Why should you install the COVIDSafe App? Part 2.

Part 2 of a 3-part series looking into the questions you should ask before downloading the Australian Government's new COVIDSafe app.

In part 1 of this series we laid out a sensible approach to thinking about the app:

  1. Do we have a current, or future, problem with contact tracing?
  2. Does digital contact tracing solve it?
  3. Does this app in particular solve it, without creating more harm than good?

Let's move on to the second question.

2) Does digital contact tracing solve contact tracing problems?

A key paper which advances the argument that digital contact tracing can contain an epidemic is Quantifying SARS-CoV-2 transmission suggests epidemic control with digital contact tracing.

The essence of the idea is simply that if you can reduce the time it takes to contact potentially infected close contacts from days, right down to instantaneous contact and mandatory 14-day isolation, then the spread can be reduced to the point an outbreak is contained:

An optimistic (naive) visualisation of instantaneous contact tracing benefits.

Instantaneous?

Yes, the paper deals with a (very) naive hypothetical app which immediately notifies registered close contacts to go into isolation. The reality is this imagined technology solutions will generate many false positives (people who shouldn't be flagged as a close contact) so you will always need a human in the loop (a contact tracer doing the thinking about each and every case).

Here's former  US FTC CTO and Obama Whitehouse Senior Advisor Ashkan Soltani on the issues of false positives and negatives in a recent Twitter thread:

For the sake of clarity let's describe those terms again:

  • false positive - someone who was detected as a close contact, but shouldn't have been
  • false negative - someone who should have been detected as a close contact, but wasn't

These are simple, but very important terms to understand. They have significant implications for you, as you'll find out.

So digital contact tracing will never replace manual contact tracing?

No, and this a really important point to keep in mind.

Contact tracing teams still need to exist, and do the human intelligence work of contacting every infected person, working through their contact history, discovering additional contacts, interviewing every one of them and reasoning about whether they should be considered a close contact.

How much benefit does a digital contact tracing system bring then?

Let's go back to an image from the paper:

An epidemiologist's algorithm.

Home and Personal Events

We can immediately determine a contact tracing app adds no value in the context of a home - you know who you live with and can immediately notify them.

In fact many (if not most) personal events such as dinners, parties, birthdays, weddings and funerals aren't served by a contact tracing app - they often involve people spending more than 2 hours in a room, which explicitly requires manual contact tracing of all people at the event.

Remember the stories of an infamous wedding where dozens were infected? In a hypothetical future situation where everyone had an app it's unlikely guests would spend more than 15 minutes in many discussions - unless of course you're unable to escape that particular relative. Even then, considering the relaxed, alcohol-infused, exuberant discussion that often occurs you might be inclined to say 2 minutes of that chatter is equivalent to 15 minutes elsewhere.

Even in a situation where face-to-face contact is relevant, people regularly leave phones in cars or in bags and purses, which are often left on side tables, coffee tables, and kitchen tables and so forth – demonstrating that in many social situations this technological approach does not reliably model your proximity to others over time.

As one piece of critical reporting observes:

But for now it seems worth saying that there’s little evidence that phones are good at contact tracing — and a lot of evidence that human beings are.

Workplace

While it might be tempting to think an app is useful here, it quickly becomes apparent that human intelligence is the dominant and essential component.

Why? There's a rich diversity of workplaces and it can be complicated to mentally model all their manifestations. That may seem like an argument in favour of an app, but the opposite is true.

Consider that a significant number of jobs are in enclosed rooms and open-plan offices with others – triggering a manual identification and contact trace anyway due to the 2 hour rule, along with an organisation's own processes which would be enacted when an employee has been identified as infectious.

This complexity of interaction requires much finer-grained processes in this work context, relying on both individual and institutional memory to get a detailed picture of legitimate close contacts, especially when you consider the high costs (both personally and organisationally) of false positives and negatives.

