10 Do and Don’t in Managing the Curve

COVID curves literally have their ups and downs. It is hard to infer how good or bad a situation is, as the severity of the problem depends on the multiple factors intrinsic to how the disease behaves in a region, the economic reserves of that region, and the ability of accurately measuring the situation. In light of the current wave of COVID in Sri Lanka there have been many view-points of different management strategies. I was not able to find public documents detailing the GOSL authorities management strategies (apart from high-level statements). Hence it is hard to comment on what works and not. However, as an academic exercise, I thought of listing down 10 general Do’s and Don’t.

Note: some of these might already be implemented .

The Dos

1. Have a Centralized Communication Center (A website!) and one Accountable Person Heading the COVID Management Effort: Currently, I have multiple sources of information for COVID in SL— ranging from online newspapers to journalists on social media. Imagine if there was a website like gosl.covid.lk, which was the central source of all COVID related information with accountability. This would stop people from coming-up with all forms of conspiracy theories. Here’s an example of the Dutch centralized website. Also, would be nice to have a COVID response director who gives bi-weekly updates and is basically responsible for the management of the entire problem. This could even be the president or prime minister even.
2. Communicate Checkpoints vs Status: much of the anxiety for me is caused when I don’t have information on what is going on and how good or bad that is. For example, it would be nice to have checkpoints like status 0: COVID totally under control, status 1: COVID is spreading be more vigilant, status 2: Curfew imminent, etc. Yes, there is the worry that people will start panic buying, yet transparency is better than panic buying based on hear-say.
3. Communication on Social Distancing, Cleanliness, Travel, and Emergency Situations during Curfew: Communication about cleanliness, travel and social distancing has been great AFAIK! During curfew there are many who need access to essential services. For example medical patients, victims of violence, and essential service workers. It would be great if the centralized website gave guidance to those who need to travel within curfew zones. And in addition contact numbers, email addresses or chat bots if there are further questions.
4. Have a Goal and know your Constraints: COVID related goals are generally around health and the economy. For example a goal could be: We Don’t kill more than 500 people a year both due to COVID and due to COVID handling consequences. Why 500? We let 500 people die each year in SL due to Dengue. Or a constraint could be the number of ICU beds, or that we don’t spend more than 1/10th of SL’s foreign reserves — 0.7 billion USD.
5. Striving for the best Analytical Predictions: COVID spreads fast, hence we need to forecast, to prepare for the future — or think backwards. There are numerous mathematical and stochastic models used to model the spread of COVID. See here and here for examples of academic publications for Sri Lanka itself. The issue here is that very simple SIR models have very high variance (can be very wrong sometimes), while very complex simulated models though more accurate than simple SIR models, can yet be very wrong if assumptions are wrong. In addition, as governments try to curtail the spread, these models need to deal with these unusual ‘events’ like lockdowns. I personally like more complex simulated, scenario-based models. However, we need to keep updating these models daily and try and improve continuously, since, if don’t forecast, we can’t prepare for the future. A good practice is to think of upper and lower bounds — or best and worse case scenarios, and then prepare for those.
6. Test, test test (but intelligently): Lockdowns have a massive economic impact (see my simulation below). Testing with contact tracing (in counties where it is possible to do so) seems to have worked well. Here is a great paper from nature with an interesting simulation study too! This was Sri Lanka’s strategy — which is great! Yet I’m not sure about two things (1) If the tests were enough (2) was some component of random testing carried out to catch stray cases. According to this paper, random PCR tests were planned islandwide only in August. Of course assuming we have the money to do all this. Either-way I feel a lockdown is still more expensive — so might as well pay for the tests.

The Don’t

1. Don’t Get too Comfortable: If COVID cases are virtually zero for a very long time, we might start getting lax in things like testing and public safety measures.
2. Don’t Compare ourselves to other Countries: It is futile to compare countries as the rate of spread in different counties and regions are driven by differing factors such as population density, climate, peoples’ health, age and other demographics. Hence, an estimate of a 1000 infected people in Sri Lanka could be OK, while the same number in Sao Paulo could be catastrophic. In addition, some counties have more resources than others, hence what they could do, you might not be able to. Have our own benchmarks.
3. Don’t Forget the Complete Picture: We sometimes get wrapped around figures like deaths (which are super important) and number of COVID cases. However, measures such as curfew, have detrimental effects on people’s health and finances, especially the poor. It is making the poor poorer, and some of the rich, richer. In this situation, there is no good decision. Either-way people are going to die and suffer. The question is what option will minimize it.
4. Don’t believe the numbers too much — intuition is useful! (lies lies and statistics, yes): The issue with COVID is that the best test we have is the PCR test. And yet we need to be careful on who we test, as tests are limited and the priority would be to test potential COVID patients as the disease spreads. On top of that, the PCR tests (like all tests) have non-zero False Positive Rates and False Negative Rates. Hence, in reality many statistics could regarding the spread could be flawed! A combination of intuition and upper and lower bounds would help.

The Simulations

I explore three strategies, using a very basic simulation with the following features (yes it is flawed, but will give you the idea):

• Average R = 2.4. That is the number of people one person infects. This is not a static number, and depends on multiple factors like population density. I’m assuming it is static.
• Population of SL = 22 Million
• Contagious Period = 14 days
• Probability of Death = 1% (of course these numbers vary with age and risk factors such as Diabetes Mellitus)
• Lockdown Cost per Day = 22 Million USD (assuming each Sri Lankan will lose 1 USD a Day — which is 10% of the daily average wage).
• Test Cost = 30 USD (Indian estimates)
• Contact Tracing Cost per Contact = 10 USD (No source)

Method 1: Somewhat Targeted Random Testing, with Imperfect Contact-tracing with Lockdowns

Here, I‘ve simulated a scenario where:

• If the cumulative number of active cases is more than 5000 we lockdown
• Contact tracing is carried out and they are quarantined, but 25% of the cases are missed.
• I ran the simulation for 5 different scenarios for the number of PCR tests per day

In this sketchy simulation, even with 50k tests per day, unless contact tracing is carried out more vigilantly COVID can spin out of control.

Method 2: Almost Perfect Contact-Tracing

Here, I assume that contact tracing will find almost all potential contacts of an infected person, but on average 2 people will be missed (with a standard deviation of 2), hence, not 100% perfect. Yet, even with this error, we can see that anything more than 1000 PCR tests per day along with contact tracing can contain COVID. Guess, as expected, more points to contact tracing with high testing.

Method 3: Only Random Testing

Does only random testing work? I’ve often heard many comments saying if Sri Lanka did random tests, it would suffice. I guess by now it is almost obvious that, unless we pretty much test the whole country, by only randomly testing people sans contact tracing is futile.

The Cost

Now let’s compare our three skeleton simulations. Perhaps not surprisingly, the lockdown approach with not so great contact tracing is by far the most costly, as a lockdown has massive economic impact.

Contact Tracing is more expensive than just randomly testing (which of course is futile). But my point was to show that contact tracing may not be as expensive as it sounds.