Julia Meeting Regulatory Standards?

Not sure where this post belongs - or where it might go - but I have an interesting question maybe for @StefanKarpinski or someone really involved in the language.

So a lot of people in biostatistics, pharma, etc, have to use tools accepted by certain regulatory agencies. In the USA that’s the FDA and probably a dozen others for other countries. So this leads to a bunch of really smart people, stuck using things like SAS (specific formats for datasets yadda yadda, don’t get me started on syntax).

Is there any goal/hope/effort to get Julia into these kinds of development spaces? If so - what might it look like? I think Julia has serious potential to improve the quality of the lives to the people stuck in those confines, also their products, and potentially brandish a new breed of people who would contribute back to the community here.

This would imply some kind of certification for the developing process of Julia (and eco system of packages), e.g. certification according to ISO 8402:1994 (taken from https://www.fda.gov/regulatory-information/search-fda-guidance-documents/general-principles-software-validation).

My guess is: this is impossible for Julia in general.

What is thinkable is a company which would go through the process of certification for a specific Julia version and a set of specific packages, verify and validate (and more) exactly this setup and provide it as a product for significant amount of money. If this product/setup is able to compete on the market with e.g. SAS is highly questionable, but maybe I am wrong here.

2 Likes

Are you at ACoP right now? We can come talk to you directly. The Pumas project (Pharmaceutical modeling and simulation, which includes NLME, NCA, QsP, PBPK, etc.) is working directly with Joga, and JuliaPro has already been approved and installed in the FDA for testing. So this is already happening, and if you want more details it might be easiest to just talk in person if we’re at the same conference :slight_smile:.

11 Likes

Oh man this is awesome! Great news Chris! I’m not at ACoP - haven’t worked out when and what conferences I can attend for my new employer, yet. Keep us in the loop, a lot of people would really benefit from this. JuliaPro is a perfect avenue for this effort.

Let me know if there’s anyway I can help. If any deadlines come up and you all need some extra hands (for free I can’t accept $) maybe I can help.

2 Likes

Sure thing. You may want to take a look at:

https://pumas.ai/

3 Likes

My understanding is that situations where one might use SAS are not nearly as strict because the software isn’t operating on a medical device. See this https://blog.revolutionanalytics.com/2012/06/fda-r-ok.html#targetText=In%20Jae’s%20experience%2C%20"R%20use,%2C%20reporting%2C%20and%20data%20visualization.

1 Like

Any insights into quality control/assurance processes?

It would be great if something like this would be available for Julia https://www.r-project.org/doc/R-FDA.pdf .

1 Like

If youre referring to quality control of the statistical component in evaluation of a clinical trial then I couldn’t imagine it would be hard for Julia to demonstrate this is because most stats at that level are very well standardized. They do a lot of vaccine testing at my University and apparently some of these trials are so standardized that the paper is almost already written. You just have to plug in the numbers and it will tell you if the flu shot is valid this year.

1 Like

I’m more so referring to the analysis of the state of pharma products/materials as they are being manufactured or shipped. But clinical trials are also super fascinating.

If you read most of that it’s clear the focus is mostly on medical devices. It’s easy to feel paranoid about what you use for an FDA trial because of all the regulation but you have to think about the context they care about. If you’re designing an artificial pancreas that monitors glucose and determines how much insulin to release then you’re under a microscope. But if you’re calculating the relative risk after a trial there’s a lot less risk involved.

1 Like

Propably not, but OP refered to SAS and depending on what is to be done with Julia it could be quite expensive to achieve required certifications or, as others pointed out, Julia is already ready to use.

Actually we do exactly that: clinical studies and diagnosis. We are certified after DIN ISO 17025 and it is fine using R (and Julia) in the analysis pipeline which is base of the diagnosis classifications done after base analysis.
As long as the software is verified after above norm.

4 Likes

I don’t want to steer this thread off topic, but if a drug gets introduced to the market because the statistical calculations were wrong, you’re putting lifes at risk too. In the end the developers/statisticians who build the applications or analyse the data need to make sure that their work is solid - no pre-written document will change this need.

Yes of course! Integrity and accuracy is much more important than using a great programming language. I’m just making the point that the requirements to validate a tool for FDA analysis isnt nearly as difficult as one that will be actively used in a medical application everyday. You have to prepare for every possible bizarre event if it’s actively used in on people. That’s why we now have up to 4 phases of clinical trials.

Any updates? :smiley:

It’ll be a long process. It is installed in the FDA but more will take awhile.

2 Likes

@ckneale you might be interested in @sbromberger talk from JuliaCon

3 Likes

Thanks Chris! I know it’ll be a while I’m just an eager beaver.

@Chakravala super relavent talk I’ll have to watch later!