Floor plan of the 11th floor of a building in South Korea, site of a coronavirus disease outbreak. Blue colouring indicates the seating places of persons with confirmed cases. From a new CDC report: Coronavirus Disease Outbreak in Call Center, South Korea

Take for instance the figure above from a CDC report on a coronavirus disease outbreak in a call centre in South Korea. The contact tracing effort above necessitated throwing a much wider net than what would be registered by even very high deployment of digital contact tracing apps.

Also, while the CDC report indicates that "duration of interaction (or contact) was likely the main factor" it also stressed that the "role of asymptomatic COVID-19 case-patients in spreading the disease is of great concern."

Transportation

We see that a contact tracing app provides very little benefit (possibly none) in a home and workplace context.

If the COVIDSafe app is our "ticket to freedom" as Scott Morrison has repeatedly reminded us, then surely transportation is where a contact tracing app really shines?

Consider anyone travelling in taxis or on buses and trains where they'll spend more than 15 minutes within 1.5 metres of someone (but not planes, where airplane mode disables bluetooth). This is a rich environment for interactions with strangers, who would be hard or impossible to track, so this looks like a promising use-case for a tracing app, doesn't it?

Creepy guy using COVIDSafe and not social distancing.

Unfortunately there are a few small problems.

The first is political: what happened to social distancing? If we are meant to adhere to 1.5 metre distancing then theoretically there will be no (remember 15 minutes duration, not walk-by) close contacts registered on the app for public transport journeys.

However the reality is, in dense metro areas with business 'back to normal' any attempt at wide-scale social distancing overwhelms the capacity of public transport. So is Morrison working on the basis there will be no social distancing on public transport? In the absence of publicly-shared modelling of public transport capacity (based on, for instance, business surveys of remote working intentions) it's hard to conclude otherwise.

In this scenario it's inevitable social distancing rules will be ignored. Faced with eternally waiting for a social-distancing 'slot' on public transport, many workers will ignore rules to avoid the personal cost of endlessly explaining why they were late to work.

The second is scientific: the 1.5 metre rule is based on a 1942 study of viral transmission, which has been superceded by numerous recent studies which suggest much wider proximity circles are important (far beyond Australia's golden 1.5 metre rule, which is itself less than the 2 metres specified in countries like Singapore and the US).

So in a way this tracking app - sold as a sophisticated, cutting-edge solution - is in truth an implementation of a primitive, superceded theory of infectious disease transmission.

Air flow modelling is more sophisticated in 2020 compared to 1942.

How much value does it really offer from an outbreak containment perspective in situations where an infectious person travels on a packed bus, tram, or train for an hour? It's an interesting one to think about.

You might even be inclined to think about how much more effective mandatory (or even voluntary, but government-sanctioned) wearing of masks are in these scenarios, and wonder why little emphasis has been given to these kind of low-tech, pro-active and relatively proven viral suppression strategies.

The third involves a little game theory: if the app's primitive modelling of disease is that close contacts of more than 15 minutes risks you getting 'sin-binned' for 14 days of isolation (potentially unnecessarily), does that incentivise more people to move about environments (like trains) more frequently to avoid negative personal consequences?

In fact, if the health advice is 15 minutes of exposure is dangerous, it's entirely rational for everyone to play musical chairs on all but the shortest crowded public transport trips. What are the impacts of increased mobility on infectious spread when infectious individuals also act in that rational, seemingly responsible, manner?

It's understandable that public health teams might like the idea of quickly being able to identify a fuzzy circle of contacts near an infected person travelling on public transport (an undeniable capability in a hypothetical world of 100% app uptake).

But is this reactive intelligence useful enough in the scheme of things to justify a nation going back to crowded, daily public transportation before local elimination?

With such uncertainty and high risk, it seems these decisions go far beyond the limitations of technology and epidemiological science, and into political, social and ideological judgements.

It appears to be an alarming, grand social experiment, conducted in real-time.


Given its popularity, we'd be remiss not to consider private commercial transportation: taxis, mini-vans as well as, Uber and it's ilk.

