Date:         Sat, 11 Aug 2007 11:48:44 +1000
Reply-To:     d@dkvj.biz
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David Johnson <d@DKVJ.BIZ>
Subject:      Re: Zip Codes
In-Reply-To:  <46BC9F98.1020001@alumni.stanford.org>
Content-Type: text/plain; charset="iso-8859-1"

The groupings make some sort of sense, and referring back to an earlier
posting about formats, one might have a generic format of infectious
diseases for all values in that range. However, they are a classification,
not a quantification.  To quantify something we would expect a change in
degree between 1 and 139, and that doesn't hold true.

I imagine we couldn't expect it to be true, since even without the E and V
codes, the existence of complications and comorbidities can have significant
impact on the weight and ALOS.  Indeed I remember having a difference of
more than an order of magnitude in the cost weight between one permutation
of PDX and SDX(s) and another.  If they existed as some measure of quantity,
then we might expect that the function( A, B, C) would be similar to the
function( B, C, A) as long as the function is additive or multiplicative.

Since the cost weights and ALOS are based on clinical experience, we can't
expect that humans will conform with simple mathematical models. In any
case, age is also a very significant factor in some diseases.  While the
mean age of persons with WHO Class 3 or Class 4 brain tumours might be in
the high 60's, and mean survival time for Class 4 is around 14 months, the
age influence is quite strong and the short survival time is influenced
strongly by the patients age.  In these circumstances, some strains of
influenza can have higher mortality rates.

Kind regards

David


-----Original Message-----
From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU]On Behalf Of Jack
Hamilton
Sent: Saturday, 11 August 2007 3:26 AM
To: SAS-L@LISTSERV.UGA.EDU
Subject: Re: Zip Codes


David Johnson wrote:

> ICD-9 codes were my earliest trouble in dealing with numbers, since 123.46
> is a variation on 123.4.  It made sense that the precision of the number
> identified the amount of information on the diagnosis, and that everything
> that rounded to the same integer was related.  However, adjacent integer
> values might be totally unrelated,

The values are somewhat related, but (as you know, but others may not),
they are related only within ranges.  For example, all diagnosis codes
starting with 001 through 139 are for infectious diseases of some kind.

> and before we dealt with the E and V codes

Oh, those are fun!  Especially the E codes -- E882 is for falls from
buildings, but if the building is burning, it's E890 or E891.  There are
special codes for being hit by objects falling from a moving car, but
different codes if the car is parked.

> it was clear that the implicit quantum in the numbers did not give us
> anything of value.

Except that, in general, change from one integer to the next means that
you're looking at a different disease or diagnosis.

> That was pre SAS/L and I could have saved myself some
> work by discussing the issue with my old friend Bill Viergiever who was
> addressing similar problems.
>
> The fun in this game is that you need to spend some time understanding
your
> data and its intricacies before you can make good sense of it.

Absolutely!  That's why our jobs aren't like working on an assembly line.