Validation without a golden standard?

Hi everyone! Do you have any ideas of the way, by which one may validate a medical machine without a golden standard? Thank you in advance!!

Sorry for the delayed reply, been away for a bit.

Can you provide a bit more information? By “medical machine” do you mean a medical device? What kind of requirement are you trying to verify?

Yes, it is a medical device. More specifically, it is a bioimpedance device. It is similar to biofeedback devices.

Ooh… that does sound like a tough one. Without knowing details (which are probably proprietary), I’ll just toss out a stream of consciousness…

So I take it that there’s some kind of stimulus (device output) and some kind of sensor reading (input back to the device) and you need some way of determining if the output + a hugely variable application (skin?) gives the expected reading?

Would it be possible to come up with some standards that when given a known stimulus (output) value result in a known sensor reading (input)? Maybe come up with a range somewhat aligned with the expected intended applications? At least that would allow you to test the range of values and by inference, you could say that anything in the range is good?

Otherwise, maybe just validate that the outputs are in the intended range, the sensor can receive the expected input range, and do some kind of study to say that it works as expected.

From a formal Verification perspective, the stronger case would be to have some kind of test fixture(s) to functionally show the device does what is expected. From a Validation perspective, though, if you can gather clinical data that shows it is (safe and) effective, maybe you can take a less rigorous approach on the verification side.

If you can share more, maybe I can toss out some other ideas (especially if I’m way off base with the initial assumption).

There is a standard PA (Phase Angle) value.(That is according to the manufacturers). If PA>1, it means that there is a metabolic disturbance. Does it help?

The real question is : Is cluster analysis the only way to validate it?

Just can’t say without a better understanding of the device and the technology. There’s usually more than one way. You have to be able to defend the test results (objective evidence of conformance to requirements) so keep that in mind with whatever you do.