COVID-19 spring redux: more cowbell

Although we’re more than a year into the COVID-19 pandemic, our ability to protect vulnerable people hasn’t improved much. On a medical front, we’ve advanced with better triage, better treatments, and now a vaccination program – but our public policy looks pretty much the same as it did last spring, expect in Ontario, this time with enhanced police powers.

Ontario is seeing a tidal wave of new infections. While new variants apparently increase risk among younger demographics, the 60+ age group and those with comorbidities experience the majority of the severe illnesses and deaths. Earlier this month, Dr. Micheal Warner pointed out that a person under 50 dies from COVID-19 in Ontario ICUs every 2.8 days. That rate among the young is higher than in the past, but with a current Ontario COVID-19 death rate of about 20 per day, that means about 60 old people are dying for every person under 50, who almost certainly has comorbidities. It’s pretty clear that COVID-19 and its variants are vastly riskier for the elderly and those with comorbidities. Unfortunately, our vaccination program has a long way to go to protect the vulnerable.

Since this pandemic began, our governments at all levels have focused on lockdowns, school closures, travel restrictions, and curfews as the primary mechanism to control COVID-19 community spread. For the last year these policies have shifted most of the suffering onto the poorest people in our society – our essential workers – who are employed in everything from meat-packing plants to supermarkets. Even the homeless haven’t escaped. In January, Ottawa homeless shelters paused intake due to an outbreak among residents and staff. Not only have the poorest faced the brunt of COVID-19, but the vaccination programs have had the greatest penetration with the well-off. For all the rhetoric and all the treasure we’ve spent, we have not protected the vulnerable.

Is this the best we can do?

Last spring, our governments implemented emergency measures in a state of overwhelming panic. This time around, we appear to be recycling the same emergency policy playbook, and in the process, discarding a year’s worth of hard won knowledge about COVID-19. Public Health Ontario has publicly stated that focused measures to protect the vulnerable are not possible:

While age may be the major driver of differential mortality risk, there are important equity-based considerations in understanding risk of dying from COVID-19 in non-elderly populations if public health measures were to focus on the risk to the elderly only….Public health agencies globally have supported the use of physical distancing measures and lockdowns to control community transmission of COVID-19, recognizing that high community transmission renders protection or shielding of higher-risk populations virtually impossible due to the connections within the broader community in which they live. Further, an approach to strict isolation of at-risk populations can also be considered inhumane and unethical.

I am not sure how Public Health Ontario comes to the conclusion that shielding higher-risk populations is virtually impossible, but it certainly points to a lack of creativity and no sense of resolve. The situation reminds me of the Berlin Airlift.

In 1948, as the Cold War was getting started, the Soviet Union prevented all surface traffic from non-Soviet zones from entering West Berlin. Stalin hoped to drive the Western allies out of Berlin, undermine the Marshall Plan for rebuilding the German economy, and place the Soviet Union as the dominant power in Europe. At the time of the blockade, West Berlin had about a month’s worth of food and fuel. The Soviets figured it was impossible to keep West Berlin supplied and the West would quickly capitulate. We didn’t. We organized an airlift that kept West Berlin provisioned for 15 months. In the end it was the Soviet Union that relented. At its peak, the airlift had a plane landing in Berlin every 30 seconds, planes filled with everything the city needed from coal to potatoes. Our response to COVID-19 looks nothing like this.

Constructing an optimal policy should use all available information. We know who the most vulnerable are: the elderly and those with comorbidities such as obesity, hypertension, and diabetes. We must use all we know about COVID-19, pandemics, and public health more generally. Two points stand out:

1) We now know that, for those over age 70, the survival rate from COVID-19 is approximately 95%. That is a terrible problem. For the vast majority of us who are at little risk, the survival rate is at least 99.95% and most likely higher. Comorbidities also increase risk. Healthy people under age 25 have a survival rate of effectively 100%. That is, the high risk group is thousands of times more likely to suffer a severe outcome. Given the extreme differences in survival rates between different population subgroups, it would be surprising if the optimal policy did not include that information.

2) As a biological fact, all pandemics end by herd immunity either by natural infection, vaccination, or some combination of both. The optimal policy that ends the pandemic gets us to herd immunity with the least total harm. Covid-zero is not possible with such a widespread pathogen and the social determinants of health stretch well beyond modern medicine.

Implicitly, our governments are telling us that the first point is irrelevant. They are saying that the optimal policy that offers the best protection to the vulnerable from COVID-19 is to treat the disease as though the risk of severe illness or death is exactly the same for all people in society. Our governments have said that for all the money we’ve spent and for all the economic burden that we have placed on our children and the least financially well-off in our society, there is little that we can do differently. In their view, our polices are close to optimal, and that our knowledge of who is most affected by the virus is of no practical significance for public health policy development. Is it really true that no other policy which uses this information is feasible, politically or otherwise?

