Further reflections on genomic screening

Scales_of_justiceNow the dust has settled, I’ve been reflecting on the controversial recommendation from the American College of Medical Genetics and Genomics (ACMG) that all clinical genomes should be screened for a specific set of conditions. Following the release of the guidelines, the European Society of Human Genetics (ESHG) published its more conservative recommendations, and vigorous debate has continued internationally regarding the wisdom of introducing genomic screening. While I still have some major reservations about the policy (outlined in previous posts), upon reflection there are certainly things some aspects that make a lot of sense…

Logistics of clinical feedback

From a logistical perspective, screening genomes for actionable variants within a clinical setting makes much more sense than in research. Feeding back any kind of individual result from a genome sequence to a patient has a number of logistical requirements, all of which already exist in a clinical setting: the diagnostic laboratory must be accredited; there is robust route for passing information about individual patients from laboratory scientists to clinicians; the clinician already has a duty-of-care relationship with the patient and so is able to provide appropriate counselling about the results; and a method for referring the patient to other clinical specialists already exists. Most research studies lack all of these elements, and creating them would require substantial investment of time, people and money.

Ensuring consistent service

Providing a shortlist of genes to be screened minimises the number of variants under consideration, and ensures a consistent service across different sites. Although one can debate the exact content of the gene list suggested by the ACMG ad infinitum, the principle of a minimal list of clinically actionable genes/conditions is a good one. It puts a stop to endless concerns that health care providers will be overwhelmed with a data tsunami, and limits the conditions to ones that might offer direct benefit through preventative actions. The existence of an approved list also ensures that a patient would receive the same information regardless of where they are seen (which contrasts starkly with the implementation of array-CGH, where a lack of consensus guidelines has led to a mixture of policies amongst different labs and clinical teams with regards to how incidental findings are handled). The current list also provides a starting point – a line in the sand – that people can now work on and improve.

Benefits, harms and personal biases

While much of the debate around whether to offer opportunistic genomic screening relates to the evidence base (which most people agree is currently woefully inadequate ), I think it also reveals different underlying personal and cultural biases. Whatever we do, there will be mistakes. The question is therefore what type of mistake I, we, you, or any individual would prefer to make.

If we don’t screen genomes for clinically actionable variants, some individuals will be harmed by diseases that could have been prevented; on the other hand, if we do screen genomes, some individuals will be harmed by treatment for diseases that would never have developed. This seems like a variation on theme of the classic ethical conundrum: is it better to make errors of omission (failing to do something when you should) or errors of commission (doing something when you shouldn’t)? Personally I err on the side of caution and, in the absence of strong evidence of benefit, prefer not to harm healthy individuals through incorrect predictions at the expense of failing to help individuals who will develop disease. Understandably, others prefer to take action to prevent disease where possible, albeit at the expense of harming a few healthy folk. Without detailed evidence from population studies to enable us to weigh the harms and benefits of each course of action, we are left with personal preference between these two different types of errors.

Perhaps the greatest achievement of the ACMG recommendations is to have raised the level of debate about ‘incidental findings’ in genomics, and to inspire and facilitate the creation of an evidence-base to aid future decision-making.

  • Digg
  • StumbleUpon
  • del.icio.us
  • Facebook
  • Twitter
  • Google Bookmarks
  • FriendFeed
  • Reddit

12 Responses to “Further reflections on genomic screening”


  • Here we go again do a limited number of people have the right to stop me as an individual to have my personal property withheld form me? It is my body you are taking information from and I have the right to know. I also have the right to NOT know. But it is not up to you as a scientist, teckie, or as a Doctor. I have personal knowledge of what this type of thing does. My mother was about age 42 and had lost children several times and her son had died at age 3 not of a thing medical help could have changed in the early 1930′s. She became pregnant again and her Doc had either retired or died. Her new Doc ask her why she was pregnant again with this tumor? What tumor? My dingy father and the Doc had made the decision that a woman did not need to know about her body. So she lay in bed with me from a benign tumor which caused her to bleed if she even got up. 7 months she lay there at age 42 till I was born. She was ask to wait for 6 months to gain strength to have the surgery. This was wrong and no one had the right to with hold information from her if she wanted it. Not even the husband has that right.

  • Just with regards this:

    If we don’t screen genomes for clinically actionable variants, some individuals will be harmed by diseases that could have been prevented; on the other hand, if we do screen genomes, some individuals will be harmed by treatment for diseases that would never have developed.

    your excellent article’s title gives the hint that there is somewhere to look for guidance – the literature of disease screening. See e.g. “Disease qualification” in:

    http://en.wikipedia.org/wiki/Newborn_screening

  • First, I would like to thank Caroline for another lucid and balanced piece of writing. Count me as a fan.

    Second, just a comment that the clinical questions vary according to the clinical setting. If testing is conducted in a highly personalized setting, the opportunities for choice are greater. If it is done in a public health scenario, like newborn screening is currently done, standardization becomes necessary. Most will occur in the broad grey zone between, and some constraints will be necessary unless the payor (the person being tested or her insurer) agrees to the added expense of adding a large helping of personal utility to the validated clinical utility of genomic sequencing. No wonder there is more and more interest in sidestepping the mess by focusing on disease or phenotype-defined NGS panels.

  • I also agree with the err on the caution side. However, if people want to know, they should know. I can agree with that. The one thing I would say is that in the absence of strong evidence of benefit, the test and any resulting treatment could be out of pocket. Either that or the payer (whoever that is) should [co]pay while collecting evidence and reevaluate its policy at a fixed date. Regardless, the issue seems to be about the strength and quality of evidence. The communication of this to both patients and providers is paramount.

  • Caroline I will raise one counterpoint as food for thought.

