topic Re: How to draw a Parkes (Consensus) Error Grid? in Graphics Programming

How to draw a Parkes (Consensus) Error Grid?

motiv3 — Wed, 01 Jan 2014 02:01:12 GMT

Hey everyone!

This is my first time on this forum so please let me know if you require any further information on the matter.

Basically, I am a hospital physician conducting an experiment looking into whether a glucose meter used heavily within our Department complies with the newly released guidelines. Now part of the new guidelines states the requirement of a Parkes Error Grid for all new meters i.e. they must demonstrate accuracy using this grid (Consensus error grid - Wikipedia, the free encyclopedia). Now I've looked into the grid itself e.g. read the original article and it seems to be an alteration of the original Clarke Error Grid. However, the new Parkes grid has different regions and so I am struggling to make such a graph using SAS; this is only worsened with my lack of experience using this tool! Therefore, I was wondering whether any of you kind people could help me with this matter. I have the paper that provides an outline of the co-ordinates for each region yet, once again, I am kind of stumped on how to approach this.

Thank you for your help.

Re: How to draw a Parkes (Consensus) Error Grid?

Jay54 — Wed, 01 Jan 2014 02:47:19 GMT

At SAS Global Forum 2013, the paper #113-2013 - Creating Clark Error Grid using SAS/GRAPH and Annotate... by Yongyin Wang and John Shin of Medtronic Diabetes discussed using Annotate to create the Clarke Error Grid. Later, I wrote a Blog article on creating a similar Clarke Error Grid Graph using the SGPLOT procedure, without any Annotation. Between these two resources, you should be able to find the information you need for your use case of the Parkes Error Grid. Personally, I would go with the SGPLOT case.

In the SGPLOT case, the grid itself is created simply using the Series Plot statement. The points for the series are defined in the data as vertices of the plot.

Re: How to draw a Parkes (Consensus) Error Grid?

motiv3 — Thu, 02 Jan 2014 00:12:54 GMT

Hey Sanjay,

Thank you for the help, I will try this out later on tonight.

Just another question. How would I go about formulating a difference plot (Bland-Altman plot) using SAS where I can show clearly the limits defined by the guidelines as shown on the attached image (the black and shaded lines). Although I know that it is essentially a Bland-Altman plot, am a bit confused as to how to place the upper and lower limits defined in the EN ISO as a shaded line.

Re: How to draw a Parkes (Consensus) Error Grid?

Jay54 — Thu, 02 Jan 2014 00:46:54 GMT

This seems like an unrelated question. Maybe you can separate it for convenience of other readers.

This graphs seems very straightforward. In the graph above, the upper and lower limits have patterns and different colors, but not realy "shaded". It could be shaded using fill colors if you want. This can be done using the BAND plot, using the X, UPPER and LOWER parameters. You can use GROUPed band in one statement, or overlay two separate bands. You can use FILLED or OUTLINED bands. You can also use a SERIES plot if you want but I would not recommend it. Use STYLE=JOURNAL for gray scale graph with line patterns.

The labels can be placed at the ends using the LABEL option. Or, if you want the label in the middle, you can put the label in another column at a vertex in the middle with others as missing. Then use the DATALABEL option. Or, use a separate overlaid scatter plot with markercharacter. Which release of SAS are you using?

Re: How to draw a Parkes (Consensus) Error Grid?

motiv3 — Thu, 02 Jan 2014 02:28:55 GMT

Thank you for advice Sanjay. However, I am relatively new to the whole SAS scene and am struggling to get to grips with how to go about devising the graphs. Is there any way that a template could be devised which I could simply access and plug my values into? If not, is there any means of discussing this further e.g. over private communication so we can see how to best combat this problem.

With regards to what release of SAS I am using, I am currently utilising version 9.3 of SAS.

Thanks

Re: How to draw a Parkes (Consensus) Error Grid?

Jay54 — Thu, 02 Jan 2014 05:20:34 GMT

Here is a graph with simulated data and bands as shown in your graph. Code is attached.

The second scatter plot is used to add the band names. You could also use a legend instead.

