Comparing statistical methods

Edward Hoffer ehoffer at GMAIL.COM
Fri Aug 10 01:39:48 UTC 2018


One of the biggest problems with our focus on p values is that
"statistically significant" may be clinically meaningless.  If you have a
large enough sample size, extending life by a month with treatment X
compared to Y - at a cost of $500,000 and horrible GI side effects - may be
trumpeted as a "significant prolongation of life.
Edward Hoffer MD.

On Thu, Aug 9, 2018 at 1:17 PM, Peggy Zuckerman <peggyzuckerman at gmail.com>
wrote:

> As a patient advocate who wants to provide meaningful education for other
> cancer patients, this issue is very important to me. A real challenge is
> that the use of the stats that come from a trial rarely give enough
> information so that a comparison of treatments can be made.  This is true
> whether it is in regard medications or timing of surgery, sequencing of
> medication.  One trial may include patients who have had multiple previous
> treatments, but not show the length of response those patients received.
> The patients who had some success with Medicine A, now being tested with
> Medicine NEW, may respond differently than those who had no meaningful
> response with A.  One could ask whether Medicine A acts as a primer for
> greater response with Medicine NEW.  If the patients in the studies have
> metastases than emerged within months past a 'curative' surgery, can those
> patients truly similar to those whose mets did not emerge until 3-4 years
> later.
>
> The older trials often accepted patients only on the basis of the landing
> site of the primary tumor, without regard for the pathology of that tumor.
> Newer trials fail to differentiate between patients with or without certain
> molecular characteristics, or choose an inappropriate level of that
> measurement to group patients.
>
> And with the understanding that cancers can be subdivided into more
> subtypes than previously recognized, some caused by hereditary tendencies,
> others by exposure to heavy metals, other just due to aging, we need to
> recategorize trial data accordingly.  Of course, that raises other issue
> which aggravate the statisticians, mainly the size of the groups needed for
> comparison.  When only 30% of patients in a study respond to a medication,
> naturally the question must be raised as to why those 30% did respond.
> Every study needs a way to characterize that subpopulation, as the failure
> to do so has cost us effective treatments and billions of dollars.
>
> Peggy Zuckerman
>
> Peggy Zuckerman
> www.peggyRCC.com
>
> On Thu, Aug 9, 2018 at 8:37 AM, Harold Lehmann <lehmann at jhmi.edu> wrote:
>
>> I am on a kick these days about causal reasoning (cf Judea Pearl, The
>> Book of Why). He points out that there are three layers of increasing
>> abstraction: statistical, causal, and contrafactual ("what if"). (The
>> anti-Bayesians on this list will be happy to know that Bayesian statistics
>> is relegated to the lowest level, although suited to represent the other
>> layers, if directed by them.)
>>
>> It seems to me that decision making is all about "what if," and
>> therefore, statistics (as traditionally conceived) are inadequate.
>>
>> Harold
>>
>>
>> On Aug 9, 2018, at 9:40 AM, Elias Peter <pheski69 at GMAIL.COM> wrote:
>>
>> I was directed by my son (math person, statistician, and professional
>> data manager)to  a post in a forum where data/stats people discuss methods.
>> It asks for information about *outcome data* comparing different
>> statistical models. It might be interesting for this group to follow this.
>> Or perhaps contribute. I have not seen any answers yet. Here is the flavor
>> of the question:
>>
>> "The overarching goal of statistics is to make decisions in the face of
>> uncertainty.” …list of statistical approaches…  "What seems to be
>> missing is head-to-head comparisons of approaches to see which ones
>> optimize utility/loss/cost functions where such functions reflect real,
>> concrete goals.” …  "Does anyone know of comparative studies that inform
>> us of the value of two or more statistical approaches when the goal is
>> making the best decisions?"
>>
>>
>> https://discourse.datamethods.org/t/choosing-statistical-par
>> adigms-by-studying-the-quality-of-decisions-to-which-they-
>> lead/372?u=f2harrell
>>
>> Peter Elias
>>
>>
>>
>> ------------------------------
>>
>>
>> To unsubscribe from IMPROVEDX: click the following link:
>> http://list.improvediagnosis.org/scripts/wa-IMPDIAG.exe?SUBE
>> D1=IMPROVEDX&A=1
>>
>> or send email to: IMPROVEDX-SIGNOFF-REQUEST at LIST.IMPROVEDIAGNOSIS.ORG
>>
>>
>>
>> Moderator:David Meyers, Board Member, Society for Improving Diagnosis in
>> Medicine
>>
>> To learn more about SIDM visit:
>> http://www.improvediagnosis.org/
>>
>>
>>
>> ------------------------------
>>
>>
>> To unsubscribe from IMPROVEDX: click the following link:
>> http://list.improvediagnosis.org/scripts/wa-IMPDIAG.exe?SUBE
>> D1=IMPROVEDX&A=1 or send email to: IMPROVEDX-SIGNOFF-REQUEST at LIST
>> .IMPROVEDIAGNOSIS.ORG
>>
>> Moderator:David Meyers, Board Member, Society for Improving Diagnosis in
>> Medicine
>>
>> To learn more about SIDM visit:
>> http://www.improvediagnosis.org/
>>
>
>
> ------------------------------
>
>
> To unsubscribe from IMPROVEDX: click the following link:
> http://list.improvediagnosis.org/scripts/wa-IMPDIAG.exe?
> SUBED1=IMPROVEDX&A=1 or send email to: IMPROVEDX-SIGNOFF-REQUEST@
> LIST.IMPROVEDIAGNOSIS.ORG
>
> Moderator:David Meyers, Board Member, Society for Improving Diagnosis in
> Medicine
>
> To learn more about SIDM visit:
> http://www.improvediagnosis.org/
>






Moderator: David Meyers, Board Member, Society to Improve Diagnosis in Medicine


HTML Version:
URL: <../attachments/20180809/13788d2e/attachment.html>


More information about the Test mailing list