Abstract for Poster Presentation

AGREMENT ANALYSIS IN CASE CONTINOUS VARIABLE

In   Clinical   and   Epidemiological   studies  researcher  are  very   much   interested   to  know  the  Inter – Observer  variation  in   continuous  variables  when   same  variable   measured  by   two  different  Techniques  or   by   two  different   observer.

The   conventional  statistical  technique  for  studying   the  agreement   between   two   methods   of   measuring  a   continuous  variable  is  to   compute  Correlation  Coefficient ( r )  ,  However  the  use  of   r  is  Misleading  in  many  cases.

 ‘r’  measure  only   the  strength   of   relationship   between  two  variables  or   observation  of   a   variable  between  two  observers.  But it   does not measure agreement between them. More  over  with  change  in  the  scale  of  measurement  does  not  alter  r  but  affect  the  agreement.  

 

Table1.   Rating of 5 subjects   by   two Raters (I & II)

                    Sample 1                   Sample2               Sample3

                        A        B                 A        B                 A        B

Subject   1        1        1                 1        5                 1        2

              2        2        2                 2        6                 2        4

              3        3        3                 3        7                 3        6

              4        4        4                 4        8                 4        8

              5        5        5                 5        9                 5        10

                         r = 1.0                 r = 1.0                    r = 1.0

In Sample 2 and Sample 3 only scale of measurement changed for variable B.

Scattered Plot between Sample 1, 2 and 3

There   will   be   a   perfect agreement exists    only   if   all   the   point   plotted    in   a scatter    diagram   lie   along   the diagonal   straight   line   ,   but     will   be   a perfect   correlation   if   the   points   lie   along   any    straight   line.  From   above graph   one   think is quit   visible   that   with   changing   scale r   remain same   but   agreement   go   on changing.  That is why   measuring   r   is   not   sufficient   to   see   agreement   between   two   observer and   techniques. To   overcome   this   misleading   approach   compute  r’   between    the   differences  (   X – Y )   against    mean    (  X +  Y  )  /  2.

i.e.   Study   the   relationship   between   the   measurement   error    and   true   value.  For   good   agreement   the   value of r’ should    not   be   significantly   different   from   zero.   

 

Stroke Volume of 21 urology patients measured by two Techniques known MF and SV

Patients  MF               SV                        Difference     Average

1                 47               43                        -4                45     

2                 66               70                         4                68

3                 68               72                         4                70

4                 69               81                         12              75

5                 70               60                        -10              65

6                 70               67                        -3                68.5

7                 73               72                         -1                72.5

8                 75               72                         -3                73.5

9                 79               92                         13              85.5

10               81               76                         -5                78.5

11               85               85                         0                85

12               87               82                         -5                84.5

13               87               90                          3               88.5

14               87               96                          9               91.5

15               90               82                         -8                88

16               100             100                        0               100

17               104             94                         -10              99

18               105             98                         -7                101

19               112             108                       -4                110

20               120             131                         11             125.5

21               132             131                       -1                131.5

Mean           86.0            85.8

SD               20.3            21.2

 

Scattered Plot between True Value and Measurement Error

Scattered Plot between two techiques MF and SV

Advantage of Bland Altman Graph

Intra Class Correlation (ICC)

 

Measure Limit of agreement

Measure the Regression Coefficient (b)

 

To see the agreement between two continuous variables they measured by two different observers or by two different techniques. We perform 5 statistical tests in case 3 come out true we can say there is agreement existing between them.

 

• 1.     r   -   should be  very  high   [  r >  .80  ]

 

• 2.     r”  -   should   be   very   low  [   r”  <  .20 ]

 

• 3.     ICC -   should   be   very   high [ICC > .80]

 

• 4.      b    -   should  not   be  different  from  1.

      5.     d    -  Bias  should  not  be  different  from  zero  and     

                       Limit Of agreement and their 95% C.I.  Should be   

                       Within acceptable range.