read

Life Contingencies: An Actuarial Look at COVID-19 Mortality

Essay by Peter Neuwirth FSA, FCA

“The work of science is to substitute facts for appearances, and demonstrations for impressions.” — John Ruskin (motto of the Society of Actuaries)

The Actuarial Perspective

As our economy opens up and the issues around the risk of becoming infected and dying from COVID-19 get more complicated and confusing, it becomes harder and harder for individuals and organizations to make well-informed decisions. I believe, at least with respect to one key aspect of these decisions, the actuarial perspective can be extremely useful.

I haven’t stopped wearing my mask, I still socially distance, and I try not to break any of the other increasingly elaborate protocols that we are being asked to adhere to in order to keep ourselves and each other safe.

But being an actuary, I started to ask myself some questions – questions that nobody wants to think about and ones that are almost impossible to discuss in public without emotion or being asked to choose sides. Nevertheless, they are questions that I think are vitally important for us to answer as we figure out how to live our lives in the age of COVID-19

It started innocently enough when a friend sent me an article about what had happened at Woodstock more than 50 years ago. The author discussed how that event, as well as numerous other concerts during 1969 all occurred during the 1968-69 “Hong Kong flu” pandemic. Because the article itself was written with an obvious political agenda, I was tempted to simply ignore it as just part of the angry noise which has become all too prevalent these days, but because I grew up in that era, I couldn’t help but continue reading and then conduct my own fact check – reviewing the historical documents and my own memories of the times to verify the truth of what he said.

What I discovered was that despite some exaggerations, leaps of logic and conclusions I radically disagreed with, the article described some indisputable historical facts about that epidemic, which was undeniably a public health crisis, albeit one of an entirely different order of magnitude than the one we are facing today. In particular, it seems that in 1969 there was widespread ignorance, or at least a casual disregard for the health risks we took—both on an individual and on a collective basis.

Granted the world had quite a few distractions at the time — Vietnam, Civil Rights and the Woman’s Movement to name just a few. But still —- nobody proposed closing schools to limit the epidemic, few wore masks, and there was no “social distancing,” though used for centuries, the term itself was only coined in 2003.

No one cancelled concerts either —- not even a 400,000-person event in the middle of a rain-soaked inaccessible field in upstate NY or a 300,000-person event (the infamous concert at Altamont) that occurred at the height of the second wave of the epidemic in December.

Here is a relatively balanced article on life during the 1968-69 flu epidemic and here is a research paper suggesting that the media should have become more alarmed.

One could argue that we were less fearful back then. I believe what was really going on in 1969 was that we just didn’t have good data in real time and we didn’t have the internet to help actuaries and other researchers collaborate globally at the speed of light – to collect, analyze and digest the data and provide their findings to policy makers who could then implement protocols and propose action steps that would mitigate and curtail the suffering and death that epidemics and pandemics can cause.

In theory we should be in much better shape today. In practice we are not. We have the data and the internet, but we are having a noise problem.

In particular, we have too much data. Specifically, we have too much bad data that confuses us and makes analysis – at least actuarial analysis — dramatically more difficult. Data can be bad for all kinds of reasons. Deliberately “faked” data is only one of many potential problems. Flawed and/or poorly designed collection methods, errors made through inattention or stress, honest bias on the part of the “collector”, and inconsistent or unclear criteria used for different information sources are just a few of the many ways that data can degrade.

For me as an actuary bad data is simply that – data that needs to be cleaned. It doesn’t matter how it got dirty; it still needs to be scrubbed to make it usable. More precisely, you can use bad data, and in fact sometimes you have to. But when you have bad data, you have to be careful how you handle it and your conclusions need to be more tentative.

In light of the above it was both understandable and prudent for us to use the precautionary principle when facing an unknown risk like COVID-19. In fact, as Nassim Taleb points out in this interview the precautionary principle should always be applied when risks are systemic and unknown – a situation we faced a few months ago, but now that we know much better what we are dealing with, we should be able to quantify the risks we are facing.

Earlier this year, the country didn’t follow the precautionary principle and was caught unprepared. As a result, there has been an enormous amount of suffering that likely could have been avoided – far more suffering than many thought possible. Extreme caution may still be justified, but now I think we need to pause and take a hard look at what has happened so far and what might happen in the future.

Many experts, politicians and commentators are purporting to do just what I am suggesting, but among all those voices, I think there is one group of experts that has not been heard from enough – and that is the actuaries. The core expertise of the actuarial profession is to evaluate risk and the core risk that we evaluate is the risk of dying.

That’s what the rest of this essay is about. It is important to note that I am not writing this on behalf of the actuarial profession. Rather I am simply expressing my own personal view – looking at the situation through the lens of an actuary who both survived the Hong Kong flu as well as the Vietnam War and all the other hazards of the late 60’s and early 70’s.