    You say “Although one can debate the exact content of the gene list suggested by the ACMG ad infinitum, the principle of a minimal list of clinically actionable genes/conditions is a good one. It puts a stop to endless concerns that health care providers will be overwhelmed with a data tsunami, and limits the conditions to ones that might offer direct benefit through preventative actions.”

    I would posit that it is less than clear that the minimum list, with no maximum, or ceiling, puts a stop to concerns about data overload or (as you say later) that a patient will get consistent information regardless of where they are tested.

    Cystic Fibrosis testing in the US provides an example. The ACMG issued very specific recommendations about how CF testing should be carried out in 2001, and revised them in 2004. They outlined a specific set of 23 common, well understood variants that should be tested and specifically stated that other rare, less well understood variants should not be assayed (unless they were more common in the patient population in which you were conducting screening).

    However, a 2008 survey of US laboratory directors found that 10% reported on a 23-mutation panel; for a host of reasons the remainder tested for more variants. At that time, 4% were coding the entire region. ACMG’s explicit limited mutation list (never mind a gene list) did not contain the number of variants that were tested, and did not result in uniform practice. I’m not sure we can expect a minimum list of genes to contain or standardze the information that laboratories decide to send back.

  • oops I meant “sequencing the entire coding region”

  • Bob Cook-Deegan

    Caroline,

    You have a gift for clarity and spare prose. Wonderful post.

    Re Dave Kaufman’s CF example. The question there is whether the broader testing after ACMG recommenations is better or worse than abiding by the recommendations. It’s the debate we’ll have over and over for the much more numerous variations that will come to light via all-exome or whole-genome analysis. Dave’s surely right that having a list won’t resolve the debate, but surely we can also agree that it’s better than not having benchmarks at all.

    BCD

  • Caroline Wright

    Thanks for your comments everyone.

    @ Dave and Bob – Interesting point about people not adhering to standards even when they exist. Surely it is still better than having no agreed standards though? I think there are two other major problems with having a list: (1) agreeing what goes on it (big topic, perhaps for another post!) and (2) managing changes. If an old gene has been removed in the next version of the list, what does this mean for patients who previously tested positive? If a new gene is added, do labs and clinicians have a duty to check former patients for variants in this gene? This can’t be a new problem though.

    @ Q Wilkinson – I don’t think with-holding information from people is really the issue. I have no problem with making people’s genomes available to them if they want them. But I would argue that this will be of very little use to most people. The question surely is whether we have a duty to actively search for disease predispositions in someone’s genome, which they might never think to ask about, and then (rightly or wrongly) act on that information.

  • I ran my genome (PGP89) through this list as an exercise. I found that restricting the reported alleles to “known pathogenic” seems too conservative, but that it is difficult to know where to draw the line on “expected pathogenic.” I ended up using my current health status as a healthy 58-year-old male to reassure myself that my variant alleles weren’t pathogenic. I think that this shows that we have some ground to cover before this can be seamlessly integrated into clinical practice. It’s not a bad start, though. Please see:

    http://www.openhumangenome.com/2013/05/16/a-whole-genome-checkup

  • Caroline, enjoyed your calm and very thoughtful post… I especially appreciate the distinction between clinical and research (which has been sometimes overlooked in the debate) and the idea of balancing errors of commission/omission – perhaps this would have been a good way to frame this when we wrote the recommendations.

  • Les Biesecker

    Caroline,
    You provide a thoughtful and coherent synopsis of our wonderful discussions in Cambridge this summer – thank you so much. I also appreciate some of the comments that have been made in response.
    I am sympathetic to the arms race of more and more results alluded to by my friend Dave. This is a question we have to think hard about. In some ways, the ACMG did the easy work – skimming off the worst and most actionable disorders. If there was doubt about a disorder, it was left off, but others could make good arguments that the cutoff should be in a different place than where it was made. I think we need a process like newborn screening where expert review of available data are used to make a sensible cutoff.
    But can there be a single cutoff or should the cutoff be set for each patient? As I often joke, it would be unfortunate if we created a system of one size fits all personalized medicine. A really challenging issue that we need to have a lot of discussion about.
    But to address the comments of Mr/Ms Wilkinson s/he raises a tough issue – that of freeing all information and having the patient decide. No one can defend the behavior of the family and the doc in the story as s/he wrote it. But does this mean that the patient should get all of their sequence data and do with it as they please? I have a hard time with that because we share a health care system (more so in the UK than in the US, which is not much of a system at all) and if we have a patient-directed free for all, then the patient may be concerned about a variant for which we have no data, and insisting upon a CAT scan to rule out something that a thoughtful and knowledgeable clinician would rightly assert is a total waste of time – that in fact the CAT scan was more likely to cause cancer than is the variant the patient is concerned about. But I don’t pretend to have an answer for that. We all need to work hard to find a balance in how these decisions can be made that is medically and economically sensible and preserves the core values of patient autonomy. This will not be easy.

  • Caroline, late to the party commenting on your thought-provoking piece

    “Perhaps the greatest achievement of the ACMG recommendations is to have raised the level of debate about ‘incidental findings’ in genomics, and to inspire and facilitate the creation of an evidence-base to aid future decision-making”

    On this point it often feels that incidental findings in genetic testing are considered to be an entirely new beast on the block, whilst in reality they have clearly existed since Messrs Tijo and Levan discovered that number 46 may be of some significance, over 50 years ago.
    However, the resultant data explosion from NGS has magnified the issue and will likely catalyze revised and evidence-based guidelines across the board in diagnostic centres across the world; as you also point incidental findings are often not addressed in existing guidelines for array-CGH and many other lower resolution technologies.

Comments are currently closed.

Page optimized by WP Minify WordPress Plugin