About the talk
R/Pharma 2020 Day 2
Baldur Magnusson (Novartis)
With Great Graphs Comes Great Power
Next, who is coming from the word, he's balder on Munson who will give us a talk on a graph comes, great power, very excited to hear and all their these Pages yours. Thank you very much. I hope you can hear me and see me this to the Flies as well. And and thank you for inviting me to speak at the conference. I'm very excited to be here. Hello everyone. My name is Baldur and today I'm going to talk to you about a topic. That's I find one of the most exciting as a statistician in the farming industry and
effective visual communication. So it connects nicely to Julia's, excellent talk earlier, but I'm, but I'm going to be talking about, it's the graphs that you use to explain what you've done to help others decide. Because I think that our job as a twenty-first-century quantitative scientist, be that a statistician, a model, or a data scientist in fermentation. Etc, is to influence our stakeholders through our analysis of the data. And I think the skill to be able to do that by communicating effectively for the visual, medium is really one of the core competencies for our roles.
Because as the title of the top states with great grass comes, great power. So, I'm passionate about raising the bar for myself in this regard and also doing my bit to try and move the needle for Effective visualization at Novartis and Beyond. So given that I figured it'd be good to remind a little bit and take a look at how I got started in the station. And I'm not sure if this is something, I should be admitting openly. But here we go. About 10 years ago. I managed to complete a PhD in statistics that somehow involve. No, graphing of real data.
It didn't actually involve any real data at all. So you got to love that Siri. That's fine. But then the next day I all the sudden find myself in this highly exploratory environment in early clinical development at Novartis and and for those that aren't sure what that stuff is expensive. Transitioning compounds into the clinic for the first time. So first-in-human healthy volunteer studies and 1st in patient that's because he studies and in that situation it is critical to look at the data literally look at it and let's just say I sense the Gap in my skill-set
and like any self-respecting professional, I turned to Google and I was already in are you sore? So Google suggested tried ggplot2. So I was instantly hooked and actually preparing for this time I think maybe the first code snippet or surely one of the first code snippet that I brought us I was trying to get you flat out and looking at that I think we got me hooked was the combination of flexibility and simplicity that you get from Layered approach, and just a small amount of code, to do something quite decent.
Even for someone, that's just starting out in visualization. So, like I said, I was hooked and not just a GT, but two, but the date of visualization because as it, turns out plotting is fun. but I also quickly learned that it's not enough just to part with abandon We all know about a situation either because we experience it ourselves or because we saw it happened to someone else, wear a presentation, Phelps lunch, a pitch didn't land. We're meeting got completely derailed because of poor data visualization
unfocused plus illegible plots with an unclear message. A bad thought can be worse than no plot and producing. A lot of grass is not the same as effective visual communication. So in this talk, I'm going to share with you a few basic principles around effective visual communication that I think could help Elevate, everyone's graphical game and help us make an even bigger impact with our stakeholders. The principles are good. But doing this well, take some practice Great Grass
while powerful are not trivial. One does not simply just plop the data. But it's clearly possible to develop this skill because effective visualisations are in fact all around us there. The ubiquitous election prediction pots from 5:38 they are the many various types of thoughts are on the crew coronavirus pandemic produced by John, by Murdock and his team at the financial times and then even thoughts from the pharmaceutical contacts. I think this is a particularly effective kaplan-meier. Papa, clearly draws our attention to the
therapy. That works the best in this contest. So what's the issue? Let's look at an example to make things a little bit more concrete. Suppose our project team is interested in, knowing the difference in exposure, between Japanese patients and Caucasian exposure to a drug Where to find the person on the team. So we want to help out. We get our hands on the data. We calculate the summaries of the exposures in two, populations from dosing until sometime. 48 Hours, let me share this with the team. Surely, this is sufficient.
