I received an email recently that alerted me to three seminars from the 2016 AACC meeting that are online. One is by the Westgards, so I had a look. This is quite an interesting presentation and shows the breadth of the contributions that the Westgards have made to quality in laboratory medicine.
Yet, one thing caught my eye and so here are my comments. Thus, the Westgards complain that in risk management as espoused by CLSI EP23, detectability has been omitted.
What they mean is that for each failure event, EP23 wants one to estimate the severity and probability of occurrence of that failure event. The Westgards suggest that the detectability of the failure event needs to be assessed as well and state that this is how industry does it.
Well maybe some industries, but I worked in industry and our company did not use detectability (we used severity and probability of occurrence).
Now in the context of EP23, I agree with the Westgards use of detectability. The problem is that EP23 itself is a poor adaptation of risk management. I commented on this before but here it is again.
As an example of a failure mode of a process step, assume that the failure is sample hemolysis which occurs during the process step to turn a whole blood sample into serum. As you go across the rows in an EP23 style risk analysis, you might see that a mitigation for this failure mode is to visually check whether the sample has been hemolyzed and how effective this check is. In this case – for this row item – you could add detectability to severity and probability of occurrence.
Here are the problems with this approach, whether you have added detectability or not.
For most labs, this (example) is already established laboratory practice. That is, labs already check to see whether samples are hemolyzed. All that has been done is to document it. Not much in the way of formal risk analysis has been done although there will be some benefit to this documentation.
The problem is that the row is “collapsed.” It really has two additional process steps embedded in it. Here it is uncollapsed:
Process step – process whole blood into serum
Detection step – examine serum for the possibility of hemolysis
Recovery step – if the serum has been hemolyzed, request a new sample
One can see that it makes no sense to ask for the detectability of a detection step.
I note in passing that one of the most important detection process steps for any assay is running quality control.
Note that each of these steps above are process steps and each can fail. Whereas the severity will be the same for the failure for each of these steps, the probability of occurrence may differ. Because each step can fail, one needs to assess whether a mitigation step is required.
BTW, one should not discount failures in the recovery step. In the Challenger accident, engineers warned about the potential problem (detection) but delaying the launch failed (recovery). And of course, recovery steps are only performed if detection steps detect something.
Disclaimer – I may not have the latest version of EP23, but another problem in developing the potential failure modes (EP23 call these hazards) is that the process is not fully delineated – it is too high level. In a more traditional FMEA, the list of process steps is long and reflects what is actually done, not some high level description.
And each process step can fail in multiple ways. EP23 is a hazard based list. A process based list is better since one can ask how else each process step can fail. Although EP23 does some of this, it’s embedded within a row and makes things too complicated. Here’s an example of a few ways the above detection step of examining the sample for hemolysis can fail:
- Technician simply misses seeing a hemolyzed sample (non-cognitive error – we all make them)
- Technician misses seeing a sample due to inadequate training (cognitive error)
- Technician misses seeing a sample due to distraction (phone call, or talking to colleague).
- Technician ignores hemolyzed sample due to pressure from management to process samples.
On a separate note, how does IQCP help one modify QC frequency?