Date: Sat, 28 Apr 2001 13:30:24 -0400
Reply-To: dubrowg@dolphin.upenn.edu
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Greg Dubrow <dubrow@UNDERGROUND.IRHE.UPENN.EDU>
Organization: University of Pennsylvania
Subject: Re: problem with area under the curve
Content-Type: text/plain; charset=us-ascii
Michael Bramley wrote:
>
> Greg:
>
> Using the first patient (#22) as an example, are you saying that you're
> calculating AUC for length & Critmet (where length=X-axis and Critmet=Y-axis)?
> Assuming this is so, the approximate calculation for AUC = sum( abs((X_i -
> X_i-1)*Y_i)), (where X_0=0), so that in your case, AUC = (41-0)*3 + (90-41)*2 +
> (125*0) for patient #22.
>
> Note that AUC is a monotonically non-decreasing function, that is, as you add
> more time points (X), the AUC function can only increase or stay the same as
> area is never negative...perhaps your computation is slipping in a negative
> value?
>
> Since you're using Proc Expand, maybe you should be using the cumulative sum
> transformation (CUSUM) ?
I tried that, and still get decreasing totals. Inf act, I got the same
sums using cusum that I did using sum.
The code is this:
proc expand data=mtxra20 out=mtxra20z method=spline;
convert critmet=total/observed=(beginning,total) transformout=(cusum);
id length; **assuming this is the y-axis;
by regnum;
run;
(where regnum=patient number, length=amount of days between visits,
critmet=number of criteria met per drug protocol)
And here are a set of totals. Again, I'm mystified why some decrease.
visit # is not in the computation
patient visit length critmet area
757 1 0 0 -1.38
757 5 28 0 24.95
757 13 85 1 101.22
757 22 147 1 56.3
1138 1 0 0 .072
1138 5 28 0 -4.11
1138 13 85 0 63.07
1138 22 147 3 512.05
1157 1 0 0 -7.06
1157 5 28 0 115.3
1157 13 82 5 511.94
1157 22 140 7 599.51
2330 1 0 0 -2.81
2330 5 28 0 48.75
2330 13 84 2 208.53
2330 22 148 2 95.7
Again, any insight would be most appreciated. I'm stumped.
greg
| 