Re: How to draw a Parkes (Consensus) Error Grid?

motiv3 — Fri, 03 Jan 2014 18:49:34 GMT

Hello Sanjay,

Thanks for all the help but I need a bit more help if possible. This post will be related to the Bland-Altman plot and I will make a separate post with regards to the Parkes Consensus Grid:

With the Bland-Altman plot, I was wondering whether you could help me with the following:

1) How to insert my own data. Although I know you used simulated data, I am struggling to insert my own data set within the graph. Is there any way you could alter the SAS code so I can start adding my own data. I know I am relatively new so if doing this was obvious then I apologise. If you in the code use a point such as YSI Plasma Result = 500 and Bias from YSI = 40 and another point being x= 300 and y = 100 (both made up results) then I will be able to clearly see where to continue inserting my own data.

2) Is there any way of changing the colour of the lines representing the criteria for 2003 and 2013? Similarly, how would I go about changing the colour of the text too.

3) Is there any way of changing the location of the points from a circle to say a cross?

4) Is there any way to change the font size and font text for both the axis and the labels (EN ISO 15197-2003 and ISO 15197-2013)? I would like to standardise mine to Times New Roman if possible

5) On the axis, the grid seems to go both above 120 and below - 120 on the y axis and slight past the 600 mark on the x axis. Is there any way of removing this so that the graph margin ends at 600 on the x-axis and the origin of the graph is (0,-120) and upper border of the graph stops at (0,120)

Thank you again for the help Sanjay!

Re: How to draw a Parkes (Consensus) Error Grid?

motiv3 — Fri, 03 Jan 2014 19:13:15 GMT

Once again Sanjay, thank you for the help but I am struggling to get to grips with constructing the Parkes Error Grid. Being so new to using SAS, it's a real struggle trying to construct this grid and I was wondering whether you could possibly help me formulate the SAS code.

1) One issue I am having is with the new region co-ordinates. I've attached the actual paper to this post (page 2 and 3 of the PDF - refer to the co-ordinates to make the grid for Type 1 Diabetes Mellitus). I just can't seem to get it to work.

2) If we can devise a way to get the above co-ordinates into your SGPLOT method (as on your blog) will all the other attributes you had in your Clarkes grid be transferred over e.g. colour scheme, percentage of points within each grid? If so, could you possibly integrate this within the code for the new Parkes grid?

3) I also can't seem to get my data across into the graph as points. Could you possibly in the code put in some values ive made up e.g. reference glucose = 200 and test device glucose = 400. If i can see where this data has been placed within the code, I can go from there and use my own data.

Apologies for all the questions but I really do appreciate the great help you've given me!

Re: How to draw a Parkes (Consensus) Error Grid?

Jay54 — Fri, 03 Jan 2014 19:24:30 GMT

I think some self-study is in order. I suggest you review the documentation for the SGPLOT procedure to get a handle of the features available to you. I would recomment my book SG Procedures by Example to get started. You can also refer to many articles on my blog. While I can certainly help out on specific questions, writing up whole programs is beyond what I can do. Maybe someone else will chime in. Good luck.

Re: How to draw a Parkes (Consensus) Error Grid?

motiv3 — Fri, 03 Jan 2014 19:29:16 GMT

Ah that is fair enough Sanjay,

If you could possibly help with with my queries on the Bland Altman Plot and any you feel you can answer on the Parkes Error Grid then I would be extremely grateful.

Thank you.

Re: How to draw a Parkes (Consensus) Error Grid?

Jay54 — Fri, 03 Jan 2014 20:06:49 GMT

The data set I created for the points only has X & Y columns. I assume you data is similar. You can merge your data to the Band and labels data exactly the same way. If you are using a group variable with band plot, the colors come from GraphData1 and GraphData2 style elements. With SAS 9.3, you will need to derive your own style, and change these colors to the ones you want. To use a different marker, just use MARKERATTRS=(SYMBOL=PLUS) in the scatter plot statement after the /. Or any other supported symbol. The space beyong 120 is the offset needed to draw a marker that may land at the edge. Aet XAXIS or YAXIS OFFSETMIN of OFFSETMax to zero. Note, any markers falling on the edges will be truncated.

Re: How to draw a Parkes (Consensus) Error Grid?

GraphGuy — Mon, 06 Jan 2014 16:21:56 GMT

Here are 2 examples created using SAS/Graph gplot - click the links to see the SAS code ...

http://robslink.com/SAS/democd61/hemoglobin_error_grid_info.htm

http://robslink.com/SAS/democd61/clarke_error_grid_info.htm

Re: How to draw a Parkes (Consensus) Error Grid?