A quick review of popular vehicle model dimensions reveals many seating arrangements are outside 1.5 metres (when considering the driver's phone is generally on the dashboard, often in the far-right corner), so it's entirely possible to spend a long trip home after work, or an evening out, in close confines to an infected driver (or passenger), without a contact being registered, even though it should be considered one.

Automobile schematics are suddenly interesting.

It's another example where human-powered manual tracing is essential, especially with larger vehicles.

Real-world analysis of app behaviour in different transportation modes and seating arrangements would seem to be relevant.

False negatives

Before we talk about false positives in detail, it's worth discussing false negatives, or 'the ones that get away'.

Touching shared objects has been a feature of transmission of Covid-19, so there are many potentially infected people in an outbreak that are infected by someone without being in proximity to them from the perspective of a digital tracing tool.

This isn't a criticism of contact tracing, but reflects the fact that if you concede the goal of elimination in a region, then there are modes of transmission that cannot be stopped with tech solutionism, only with simpler, low-tech solutions like lockdowns, masks, gloves and increased hygiene practices.

Within false negatives it's also worth re-iterating the point that mobile phones are a poor model of real human proximity in many social settings: people regularly leave mobiles charging on desks, or  in handbags, bags and on side tables as they engage in close interaction like coffee-catchups, lunches, and meetings.

False positives

There are lives at stake. False positives and false negatives have real-life (and death) consequences - Jason Bay, Product Lead, Singapore TraceTogether app.
An epidemiologist's algorithm.

Returning to the visual representation from the paper mentioned above, it imagines an instant alert being sent to the phones of automatically-detected close contacts, but this will never happen because of all the false positives that will invariably occur.

False positives? Yes, think of:

  • adjacent offices which could be in the same building and floor, but in an entirely different workplace
  • neighbouring apartments or terrace houses that unknowingly use the same shared wall to place a side-table (in their lounge or bedroom) which is the favoured place to keep their mobile
  • meeting in a lobby that includes a divider or indoor garden with an adjacent table out-of-sight at which an infected person sits
  • a co-working space of small, isolated and adjacent meeting rooms
  • sitting on a bus, windows open,  in a seating configuration where you and an infected person are close but facing directly away from each other
  • sitting on a train in a section separated by a glass divider, while an infected person is standing away from you gazing wistfully through the doors
  • sitting inside a cafe while an infected person sits inches away on an outdoor table, through glass.

These are examples of false positive cases which will inevitably create a lot of investigative work for contact tracers. While some may be found not to be close contacts, there is a possibility that many more won't be possible to determine.

Especially when you consider the contact tracing policy is to try not to identify the infectious person in the contact process.

The allure of more data risks the same type of 'analysis paralysis' concerns the NSA discussed internally in the face of their 'collect it all' strategy. Recall that the project lead of Singapore's effort has already flagged issues with false positives (in a scenario where less than 20% of citizens have downloaded the app).

It's also worth emphasising that none of this relieves the infected person, or supposed close contact, of the burden of remembering their movements. Faced with a call from a health office, you're provided with times and dates of registered contacts, so it's still on you to attempt to remember every last detail of your moves. Without this environmental context health officials are not able to accurately determine whether it really is considered a close contact. How will this uncertainty be dealt with?

(Maybe New Zealand Prime Minister Jacinda Ardern's approach to contact tracing - urging New Zealanders to keep a simple diary - isn't such a bad idea after all.)

Even when the health official uses their judgement to decide, you might be a close contact you're faced with an incredibly difficult situation.

Threat of unnecessary economic and social isolation

Once you receive a call advising that you've been deemed a close contact of an infected person, the legal requirement to self-isolate for 14 days looms large.

With only the time and date of the registered contact provided, will you remember exactly where you were sitting on the train last week? Even as a responsible citizen it will be difficult to remember, especially with limited information about the infected person. What will a less responsible citizen do?

These situations of fuzzy false positives could see contact tracers either giving up on pursuing particular public transport contacts (in which case the app provides little use in these contexts) or erring on the side of caution and advising that you must isolate (triggering potentially unnecessary economic and social isolation).

