Present: Abby
Crocker, Amanda Kennedy, Rodger Kessler, Ben Littenberg, Connie van Eeghen
1.
Start Up:
a. What
are the predictors of successful collaboration within a practice? It’s more than excitement and clearly
includes a champion. To be continued…
b. New
CTS applicant: surgical resident (Charles Parsons) to start today; will attend
the Design class starting this afternoon
c. Questions
about opportunities from COM:
i.
Laurie St. Gelais: yes, a good opportunity and further
discussion is warranted but I deleted the email and can’t for the life of me
remember what it is about. Vacation
brain syndrome… sorry! Would someone
please come to my rescue?
ii.
Hilda Alajajian: Emerging Researcher NIH training opportunity
to get an insider’s view of NIH study
section in the Early Career Review Program.
Can be done through training on website.
iii.
Ruth Farrell: HHS has new conflict of interest
regulations. Information sessions are on
Monday, July 9th, 3:30 p.m., Carpenter Auditorium and Wednesday, August 1st,
9:00 am - Carpenter Auditorium. Go if
you want to go.
2.
Presentation:
Abby is working on research articles this summer, one in particular, including
the use of STATA. She is looking ahead
at her plan for the next few weeks of work.
Goal for today: skeleton of the paper and what to move forward without a
complete analysis.
a. Check
out the “Report in Clinical Usage” book
b. Goal:
understand the predictors for neonatal abstinence syndrome that indicate follow
up. Essentially: build a model
c. Target
publication: Pediatrics. Use the instructions
on this web-site to draft the article.
d. Title:
A Predictive Model for Neonatal Abstinence Syndrome
e. Background:
The 50,000 foot hook (with a number) of why important. What NAS is and how identified. Why the current identification method is sub-optimal. The need for better prediction. “Therefore, we sought to… “ or “Therefore,
the purpose of the study is to…” Shoot
for three paragraphs.
f. Methods:
Description of design, with statement of IRB approval. (This is the place to
start, in writing the article)
i.
This is a secondary analysis of an existing data
set.
ii.
Describe the data set as a prospective cohort of women,
etc. “To develop our model, we took
advantage of an existing data set of pregnant women created for the purpose of… The data were collected through the ICON
clinical registry that is a QA registry that includes (# live births for #
delivering women)” “Includes all
deliveries between 2005 - 2012 of opiate exposed women at FAHC…” “Prospectively collected, with variables
including social data about mother, clinical data about pregnancy, delivery,
and immediate post-partum period…”
iii.
Subject selected (everybody?) with inclusion/exclusion,
e.g. mothers not in treatment for substance abuse. Refer to 1st table here? Decide later.
“All women in database were potentially included, but we excluded those
that (died, didn’t have a chance to develop NAS)” IRB statement follows this description.
iv.
Variables
1. Outcomes
2. Predictors
(there are many; may need to use a table).
Important to identify the starting point of the study: what did you
consider.
a. Definitions:
what is substance abuse? Defined by
opiate dose? Use morphine
equivalence? Ask Amanda.
v.
Analytic approach.
1. Descriptive
statistics: how narcotics were categorized.
Use of means and standard deviations for total population of eligible
women for this study (770); medians and quartiles; proportions... Box plots…
Use of morphine equivalence can also go here.
2. Univariate
analysis: Predicting a binary outcome (with and w/out NAS): continuous (t test
if it’s normally distributed or use rank based test if not normally
distributed) and dichotomous (Chi square, exact test…) Each predictor was analyzed separately with
respect to the outcome.
a. Purpose:
test each predictor vs. NAS
b. Method:
x methods for this analysis
c. Table
1: variables by total with P value in last column
vi.
Model building process: P value <=0.10. Binary outcomes can be predicted by one of
the following methods: logistic regression, recursive partitioning, and neural
networks.
1. Logistical
Regression used when all sub-groups are equally predictive, i.e. age of mother
is as important a predictor in male vs. female babies.
2. Ben
strongly recommends one of the others.
To be continued…
g. Results
(the rest of the outline will be continued at a later date)
h. Discussion
& Limitations
i.
Conclusion
j.
Next step:
i.
Get organized around the model building process. Meeting with Ben tomorrow.
1. How
many variables can this study tolerate before “over-fitting” the model? There are 770 subjects in the data base. About 15.
(Over-fitting: fitting random noise into the model. There are ways to minimize this noise, but
they have trade-offs.)
ii.
Abby will get editing support from Amanda.
3.
Workshop
Goals for 2012:
a. Journal
club: identify UVM guests and articles; invite to CROW ahead of time
b. Research
updates: share work-in-process
a.
July 12: Rodger – outline of new research paper (Connie
by phone, no Abby or Kairn)
b.
July 19: (no Abby or Ben)
c.
July 26: (no Ben)
d.
Aug 2: Abby – “How we picked the predictive model for
the NAS article”
e.
Aug 9: (no Abby)
f.
Aug 16: (no Abby)
g.
Aug 23:
h.
Aug 30: (new schedule?)
i.
Future agenda to consider:
i.
Ben: budgeting exercise for grant applications
ii.
Journal Club: “Methods and metrics challenges of
delivery-system research,” Alexander and Hearld, March 2012 (for later in the
year?)
iii.
Rodger: Mixed methods article; article on Behavior’s
Influence on Medical Conditions (unpublished); drug company funding. Also: discuss design for PCBH clinical and
cost research.
iv.
Amanda: presentation and interpretation of data in
articles
v.
Sharon Henry: article by Cleland, Thoracic Spine
Manipulation, Physical Therapy 2007
Posts
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.