Return to Discussion Forum

Title: Topics - Face-to-Face: Omics and Biomarkers and BROADER

On the bottom of this page, you will find the topic for discussion and the name of the contributor.

Please add comments and then click on the "Add comment" button.

%COMMENT{type="belowthreadmode"}%

MaryBanach - 16 Nov 2011 - 14:49

As the discussion proceeded there was a broadening in the topic to look at what students learn and apply from their biostatistics courses. Felicity Enders has already put out a request for help with a JSM roundtable on teaching statistics in the health sciences. It sounds like Madhu Mazumdar, Bob Oster and the Education Working Group, and Andy Cucchiera and the Online Resources Working Group are also interested in this topic.

MadhuMazumdar - 13 Nov 2011

Hi Everybody,

I use Dr. Windish's JAMA survey for our Intro to Biostat course for assessing the aspects students find the hardest to grasp. Our students struggle with 'confidence interval' the most. Using this tool 3/4 th way of the course helps us in modifying our review session. It would be of interest to see how many ways we can teach the concept of CI. Tx.

Madhu

FelicityEnders - 7 Nov 2011

Kyle, your idea sounds broad but quite meaningful. If we began this discussion at the face-to-face and determined which areas to pursue, we might also use conference call time to delve into some of these in greater depth.

Felicity

Kyle Rudser - 7 Nov 2011

Hello Chris and everyone,

I have been in the process of composing this message since the last BERD call, which is in a direction of what Felicity sent and goes further.

Some topics for discussion for the face-to-face meeting in the form of questions to ask ourselves as a group:

How can BERD best support and facilitate the overall mission of the CTSA consortium?

Are there aspects to include in our focus beyond discussion of statistical methods (loosely speaking: aspects beyond talking amongst ourselves)? - On the previous BERD KFC call there were (initial) reactions of shock when hearing what passed acceptance at an IRB, though many in turn have similar stories of their own. Do we have a role for support to IRBs (at least those directly related to studies conducted by members of the 60 institutions comprising the CTSA consortium.)? Would it be appropriate for us (BERD) to have a statement/guidelines for distribution to IRBs pertaining to the statistical analysis plan portion of their review, that it could be improved? - Follow-up comments on the call noted that the senior PI (presumably an MD) took a stance of complacency with the level of detail in the analysis plan and extent to which the plan was pre-specified. Do we have a role for making an effort towards changing the perception of clinician investigators, both the upcoming future investigators and their mentors who are the current senior clinical investigators?

There was a suggestion to have a panel of CTSA PIs come and tell us what they need from us. Would it be useful to in addition, at some point in the future, communicate to the PIs what we feel are top priority action items to help facilitate and support the bench to bedside consortium endeavor in an efficient manner that preserves quality (e.g., that we distribute such across the CTSAs in the form of a white paper or something).

If we do feel compelled to communicate outside our KFC, what should we say and to whom? Perhaps this would involve the Education and Career Development KFC and/or Comparative Effectiveness Research (CER) KFC and the "Education/Training/Workforce Development Workgroup" therein. One goal would be to not have them feel we are outside our purview and on their "toes"/"turf". There are other efforts that might parallel this: e.g., Association for Clinical Research Training (ACRT).

Best regards, Kyle

FelicityEnders - 7 Nov 2011

love this idea - the robust discussions we have been having really show that there is a need for educating our clinical colleagues about how to interpret our product. I sometimes wonder if we spend too much time teaching researchers and clinicians how to do a t-test in every piece of software imaginable, but fail to challenge them to think about how to make sense of what the results mean. I can't count the number of times I have had PIs show up in my office with reams of statistical output asking me to point them to the p-value that counts. In essence, we focus too much time on doing rather than thinking.

The planned breakouts on education should be a great place to discuss causes, consequences and solutions. Do you have suggestions on a focus if this were to form a theme in a didactic or the keynote address?

Chris

FelicityEnders - 7 Nov 2011

Chris,

I propose something a bit broader - what do students actually learn in our introductory courses? Do they really understand confounding? Effect modification? Can they apply what they've learned to decide which statistical methods are appropriate or inappropriate? I have preliminary data on this for both a first course and a course on linear regression, and I suspect many others have thoughts as well. We could also draw on the work of Donna Windish (JAMA 2007). The overall goals would be to 1) increase attention on likely problem areas for our students and 2) work as a group to consider potential solutions.

thanks, Felicity

FrankHarrell - 6 Nov 2011

Knut I like your emphasis on analytical strategy. Too many researchers think that screening genes for "winners" (as opposed to high-dimensional regression) is the way to go, and too many researchers ignore exceptionally high false negative rates.

Frank

Hongtu Zhu - 16 Nov 2011

I totally agree that there are tons of issues in many neuroimaging studies.

If you could read many data analysis papers published in neuroimage, HBM, nature neuroscience, etc., many data analysis methods used in these papers are not correct and full of errors.

unfortunately, many people just ignore them even though they know the problems.

best wishes hongtu

LauraLeeJohnson - 6 Nov 2011

fMRI data, while not \x91omics\x92, has many similar issues.

Laura Lee

KnutWittkowski - 6 Nov 2011

Chris,

I like your focus on omics. The recent events have demonstrated the value of statisticians in identifying misuse of software tools, but we should also be more proactive in getting involved in the design and analysis process much earlier, so that we can be a resource, rather than merely watchdogs.

IMHO, the recent controversy has highlighted two questions:

- can we really identify the risk factors of common, complex diseases based on p-values computed one SNP at a time? And if not, how do we address epistasis between SNPs, intragenic regions, and genes?

- can we really do "confirm" the results from screening thousands of SNPs or genes? Is it bad to have reduced the number of candidate genes for a particular risk (assuming there is only one) from 30,000 to 10 (false positive "rate" = .90)?

As BERD members, we could - and should - educate ourselves and others about the concepts underlying the statistical approaches being available to omics research, to avoid the widespread misuse of statistical methods (and other tools).

I'd be happy both to contribute to the educational process and to share the computational statistics methodology and tools developed under the CTSA.

Knut Sent via BlackBerry by AT&T

This topic: CTSpedia > WebHome > DiscussionForum > DiscussionBERD004
Topic revision: 09 May 2013, MaryBanach

---