Open Thread - Most Important Questions

[jwildfire]

What do you think are the most important and challenging questions in safety analytics that our group should address in the coming year(s)?

[jwildfire]

One topic of interested is figuring out how to submit interactive graphics to the FDA.

[jwildfire]

@samussiah It's very much a work in progress, but that's what we're going for in safetyCharts. This draft vignette for v2 goes through some of the details. Any questions/feedback related to the vignette would be great!

[MikeColopy]

This is my first attempt after Jeremy's demo of GitHub. I have a QLIK safety dashboard that handles all lab data and I will be helping to develop its twin using RShiny. How do I avoid R not choking on the huge volume of lab data? Is the trick to focus on specific toxicities such as liver or kidney or CV?

[zskrivan]

We already know how to submit a stand alone interactive HTML to the FDA by including this directly in the submission package (in a clever way so the experience is seamless). We have included interactive data viz in two separate submissions as a pilot for this approach, working with PHUSE. The PHUSE group is currently seeking an FDA sponsor to pilot a submission using containerization.

Do you want to replicate the efforts of the Phuse group? It seems that it would be more productive if we focused on a gap like 'how to validate our interactive visualization used in submission". That could complement PHUSE's effort, rather than duplicate it.

[jwildfire]

How do I avoid R not choking on the huge volume of lab data?

This is a very good question @AlbrechtDurer, and in my experience the answer is complex and varies for different use cases. Focusing on specific toxicities helps, but probably isn't enough in really big studies. In those cases, I think the most important thing is to design a good data pipeline that includes both a database backend (as opposed to loading all of your study data each time you initialize the app) and visualizations that summarize the data in a reasonable way (as opposed to just plotting every single data point in a scatter plot no matter what). Fortunately this is all doable in R, and improvements in this area are our radar for safetyGraphics after v2 goes live.

Note that I've added your question and my answer above to the FAQ section in the data pipeline and security vignette. Hopefully others can weigh in with additional guidance for this area.

[jwildfire]

It seems that it would be more productive if we focused on a gap like 'how to validate our interactive visualization used in submission".

Agreed, @zskrivan! We've discussed reaching out to the folks at R Validation hub to see if they're interested in collaborating on this topic once we're a little bit closer to release.

[anzianorj]

A non-technical topic but still important to ensure these tools get used, would be to discuss our communication strategy.

[jwildfire]

I believe there is a communications team from the full working group that we could collaborate with. Maybe @xni7 or @mli1 could say more?

[xni7]

Correct, @jwildfire , we do have an Outreach and Communication Team (OCT) for the overall Safety Working Group. I'm a member of the OCT.

See http://bit.ly/asa-safety "Who We Are" tab.

Agreed that this would be a good topic, jointly with the OCT team, to come up with the outreaching strategy.

[mli1]

@anzianorj Rich do you have any suggestions or thoughts on how we can do better communication? It is always good to hear fresh thoughts and perspectives:)

[anzianorj]

Apologies, being a newbie to the group I wasn't aware that there was a comms team. Very glad to hear there is, as I think that is where we fell short in previous groups I've been involved with. Ultimately there needs to be a critical mass of users to create a sustained 'pull' otherwise our 'push' will fade over time. No specific suggestions but always good to have posters or presentations at the bigger meetings.

[zskrivan]

One general safety topic is how to visually analyze 'digital health' data. We are getting more and more safety related data that comes for iOT (internet of things) devices. We should lead the industry in how we visualize this data (which will often involve modeling and data reduction techniques).

[jwildfire]

Interesting (and BIG) topic, @zskrivan. I've dabbled with wearable data on a few projects and found that data architecture (database design, etc) ends up being just as important as the visualization techniques, so we may want to keep an eye out for people with that skill set.

[mli1]

How to visualize the three-line (RBC/Hb, platelet, WBC/neutrophil) change with regards to drug interruption/dose reduction? In reality, when a drug is known to cause myelosuppression, it is hard to believe there would be no action taken at all. From an aggregated level, is there a better to visualize the efforts by the drug and the efforts of intervention?

[aslimem]

What do you think are the most important and challenging questions in safety analytics that our group should address in the coming year(s)?

I believe it would be useful to have a base tool developed in R that is also customizable and expandable. Our safety teams have settled on using Tableau dashboard for much of their safety reviews. Although Tableau was not intended for safety reviews, the Tableau folks in our company can very quickly develop dashboard and patient profiles. Using R it takes us a lot longer and it takes some time to make it look slick. The drawback of Tableau is that it isn't linked to powerful statistical tools like R is. A base R app that can summarize safety information and patient profiles but also be expandable so the user can add additional variables would be cool. JM2C

[mli1]

Jerry @tbajn and I just added a chart example (#496 ) where we wanted to address the question regarding visualization of difference between/among the giving categorical factor (subgroups) with regard to the time to a particular AE of interest (or AEs), plus proportion of those who had the AE(s). We understand the AE volcano plot can do a lot, especially with regards to proportion and the hazard ratio or RR , but the current version is not able to address "time to onset" issue/ subgroup analysis?

@xni7 @DOBrien1 Xiao and Dennis I will nudge you here because I know you are working on AE volcano plot. @JimBuchanan @hendricksonb feel free to weigh in as well.

[DOBrien1]

Hi Mengchun Not sure if this is the question is but I can comment that I definitely find the KM value plots valuable. I agree with you. We use then frequently visually comparing for example TTO and number at risk for a given event of interest for tx groups. Can be plotted by subgroups. I regard these analysis as standard for focused analyses. Dennis From: Mengchun Li <notifications@github.com> Sent: Wednesday, February 10, 2021 12:31 PM To: SafetyGraphics/safetyGraphics <safetyGraphics@noreply.github.com> Cc: O'Brien,Dr.,Dennis (MED PV) BIP-US-R <dennis.obrien@boehringer-ingelheim.com>; Mention <mention@noreply.github.com> Subject: Re: [SafetyGraphics/safetyGraphics] Open Thread - Most Important Questions (#479) CAUTION! Please do not click links or open attachments unless you recognize the sender, this email is from an External Sender (outside Boehringer-Ingelheim) Jerry @tbajn<https://github.com/tbajn> and I just added a chart example (#496<#496> ) where we wanted to address the question regarding visualization of difference between/among the giving categorical stratification factor with regard to the time to a particular AE of interest (or AEs), plus proportion of those who had the AE(s). We understand the AE volcano plot can do a lot, especially with regards to proportion and the hazard ratio or RR , but the current version is not able to address "time to onset" issue/ subgroup analysis? @xni7<https://github.com/xni7> @DOBrien1<https://github.com/DOBrien1> Xiao and Dennis I will nudge you here because I know you are working on AE volcano plot. @JimBuchanan<https://github.com/JimBuchanan> @hendricksonb<https://github.com/hendricksonb> feel free to weigh in as well. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#479 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJUQS56YFJRZQ32F77XFUFDS6K7EJANCNFSM4WJODMFQ>. This e-mail is confidential and may also be legally privileged. If you are not the intended recipient please reply to sender, delete the e-mail and do not disclose its contents to any person. Any unauthorized review, use, disclosure, copying or distribution is strictly prohibited.

[hendricksonb]

Agree with Dennis about the KM plots being the standard way we would look at time to onset. The issue with subgroup analyses at filing is that you have to have enough of the events of interest in each of the subgroups you are comparing to be able make a meaningful conclusion. Otherwise you have to wait and look at larger post-marketing database studies

[mli1]

A QT subgroup is having ongoing conversations. Key notes are posted at #511.