Joanlee — Mon, 08 Aug 2016 05:00:14 GMT

Hello

please see pfutzner et al. J diabetes science and technology. 2013 7(5) 1275 for coirdinates

thanks.

Joan Lee Parkes

Re: How to draw a Parkes (Consensus) Error Grid?

prem_sahu — Mon, 21 Nov 2016 10:44:48 GMT

thanks a lot for providing the reference. The code pasted was worked well.

I am New to SAS and facing difficulty while constructing this.could you please tell me how to get percentage for different zones.in the mentioned code.

Re: How to draw a Parkes (Consensus) Error Grid?

Joanlee — Mon, 21 Nov 2016 13:41:50 GMT

Hello
I'm not sure what u r asking. I also forwarded your message to Scott Pardo also on the original paper. He. Is the statistician.

Re: How to draw a Parkes (Consensus) Error Grid?

prem_sahu — Tue, 22 Nov 2016 11:25:10 GMT

Ok Thanks a lot

Re: How to draw a Parkes (Consensus) Error Grid?

Joanlee — Tue, 22 Nov 2016 15:48:09 GMT

Joan,

He seems to be confused. He first asks how to draw the zone boundaries, and I guess he figured that out. But he wants to know how to compute the percentages of points falling in each zone. He will need to have code that checks every point to determine in which zone it lies. Then he will need code to count the number of points in each zone. It’s not an easy programing task; you need to check a lot of boundaries. It would be particularly difficult in SAS.

I do not know if this helps.

Bye for now,

Scott P.

> On Nov 21, 2016, at 8:39 AM, Joan wrote:
>
> Do u understand his question
>
> Sent from my iPad
> Joan Lee Parkes, PhD
>
> Begin forwarded message:
>
>> From: SAS Support Communities
>> Date: November 21, 2016 at 5:44:55 AM EST
>> To: joanleeparkes@gmail.com
>> Subject: Re: How to draw a Parkes (Consensus) Error Grid?
>>
>> Hi Joanlee,
>>
>>
>> prem_sahu (New User) posted a new reply in SAS/GRAPH and ODS Graphics on 11-21-2016 05:44 AM :
>>
>>
>>
>> Re: How to draw a Parkes (Consensus) Error Grid?
>>
>>
>>
>> thanks a lot for providing the reference. The code pasted was worked well.
>>
>> I am New to SAS and facing difficulty while constructing this.could you please tell me how to get percentage for different zones.in the mentioned code.
>>
>>
>>
>> Reply | Like this message
>>
>>
>>
>> SAS Support Communities sent this message to joanleeparkes@gmail.com.
>> You are receiving this email because a new message matches your subscription to a topic.
>>
>> To control which emails we send you please go to, manage your subscription & notification settings or unsubscribe.
>

Re: How to draw a Parkes (Consensus) Error Grid?

Ksharp — Wed, 18 May 2022 02:41:00 GMT

Consensus Error Grid

/*产生样本数据*/
data have;
  pi=constant('PI');
  call streaminit(123);
    do i=1 to 500;
      x=40+30*ranuni(2);  y=40+40*ranuni(2); output;
    end;

    /*--Add data all over--*/
    do i=1 to 100;
      x=550*ranuni(2);  y=550*ranuni(2); output;
    end;

    /*--Add data in the middle fan--*/
    do i=1 to 2000;
      x=40+360*ranuni(2);  
	  y=x+(x-50)*2*tan(pi/4 *ranuni(2) -pi/8);
	  y=ifn(y>550 or y<0, ., y);
      output;
    end;
 keep x y;
run;












