Date:         Mon, 31 Mar 2008 20:40:17 -0400
Reply-To:     Lou <lpogoda@HOTMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Lou <lpogoda@HOTMAIL.COM>
Subject:      Re: Does (Clinical)Tables need validation ???
Comments: To: sas-l@uga.edu

"SAS_learner" <proccontents@GMAIL.COM> wrote in message
news:c2192a610803302048p352dc796oc3ec4e3ba9496fea@mail.gmail.com...
> Lou,
>
> It was Just difference in opinion about the double Programming. He thinks
> that all the variables that are needed to be used in the tables should be
> in
> the analysis dataset and the table should be Just a Proc's away.

Theoretically, he's correct.  As a practical matter, I've never seen
anything that's just a proc away, except maybe an include/exclude listing.

But that's not to say it couldn't be done - if your boss want the analysis
data sets to contain subtotals and totals and all mins and maxes etc. it
could be done that way, though I think it would be a lot of extra work
leading to some pretty wide datasets.  I think checking would still be
needed to make sure that the right population was being selected for display
on any given table.

On the other hand, if your place is using standard macros to produce
standard tables, and the macros have the demonstrated ability to produce
those tables given the input data is in the proper form, it wouldn't be the
first tine I've heard of that either.

>
> At OLD Place which is a CRO they used to Insist that there is 100%
> independent validation was done during that Process we used to come across
> lot issues that we used to come across that we used to let the
> statistician
> know who in tern used to change his Specs depending on feedback that we
> used
> to give.
>
> There were couple of times that main Programmer can go wrong like data
> _null_ pointed out using wrong denominator or using wrong group by
> variable
> in Proc SQL or usage of Nodupkey or different usage of First. or Last or
> Picking of wrong P-value .  .
>
> Even in the summary tables we do little bit data manipulation to get till
> Proc means and Proc Freq or Univariate right.Some mistake done in that
> Process might affect the numbers or Percentages. My school of thought was
> a
> second person going over the numbers should give a better numbers and
> confidence. I do not know I was bought up in a school where validation was
> almost like main Programming expect for Proc Report.
>
> Lou suggestion of standardized macro's for summary tables also make sense
> Once they are standardized you might want to use but they we need tweak
> them
> as per the data right.

If you're using truly standard macros, you don't tweak the macros, you tweak
the data until it's in the form the macros expect.