Okay so I was always taught to use the rule of 3, 3 positive reactions and 3 negative reactions for peforming an antibody ID. I was also taught to always use homozygous positive and negative cells whenever possible. Sometimes of course it is not due to low incident/high incident antigens. I do know you need to use a homozygous cell when performing "rule outs". What is everyone else's practices and thoughts as I need to clarify our current antibody identification policy. Thanks in advance.

I was taught to use 'p' values to establish confidence in determining antibody ID's, using a 7 and 1 or 3 and 2 arrangement. (7 positives and 1 negative or vice versa OR 3 positives and 2 negatives or vice versa).

I require techs to rule out using cells that are homozygous for the expression of the antigen they are trying to rule out, except for K. 

Do you have a copy of the Technical manual, 16th, 17th, or 18th edition? This discussion can be found under Chapter 16: Identification of antibodies to red cell antigens. Under the section titled 'Probability.' Table 16-3 shows a breakdown of p values based on different mixes of positive/negative and different statistical methods. Our procedure looks for at least two reactive & two nonreactive.

In general, we also follow the 3 x 3 rule. 

 

Obviously, the more cells you use to r/i or r/o, the higher the probability that your interpretation is correct.  On the other hand, since reaction variability is variable (if you will) depending on the antigen, a precise calculation of p values for a given frequency in a population is probably not worth worrying too much over.

 

Scott

AABB changed the standard a few years ago to 2 and 2 to accommodate reference labs - due to low/high incidence ags and the scarcity of either positive or negative cells.  I try to use 3 and 3.  I can r/o Kell and Lu system ags with cells heterozygous for ag expression (unless I enzyme pretreat, then I will r/o with heterozygous cells for those ags not sensitive to enzyme - Rh, Kidd).

Note that we are all ruling out with only 1 cell every time we interpret AS = Negative.  If we had to use the 3+3 rule for that, Antibody Screens would have to contain a lot more cells.

 

I see the 3+3 rule as applying to Antibody Identification, e.g. Don't call it an Anti-Cw unless you have 3 positives.  Don't call it an Anti-Jsb unless you have 3 negatives.  (Ruling out others, of course.)

OK, the only magic to the "3+3" rule is that if you plug those numbers into the horribly unwieldy formula for Fisher's exact method of calculating probability, you will get a probability (p) of 1/20, or a 1 in 20 chance that the reactions could have occured by chance, or a 95% confidence level that your conclusion is correct. (This is a totally arbitrary number by the way.) Lower p values (1/15, 1/9 etc) allow for too much chance of random association. Higher values (1/28, 1/56) show that there's a much smaller chance that your conclusion is incorrect.

 

But there are other magic combinations that will give you an acceptable p: 5 and 2 (1/21), 4 and 3 (1/35), 6 and 2 (1/28) and so on. You do not necessarily need 3+3. See Goodchild's reference.

 

You run into problems when you only have one positive or negative cell: 7 and 1 (p of 1/8), 8 and 1 (1/9), etc. You would have to get to 19 and 1 to get the magic p of 1/20. If you think about it non-arithmatically, what if your one reactive panel cell is also positive for an unlisted low frequency antigen? What if your one negative didn't have serum added or isn't reacting for some other technical reason?

 

So you don't necessarily need 3 Cw+ cells; 5 or more neg and 2 pos would suffice. And you don't need 3 Js(b-) cells; 19 pos and 1 neg would be OK statistically. The problem I see with the high incidence antigens like this would be that with only one negative cell with which to rule out, you would still have a bunch of other antibody choices you would like to rule out, hence the need to test more negative cells.

 

So, pedantry aside, the bottom line is "don't base your ID just on the reaction with one cell". A second cell of similar makeup coupled with the pos or negs from the rest of the panel should bump your p past 1/20.

Edited October 27, 20159 yr by Dr. Pepper

On 10/27/2015 at 9:02 AM, Dr. Pepper said:

OK, the only magic to the "3+3" rule is that if you plug those numbers into the horribly unwieldy formula for Fisher's exact method of calculating probability, you will get a probability (p) of 1/20, or a 1 in 20 chance that the reactions could have occured by chance, or a 95% confidence level that your conclusion is correct. (This is a totally arbitrary number by the way.) Lower p values (1/15, 1/9 etc) allow for too much chance of random association. Higher values (1/28, 1/56) show that there's a much smaller chance that your conclusion is incorrect.

 

But there are other magic combinations that will give you an acceptable p: 5 and 2 (1/21), 4 and 3 (1/35), 6 and 2 (1/28) and so on. You do not necessarily need 3+3. See Goodchild's reference.

 

You run into problems when you only have one positive or negative cell: 7 and 1 (p of 1/8), 8 and 1 (1/9), etc. You would have to get to 19 and 1 to get the magic p of 1/20. If you think about it non-arithmatically, what if your one reactive panel cell is also positive for an unlisted low frequency antigen? What if your one negative didn't have serum added or isn't reacting for some other technical reason?

 

So you don't necessarily need 3 Cw+ cells; 5 or more neg and 2 pos would suffice. And you don't need 3 Js(b-) cells; 19 pos and 1 neg would be OK statistically. The problem I see with the high incidence antigens like this would be that with only one negative cell with which to rule out, you would still have a bunch of other antibody choices you would like to rule out, hence the need to test more negative cells.

 

So, pedantry aside, the bottom line is "don't base your ID just on the reaction with one cell". A second cell of similar makeup coupled with the pos or negs from the rest of the panel should bump your p past 1/20.

 

You wouldn't happen to know the Fisher formula or where I can find it? I am just curious about the actual math/statistical tool(s) involved in the rule of 3. Thank you.