statistical cut point

[Swank, Paul R]

I have been asked to suggest a statistical solution for a problem. The problem is that we are taking ER patients with a mild traumatic brain injury for a research study. Someone has raised an issue about patients who may be intoxicated upon admission. One hospital says they routinely do alcohol and drug tests but the other does not. It has been suggested that we use clinical judgment and then do a follow-up screen after we get consent. Then if we find that a "significant" number of patients are getting recruited who were intoxicated (say 10% or more of the first 20 patients) then we rethink this. Someone else suggested the statistician (that's me) come up with a statistical solution. I know that the confidence intervals for 2 out of 20 (10%) are pretty wide, like 1%-30%, so this doesn't seem like a good statistical solution. I think it should be clinical decision but I am open to a reasonable statistical solution if someone has one.

Paul

Dr. Paul R. Swank,
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center-Houston

[Steve Simon, P.Mean Consulting]

Swank, Paul R wrote:

> I have been asked to suggest a statistical solution for a problem.
> The problem is that we are taking ER patients with a mild traumatic
> brain injury for a research study. Someone has raised an issue about
> patients who may be intoxicated upon admission. One hospital says
> they routinely do alcohol and drug tests but the other does not. It
> has been suggested that we use clinical judgment and then do a
> follow-up screen after we get consent. Then if we find that a
> "significant" number of patients are getting recruited who were
> intoxicated (say 10% or more of the first 20 patients) then we
> rethink this. Someone else suggested the statistician (that's me)
> come up with a statistical solution. I know that the confidence
> intervals for 2 out of 20 (10%) are pretty wide, like 1%-30%, so this
> doesn't seem like a good statistical solution. I think it should be
> clinical decision but I am open to a reasonable statistical solution
> if someone has one.

How much does it cost you when an intoxicated patient gets past the
first screen? When that cost times the probability of occurrence times
the total number of patients reaches a pre-defined intolerable level,
then you should re-evaluate. Get the researchers to define cost per
patient and what overall cost is intolerable and then back-calculate to
estimate the probability of occurence. When you have convincing evidence
that the proportion intoxicated is above this value, then re-evaluate.
This could be done with a Bayesian informative prior to incorporate what
people already know about the issue.

You might factor in the costs associated with the re-evaluation as well.

I hope this makes sense.
--
Steve Simon, Standard Disclaimer
The Monthly Mean is celebrating its first anniversary.
Find out more about the newsletter that dares
to call itself "average" at www.pmean.com/news

[Swank, Paul R]

Once the subject is included in the study (ie., consented) then it is somewhat difficult to exclude them. The problem is two-fold. What impact does intoxication have on the tests that are planned and to what extent is the generalizability of the results (to a military population) compromised by inclusion of such patients. After a long discussion at out meeting today, it was decided to exclude anyone who had documented (in the medical chart) intoxication or a blood alcohol level of .08 or higher. We would also exclude anyone with a history of chronic drug/alcohol use. Then we give them a breathalyzer at baseline (an hour or two after recruitment) and record this data as a possible covariate and/or moderator. Not perfect, but at least we won't have to change our recruitment strategy after the first 20 cases!

Thanks for your questions! Unfortunately, I know nothing about Bayesian methods and I am too old and close to retirement to want to retool now!

Paul

Dr. Paul R. Swank,
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center-Houston