IEC 62304 safety classification, External Controls and off-label use related risks

[Ash Watson]

We have some software in development.
Initially we weren't developing this software along 62304, but had performed some of the early 62366 use related risk anticipations to help place the product concept / use specification in less risky waters.
I am wondering about how, when and whether to count some of these in regard to 62304's software safety classifications.
Basically all of these early "controls" I'm considering were applied by limiting features and changing the intended use and indications for use. In 62366 speak they are applied all the way down at the use specification level.
The hazardous situations we workshopped would only arise and convert into harms with some gross, deliberate off-label misuse or use in the wrong environment(actually th wrong institution) and even then by breaking with institutional procedure.

Would a 62304 classification exercise take these preliminary, pre-user need controls into account? Are Off-label uses considered?
I expect the following:
Off-label misuse could ostensibly place software in C
On-label use places the software in A or B

How and what stages do people generally sensibly synergise their 62366 and 62304 processes?

Hope this makes sense and follows forum rules. I did look for an answer for a long time, I promise!

 

[akp060]

Hi, I need a bit more clarity for "controls" and "pre-user need controls".
Q1.

Are Off-label uses considered?

A. Yes, please see subclause 5.6.4 NOTE 1, fifth dashed-point
Q2.

How and what stages do people generally sensibly synergize their 62366 and 62304 processes?

A. For standalone software it would be the beginning of the product's lifecycle
For integrated software, it would be as soon as you begin specifying the technical requirements of the software

Let me know if anything remains unanswered. I would suggest simplifying the context of your questions especially the controls part, and the part you mention about limiting features, changing the intended use, and indications for use; the relevance of this part to the question is ambiguous for me.

 

[Tidge]

The latest revisions to 62304 added some (important, IMO) nuance (*1) to software system safety classification. However, it should be recognized that areas of nuance are those where some potential issues can hide.

To answer this question, in the context of the latest revision of 62304...

Would a 62304 classification exercise take these preliminary, pre-user need controls into account? Are Off-label uses considered?
I expect the following:
Off-label misuse could ostensibly place software in C
On-label use places the software in A or B

... I think you should consider what functions you are actually allocating to software. I am assuming that this is not SaMD; if it is SaMD, all functions of the device are allocated to software, and as such if misuse leads to class C, then class C it shall be!

If the software has been allocated functions that can lead to the most serious harms, or if the software is intended to mitigate the most serious harms, then the software is almost certainly class C.

(*1) The nuance of 62304 (as I see it) is that now it is explicitly possible to allocate device functions to software that don't participate in the risk profile of a device in any meaningful way and not have to default to the most serious level of development effort (as if the software was "doing more"). This option was there previously, via software system decomposition, but under the previous revision it was always expected that there would be some level of software development commensurate with the risk profile of the finished product.

 

[Ash Watson]

Thanks akp060, Tidge, for your great answers.

Hi, I need a bit more clarity for "controls" and "pre-user need controls".

Yeah sorry, I was having difficulty figuring out how to explain this.

What I meant by this is that there was a very early Hazard Scenario brainstorming session which identified some potential use related risk scenarios.
In practice we decided to explicitly exclude certain environments and certain uses from the Indications for Use, Intended Use, and Use specification.... "Pre-User needs" really refers to the order that 62366 prescribes... This control was applied by rewriting the use specification.

As far as I can now see, the risk is mitigated so long as the product is not used in hospitals (it would have too be sold to a hspital for them to get it), and it's not used as a physiological monitor.

I don't see how I can estimate the risk there as the control was applied from the very beginning. I think it's exceedingly unlikely.

For these specific risks I don't see how there could be any feasible software based controls to further mitigate beyond interactive IFU, Instructions etc.

Let me know if anything remains unanswered. I would suggest simplifying the context of your questions especially the controls part, and the part you mention about limiting features, changing the intended use, and indications for use; the relevance of this part to the question is ambiguous for me.

