Ironbarque

Medical Device Consulting Regulatory Quality Complaince

The RPN Expansion

We have all used them.  We jump through the hoops of the FMECA process. Misusing RPN scales can lead to dangerous errors and risk blindness.

The Basics

The Risk Priority Number is used in Failure Mode Effects and Criticality Analysis to help assess the concern associated with a particular failure (the criticality of the failure).  The higher the RPN number, the greater the concern. 

Some RPN scores are obtained by considering the product of the relative probability of a failure (Probability), the relative severity of the hazard of a failure (Severity) and the relative chance that the failure will be detected prior to causing harm (Detectability).  This type of RPN expansion commonly assigned values between 1 and 5 for each score with increasing score indicating a worsening case.  The resulting product can then be in a range of 1 to 125.

RPN = Probability x Severity x Detectability

So Far So Good

Once you have the score, then the higher the score the more concern.  This technique came out of World War II military procurement programs[1].  It seems eminently logical:

big score = big risk so do more work on mitigating the failure;

little score = little risk so do less work on mitigating the failure.

Issues

When using RPN Expansion for the development of medical devices there are several issues that must be addressed.

Misusing RPN scales can lead to dangerous errors and risk blindness.

Non linearity

This scale is not linear.  Take the example of a 5x5x5 RPN expansion; one in which Probability, Severity and Detectability are all scaled from 1-5. We do not get a simple one instance for one value for each value across the range of 1 to 125. We cannot because it is a cubic expansion: 5x5x5. All multiples of prime numbers above 5 (7, 11, 13, 17 etc).are not represented at all.  The distribution of possible scores for the full expansion is presented in the image below.

As per the central limit theorem, the histogram is left shifted. Note the many missing values.

One simply cannot say that a value of twenty eight is twice as risky as a value of fourteen in a 5x5x5 RPN expansion. Neither number is represented; both value cannot be reached in the expansion.

Gaming

No one really goes home to tell their family - "Hey. today I made a decision that will really hurt patients and the Company in a few years." However, we do find ways to bring late programs back onto schedule or to reduce program testing costs. To do this, we reduce validation samples sizes justifying the change based upon "Risk". Reduce one or two of the components of an RPN score and the compound effect on the score moves a device from an high risk category into a not so high risk category.  This is "gaming" the RPN system. Especially when using attribute based test statistics, the drop in sample size can be signifiacnt.

Reducetion in Reliability at 95% Confidence
Change in sample size
Work Load Reduction
99% -> 95% 300 -> 59 80%
95% -> 90% 59 - 29 50%

Even when using normal data sets, changes in required certainty allow device design elements to pass their required validations, problem solved.  In both cases, the reduction in work is the avoidance of a design iteration.

A question for you: which would you rather see 99% confidence of 95% reliabiility, or 95% confidence of 99% reliability? And why?

Medical Judgement

RPN scores are best when based upon well documented literature reviews conducted with strong, and unbiased, medical review.



1 For exampel see: milstd1629 - 1980