Tuesday, May 28, 2013

Clinical Research Oriented Workshop (CROW) Meeting: May 23, 2013

Present:  Abby Crocker, Kairn Kelley, Rodger Kessler, Ben Littenberg, Connie van Eeghen
1.                  Start Up: Ben’s dad is turning 89 – hooray!

2.                  Journal Article Review Kairn: Discussion of binomial model in 1978 journal article by Thornton Raffin “Speech Discrim Scores Modeled As Binomial Variable”
a.       Article focuses on a specific test: W22; Kairn uses a similar test (single syllable words): NU6
b.      One part of the article is focused on adjusting the confidence limits of this use of a binomial distribution.  There are several ways to do this transformation; the various methods aren’t in complete agreement with each other but they’re close.
c.       The article does not use a standard binomial distribution; it uses a transformation to account for distributions that can’t be normally shaped because their means are close to extreme values (i.e. close to 0 or 100).  Per Abby, the appendix does a good job illustrating how the transformation variable theta is derived. 
d.      These adjusted confidence intervals estimate a range of uncertainty but don’t indicate whether 2 scores are different.  To figure this out, we need to know the proportions of the two scores (# right/total), the difference between them, and the standard deviation of the difference.  Abby explained this as the combined, pooled estimate, which is described here as the square root of the sum of each “pooled” proportion complement, each divided by sample size.  Or, SQRT((PQ/N1)+(PQ/N2)) as an Excel formula. 
e.       We tested this out manually and then validated with a chi square and Fisher’s Exact on Stata.  It worked!  The point of the article, though, was that the binomial distribution needs to be adjusted for test results at the extreme (when a tester gets all or none of the responses right).  In these cases, the distribution is skewed because the number of wrong responses can’t be less than 0 or greater or 100.
f.       Abby went on to show us the shape of the binomial distribution and how it changes using our proportion estimates, mirroring the distribution we had been working on.  The distribution will look like a normal model for probabilities that are not close to the extreme ends of the scale. 
g.      Abby pointed out that when the small proportions are very small (i.e. the denominator is very large or incalculable), the preference is to use the Poisson distribution, rather than a very skewed binomial distribution. 
h.      The reason this discussion is important is because many audiologists don’t have a strong understanding of probability and it could be possible to interpret changes in test scores incorrectly.  This work can both help develop strong research methods and help develop a method for communicating the results clearly to audiologists.  Use the chi square for the difference between two results and use a transformation (a table or sin arc formula) to the normal model to create confidence intervals.

3.                  Next Workshop Meeting(s): Thursday, 2:00 p.m. – 3:30 p.m., at Given Courtyard South Level 4. 
a.       Thursday, May 30:  (no Abby) LAST THURSDAY MEETING Ben: Journal article by Feldman, 2013, “Impact of Providing Fee Data on Laboratory Test Ordering, JAMA
b.      Wednesday, June 5: NOTE: New summer schedule will start: Wednesdays, 11:30 – 1:00.  Abby: Journal article (no Connie)
c.       Wednesday, June 12: Marianne: Feedback on ideas for web site that will help CTS students (faculty, and fellows too)  find/know /access, evaluate/apply content and literature they need for courses and research
d.      Thursday, June 13: 2:00 – 3:30 Wilson Pace, Director & Dave West from the largest practice based research networks from American Academy of Family Physicians; access to Medicaid data base.  Kairn, Abby, Charlie can all make it. 
e.       Wednesday, June 19:
f.       June 26:
g.      July 3:
h.      July 10: Marianne: Review of literature review
i.        July 17:
j.        July 24:
k.      July 31:
l.        August 7
m.    August 14
n.      August 21
o.      August 28
p.      Future agenda to consider:
                                                  i.      Abby and Charlie: data analysis of Exploration of analytical plan for Natural History of Acute Opioid Use
                                                ii.      Peter Callas or other faculty on multi-level modeling
                                              iii.      Charlie MacLean: demonstration of Tableau; or Rodger’s examples of Prezi
                                              iv.      Journal article: Gomes, 2013, Opioid Dose and MVA in Canada (Charlie)
                                                v.      Ben: Tukey chapter reading assignments, or other book of general interest
                                              vi.      Summer plan: each week, one person will send out an article or prezi ahead for review or discussion by all.  Alternatively, if a participant is working on a key document for their professional development, this is also welcome (e.g. K awards, F awards, etc.)

Recorder: Connie van Eeghen

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.