The second point has become hopelessly and needlessly politicized. Herd immunity has somehow been equated to a political statement. It is not; it’s biology. As the Harvard epidemiologist Martin Kulldorff says, “it’s weird and stunning to have this discussion about herd immunity—flockimmunitet in Swedish. You wouldn’t have physicists talking about whether we believe in gravity or not. Or two airline pilots saying, ‘Should we use the gravity strategy to get the airplane down on the ground?’ Whatever way they fly that plane—or not fly it—gravity will ensure eventually that the plane is going to hit the ground.” However we get there, herd immunity is the only thing that will end the pandemic.

So what would an optimal policy look like?

First, anyone who is at high risk and who has to work in settings ripe for spread should immediately be put on long term disability that replaces their wage. We should not be sending pre-diabetic, hypertensive, heavy-set 50+ year-olds to essential workplaces. We need legions of young healthy people to step up, train quickly, and fill in. If 20 year-olds with minimal training – often less than a few hours in a Spitfire or a Hurricane – won the Battle of Britain, we can find a way to use our young healthy population to replace high risk workers in everything from meat-packing plants to hospitals. Second, for those in the high risk group who cannot self-isolate, such as poor elderly immigrants in multi-generational homes, re-purpose our hotels. Again, have legions of young and low risk people isolate with them, on rotation, to run operations. Ensure high levels of testing. For those who can self-isolate, our army of young people can still help by delivering groceries, medications, and other necessities. Third, let everyone else go about their normal lives to the greatest extent possible – especially children in schools. As we vaccinate our high risk population with an ever expanding vaccination program, and as the virus spreads among the low risk group, we will reach herd immunity and the pandemic will end. The current policy looks very much like let it rip in slow motion.

Perhaps these ideas are infeasible; perhaps our current policies of lockdowns and curfews are the best we can do. Maybe ideas like mobilizing our young is impossible. Maybe it’s too late. But I challenge anyone to rigorously show that we could not have done better or that we can’t do better now.

With bold leadership and a call to civic duty, our accomplishments during WWII and at the beginning of the Cold War showed how we can do great things under enormous stress, against terrible odds, at extremely high stakes. From the Dunkirk Evacuation to the Berlin Airlift, we never accepted the answer “it’s virtually impossible”.

We’ve been told that our civic duty is to stay home, wear a mask during essential trips, and wait for a vaccine all the while we push Covid-19 onto vulnerable essential workers and their relatives.

I bet we can do better. Let’s focus our protection.

2 thoughts on “COVID-19 spring redux: more cowbell”

  1. David,

    As always, you raise some interesting points. It does seem odd that policymakers do not seem interested in making maximal use of all available information when tailoring their response.

    While I have no good answer on why, I have spent more than a little time over the past year contemplating what an “optimal” solution might look like. I share a few of my observations below.

    From a mathematical standpoint, I view the problem as a large dynamic program. Given a value function and a set of policy levers, for any given state of the world, at what levels should we apply our available controls? The problem is further complicated by the fact that we have imperfect knowledge of the system dynamics – for a given state and action pair, we have only a noisy understanding of the transition matrix that will determine in which state(s) we may end up. Our ultimate policy response will therefore consist of a balance of exploration – trying to learn more about how the system responds – and exploitation – taking actions we have previously applied that seemed to advance our objectives, and reapplying them. The balance between these two activities will depend on our discount factor and risk preferences. All in all, quite an interesting little math problem, but there exist a number of approximate techniques to find near optimal solutions quickly.

    Of course, before we can solve the math problem, we have to develop our abstraction of reality into a mathematical format. What variables should we use to specify the state of the world and our current imperfect understanding of the system dynamics? And, perhaps most critically, what is our value function – what are we actually trying to optimize?

    Our value function is almost certainly a multi-objective one; we seek to strike an appropriate balance between public health and economic objectives, weigh the relative importance of individual freedoms against broader social responsibility, decide if/how we address issues of “equitable” distributions of risks and benefits across society, and so on. Key results in social choice theory, such as Arrow’s Impossibility Theorem, highlight that construction of a value function that reflects (on average) the values of society writ large, is much easier said than done.

    Determining the relative importance of non-commensurate objectives is hard enough, but even within a narrower portion of our value function, there’s still issues surrounding selection of appropriate metrics. So, for example, are we looking to minimize the number of COVID deaths? Well, yes, that’s certainly part of the problem. But any intervention we might make in the system has the potential for substitutions that may not be strictly Pareto improving. So if we apply policy X, and prevent Alice and Bob from dying from COVID, but now Carenza dies from some other cause as an unintended consequence of our intervention, how do we capture the substitution effect? So perhaps we should be optimizing on excess deaths instead.

    But wait a minute, are all deaths created equal? A generic 95 year-old Canadian has a residual life expectancy of 3.4 years, according to Statistics Canada. By contrast, a 65 year-old has a residual life expectancy of 20.9 years. If we tailor our response around simple death figures, then we allocate more of our scarce resources to protecting the 95 year old, as they are at the highest risk of death from COVID. But if we measure the effect in life years lost, then we may favour allocating more resources to protecting the 65 year old, as we will likely save more life years for the same “investment”.

