Detection in FMEA - why it should not be included in a Pareto ranking - 2/2005

An early FMEA guideline (1), states that numerical rankings should be given to two properties of a potential error event: 1) the severity and 2) the probability of occurrence. These two numbers are multiplied together for each event in what is called a criticality analysis and the resulting list of potential error events is sorted by descending criticality in what is commonly known as a Pareto chart (or table).

In some recommended ways of conducting a FMEA (2), the likelihood of detection of the error event is ranked as a third property of an event. Thus, the most severe, most probable, and least likely to be detected potential error event is the highest ranked event. This ranking is called the risk priority number.

The inclusion of likelihood of detection is not recommended. First, in an error model, one can postulate that severe events are the result of a cascade of prior events (see near miss essay). This cascade may be terminated by a detection / recovery combination. So if one were to include detection, one should also include recovery. Whereas, one might think that recovery always occurs and is successful – this isn't the case. Another reason for excluding likelihood of detection from the Pareto ranking is that both detection and recovery are often process steps, meaning that they themselves are potential error events. In any process step, one has already asked about the probability of occurrence. Hence for a detection process step, asking about the probability of occurrence of a detection error event makes things very confusing since one would be asking:

• the severity of an error in the detection event

• the probability of occurrence of an error in the detection event and

• the probability of occurrence of detecting error in the detection event

There are other problems. One can envision a case where an event has a higher risk priority number due to the detection ranking. This could result in lower severity events to be ranked higher than higher severity events, which is illogical (e.g., given the same probability of occurrence, injury could be ranked higher than death):

Recommendations

All of this does not mean that detection isn’t important – it is – it just shouldn’t be included in the Pareto ranking. One can and should ask about whether a potential error event is detectable. However, looking at an anatomy of a potential error event (see near miss essay), error events have a detection / recovery sequence and both detection and recovery can be process steps. Reducing risk means reducing the likelihood that the effect of an error event will occur. This can be accomplished by:

• reducing the likelihood of an error event

• adding or improving a detection step for that error event

• adding or improving a recovery step for that error event

References:

1. Available at http://www.sre.org/pubs/Mil-Std-1629A.pdf
2. See for example: http://www.ihi.org/IHI/Topics/PatientSafety/MedicationSystems/Measures/Risk+Priority+Number+%28from+Failure+Modes+and+Effects+Analysis%29.htm

Reference added 4/9/05

3. Schmidt MW. The Use and Misuse of FMEA in Risk Analysis. Medical Device and Diagnostic Industry 2004 p56 (March), available at http://www.devicelink.com/mddi/archive/04/03/001.html