Seven Smart Ideas for COVID-19
The COVID–19 pandemic has surprised us all with the speed at which it has become a global phenomenon, ravaged national economies, and caused widespread suffering. For the first time in a long time, most of the world is caught up in the same calamity. It has tested the ability of nations to respond to a major crisis and most have been found wanting. The vulnerability of rich, powerful, technologically advanced countries like the USA and UK has been a huge surprise. Poor political leadership is certainly to blame, but there are other structural factors that can explain the failures.
Humans may have learned to harness nature, but tiny invisible microbes have always posed a problem. Images of the Coronavirus have a suitably threatening appearance — flaming balls with many clublike appendages. The clubs — glycoprotein spikes — in the envelope give the viruses a crownlike, or coronal, appearance. Viruses straddle the domains of the living and non–living and it is this aspect that has complicated our ability to fight it. Nevertheless, we could have done better. As a society, we have great ideas on how to use limited testing resources, mitigate supply shortfalls and decide between public safety and economic hardship. A few of these ideas are finally being applied.
The daily counts of new cases, new deaths along with the forecasts of pandemic models have been making it to the daily news. Dozens of modeling efforts are ongoing. Each model aims to answer some question about the pandemic. Some forecast future trends in infections and deaths. Others forecast the number of hospitalizations, demand for ICU beds, etc. A model developed at Imperial College in London led by the epidemiologist Neil Ferguson has been particularly influential. In mid–March, this group published a report projecting upto 500 thousand deaths in Great Britain and upto 2.2 million deaths in the US if no measures were taken. Due in large part to these projections, stronger measures were ordered in Great Britain and in the US. Regrettably, this model did not meet the criteria models used for important policy decisions should be expected to meet — transparency of modeling assumptions and techniques, and the availability of model code for close scrutiny. Later, the Imperial College team made their code available to others. One reviewer (posting on an anti–lockdown site) found lots of bugs in the code, but a different reviewer was able to reproduce their results.
Models are important tools for policy making, but they rely on data and assumptions about how the virus spreads and how effective policies are in preventing spread. Policies, especially severe ones such as lockdowns can have significant negative economic and social consequences. Unfortunately, in the early days of the pandemic, very little data was available, and what was available was believed to be incomplete. Modelers rely on data reported by national agencies, which in turn rely on reports from hospitals and other medical establishments. Deaths in places such as homes and smaller clinics may not get reported. Countries do not report cases and deaths in a standard way. Moreover, China is suspected to have fudged its numbers with official numbers as much as 10 times lower than estimates of actual cases. Besides, the effectiveness of measures such as social distancing and use of masks was not well understood — modelers relied on ad hoc assumptions about the effect of interventions. With the poor quality of data available and ad hoc assumptions about the effect of interventions, forecasts of infections and deaths have a high–degree of uncertainty with wide confidence intervals.
While data was sparse in the early days of the pandemic, there were a few things we learnt from the experience in Wuhan. We knew that the virus spread very quickly and caused severe symptoms in some patients (especially the elderly). Case in point: a single superspreader event in South Korea caused the coronavirus case count to quickly jump from 29 cases on February 15 to more than 2,900 two weeks later. Several features of the modern world — the rise of mega–cities, growth in airplane travel, and modern farming practices — make us more prone to pandemics than in the past. Besides, there was no vaccine or antiviral against the virus. The Chinese government quelled the virus through means that may not be feasible for democratic governments. An important modeling question we should ask is: how high can the number of infections and deaths get if we did nothing or took only the basic precautions? This sort of question is commonplace in the insurance and investment industry where it is important to estimate the likelihood and severity of natural disasters like floods, or market crashes. A branch of statistics known as Extreme Value Theory (EVT) has been developed to estimate the distribution of extreme values of natural events. The Netherlands government used it to compute the optimal height for their dikes by estimating the probability that sea levels will exceed a certain level. Nassim Taleb and researchers at the New England Complexity Institute and the Delft University of Technology have applied EVT to model the outcomes of pandemics. In January this year, Taleb and colleagues at the New England Complex Systems Institute published a note, where they argued that, due to the highly connected nature of our modern world and based on the contagion rates and the severity of the symptoms seen in Wuhan, China, this disease should be viewed as having a fat–tailed distribution which has much higher probabilities for extreme outcomes, compared to a normal distribution. Fat–tailed distributions impose a non–negligible risk of a very large number of fatalities. Such distributions do not have a defined average or variance — the sample mean and variance do not converge as the sample sizes increase. Hence, early estimates of R0 and infections rates are very likely to be underestimated (even if accurate data were available). In a subsequent paper, using data from past pandemics of a certain size, Taleb published a paper with Pasquale Cirillo confirming the fat–tailed nature of pandemics. The tail distribution can be estimated from prior pandemics (i.e. pandemics above a certain threshold of victims). They estimate a mean of 7 million deaths from the past 2500 years of pandemics. We can infer from this that left unchecked, COVID–19 can cause a large number of fatalities, possibly in the millions.
