Reconciling FMEA RPN ratings with Risk Acceptability

[Bev D]

There are many flaws with the RPN. Although there is little written about them. I suppose because most of the flaws can be understood through applying the fundamentals of mathematics and science. In the resources section under internal resources you will find an article that I wrote concerning FMEA and the 'score' or 'bright line' yes or no mentality. It also covers the various flaws of the RPN. In addition there are several citations for the source material.

Here a few direct links to the best articles regarding FMEA & RPN:

Wheeler, Donald, “Problem with Risk Priority Numbers, More Mathematical Jabberwocky”, Quality Digest, June 2011. Problems With Risk Priority Numbers

Youssef, Nataly F. and Hyman, William A., “(broken link removed) Applying risk assessment may not give manufacturers the answers they think they are getting”, Medical Device & Diagnostic Industry, October 2009


Flag, John, “Rethinking Failure Mode and Effects Analysis”, Quality Digest, June 2015

Crosby, David, “Words that Kill Quality and Spill Oil”, Quality Digest, July, 2010


Haynes, Alfred C., “United Flight 232: Coping with the ‘One-in-a-Billion’ Loss of All Flight Controls”, Flight Safety Foundation, Accident Prevention, Vol 48. No. 6, June 1991.

Imran, Muhammad, “The Failure of Risk Management and How to Fix It”, Book Review, Journal of Strategy & Performance Management, 2(4), 2014 pp. 162-165


and finally a good Christmas present for yourself:
Hubbard, Douglas W., The Failure of Risk Management; Why It’s Broken and How to Fix It, John Wiley and Sons, 2009

 

[Zero_yield]

Thank you for the resources!

 

[Zero_yield]

Just wanted to say that since reading these, I've had a couple conversations with my team where we've been able to remove RPN as a required tool from a couple procedures. Thanks again!

 

[d_addams]

Bev D,

Thoughts on if this article would be an appropriate reference for a Notified Body reviewer who is asking 'how is it possible that two RPNs values can be treated differently?'

I plan on explaining the RPNs are not intended to be considered an explicit quantitative risk estimation, but I expect I'll want to provide at least one external reference as well.

Youssef, Nataly F. and Hyman, William A., “(broken link removed) Applying risk assessment may not give manufacturers the answers they think they are getting”, Medical Device & Diagnostic Industry, October 2009

 

[Bev D]

Donald Wheeler is a pretty well recognized source. He is a PhD in the topic. His article is the most cogent one out there. While the article you cite is excellent, there is no reason that a believable article has to be published in a medical device journal to be believed. You might cite both articles.

I struggle with this one as it really is simply a matter of mathematics and knowing the actual rules. Who invented math? I know there are books out there that describe the basic rules that support mathematic operations - any basic mathematics book should have them. Donald Wheeler clearly outlines these rules in his article - he didn’t make them up. Youssef and Hyman also lay out the case fairly well. Both use irrefutable examples…

 

[d_addams]

Donald Wheeler is a pretty well recognized source. He is a PhD in the topic. His article is the most cogent one out there. While the article you cite is excellent, there is no reason that a believable article has to be published in a medical device journal to be believed. You might cite both articles.

I struggle with this one as it really is simply a matter of mathematics and knowing the actual rules. Who invented math? I know there are books out there that describe the basic rules that support mathematic operations - any basic mathematics book should have them. Donald Wheeler clearly outlines these rules in his article - he didn’t make them up. Youssef and Hyman also lay out the case fairly well. Both use irrefutable examples…

Thanks. I agree with your thoughts. I didn't suggest the Wheeler article in this case since it was behind a registration wall (which a reviewer may not appreciate) vs the MDDI article.

 

[Bev D]

Hmm. OK. The registration is free. The MDDI article is also free. Neither is peer reviewed. Although peer review of any article supporting basic mathematical fundamentals is probably laughable.

 

[Tidge]

Bev D,

Thoughts on if this article would be an appropriate reference for a Notified Body reviewer who is asking 'how is it possible that two RPNs values can be treated differently?'

I plan on explaining the RPNs are not intended to be considered an explicit quantitative risk estimation, but I expect I'll want to provide at least one external reference as well.

I recommend against dropping a third-party article on an NB reviewer, certainly NOT one on basic math. RPN itself is a simple multiplication(*1), and an NRTL (for 60601 at least) will need to see things explained per your established risk management process, not something found in someone else's library.

I spent a LOT of time with representatives of NRTLs (in medical device industry) that don't grok modern concepts of risk management, but the QMS registrars are usually better about these sort of concepts.(*2) RPN is a crude, zeroth-order mechanism for risk controls, especially when human health effects are under consideration.

(*1) A typical RPN = (S)everity x (D)etection x (O)ccurence; RPN, S, O, D are typically all positive integers. Basic math reveals two peculiarities:

• Some integer values of RPN are unattainable
• It it possible to get the same values of RPN for different values of S, O, D

Risk management implements controls to modify (typically "improve by lowering") individual values of O, D (and sometimes S), which result in new (typically lower) RPN. If a risk management policy doesn't provide direction on what "improving an RPN" really involves (or means) then the NB has asked a legitimate question IMO.

(*2) Risk Management Files require a risk controls options analysis (RCOA). In a QMS that includes FMEA only as subordinate documents (typically to a Hazard Analysis) I disagree with the common NB approach that each FMEA requires an RCOA, but in the scenario described in the quoted text the RCOA for the document with the RPNs would answer the question 'how is it possible that two RPNs values can be treated differently?'

 

[Bev D]

Which is why I always recommend against any kind of RPN or even S*O. It is simply mathematically incorrect. You cannot multiply ordinal data. And ordinal data that are based on a guess and not a test (actual data) are even worse. It is fake math. Period.
to show improvement in reducing the occurence or improvign the detection show the before and after data from your experiments and MSAs…simple

 

[Ed Panek]

If you were trying to obtain insurance on the use of your product, the insurance agent would not place much significance to your RPN numbers. RPN values dont communicate much except an internal type of ranking of risks and hazards but from a real world implementation they are completely abstract.