Editorial published and freely available

June 20, 2013

article

I had mentioned an upcoming editorial about glucose standards. That editorial has now been published and appears to be freely available.

I have argued that glucose specifications (ISO 15197 and CLSI POCT12-A3) should have goals for 100% of the data, not 98 or 99%. A manufacturer might argue that this is not needed because …

  1. The actual occurrence of dangerous glucose meter results, due to a failure of a glucose meter may not occur at all.
  2. To have goals for 100% of the data, meaning that 0% of the data should not be allowed in certain error regions would be too burdensome.

My responses would be: to 1), yes this may be true but does not mean the goal should be abandoned. As for 2), one should realize that one cannot prove the occurrence of zero failures – the FDA knows this – thus there is no reason that a zero failure rate goal for dangerous errors would be burdensome.

So the way the glucose meter standards are written looks bad – it’s time to change things.

Advertisements

Glucose Meter Error Statistics

June 5, 2013

size

The CLSI and ISO glucose meter guidelines imply the use of Normal statistics in their error limit guidelines. For example, the CLSI guideline states that meter performance is acceptable when 95% of the results are within 12.5% of reference at or above 100 mg/dL and 98% must be within 20%.

Assume that a glucose meter has a CV of 5% (or an SD of 5) and the reference is 100 mg/dL. Normal statistics say that 95% of the results will meet the guideline. If all of the errors follow the Normal distribution, where will they be? The following table shows this.

n std dev Glucose No more than 1 in … Percent in

2

90

20

94.99995

3

85

370

99.73002

4

80

15,787

99.99367

5

75

174,428

99.99994

6

70

507,000,000

 

7

65

3,910,000,000,000

 

Thus, 19 out of 20 results will be no lower than 90 (10%), but 1 of out 20 is still a lot of glucose errors. As shown by the table, most of the errors beyond 2 standard deviations are associated with the lower standard deviations (e.g., 3 and 4). And to get a hypoglycemic result (<70) requires 7 standard deviations and would be expected no more than once in almost 4 trillion times.

But larger glucose errors are more frequent than this table. The reason is that these larger errors are coming from different glucose error processes; hence the table no longer applies.

The point is that the table should not be used to judge the frequency of larger glucose errors.


Consensus Conshmensus

June 3, 2013

compromise

The new glucose meter standards allow for 2% (CLSI) or 1% (ISO) of glucose meter errors to be limitless. In an upcoming editorial in Clinical Chemistry and Laboratory Medicine, I say why this is a bad idea.

In this entry I talk about the non-technical aspects of this situation. My second hand information says that the 2% CLSI spec was a compromise and part of the consensus process and that some members of the subcommittee wanted to keep the old 5% of the errors could be limitless. Compromise is of course a way of life. You buy a new car and compromise on the price. The problem is with some things you don’t compromise. No one would say that 2% wrong site surgery is acceptable even though having a spec for 0% wrong site surgery will not prevent wrong site surgery from occurring. So to have a bad spec and pass it off as part of a consensus process is a problem. And if nothing else, it makes you wonder if other parts of the spec are the result of compromise.