This article was first published on Mashed Calculus and Differential Potatoes on 6 November 2021.
Dr Andrew Chen
Earlier this week I presented to some government officials about technology for contact tracing from an outside perspective. I’m not in the health/govt system, and so can only provide comments based on the evidence that I can find from OIA requests and publicly published data sources. There is a reasonable chance that my perception of what is happening is different to what is actually happening in the health system and in the contact tracing / case investigation process, so take my comments with a grain of salt.
The question that I’ve been asked is “how could technology better support contact tracing [as the case numbers increase and we head towards a new public health strategy]?”
Requirements
An engineer starts with the requirements – firstly what is it that we are wanting to achieve? We need to physically isolate infected persons to cut off the chains of transmission. To do that, we need to know who an infected person has been physically proximate to. Since people are sometimes around strangers or can’t remember who they were near, we can use location as a proxy.
What else informs the requirements? The overall strategy is important – what is your tolerance for false positives (isolating a person who doesn’t have COVID-19) and false negatives (leaving a person with COVID-19 in the community)? The context is changing as we move to there being more daily cases while opening up the country as we reach high vaccination rates.
Time appears to be the most important factor – time-to-isolation is modelled to be one of the strongest influencers on the reproductive rate of the virus. But what are requirements need to be considered? Which might be dealbreakers? Flexibility? Equity? Interoperability? Privacy? Social Licence? The technology is theoretically available to track every person with a cellphone in real-time, but that would strongly infringe on the public’s rights. We cannot optimise only for one factor.
Automation is key to speeding things up, which relies on a good understanding of how processes are deployed today. To help us with this, let’s split the tools for better contact tracing into two categories: contact tracing process automation (i.e. “business processes”) and consumer-facing digital contact tracing (i.e. “the app”).
Business Processes
The contact tracing processes are opaque to the outside world. Even after all this time the public does not have a great understanding of exactly how contact tracing works and what the protocols look like. In addition to this, we don’t know how much deviation there is between the protocols and the real-world case investigation process. How much can we speed things up without losing too much accuracy, noting that we are likely already losing accuracy because humans are not perfect?
It appears to me that the capacity constraints in the contact tracing system are critical now. We have seen this in recent days with statements that prioritisation of locations of interest by risk-level is now necessary. Even where digital contact tracing tools are providing data from the case, it is hitting the bottleneck of human contact tracers making phone calls. We could try to add capacity to the system, but even there we find bottlenecks. Training is an important part of adding more contact tracers, but there are people who are volunteering to become contact tracers who seemingly can’t get training right now.
I am not advocating for a fully automated contact tracing system – the human element is very important for making people feel at ease and to ensure that their needs are met, especially at the point when a case is first notified that they have tested positive. But there are parts of the system that could be automated. For example, right now a contact tracer has to talk a case through the process of going into the NZ COVID Tracer app and typing in a code to upload their diary to the contact tracers. In some other jurisdictions, this is sent out as a text with a link to further instructions, which takes up less contact tracer time.
To my understanding, the contact tracing / case investigation protocols are currently being redeveloped. I think part of this needs to include an honest look at which steps could be automated, in consultation with clinical advice, to make contact tracing more efficient and therefore scalable.
Consumer Digital Contact Tracing
We can look at the digital contact tracing tools that we try to put in the hands of all New Zealanders. When we look overseas, our tools are pretty much at parity – more jurisdictions are moving towards contact tracing records (e.g. QR codes) and moving away from Bluetooth (because they can’t get participation rates up high enough). But there isn’t a lot of use of other technologies like GPS or cell tower geolocation.
But instead of looking at the other technologies (which require a different trade-off with other factors), we should look at the technologies we already have. We are not using Bluetooth Tracing (BT) nearly as much as we could or should. There is something happening in the contact tracing / case investigation process that we need to understand.
The public is able to see the number of BT keys that are uploaded to the MOH server, which correspond to individual cases who have provided data (noting that multiple keys may relate to the same person). Between 17 August and 1 October 2021, there were 6 devices providing BT keys, and 11 contacts were notified based on those keys. There were over a thousand cases over that time period.
We have seen more keys being uploaded in October, but still not enough given that there are around a hundred new cases a day. We have a relatively high uptake/participation in BT relative to other jurisdictions. I was on a videocall recently that included representatives of a number of European countries, and I asked what their uptake rates were.
After the provisos about how hard it is to measure uptake rate, they all landed somewhere around 20-25% of the adult population. They thought this was great and they were finding contacts and saving lives. In New Zealand, we currently have devices equivalent to 55-60% of the adult population participating in BT each day. They were amazed that we had such a high participation rate. I was amazed that we were getting so little data out of it.
We also don’t track the data all the way through the process – we currently don’t know how many people got a test because they got a NZ COVID Tracer alert of any type. So we know how many people were notified that they are a contact through the app, but lose the trail after that [note that I think the protocols ask for whether or not a positive case has received an app alert]. We need better metrics for understanding the usefulness of digital contact tracing before we can really make decisions about whether it is useful or not.
There is also concern that the changing approach in only releasing high-risk locations of interest (which are related to the QR codes rather than BT) will also change the comprehensiveness of alerts that are sent through NZ COVID Tracer, and how this might affect uptake/participation. This may help with avoiding a high rate of false positives (aka the “pingdemic” seen in the UK), but is also at odds with most users’ conception of the purpose and function of the app.
Conclusion
So could we use technology to better support contact tracing? Yes, but not by using new technologies to collect more data. It’s about using the tools we already have more effectively, rather than trying to add more tools that may not add much benefit at great cost. It’s about how the data that is already being collected is being used, and how we can measure that to understand its utility. And it’s about supporting human contact tracers to be the most effective they can be, by automating the simpler and more repetitive parts of their processes.
Unfortunately we might look back and realise that this is a conversation we should have had six months ago. But hopefully we can learn from this experience and be better prepared going forward.