Reversibility and Pandemic Policy Decisions
How does one make policy decisions in the face of high degree of uncertainty about the outcomes of the pandemic — one with a non–negligible risk of deaths in the millions? Should we do everything we can to protect public health or should we protect the economy by taking a wait–and–see approach. In the case of COVID–19, there are two options, a) institute a strict lockdown policy with its enormous economic costs or b) take a more measured approach that allows the economy to operate for the most part — travel restrictions, testing, contact tracing, limiting the size of gatherings etc (something like what Sweden did). Both options have costs — the strict lockdown brings a large part of the economy to a halt, while the more measured approach may fail to stop the pandemic from spreading, overwhelm hospitals, and result in a large number of deaths. One key difference is that the lockdown option is reversible — the lockdown period gives societies the time to implement testing infrastructure, increase hospital capacities and shore up essential supplies such as protective masks and ventilators. With the measured approach, if the pandemic spreads rapidly and many people fall sick, it may not be possible to reverse course. The measured approach can work if testing can be ramped up quickly to test as many people as possible — then those testing positive can be isolated, rather than an entire population. However, the US and some other countries were not ready to do this. As Joshua Gans argues in a recent book, a strict lockdown policy is the only one that gives you the option of making a choice once you have learned more information regarding what the pandemic’s effects on your options actually looks like. China took the strict lockdown approach and quickly contained the virus. Other countries like South Korea and Taiwan managed to contain the virus without a strict lockdown. However, these countries were able to immediately implement a strict regime of testing and tracing and isolating those with positive tests.
Strict lockdown was supposed to last only a few weeks, until US states ramped up on their testing infrastructure. Unfortunately, this didn’t happen, and states that have come out of lockdown are seeing an upsurge in infections and deaths. Lockdown remains a controversial policy.
Efficiency vs Resilience
COVID–19 has caused major disruptions to supply chains, especially in the US. We have experienced shortages in face masks, hand sanitizers, toilet paper, and some food products. Hospitals have experienced shortages in ventilators and PPEs. Michael Pollan recently wrote about the perplexing concurrence of shortage and excess in our food supply. While dairy farmers were dumping millions of gallons of milk, supermarket store shelves were empty. While contract growers were plowing under perfectly good vegetables, and industrial livestock farmers were euthanizing animals meatpacking plants, people were lining up at food banks.
In retrospect, this should not have come as a surprise. US industry has pursued efficiency as its top objective. Efficiency is about minimizing costs and delays. It is about eliminating redundancy. Resilience against shocks requires emphasis on a different set of criteria — many of which are in opposition to efficiency.
Efficient supply chains are marked by specialization, consolidation and leanness. Specialization and consolidation achieve efficiency through economies of scale. Leanness reduces cost by reducing inventory (favoring Just–In–Time inventory systems), thereby reducing waste and costs of storage and obsolescence. The food supply chain in the US is highly specialized. In the industrial part of the food supply chain, milk is sold to restaurants, schools, and corporations in giant containers. Whereas the retail supply chain is used to supply milk to supermarkets and grocery stores in gallon, half–gallon and quart sized containers. Farmers and their distributors don’t have an easy way of moving supplies from one supply chain to the other. Similarly in the supply chain for meat and other produce, most farmers become so specialized when they become large that they only have contracts to supply large retailers such as Walmart or Kroger or Sysco. Walmart and Kroger require food to be delivered in small packages that consumers prefer, while Sysco, a food–service supplier, is selling to restaurants and institutions, such as hospitals and schools that buy in bulk. For Sysco, demand tanked when stay–at–home orders closed schools, restaurants, and other facilities that would use a case of lettuce or a 50–pound bag of onions. Whereas demand spiked in the retail supply chain due to more people cooking and eating at home, and retailers could not easily ramp up supply.
