Page 25 - qp_17

d. Consider outlier if the individual value in the respective column exceed:
• 4x mean absolute difference; AND
• 4X Mean normalized (relative) absolute difference

e. Investigate & document outlier before deleting the data from the data set.

f. If more than one sample has to be deleted, carry out an expended
investigation into the cause of discrepancies. If the cause is due to offending
samples alone replace those samples in the data sheet. Document the cause
of the problem. If it can be corrected but not traced to specific samples, the
entire data must be recollected.

g. If problem cannot be found and cannot be corrected, evaluate size of
maximum different between duplicates relative to allowable medical decision
limit for precision of the method, used Relative changed Value (RCV) for this
purpose

NOTE:
RCV=2.77*√(CVi² + CVa²)
 within method assessment used CVa of the method being assessed
(current or/and new)

6. Test for between method outlier
a. Compute the absolute differences between methods and their average (similar
as in step 4).
• Absolute difference for both Run 1 and Run 2 on new method from mean
of replicates of old method
o EgRun1(new)-mean old, Run 2(new) –mean old
• Mean absolute difference, ∑ total absolute diff/2n

b. Compute a relative absolute difference between methods and their average (also
similar as in step 4).
• Relative absolute difference, eg (Run1(new)-mean old)/mean old
• Mean Relative Absolute difference, Ẽ’=∑ total relative absolute diff/2n

c. Consider outlier if the individual value exceed:
• 4x mean absolute difference; AND
• 4X Mean normalized (relative) absolute difference

d. How many outlier can be excluded?
• Only 2.5% of total samples and IF
• >2.5% outliers - investigate possible interference (human error and device
malfunctions). If obvious causes cannot be determined, and if the
differences between the values exceed the bounds of medical
significance then
 Stop the evaluation or
 Add 40 new samples or
 If the outlier do not exceed a medically significant difference, based
on RCV &TEa, retained and use the data. If expanded investigation