All the data is there, right? Anybody can see We're not meeting, just imagine. This will show the parts and there is silence. And after a while a colleague gently probes excuse me so it's the exposure difference. As we hear that question we realize the team can't tell, they want to know if the exposure is different from that spot, they can't see it. And looking at it it's hard to blame them. The Peaks are a jumbled mess which might suggest that they are similar, but we can't really tell the elimination seems faster in Japanese patients. But
again it's very hard to tell by how much and whether that's meaningful. So what is the team now supposed to conclude? This is a failed opportunity. We had a chance to influence our team potentially Drive immediate decision or action, but our visual it was poorly fall through. It didn't address. The team's question I did left, everybody confused and frustrated and this isn't a rare occurrence. I don't have to look hard at least at Novartis to find more of these examples of grass gone bad. There is an interesting use of the y-axis, interesting, use of the x-axis. And,
and here, I don't really know what we're supposed to conclude. But for me, I find this leaves a lot to be desired for full disclosure. I'm responsible for the first thought on this slide, I signed off on that. It one of my early trials. So you live in, you learn We generate more and more pots every day, but most are generated without thinking deeply about the underlying question and Austin. We are on the side of the house, put the data in there and I'm at the team figured out. And that's a real shame because we all
want our work to have purpose. We all want our work to be understood and we all wanted to make an impact. What if we could do better? And that's a rhetorical question because I believe he can do better. And so, I will rephrase what if we consistently did better? What if our graphs in addition to following good graphical principles or carefully crafted with a particular scientific question in mind, it would just sign for a particular audience and then make them answer. So obvious that there
could be no confusion as to our position based on the analysis. We conducted That sounds like a pretty cool place, right? To me, that sounds like a place. Where are scientific influences elevated guard effect of communication are affected visual communication. So, how do we get there? Now, I will introduce three principles that I speak believe service. A Cornerstone of any effective visual communication. I need to know your purpose. Show the day. Clearly. I make
the message obvious. Purpose Clarity message. Okay, the first just know your purpose. We don't analyze data just for the fun of it. I don't know what everybody does in the evenings. Of course there's tidy Tuesday's weather for Wednesday's Etc but at work at least there's usually a reason for doing it. I just said, if you will go before, doing anything, ask yourself, the following questions, first of all, what is the question? What is the scientific question with a business? Question that
we're trying to answer with his analysis? Are we asking about side effects? Are we asking about a treatment effect? Maybe is it about comparing different options for going forward or comparing exposures in the previous exam. Who wants to know? It's important to adapt to your audience, and it's even more important not to assume that your audience will adapt to you. And so your visual, your communication may look completely different. If you were just showing this to your quantitative Pierce, we are project team, senior decision-makers in the company or
external slight kol Square out the parties. And I sleep. Why do they want to know be clear on the expected outcome? Are we just sharing and Analysis to facilitate a scientific discussion? Are we delivering a message? Are we trying to convince an audience? Are we still attending a decision? Know the answers to these questions before you do anything else. Know your purpose. Next one is show the data clearly, this seems obvious but you'd be surprised at the Innovative schemes. We can come up with to show that a bad late,
just think of the slide that showed earlier. And so the first recommendation here is to simplify simplify to clarify. It's our job to make the complex simple so choose the simplest appropriate graph, keeping in mind your goals and your audience, Next is maximized the signal over the noise and then he's mostly by getting rid of anything that distracts from the purpose or the noise that is colors, shapes and lines. And other attributes. Restart, no choice unless they start a clear purpose.
Unless they show the relevant data directly in the grass. Sometimes this is raw data, sometimes it's summer. And sometimes if we fit on model for inferential or prediction purposes, may actually make the most sense to show the models arrive, Corner gets directly, and ask yourself what's the best way of summarizing the relevant features of the data, keeping in mind, your goals and your audience. Know your purpose and show the data clearly? Tying it all together. After all that work on purpose and Clarity. We do not want to show up
to our meeting and had the audience play, where's Wally, or where's Waldo for the US. People in the audience for five minutes before they kind of figure out what we're trying to say, or before, they just leave in disgust or or or confusion. And that's where the last principle is. So important makes a message, obvious think back to purpose. What's the question? We have identified a question that needs to be answered. So our communication, our visual should show the answer some of the questions about a treatment effect or a difference from Placebo. Show the difference.
