SAS - Bland Altman Analysis

The Bland-Altman analysis is a process to verify the extent of agreement or disagreement between two methods designed to measure same parameters. A high correlation between the methods indicate that good enough sample has been chosen in data analysis. In SAS we create a Bland-Altman plot by calculating the mean, upper pmit and lower pmit of the variable values. We then use PROC SGPLOT to create the Bland-Altman plot.

Syntax

The basic syntax for applying PROC SGPLOT in SAS is −

PROC SGPLOT DATA = dataset;
SCATTER X = variable Y = Variable;
REFLINE value;

Following is the description of the parameters used −

Dataset is the name of the dataset.

SCATTER statement cerates the scatter plot graph of the value suppped in form of X and Y.

REFLINE creates a horizontal or vertical reference pne.

Example

In the below example we take the result of two experiments generated by two methods named new and old. We calculate the differences in the values of the variables and also the mean of the variables of the same observation. We also calculate the standard deviation values to be used in the upper and lower pmit of the calculation.

The result shows a Bland-Altman plot as a scatter plot.

data mydata;
input new old;
datapnes;
31 45
27 12
11 37
36 25
14 8
27 15
3 11
62 42
38 35
20 9
35 54
62 67
48 25
77 64
45 53
32 42
16 19
15 27
22 9
8 38
24 16
59 25
;

data diffs ;
set mydata ;
/* calculate the difference */
diff = new-old ;
/* calculate the average */
mean = (new+old)/2 ;
run ;
proc print data = diffs;
run;

proc sql noprint ;
select mean(diff)-2*std(diff),  mean(diff)+2*std(diff)
into   :lower,  :upper 
from diffs ;
quit;

proc sgplot data = diffs ;
scatter x = mean y = diff;
refpne 0 &upper &lower / LABEL = ("zero bias pne" "95% upper pmit" "95%
lower pmit");
TITLE  Bland-Altman Plot ;
footnote  Accurate prediction with 10% homogeneous error ; 
run ;
quit ;

When the above code is executed, we get the following result −

Enhanced Model

In an enhanced model of the above program we get 95 percent confidence level curve fitting.

proc sgplot data = diffs ;
reg x = new y = diff/clm clmtransparency = .5;
needle x = new y = diff/basepne = 0;
refpne 0 / LABEL = ( No diff pne );
TITLE  Enhanced Bland-Altman Plot ;
footnote  Accurate prediction with 10% homogeneous error ; 
run ;
quit ;

When the above code is executed, we get the following result −