Over the years many sectors of US industry have experienced consolidation and a small number of large firms dominate many industries. Large firms realize economies of scale. More than 50% of Personal Protective Equipment (PPE) supplies in the US are manufactured in China, yielding cost savings for consumers. Lockdowns in China and increased demand there reduced deliveries to the United States. Similarly, there are only two major manufacturers of test swabs worldwide — Puritan Medical Products in Maine and Copan Diagnostics in Italy. When COVID–19 hit and demand for swabs skyrocketed, the companies became overwhelmed and couldn’t keep up.
To be prepared for the next big shock, supply chains for essential products need to be resilient. We may need to sacrifice some amount of efficiency to achieve this, since the features that are necessary for resilience are detrimental to efficiency. Redundancy is a key feature of a resilient supply chain — having contractual relationships with multiple suppliers and having multiple manufacturing plants in different locations. The product design should itself allow for flexibility — with interchangeable and generic parts in many products, maximum postponement of as many operations and decisions as possible. National policy should incentivize suppliers to locate some part of their supply chain within national boundaries. Increased technology adoption plays a role as well, for instance using information to adjust supply chains quickly. The good news is that businesses caught short by COVID–19 are already mobilizing to make these changes.
Businesses cannot completely forsake efficiency if they should remain competitive. Governments should step in to manage the supply of essential products in a crisis. They must maintain a minimum stockpile of essential products — many governments already maintain oil reserves for emergencies.
Until a vaccine is widely available, frequent, repeat, widespread testing is necessary to bring COVID–19 under control. Economist Paul Romer makes a case for this in his plan for restarting the economy with a national “test and isolate” strategy. His approach is simple:
- Test everyone to find out who is infectious
2. Isolate them
3. Continue testing
4. Continue isolating.
“Test and isolate” builds on the same logic that goes into the lockdown approach: mass lockdown works because it isolates people who are infected. But instead of isolating everyone, an “identify and isolate” strategy keeps the economy going and allows those not infected to lead normal lives.
When capacity for testing is limited an idea called Pooled Testing can be effective. Pooled testing uses fewer tests and allows more people to be tested quickly. When the incidence of infection is low pooled testing can be very effective. This (also known as group testing) is a well–established approach to track infectious disease, wherein instead of testing each individual sample, test samples from several people are pooled and a single test is applied to the pooled sample. If the test comes out positive, each of the individuals in the pool is tested. In contrast, if the pooled sample comes back negative, all members of the pool are cleared, at least until the next testing cycle. While there are inherent challenges in this approach, including the choice appropriate pool sizes and test–reliability concerns, estimates suggest that pooled testing could lower testing costs by half or even three–quarters. Greater savings can be realized in low–prevalence areas. A recent paper describes how Machine Learning can be used to predict the risk profiles of individuals being tested which can be used to pool samples for greater efficiency. The authors point out that group testing becomes cheaper with the frequency of testing as positives are removed from the test groups. The US FDA has finally given its stamp of approval to this idea — on July 18th, Quest Diagnostics received emergency use authorization for pooled testing. Pooled testing is an effective strategy for schools and businesses that want to re–open.
Friendship Paradox and Testing
Some of us are more connected than others. If we can identify individuals who come in contact with others more frequently, testing them earlier and more frequently than others would make the “test and isolate” much more effective. A phenomenon known as “Friendship Paradox” helps us to identify such individuals. The Friendship Paradox says that in any network, the average number of friends that individuals have is less than the average number of friends of their friends. This result holds for any network where some people have more friends than others. Stephen Strogatz has a great explanation of this paradox. Christakis and Fowler used this idea at Harvard during the H1N1 flu pandemic of 2009 to test whether the friends group being more connected got infected faster. They found that on average the friends group got infected two weeks earlier than the random group. Another group of researchers used simulations to show that targeting the friends’ group for immunization achieves herd immunity when only 20 to 40 percent of the friend population is immunized, as opposed to the 70 to 90 percent coverage typically needed when people are randomly chosen for immunization.