For some reason you can't show that directly at least place two quantities are supposed to be compared as close to each other as possible to facilitate that comparison. The second is draw attention, when do repeat yourself. So here, I mean, use contrasting colors, and annotation and other attributes in the graph to draw, attention to where you wanted to be. Add meaningful information to provide context and to tell the whole story. And then do repeat yourself but effectively. So if your particular aspects of the grass that you want
the audience to remember and convey them in more than one place, I consider always adding a title and really not just a description of what's in the point. But phrase it as a conclusion. What do you think? Amethyst pointers. It's not so much that the message is obvious that it's really impossible to miss. Know your purpose. Show the data clearly and makes a message obvious. I do want to address the elephant in the room which is the these principles they seem obvious or like I hope that they do, but the reality is that
they're not a silver bullet. There is no one-size-fits-all solution that's going to work in every situation. So the important thing is that you think consciously about each case and you make conscious choices depending on what's needed. And like I said before, this is a skill that can be developed and so the suggestion is to make it a habit. Don't be afraid to experiment. It'll be ready to sail from time to time but make it a habit test and repeat so let's go back to our previous example, where the question from the team was is the exposure difference and if so how
And supposed to, instead of bringing this graph to the meeting, which which fell flat left, everybody confused. And I'm sure we internalize the principles of purpose Clarity message, every brought this graph to the team. This crap summarizes the 48-hour exposure in terms of three, important metrics maximum concentration. Trough, and AC last, it shows the difference between the Japanese and the patient population, ratios relative to the vertical line at 1 meaning, no difference. It stinks what we think
in the heading exposure differs by ethnicity but then uses annotation to point out that the Peaks are similar like we thought but the elimination is substantially faster in Japanese patient's. Individual Point estimates in case that would be helpful for some members of the team. This is the sixth thanks message graph. Specifically tailored to address the questions the team one and answered. I'm not saying it's perfect. But it's a whole lot better than what we saw before and this one just might get the job done. I sent the beginning, our goal
as 21st century quantitative scientist. Use to influence, teams are stakeholders to our analysis of the data. An influence necessarily involves communication and the most affected medium for that. Communication is the visual It was our friends purpose Clarity and message. We maximize our impact on the elevator influence with our stakeholders and a Spider-Man definitely. One said with great power comes great power. If you want to be more like Spider-Man, I want to leave you with a few resources to check these ideas further out. So we had to get her page on these
principles, or you can check out a video on effective communication and a tutorial that we wrote about it, and he slides are also posted on that getup page. We also have a graphics principles cheat sheet, that you will hear more about tomorrow and Doug Robinson's keynote. And I also want to put you to an open-source library on exploratory Graphics. 4K 3D modeling a very nice resource. And finally, just highlighting as well, a couple of external to cross company initiatives is the wonderful Wednesdays, webinar, a friend by the PSI visualization steak and and then the beasts are in a shit. What
you will hear about in the talk after mine from charlotta. And finally, I would be remiss if I didn't acknowledge many collaborators over the years, Allison, Mark, and mark the ones with the Stars next to their names are the co-author stuff that you're really attractive. Visual communication, principles are many others that I feel very fortunate, and you don't work with over the years. People are passionate about this, just as I am and really it, it takes a village to really move this forward. Thank you. I apologize for the technical issues but I'm at if we have any time, I'm happy
to take questions. Thank you so much Bolder and thank you. Notes are going so smoothly through for you but I have to tell you the check was so people are questions I just asked find about theranos what channel and ask them there but otherwise I'll watch for you can just bring a great presentation. I was giggling. A lot of times behind the camera you can find open all the situation. Thank you very much. Thank you.
Buy this talk
Interested in topic “Medicine, Health and MedTech”?
You might be interested in videos from this event
Buy this video
With ConferenceCast.tv, you get access to our library of the world's best conference talks.