R Medicine 2020: The Power of Going Virtual

The third annual R/Medicine conference was planned as a physical event to be held in Philadelphia at the end of August 2020. However, a nationwide lockdown induced by the COVID-19 pandemic required a swift transition to a virtual conference. This article describes the challenges and benefits we encountered with this transition and provides an overview of the conference content.

Elizabeth J. Atkinson (Department of Quantitative Health Sciences, Mayo Clinic) , Peter D. Higgins (Department of Internal Medicine, University of Michigan Health) , Denise Esserman (Department of Biostatistics, Yale School of Public Health) , Michael J. Kane (Department of Biostatistics, Yale School of Public Health) , Steven J. Schwager (Cornell University) , Joseph B. Rickert (R Consortium, RStudio) , Daniella Mark (Procogia) , Mara Alexeev (Department of Biomedical Informatics, Boston Children’s Hospital) , Stephan Kadauke (Department of Pathology and Laboratory Medicine, Children’s Hospital of Philadelphia, Perelman School of Medicine, University of Pennsylvania)
2021-06-01

Introduction

R/Medicine is an active working group of the R consortium, with the goal of supporting R users in medicine, medical research, and related health fields. The group includes a mix of statisticians and physicians who use R regularly in their work, as well as members from RStudio and the R Foundation.

The 2018 inaugural R/Medicine conference was a two day event held near Yale University with Rob Tibshirani providing the initial keynote address; each day began with a two hour tutorial. In 2019 the conference was held in Boston and included, among others, keynote addresses by Terry Therneau and Frank Harrell. Two half-day workshops were held on the day prior to the conference.

In 2020, the R/Medicine Working Group organized the third annual R/Medicine conference which was originally planned to be held in Philadelphia. By early April 2020 it was clear that the conference would need to be held virtually. This article describes some of the challenges as well as unexpected benefits we encountered when pivoting to a virtual conference, with the hope of providing useful information for other conference organizers.

Conference Planning and Logistics

Initial planning for the conference began in January 2020. By February, we had identified keynote speakers and workshop leaders, and started negotiations with a hotel near the Children’s Hospital of Philadelphia which was pegged as the site of the conference. However, by the beginning of April we realized that we would need to switch to a virtual event.

The initial challenge for the planning committee was identifying technology that would allow us to run a main session of presentations, teach short courses, and host Birds-of-a-Feather (BoF) activities that allow like-minded R users to network and socialize. Because of our limited budget we had limited choices for an online conferencing platform. With help from the Linux Foundation, which provided recommendations of some of the available conference platforms, we chose to use Crowdcast (https://www.crowdcast.io). This tool was simple to use for the attendees and the organizers, had a green room to line up speakers, an interactive chat feature, and the ability to automatically provide a recording of the session for replay immediately after its conclusion. Compared to other conference platforms, Crowdcast has a rather minimalist set of features. However, feedback from conference participants confirmed that the user interface, consisting of a single-track video feed and sidebar participant chat worked well. One downside of Crowdcast was that it did not support screen readers or closed captioning.

For the short courses we chose Zoom (https://zoom.us) as the delivery platform, primarily because of its widespread use and because of a mature breakout room functionality. We found that breakout rooms are a useful tool for teaching, but should be used sparingly, because the overhead of sending participants into breakout rooms can interrupt the flow of teaching the material. For the Introduction to R for Clinicians course, we had a meet-and-greet breakout near the beginning of the course as well as a final “Hackathon” breakout in which participants were able to team up to design a dashboard.

Often, the most effective way to teach programming topics is live coding (Wilson 2019). Much effort has been devoted to create teaching environments that allow participants to live code during a short course without requiring installation of any software. In the R ecosystem, the most notable and laudable effort in this regard is RStudio Cloud (https://rstudio.cloud), which is based on the RStudio Server Integrated Development Environment (IDE). From previous experience teaching R to an audience of clinicians who have no programming experience and are unfamiliar with the R ecosystem, we knew that the RStudio Cloud environment could be challenging because it requires each participant to first register a personal account on the platform before being granted access to the training environment. To address this, we created a training environment with pre-configured generic sequentially numbered training accounts (e.g., train001, train002, etc.). After receiving training account credentials, a participant could directly access their training environment from a web browser without any further sign-ups. The training environment was implemented as a Docker-based web application based on RStudio Server which was configured with custom R packages and exercise files and hosted on a cloud provider. The application is available for free online (Kadauke 2020c).

Zoom was used for the BoF sessions, and we used Slack for back-channel communication between the organizers, participants, and presenters. Some Slack channels were created for participants while other channels were limited to the conference planners. After the conclusion of the conference, all the recorded presentations were loaded onto the R/Medicine 2020 YouTube channel which is hosted by the R Consortium (2020). Since September 13, 2020 when the videos were uploaded, there have been over 3,200 views. Registration and abstract submission were facilitated by infrastructure provided by the Linux Foundation. Google Sheets were used to centralize organization of the program and the list of volunteers. We used Google Docs for creating conference planning documents and Google Forms for course sign-ups and surveys.

Since the @r_medicine Twitter account has a large following (2,557 as of June 2021), we used Twitter as the main outlet to promote the event. We also directly engaged specific groups that we wanted to participate to improve the conference experience, including R/Ladies, Minorities in R, as well as medical professional organizations including the American Association for Clinical Chemistry (AACC), and Mass Spectrometry & Advances in the Clinical Lab (MSACL).

Volunteers

The number of volunteers needed for the virtual conference was significantly higher than for the previous in-person R/Medicine conferences. For each short course we had at least one teaching assistant for every 10 participants. Additionally, we had someone monitoring the Zoom chat to escalate issues when necessary.

For the main session we had two moderators who alternated between speakers in order to orient the upcoming speaker in the green room. We also had a Crowdcast host who helped behind the scenes and moved participants from one presentation to the next. One person was also in charge of playing any of the talks that were pre-recorded. Each BoF room had a facilitator. Because this was a virtual conference, we also identified back-up volunteers in case there were power outages or issues with internet access. Pre-conference training and discussion sessions were held for short course teaching assistants. Pre-conference technical checks were held prior to the conference for the speakers, instructors, and volunteers.

The Conference

In the previous two years, attendance at each conference was around 150 participants. In 2020, 587 people registered for the conference, including 225 students, 185 academics, 104 from industry, and 73 who were working on COVID-19 related projects. Of these, 32% were from outside the United States coming from 43 countries (Figure 1). Of those who registered, 452 attended at least some sessions live and 431 watched some sessions via the replay feature in Crowdcast. The main conference was held over two days and included 4 keynote addresses, 16 regular talks, 11 lightning talks, and one panel discussion. The schedule included three BoF time slots, each with 4-5 different topics.