/*画图*/
/*--Generate grid and point data--*/
data Grid;
input  id rfbg  sbg;
cards;
11 50 0
11 50 30
11 90 80
11 330 230
11 550 450
12 0 50
12 30 50
12 230 330
12 440 550
21 90 0
21 260 130
21 550 250
22 0 60
22 30 60
22 280 550
31 250 0
31 250 40
31 410 110
31 550 160
32 0 80
32 25 80
32 35 90
32 125 550
41 0 200
41 35 200
41 50 550
;
run;
data label;
    xl=320; yl=280; label='A';  output;
    xl=280; yl=320; label='A';  output;
    xl=370; yl=220; label='B';  output;
    xl=220; yl=370; label='B';  output;
    xl=400; yl=140; label='C';  output;
    xl=140; yl=400; label='C'; output;
    xl=420; yl=70; label='D'; output;
    xl=70; yl=420; label='D';output;
    xl=20; yl=430; label='E';output;
run;
/*--Add data points--*/
data Plot;
  set grid have label;
run;
/*--Draw the grid--*/
ods html dpi=300;
ods graphics / reset antialiasmax=5700 ;
proc sgplot data=Plot noautolegend aspect=1;
  lineparm x=0 y=0 slope=1/lineattrs=(pattern=dash color=gray);
  series x=rfbg y=sbg / group=id lineattrs=(color=gray ) nomissinggroup;
  scatter x=x y=y /  markerattrs=(symbol=circlefilled size=2 color=black) transparency=0.5;
  scatter x=xl y=yl / markerchar=label markercharattrs=(size=12);
  xaxis values=(0 to 550 by 50) offsetmin=0 offsetmax=0 label='Reference Blood Glucose';
  yaxis values=(0 to 550 by 50) offsetmin=0 offsetmax=0 label='Sensor Blood Glucose';
  run;

Re: How to draw a Parkes (Consensus) Error Grid?

Ksharp — Wed, 22 Jun 2022 12:29:23 GMT

To complete Consensus Error Grid to calculate percent in each area . I rescue to PROC GINSIDE. This PROC is awesome.

NOTE: unit is (mg/dL)

/*产生样本数据*/
data have;
  pi=constant('PI');
  call streaminit(123);
    do i=1 to 500;
      x=40+30*ranuni(2);  y=40+40*ranuni(2); output;
    end;

    /*--Add data all over--*/
    do i=1 to 100;
      x=550*ranuni(2);  y=550*ranuni(2); output;
    end;

    /*--Add data in the middle fan--*/
    do i=1 to 2000;
      x=40+360*ranuni(2);  
	  y=x+(x-50)*2*tan(pi/4 *ranuni(2) -pi/8);
	  y=ifn(y>550 or y<0, ., y);
      output;
    end;
 keep x y;
run;












/*画图*/
/*--Generate grid and point data--*/
data map;
input  id $ x y;
cards;
B_L 50 0
B_L 50 30
B_L 90 80
B_L 330 230
B_L 550 450
B_L 550 250
B_L 260 130
B_L 90 0
C_L  90 0
C_L  260 130
C_L  550 250
C_L  550 160
C_L  410 110
C_L  250 40
C_L  250 0
D_L  250 0
D_L  250 40
D_L  410 110
D_L  550 160
D_L  550 0
A 50 0
A 50 30
A 90 80
A 330 230
A 550 450
A 550 550
A 440 550
A 230 330
A 30 50
A 0 50
A 0 0
B_U 0 50
B_U 30 50
B_U 230 330
B_U 440 550
B_U 280 550
B_U 30 60
B_U 0 60
C_U 0 60
C_U 30 60
C_U 280 550
C_U  125 550
C_U  35 90
C_U  25 80
C_U  0 80
D_U 0 80
D_U 25 80
D_U 35 90
D_U 125 550
D_U 50 550
D_U 35 200
D_U 0 200
E 0 200
E 35 200
E 50 550
E 0 550
;
run;

/*判断点在哪个区域*/
proc ginside data=have map=map out=id INCLUDEBORDER;
  id id ;
run;
data id;
 set id;
 id=scan(id,1,'_');
run;
proc freq data=id ;
table id;
run;

data Grid;
input  id2 rfbg  sbg;
cards;
11 50 0
11 50 30
11 90 80
11 330 230
11 550 450
12 0 50
12 30 50
12 230 330
12 440 550
21 90 0
21 260 130
21 550 250
22 0 60
22 30 60
22 280 550
31 250 0
31 250 40
31 410 110
31 550 160
32 0 80
32 25 80
32 35 90
32 125 550
41 0 200
41 35 200
41 50 550
;
run;
data label;
    xl=320; yl=280; label='A';  output;
    xl=280; yl=320; label='A';  output;
    xl=370; yl=220; label='B';  output;
    xl=220; yl=370; label='B';  output;
    xl=400; yl=140; label='C';  output;
    xl=140; yl=400; label='C'; output;
    xl=420; yl=70; label='D'; output;
    xl=70; yl=420; label='D';output;
    xl=20; yl=430; label='E';output;
run;
/*--Add data points--*/
data Plot;
  set grid id label;
run;


