[S] Modeling Health Service Rates with Poisson GLM

Andy White @ STATS (awhite@hmsa.com)
Mon, 3 Aug 1998 12:39:32 -0700

I would appreciate any comments/suggestions on appropriate modeling
approach(es) for explaining health service utilization rates (e.g.,
emergency room visits or hospital admissions per 1000 population per year)
as a function of multiple categorical variables: age level, sex, morbidity
level, and health plan type.

Since our dependent variable is number of occurrences, we figure a Poisson
Regression is best. Since we have occurrences as a function of "population
exposure" we include an offset (i.e., log(population) ) to provide the
denominator to our service utilization rate. We have factors which we start
off in an additive model: sex + agegroup + morbidity_level + plan_type. We
have 144 categories (6x2x6x2 as above) and pop counts for each as well as ER
visit counts for each.

But here's the rub. ER visits do not happen just once per patient (unless I
change the nature of the dependent variable to "the number of patients with
One Or More ER visits"). For example, an asthmatic can have multiple ER
visits, and we believe that the sicker they are the more visits they are
likely to have, so we have a dependent count situation which is not strictly
Poisson. I guess you could call this context repeatable-event situations
(vs repeated events).

What to do? How should the Poisson Regression model be modified? (We are
NOT interested in time between successive ER events so I do not think event
history or survival analysis modeling is appropriate.) One source suggests
that repeatable-event situations of this type imply "overdispersed" Poisson
distributions. If so, then would quasi-likelihood estimation of a Poisson
model handle the context?

Has anyone else out there had experience with modeling of health services
utilization, or with similar "rate" or proportion contexts?

Sincerely, Andy White

Andrew N. White, Ph.D., Research Manager
Statistics Department
Hawaii Medical Service Association
818 Keeaumoku Street
Honolulu Hawaii 96814
ph 808-948-5344 / fax 808-948-5063
email awhite@hmsa.com

This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news