One way an employer can apply this principle is to: a) choose a random set of employees and ask each of them to name one person they have the most contact with, b) prioritize the set of people named by the first set of employees for testing — testing them more often or before the others.
Allocation of Scarce Lifesaving Resources
In the early months of the pandemic, many hospitals faced a shortage of ventilators. Ventilators are used to revive patients who have difficulty breathing on their own. There is also a shortage of antiviral drugs such as Remdesivir. Hospitals are forced to ration the critical resource. In the US, typically, states set policies that guide hospital choice. Some states like Michigan prioritize emergency care providers, others like New York have a priority score system which gives higher scores to patients that are most likely to recover due to the ventilator.
While a priority point system — whether single or multi–principle — does provide a well–defined protocol, it does not offer the flexibility to accommodate competing objectives and values. Any set of criteria however fair–minded can be biased in some way as this opinion piece argues. For instance, New York’s principle of saving the most lives has been criticized for being biased against disabled people. Economist Parag Pathak and his colleagues have a proposal which improves upon a priority point system. In their scheme they accommodate different ethical principles through a reserve system. In a reserve system, patients are identified as members of particular groups — e.g., young or old, frontline health worker or not, very sick or sick. Group membership can overlap: a young patient could also be a frontline health worker. In a reserve system, reserve sizes (i.e., the quantity of resources reserved) are set for each group, and within a reserve group, priority can be based on an explicit score or random assignment. They propose a soft cap for the reserved pool where no units are left unassigned and every one is eligible for all available resources but for some reserves there is a priority order. A reserve system provides greater flexibility over a priority point system because it resolves the tension between different ways for prioritizing candidates for critical medical supplies. For example, a fraction is reserved for frontline health workers, while the rest is unreserved for all community members. Frontline health workers are always first in line in their reserve category, and the remaining units can be allocated to the other groups. The University of Pittsburgh Hospital System has adopted this reserve system for fair allocation of Remdesivir which is in short supply. South Carolina is also considering this scheme in its hospitals.
Speeding Up Vaccine Production
Vaccine development is a high-risk endeavor. Private biotechnology and pharmaceutical firms have to spend millions of dollars on vaccine R&D without any assurance of a return on their investment. Several diseases have been neglected by drug companies unwilling to take this risk. Many of these neglected diseases afflict populations in the poorer regions of the world. A proposal to incentivize private sector R&D investments in drugs for neglected diseases is for sponsors to make “advance purchase commitments” for desired drugs, such as an HIV vaccine. Sponsors can be rich–country governments, private foundations, or international organizations such as the World Bank. A commitment to purchase these products in advance of their development creates market incentives for firms to develop needed vaccines. For COVID-19 the idea of advance purchase commitments has been extended to fund the building of manufacturing capacity in advance of drug approval by drug companies with promising vaccines in development. This is expected to cut the time to bring a vaccine to market — vaccines typically take several years to bring to market since drug companies typically build manufacturing plants only after a vaccine has proven successful. CEPI, the Coalition for Epidemic Preparedness Innovations, recently announced a partnership with AstraZeneca which will support the manufacture of 300 million doses of its vaccine candidate. This funding is expected to support the technical transfer of vaccine production technology to manufacturing sites predominantly in Europe, thereby creating additional manufacturing capacity for this vaccine.
None of the ideas we have discussed are new, but it has required a shock from COVID-19 for society to take them seriously. After this experience, we should hope we are better prepared for the next pandemic. While COVID-19 has caused a lot of suffering, it has also spurred innovation and has had some beneficial effects. The air is cleaner, crime is down and doctors are seeing fewer incidence of respiratory infections in children. No one can say whether these effects will last after the pandemic ends. At the same time, some structural changes are likely to last. COVID-19 has placed new constraints on society. Solutions that we did not consider before seem attractive now. Lockdowns have forced some businesses to confront the possibility of working from home, tools for remote working have improved and productivity appears to have increased at least for knowledge workers. People may spend less time commuting to work and work will move to where people want to live. At least in some ways we can hope for a better future.