The latest revisions to 62304 added some (important, IMO) nuance (*1) to software system safety classification. However, it should be recognized that areas of nuance are those where some potential issues can hide.

To answer this question, in the context of the latest revision of 62304...

... I think you should consider what functions you are actually allocating to software. I am assuming that this is not SaMD; if it is SaMD, all functions of the device are allocated to software, and as such if misuse leads to class C, then class C it shall be!

If the software has been allocated functions that can lead to the most serious harms, or if the software is intended to mitigate the most serious harms, then the software is almost certainly class C.

I would think these particular harms have been mitigated by restricting the customer base in this case.

Product is a software platform and a third party manufactured class II connected device. It's intended to be Non-device CDSS / Non-device MDDS in FDA view, but there is a desire to get a 62304 certification.

I suppose a qualitative risk estimation would show the likelihood of harm as exceedingly rare. There's no real way to quantify the risk of these particular misuses.

If the risk is exceedingly unlikely and the software wouldn't have a bearing...

 

[Tidge]

Product is a software platform and a third party manufactured class II connected device. It's intended to be Non-device CDSS / Non-device MDDS in FDA view, but there is a desire to get a 62304 certification.

I suppose a qualitative risk estimation would show the likelihood of harm as exceedingly rare. There's no real way to quantify the risk of these particular misuses.

If the risk is exceedingly unlikely and the software wouldn't have a bearing...

62304 is a "development standard" rather than a "product standard"; the prescribed development activities of 62304 arise are determined by risk analysis. You can always take the approach of "We opted not to quantify the risk, so we treated the software as Class C for the purposes of development."

I personally don't want to make any suggestions (in an open forum, devoid of context) in the area of Clinical Decision Support Software. Over the past three decades, the pendulum has swung wildly, and the emergence of apps (as well as the weight of industry heavies behind them) has also had effects that I'm not convinced are settled. As we have witnessed with medical devices and procedures, attitudes can quickly shift.

I am more comfortable offering this piece of advice: If the software is designed to interface with an ME device, you should consider the cybersecurity implications/vulnerabilities to both the software and the device. The 14971 basis of 62304 doesn't exactly apply for non-physical harms (*1), but (medical device, medical software) developers can apply the prioritization concepts to cyber-security risks to drive appropriately thorough development processes as outlined in 62304.

(*1) I'm going to sidestep any discussion on whether or not a 14971-compliant, risk-based approach to medical device development could or should address all cybersecurity risks and simply state that it should be clear that regulatory authorities don't want these areas ignored.

 

[Ash Watson]

62304 is a "development standard" rather than a "product standard"; the prescribed development activities of 62304 arise are determined by risk analysis. You can always take the approach of "We opted not to quantify the risk, so we treated the software as Class C for the purposes of development."

I guess my issue is for these particular risks I'm not sure how to quantitatively estimate, since their controls/mitigants - some deliberately and some discovered on closer analysis of original product concept - were baked into the product concept from the start. So this is probably more of a 14971 and best practices question at this point, and not for this sub-forum. In these cases when drawn out puts our qualitative estimation of probablity very low. Any attempt at quantitative estimates will only make sense after test releases.

I personally don't want to make any suggestions (in an open forum, devoid of context) in the area of Clinical Decision Support Software. Over the past three decades, the pendulum has swung wildly, and the emergence of apps (as well as the weight of industry heavies behind them) has also had effects that I'm not convinced are settled. As we have witnessed with medical devices and procedures, attitudes can quickly shift.

I am more comfortable offering this piece of advice: If the software is designed to interface with an ME device, you should consider the cybersecurity implications/vulnerabilities to both the software and the device. The 14971 basis of 62304 doesn't exactly apply for non-physical harms (*1), but (medical device, medical software) developers can apply the prioritization concepts to cyber-security risks to drive appropriately thorough development processes as outlined in 62304.

This actually clarifies the issue greatly from a decision making perspective and bypasses my original quandry.

Once again, thank you, Tidge