    But are all life years created equal? What if our 95 year old is in excellent shape and still lives on their own, while the 65 year old is a paraplegic as a result of a car accident earlier in their life, shows signs of early onset dementia, and requires constant medical care? There are metrics such as quality-adjusted life years (QALYs) or disability adjusted life years (DALYs) developed in public health economics that exist to attempt to address such concerns. If we base our policy interventions on this metric, we may find that we choose not to allocate our scarce resources to people in ill-health with pre-existing conditions such as diabetes or hypertension, despite these being significant comorbidities that increase a patient’s susceptibility to COVID. Try explaining to someone’s family why, despite their high risk factors from COVID, you chose to let someone die because “the resources were better spent elsewhere.”

    We can, of course, go further. Actuarial tables tell us women have higher residual life expectancies than men at any age. This suggests that we bias our interventions towards protecting women. There are also clear income effects in the data – people in the top income quintile live longer than those in the lowest income quintile. If we have limited resources, then, we gain the largest effect by biasing our allocation of them towards protecting the wealthy at the expense of the poor. But here I think we see some portion of the problem.

    While a richer information set affords the possibility of ever more efficient use of limited resources to maximize a given effect, it also calls upon the decision maker to make progressively more difficult and potentially uncomfortable value judgements. If, instead, they choose to employ a simpler value function, they can claim to be behaving optimally under the given metric, with lower perceived risk of political blowback, albeit with a corresponding sacrifice of efficiency in our resource allocation.

    But let’s say that we somehow manage to come up with a value function that does a good job of capturing society’s preference structure writ large, at a level of granularity with which everyone is comfortable. What will drive the overall structure of the solution to our dynamic program?

    Ultimately, it comes down to our resource constraints. We can, for example, process a certain number of COVID tests of a given sensitivity and specificity per day. We can invest time, money and other resources to increase this capacity, at some finite, sub-exponential rate. Similarly, we can trace a certain number of contacts each day, and invest resources to increase this capacity, again at a some finite sub-exponential rate. The same holds for hospital/ICU beds, mortuary capacity, hotels/dormitories for quarantining people, our ability to retrain/reallocate workers across sectors of the economy, resources to monitor and enforce compliance with imposed restrictions, and so on. Against this background of resource constraints, we have a disease which, in some states of the world, can experience exponential growth. This structure suggests that the optimal policy will have a phase diagram describing changes in optimal response depending on which constraints are currently binding. The optimal policies in adjacent regions of the phase diagram may change slightly, differing only in the degree at which particular controls are applied, but the exponential growth potential of the disease background can also lead to dramatic structural changes in the optimal policy in the “phase transitions” between adjacent regions.

    I think most of the control policies which have been proposed and/or implemented around the world all make a certain amount of sense – each are likely good approximations to solutions that exist somewhere in the overall phase diagram. It is less clear, however, exactly where we are on the phase diagram at present; this is compounded as differing stakeholders are all working with different value functions and different abstractions of reality. This could be addressed by layering a certain amount of robust control logic on top of the problem; i.e., “if we apply the “wrong” policy for the region we are actually in, how bad could it get?”, but that’s an entirely different rabbit hole …

    I’ll close this (rather too long) “comment” by returning to the exploration/exploitation tradeoff I mentioned earlier. The continued implementations of lockdown-based policies suggest a strong bias towards “exploiting” a known solution that “works”, as opposed to “exploring” for better options. This in turn suggests a very strong discount factor; i.e., our decision makers are planning/optimizing over timescales on the order of days/weeks as opposed to months/years. It is also consistent with a high degree of risk aversion – politicians are unwilling to try something different, lest it not work out as well as what everyone else is doing.

    My $0.02, at any rate. Although that rounds down to zero these days…

    1. Very true. It is a dynamic program with an objective function that at the very least is fuzzy. What I find surprising is that policy has largely remained the same since day one. Even though it was pretty clear as early as April of 2020 that Covid-19 has a steep age gradient, the information was still new and it was unclear how certain we were about anything. A lockdown at that time makes sense, even if just to buy time. But a year later and the policy still looks the same…? That points to some pretty strong risk aversion in any attempt to exploit information. As for exploiting a known solution that “works”…maybe…but works for who? It seems to be good at protecting the healthy laptop class from catching Covid-19, but not so much for the actual high risk vulnerable set of our population. Given the risk profile, our slow motion let-it-rip policy shows up in the distribution of hospitalizations and deaths – the observed distribution matches the empirical risk. If we were good at protecting the vulnerable, the small number of deaths and hospitalizations would only appear in the low risk group with few deaths among the elderly and those with comorbidities.

      I think there are many practical ways to traverse the exploration-exploitation problem efficiently or at least better than we have. We have different provinces in Canada and many different health units within each province. At the very least, small exploratory tweaks to policy, especially in regions with high at risk groups, would have revealed a wealth of new opportunities last summer and fall.

      I am not sure what’s the origin of our resistance to use information, but the political culture in the West seems to be in an odd place right now. Everything is becoming a politics these days, even the mundane aspects of everyday life. There seems to be no escape. Maybe that has something to do with it, I don’t know.

Leave a Reply to Gittins Index Cancel reply

Your email address will not be published.