/*--Attributes Map for zones--*/
data attrmap;
  length id $1 value $1 markercolor $10;
  id='A'; value='A'; markercolor='cx00afdf'; linecolor='cx00afdf'; output;
  id='A'; value='B'; markercolor='cx00ef7f'; linecolor='cx00ef7f'; output;
  id='A'; value='C'; markercolor='gray'; linecolor='gray'; output;
  id='A'; value='D'; markercolor='pink'; linecolor='pink'; output;
  id='A'; value='E'; markercolor='red'; linecolor='red'; output;
run;


/*--Draw the grid--*/
ods html dpi=300;
ods graphics / reset antialiasmax=5700 ;
proc sgplot data=Plot noautolegend aspect=1 dattrmap=attrmap;
  lineparm x=0 y=0 slope=1/lineattrs=(pattern=dash color=gray );
  series x=rfbg y=sbg / group=id2 lineattrs=(color=gray ) nomissinggroup;
  scatter x=x y=y /  markerattrs=(symbol=circlefilled size=2 ) group=id attrid=A;
  scatter x=xl y=yl / markerchar=label markercharattrs=(size=12);
  xaxis values=(0 to 550 by 50) offsetmin=0 offsetmax=0 label='Reference Blood Glucose';
  yaxis values=(0 to 550 by 50) offsetmin=0 offsetmax=0 label='Sensor Blood Glucose';
  run;

Re: How to draw a Parkes (Consensus) Error Grid?

Ksharp — Wed, 22 Jun 2022 12:32:49 GMT

Here is Clark Error Grid Graph by using PROC GINSIDE .

NOTE: unit is (mg/dL)

/*产生样本数据*/
data have;
  pi=constant('PI');
  call streaminit(123);
    do i=1 to 500;
      x=40+30*ranuni(2);  y=40+40*ranuni(2); output;
    end;

    /*--Add data all over--*/
    do i=1 to 100;
      x=400*ranuni(2);  y=400*ranuni(2); output;
    end;

    /*--Add data in the middle fan--*/
    do i=1 to 5000;
      x=40+360*ranuni(2);  
	  y=x+(x-50)*2*tan(pi/4 *ranuni(2) -pi/8);
	  y=ifn(y>400 or y<0, ., y);
      output;
    end;
	x=70;y=84;output;
    x=70;y=56;output;
 keep x y;
run;










/*画图*/
/*--Generate grid and point data--*/
data map;
infile cards expandtabs;
input n id $ x y;
cards;
1	E_L 180 0
2	E_L 180 70
3	E_L 400 70
4	E_L 400 0
40	E_U 0 180
41	E_U 70 180
42	E_U 70 400
43	E_U 0 400
8	D_L  240 70
9	D_L  240 180
10	D_L  400 180
11	D_L  400 70
35	D_U 0 70
36	D_U 58.3 70
37	D_U 70 84
38	D_U 70 180
39	D_U 0 180
5	C_L  130 0
6	C_L  180 70
7	C_L  180 0
32	C_U 70 180
33	C_U 290 400
34	C_U 70 400
12	B_L  70 0
13	B_L  70 56
14	B_L  400 320
15	B_L  400 180
16	B_L  240 180
17	B_L  240 70
18	B_L  180 70
19	B_L  130 0
28	B_U 70 84
29	B_U 333 400
30	B_U 290 400
31	B_U 70 180
20	A 0 0
21	A 0 70
22	A 58.3 70
23	A 333 400
24	A 400 400
25	A 400 320
26	A 70 56
27	A 70 0
;
run;
data map;
 set map;
 if n in (23 29) then x=400/1.2 ;
 if n in (22 36) then x=70/1.2 ;
 drop n;
run;



/*判断点在哪个区域*/
proc ginside data=have map=map out=id INCLUDEBORDER;
  id id ;
run;
data id;
 set id;
 id=scan(id,1,'_');
run;


proc freq data=id ;
table id/out=zone;
run;

