Louise Pryor, Colin Dutkiewicz and Krishna Kumar Shrestha reflect on the ICAT’s recent R number hackathon, and what it means for the profession’s future skillset and role in society.
As part of the Actuarial Innovation in the COVID-19 Era webinar series, the IFoA COVID-19 Action Taskforce (ICAT) invited members to take part in an R number modelling hackathon – the IFoA’s first hackathon.
What does the hackathon mean for the profession?
Louise Pryor: I was thrilled to see the hackathon emerge as an ICAT event, and it has certainly lived up to expectations. Like ICAT as a whole, it has provided opportunities for members to be involved in IFoA activities, especially students and others who are near the beginning of their careers. As a short-term, highly focused event, it was exciting and dynamic – but there’s more to it than that.
The hackathon provided an environment in which participants could try out new things without risk – there weren’t going to be problems with clients or employers if anything went wrong. Waiting to be told what to do wasn’t going to work; participants had to work out possibilities for themselves and take the initiative. They had to think of new things to do with the actuarial toolkit – things that they hadn’t necessarily come across before in their studying or day jobs – and really get to grips with practical aspects of data science. It was an ideal learning environment.
Moreover, the hackathon provided an opportunity for participants to try out new ways of tackling problems in a team. Team members had to work together to brainstorm and critique ideas, rather than rely on experienced experts. The more opportunities we get to flex our creative muscles, the more likely we will be able to use them throughout our careers.
Why the R number?
Colin Dutkiewicz: We wanted the ICAT work to be dynamic and unconstrained, and COVID-19 provided an environment of rapidly emerging information that required complex analysis and also involved significant data uncertainty. We wanted to emphasise the key actuarial skills of critical data evaluation, complex modelling, professional analysis of results, and clear communication of the consequences to a non-actuarial audience.
The hackathon question, then, needed to involve something that hadn’t been so widely studied that one team might have done the work already.
This is how we settled on ‘the determination of an R number’ – a part of our daily conversation that few people had heard of before 2020. This is a difficult thing to do. The data is skewed and biased, and sometimes plain wrong. The analysis to get to the number must deal with significant heterogeneity. We supplied contestants with a link to an Actuaries Response Group paper on super-spreaders, introducing them to the idea that an R number is an average of very different outcomes. And to make the event maximally topical, we incorporated the emerging crisis in India at the last minute.
Why did you take part in the hackathon?
Krishna Kumar Shrestha: The four of us – myself, Indira Aryal, Isha Tacchekar and Rohini Khanal – share an excitement for research and are always looking for opportunities to put our skills to the test. When we found out about the hackathon, we instantly formed a team. The hackathon had a somewhat unusual set-up, though: instead of being together, teeming with enthusiasm for the five-day challenge, we were in front of our laptops in our rooms, under a nationwide lockdown in our home country, Nepal.
How do you go about modelling an R number?
Krishna Kumar Shrestha: The topic was challenging. We had only seen the ‘R number’ in international news headlines; we were not aware of its value in Nepal, or how it could be used to inform real-life policy decisions.
We started with research. Each of us individually studied how the R number was being used by developed nations to implement lockdown measures, and subsequently shared our findings with each other. During our research, we discovered a paper that discussed the modelling of the R number, ‘Estimating the time-varying reproduction number of SARS-CoV-2 using national and subnational case counts’ (bit.ly/Wellcome_Rnumber).
The method used in this paper is implemented in the R package EpiNow2. As we were familiar with the R language, we decided to do our analysis in R, and based on this paper. We also found data in the John Hopkins GitHub repository that exactly suited the requirements of our model. The data was high quality and consisted of daily cases, recoveries and deaths.
“The hackathon provided an environment in which participants could try out new things without risk”
What did you find?
Krishna Kumar Shrestha: Our results gave a startling R number of 1.4 for Nepal. This led us to question Nepal’s lack of social distancing guidelines, and how this might have contributed to recent alarming rates of transmission. We were also concerned that 1.4 could be an underestimate, given that, in Nepal, testing was primarily for individuals showing symptoms – and many people with symptoms chose to quarantine without being tested.
After digging deeper, however, we realised that the R number did not tell us the full story, and the situation might not be as alarming as an R of 1.4 might suggest. For example, a single R number cannot tell us the difference between the infection rates in a hospital, a care home, a workplace and the wider community. A simple drop in R doesn’t necessarily mean a reduction in the number of infections, and vice versa.
What was the outcome of the hackathon?
Krishna Kumar Shrestha: We produced a communication paper, which included recommendations to policymakers. This paper focused on how the R number could help control the recent rise in Nepal’s COVID-19 transmissions. So far, the only measure considered by policymakers has been a strict lockdown. However, given the negative impacts that past lockdowns have had in Nepal, we also researched and recommended alternative measures. These included quarantine for infected people or people returning from other countries, more localised restrictions (for example smaller lockdowns, or controls on mass gatherings), more stringent safety measures and mass testing.
The submission of our project gave us a sense of relief. We did not have the slightest expectation of winning – we were delighted just to have participated in this challenge and to have had this truly unforgettable experience.
What made the winning team…the winning team?
Colin Dutkiewicz: The team from Nepal really engaged with the topic. They had to deal with problematic local data and grapple with the fact that any Nepali action was likely to be overpowered by events in India due to the volume of movement of people and resources between these countries. The team’s efforts clearly illustrated the difficulties of determining and interpreting an R number. They communicated the complexity and the results very well, and even made the next step of suggesting courses of action based on the outcomes they modelled.
Louise’s Final thoughts
As I said in my recent presidential address, we all have to keep learning throughout our careers, and the hackathon provided a good introduction to that.
Importantly, the hackathon, as part of ICAT, was also directed towards contributing to the society we live in. COVID-19 has affected all parts of the world and all sectors of society, and it has been great to see members of the IFoA from so many different countries coming together to work on similar problems with such broad applications.
Many thanks to all those who organised the hackathon, and to all those who took part. I hope we can put on many similar events in the future.
Louise Pryor is a sustainability actuary and president of the IFoA
Colin Dutkiewicz is global head of life at Aon Reinsurance Solutions, chair of the IFoA Life Board and a member of the ICAT
Krishna Kumar Shrestha is a student at Tribhuvan University, Nepal, a student member of the IFoA, and led the winning team of the ICAT’s R number hackathon