As stories of forced isolations are heard from people who feel they were incorrectly and unfairly categorised (with the limited information they are given feeding that perception), it is predictable that people will begin to become more wary of  the real potential of false positives with this system, and the underlying rationale for its necessity as a public health measure.

So how much value does a general digital contact tracing solution really provide?

One of the federal government's invited experts advising the AHPPC is emphatic:

Actually it's faint praise, and as we've outlined above, in many contexts this digital solution appears to add no value, and in those it may (such as public transport) it requires an assumption of no social distancing and potentially significant numbers of false positives which have to be methodically, manually reviewed.

The final note on human recall is odd too, as we've highlighted how a digital contact tracing app does not remove that burden from contacts caught in the web of primitive, bluetooth proximity estimation.

But what about in the real world? The project lead of Singapore's TraceTogether initiative (which Australia's CovidSafe app is built upon), is extremely open and transparent about the significant limitations of contact tracing (even as someone naturally invested in its success):

A US technology expert, Bruce Schneier, recently reflected on many similar concerns in a post on contact tracing:

In case that wasn't enough, he doubled down in a subsequent post on the same topic.

Moving on, up to this point our tour of general digital contact tracing has assumed 100% uptake, but in Singapore's case uptake is only 20%, which means a 4% chance that any two people both have the app installed and working.

Only 4%? Yes that's 20% of 20% - which represents the probability that Person A has the app multiplied by the probability that Person B has the app.

In reality it's likely even lower than this.

How low?

Tomas Pueyo has published a number of deep research pieces, and the latest one dealing with testing and tracing includes a (very rough) modelling of the considerations in drop-off rates, from the total population through to the contacts that actually get registered in a voluntary system:

His reasonable conclusion: 'Every bit counts, but this bit is an awfully small bit.' or in other words, voluntary contact tracing apps will not work.

An idea he does entertain for illustrative purposes (but Scott Morrison and civil society seem to agree is not acceptable in Australia) is that a mandatory or almost-mandatory (installed-by-default and opt-out) regime, is what will work:

If you need more convincing of the issues around digital contact tracing, consider the opinion of the Margolis Center for Health Policy at Duke University, who in A National COVID-19 Surveillance System: Achieving Containment (a report authored for US and Global policy makers) concluded that:

Cell phone-based apps recording proximity events between individuals are unlikely to have adequate discriminating ability or adoption to achieve public health utility, while introducing serious privacy, security, and logistical concerns. Instead, timely contact tracing can be achieved through strengthened public health case investigation augmented by technology and community-level collaborations.

Alternatively, consider the opinions expressed in a recent piece published by the Brookings Institute in the US: Contact-tracing apps are not a solution to the COVID-19 Crisis:

We are concerned by this rising enthusiasm for automated technology as a centerpiece of infection control. Between us, we hold extensive expertise in technology, law and policy, and epidemiology. We have serious doubts that voluntary, anonymous contact tracing through smartphone apps can free Americans of the terrible choice between staying home or risking exposure.
We worry that contact-tracing apps will serve as vehicles for abuse and disinformation, while providing a false sense of security to justify reopening local and national economies well before it is safe to do so.

So to determine whether you should install the app, answering 2) appears to involve:

  • assuming elimination through lockdown is impossible or unfeasible (as per 1.),
  • accepting a primitive modelling of infectious disease in technological form is useful
  • understanding and conceding that mandatory or almost-mandatory app install is necessary
  • betting that the scale of data (and false positives) won't itself slow down and overwhelm manual contact tracing efforts
  • accepting the very real personal risk of being identified as a false positive in a very imperfect system (remember 14 days isolation is required as a close contact). Here we might also consider how this impacts the more vulnerable in society – can they pay the rent, stay at home safely, and so forth?

How you respond to those questions finally gets you to a point where you might finally consider 3).

Thanks for reading this far. We hope we've answered some of the questions you have and perhaps raised a few more.

In our final part of this series on the COVID-Safe app we'll be looking at the last question: "Does this app in particular solve it, without creating more harm than good?"