%let A=0.00;
%let B=0.00;
%let C=0.00;
%let D=0.00;
%let E=0.00;

%let count_A=0;
%let count_B=0;
%let count_C=0;
%let count_D=0;
%let count_E=0;

data _null_;
 set zone;
 if not missing(id) then do;
  call symputx(id,put(PERCENT,8.2));
  call symputx(cats('count_',id),COUNT);
 end;
run;



data Grid;
  /*--Zone A--*/
  id2=11; rfbg2=70;  sbg2=0; output;
  id2=11; rfbg2=70;  sbg2=56; output;
  id2=12; rfbg2=58.3;sbg2=70; output;
  id2=12; rfbg2=0;   sbg2=70; output;
  id2=13; rfbg2=70;   sbg2=84; output;
  id2=13; rfbg2=70;   sbg2=180; output;
  id2=13; rfbg2=0;   sbg2=180; output;

  /*--Zone E - Y--*/
  id2=3; rfbg2=70;  sbg2=180; output;
  id2=3; rfbg2=70;  sbg2=400; output;
  id2=3; rfbg2=0;   sbg2=400; output;
  /*--Zone E - X--*/
  id2=4; rfbg2=180;  sbg2=0; output;
  id2=4; rfbg2=180;  sbg2=70; output;
  id2=4; rfbg2=400;  sbg2=70; output;
  /*--Zone D - X--*/
  id2=5; rfbg2=240;  sbg2=70; output;
  id2=5; rfbg2=240;  sbg2=180; output;
  id2=5; rfbg2=400;  sbg2=180; output;
  /*--Zone C - Low--*/
  id2=6; rfbg2=130;  sbg2=0; output;
  id2=6; rfbg2=180;  sbg2=70; output;
  /*--Zone C - High--*/
  id2=7; rfbg2=70;  sbg2=180; output;
  id2=7; rfbg2=290; sbg2=400; output;
  /*--Zone B - High--*/
  id2=8; rfbg2=58.3; sbg2=70; output;
  id2=8; rfbg2=333.3;  sbg2=400; output;
  /*--Zone B - Low--*/
  id2=9; rfbg2=70;   sbg2=56; output;
  id2=9; rfbg2=400; sbg2=320; output;
  run;

data label;
length label $ 40;
    xl=20; yl=20; label='A';   output;
    xl=20; yl=120; label='D';  output;
    xl=20; yl=300; label='E';   output;
    xl=100; yl=20; label='B';   output;
    xl=160; yl=20; label='C';   output;
    xl=100; yl=160; label='B';   output;
    xl=320; yl=20; label="E(&count_E.,&E.%)";   output;
    xl=320; yl=120; label="D(&count_D.,&D.%)";  output;
    xl=320; yl=200; label="B(&count_B.,&B.%)" ; output;
    xl=340; yl=340; label="A(&count_A.,&A.%)" ; output;
    xl=120; yl=320; label="C(&count_C.,&C.%)" ; output;
run;


data plot;
 set Grid id label;
run;

/*--Attributes Map for zones--*/
data attrmap;
  length id $1 value $1 markercolor $10;
  id='A'; value='A'; markercolor='cx00afdf'; linecolor='cx00afdf'; output;
  id='A'; value='B'; markercolor='cx00ef7f'; linecolor='cx00ef7f'; output;
  id='A'; value='C'; markercolor='gray'; linecolor='gray'; output;
  id='A'; value='D'; markercolor='pink'; linecolor='pink'; output;
  id='A'; value='E'; markercolor='red'; linecolor='red'; output;
run;


/*--Draw the grid--*/
ods graphics / reset antialiasmax=5700 ;
proc sgplot data=plot noautolegend aspect=1 dattrmap=attrmap;
  scatter x=x y=y / attrid=A group=id  markerattrs=(symbol=circlefilled size=3 ) ;
  scatter x=xl y=yl / markerchar=label markercharattrs=(size=12)  ;
    series x=rfbg2 y=sbg2 / group=id2 lineattrs=(color=gray ) nomissinggroup;

  xaxis values=(0 to 400 by 50) offsetmin=0 offsetmax=0 label='Reference Blood Glucose';
  yaxis values=(0 to 400 by 50) offsetmin=0 offsetmax=0 label='Sensor Blood Glucose';
  run;

Clark Error Grid