Discussion: View Thread

I'd appreciate suggestions /recommendations and -citations to publications- for definition of baseline measurement in a "two group parallel design" randomized clinical trial.

Some background: In the Pharma industry statistical analyses of clinical trials (parallel design) , "baseline" for nearly every  statistical analyses is typically defined as "the last available observation before patient was administered study drug".

This applies to all variables, such as age, gender or "serum creatinine", etc.

 A toy example. - say for a hypothetical patient I have serum creatinine measured on Jan 1, Jan 2 and Jan 3, Jan 10 etc.- and the patient was administered study drug  on Jan 5. Then "baseline" is the January 3 value.

 Interesting exceptions are for cross-over /Latin square designs, e.g. such as Thorough QT (TQT) designs, where a patient can have a new baseline before each treatment/period.

In my clinical trial, ( a double blind randomized, paralllel 2 group design) there are pre-study-drug daily measurements of a "PRO" (patient reported outcome) variable for up to 14 days prior to administration of study drug. The vast majority of patients have 14 days of measurements pre study drug.  

The measurements of the PRO pre-drug vary from from  to day, they are "noisy" and in some patients vary dramatically from day to day.  I am not at all surprised to see the day to day variation in the PRO.

I think defining "baseline" for the PRO, as say, an average or median over 2-3 days immediately preceding study drug may  be a more appropriate alternative than the "single last  observation before baseline" - as a way to reduce the potential "noise" in the baseline value.

comments/recommendations and especially citations, appreciated.

thanks in advance



-------------------------------------------
Chris Barker, Ph.D.
Consultant and
Adjunct Associate Professor of Biostatistics
www,barkerstats.com

---
"In composition you have all the time you want to decide what to say in 15 seconds, in improvisation you have 15 seconds."
-Steve Lacy
-------------------------------------------

Chris,
You could also consider more than one "baseline" values to be included in the analyses.
See Bristol (2007). The Choice of Two Baselines. Drug Information Journal, 41(1), 57-61.
We had a pre-randomization treatment period for an enrichment design.
The pre-treatment and pre-randomization values were both considered as covariates.
David

-------------------------------------------
David Bristol
Statistical Consulting Services
-------------------------------------------

Chris:

Averaging makes sense. It might be helpful to examine the correlation between each of ave(2), ave(3), ... ave(14) and the endpoint to see how taking more and more pretrt obs in the average affects the correlation with endpoint.

Roy
-------------------------------------------
Roy Tamura
Associate Professor
University of South Florida
-------------------------------------------

Chris,
I agree with Roy.
As you will see in my paper, the correlations play an important role.
David

-------------------------------------------
David Bristol
Statistical Consulting Services
-------------------------------------------







Original Message:
Sent: 06-13-2014 18:15
From: Roy Tamura
Subject: Definition(s) of "Baseline" - general recommendations/suggestions


Chris:

Averaging makes sense. It might be helpful to examine the correlation between each of ave(2), ave(3), ... ave(14) and the endpoint to see how taking more and more pretrt obs in the average affects the correlation with endpoint.

Roy
-------------------------------------------
Roy Tamura
Associate Professor
University of South Florida
-------------------------------------------

Hi Chris,

In addition to the great advice that you have received, I can add that it is common to average the baseline period depending on the clinical outcome assessment (e.g., PRO) and therapeutic area.  You can declare that the baseline value will be the average of the most recent X days that are available prior to randomization including an algorithm for skipping over missing or disqualified days.  The length of the baseline period depends on the nature of the underlying construct and assumes that disease severity is stable or waxes and wanes around a particular value.  If you believe that the underlying construct is changing then you should select a shorter baseline period but 5-10 days is not unusual.  For example, if you are trying to assess severity during the onset of an event (i.e., bipolar depression or COPD) then the period would be short but the baseline period for chronic disorders can be counted in weeks (e.g., insomnia or IBS).  [Also, an empirical approach using historical data, as suggested, should be very helpful in optimizing your definition in a representative sample.]  If your client has access to the MAPI PROLabels database then you can search for approvals in your therapeutic area, and look-up the medical and statistical reviews on the FDA website (below) which will contain details regarding the baseline period and analysis methods.

Best regards,
David

http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cfm