re looking at the difference between a common and a personalized signature.  A common signature is based on study and tires to guarantee the general ability of a drug or treatment to have an effect on a disease or condition. A personalized signature looks at all of the data available about treatments and filters for data that is similar to the patient at hand.  For this we need more representatives and better data so we can eliminate parameters.

He spoke of John Arbuthnot, who he described as first to formalize statistical tests.  This u_statistics is good for binary, univariate and censored data.

Because we have any number of elements we could be working with, the suggestion is that we look at pairwise orderings, which is not dependent on weights or our external parameters.  Looking at one gene may not give us clues but looking at connections between two or three may give us an entirely new view.

He suggested that if we’re interested in helping with grid architecture we sign on to muStat.rockefeller.edu and join the grid, allowing our computers to work on the process when we’re not actually using them.

Audience question: could we develop a signature that would allow us to have selective vaccination?

Answer: It would be very very difficult because we would have to allow children to get sick.  It’s not entirely impossible but doubtful, when we have vaccines, that we would do it.

Speaker:

Alan Shuldiner

Dr. Shuldiner started with a major goal to have the right intervention for the right patient at the right time.  Out goal is to create medical interventions that are the 4ps: Predictive, personalized, preemptive and participatory.

This is a major undertaking as it is a long and hard road to go from discovery to clinical.

His specific topic was Pharmacogenomics.  Many people do not respond well to drugs, having adverse and unpredictable reactions.  He estimated that 100,000 die per year due to adverse reactions.  If we could pair the right drug with the right patient at the right time, we might be able to prevent that.  Pharmacogenomics is using genomes to evaluate drug efficacy for while it is not the only factor, it is very important.

The primary example he gave was clopidigril and aspirin.  Plavix (clopidigril) and aspirin are known to be effective for prevention of myocardial infarction and stroke and Plavix has had $4 million in sales last year.  But there’s a huge difference in response, with nearly a third of people taking it being resistant.

Shuldiner spoke about studying and coordinating with an Old Order Amish community in Lancaster for multi gene studies because it is a closed population with a homogeneity of lifestyle, people who are healthy and highly compliant with drug adherence, there is low drug usage, and they can work with family units to attempt to determine inheritance.  This allowed them to identify a gene where prasugrel (son of clopidigril) was a better treatment despite it’s higher bleeding risk and cost (as Plavix is about to be off patent).

 

The challenges that Dr. Shuldiner pointed to included

  • increased focus on EBM when it’s hard, and costly to get genomic trials done, leaving them with a lack of randomized trials for that evidence
  • health care provider education and expectations is not prepared for genomics
  • the logistics of genetic testing
  • need for reimbursement, which would need evidence, see bullet point 1
  • there are ethical and legal complications

However, he did not that patients do get the need for genomic testing and do ask for it when properly informed.

Overall, he noted that adoption in clinical is slow. In the future, he suggested that the FDA will have to mandate pharmacogenomic studies and that we’ll need to find a way to shorten the time that is currently spent getting from the raw research to bedside outcomes. This will be improved with funds coming for translational science, but w also may see genetic testing beginning to be required as a way to manage pharmaceutical costs, particularly for more costly drugs.

During the Q&A after the final speaker two points caught me:

International studies will help us gain enough numbers and information to make appropriate clinical judgments.

Studies being created and done today are no longer standing alone. Instead, they are part of a future meta-grouping.

Abigail’s Thoughts:

Personalized medicine has huge gene components.  We want risk stratification to be improved so that those furthest from the averages and at highest risk get appropriate intervention and that we’re not over-treating based on “mights.” We need to analyze not just single genes but interactions of genes to determine their relationship to diseases and conditions and grid computing will help.  While we will need to pull in multivariable statistics, we’ll have to be careful how many of them we assign.  Pharmacogenomics is working to determine what drugs will work in who m and when but we will likely need more government mandation and movement towards translational science and education to get it to the bedside and to find funding for the randomized trials and EBM that medical education is presently teaching.