The Scope of the Problem

Despite the fact that COVID-19 is a new disease, and its impact has been catastrophic it might have been much worse. In fact, rightly or wrongly initial projections suggested that millions of Americans would die from COVID-19 if no steps were taken.

Yet the steps we have taken have led (regardless of who you blame) to a blindingly fast descent into economic and psychic distress as well as noise, confusion and now violence. It seems to me that given the stakes and given the fact that the behaviors of all of us as individuals will impact both the course of the pandemic as well as where the economy and society in general goes, we need to figure out just how great our risks are and what we must do to learn to live with whatever danger there is.

I decided to take a hard look at the data myself.

The core of what we actuaries do is to work with what are called “Life Contingencies”. Fundamentally, Life Contingencies is about determining your chances of dying and quantifying the implications those probabilities have for other aspects of our financial lives (the cost of insurance, annuities and future workforce demographics).

I decided first to focus on the narrow question of what COVID-19 has done to our probability of dying – in the next few months and longer term, until the pandemic is over. I believe that answering that question is essential for navigating our way through the rest of this pandemic

Data and Mortality Rates

When actuaries think about mortality rates, we start by looking at the data – how many of us are dying and why. That may seem like a straightforward exercise, but in fact it is not.

By Memorial Day weekend, Johns Hopkins University (JHU) had reported that slightly more than 100,000 Americans had died as a result of contracting COVID-19. While that may have been a good estimate, it is a messy number that is the result of a lot of judgment calls made by a multitude of coroners, doctors, scientists and employees of county and state health departments throughout the country

The actual number of people who die each year is not controversial. We are able to keep track of births and deaths in this country, and pretty much every death comes with a death certificate that is filed at the county health department where the individual dies.  That death certificate will include a cause of death. The number of those death certificates that note COVID-19 as the cause of death is what JHU is keeping track of.

As almost everybody knows by now, whether someone has died of COVID-19 is not an easy determination. The first question that arises is whether it was the COVID-19 that caused the death or did the individual die from something else and simply happened to be infected with COVID-19 at the time of their passing. Most counties have deemed anybody who dies having tested positive for the virus as a COVID-19 death, and that is why some think the 100,000 is an overstatement. On the other hand, there are also many people who have died at home from something that could have been caused by COVID-19 like pneumonia, heart attack, etc. but who were never tested. That is why other people say the 100,000 is an understatement. Even more confusing is the fact that the Center for Disease Control (CDC) who is supposed to be tracking the same data as JHU has a different total of deaths. That difference could stem from all sorts of factors ranging from transmission errors (surprisingly common) to time lags between when the deaths occur and when each organization receives and tallies the death certificates.

As confusing as the situation is, fortunately there is an “actuarial shortcut” we can use to get a better estimate of a slightly different question which is almost as important – i.e. “How many people have died as a result of the COVID-19 pandemic?” The way we do that is by not focusing at all on the cause of death, but instead just looking at how many extra deaths we have suffered relative to those that might have been expected to occur if there had been no pandemic. That number won’t tell us much about the risk of dying from the disease (we will get to that later), but it will tell us what the impact of the pandemic has been. And while that information is not that useful for an individual assessing their risk, it is important for policy makers and modelers who want to assess the future course of our economy as well as for Life Insurance companies who need to determine the pricing of their products and who have a keen interest in how many deaths claims they will have to pay out in the future.

Excess Mortality due to COVID-19

In order to track excess deaths due to the virus, we first need to determine how many people we would expect to die in a normal year. Surprisingly, in normal times this number doesn’t vary much from year to year because the America’s population is so large and as long as the average risk of dying doesn’t change from year to year, the number of deaths experienced each year from all causes combined will stay relatively stable.

In fact, every year, a little less than 1% of the population (2.8 million) can be expected to die. Things do change as health care improves and periodically other factors intrude to complicate the actuarial calculations. Actuaries try to take that into account as mortality rates change. Most recently, actuaries have had to take into account the opioid epidemic which has offset (but not completely) the year-to-year reductions in mortality rates that actuaries have come to expect as our medical technology improves. Click here to look at the kind of analysis the Society of Actuaries conducts each year.

As we will discuss later, SOA has also begun to look at the question of excess deaths directly using CDC data. This analysis is still in process and when complete will give us a much better understanding of the impact of COVID-19 on overall mortality. However, I don’t want to wait,  and therefore I took a look at the CDC’s data directly which they use to track the total number of deaths in 2020. I then compared this number to their determination of “expected deaths during 2020”. So far, the CDC has determined that through the middle of May, actual deaths from all causes are running at 103% of expected

Scroll down to the first line of the chart.

Because through mid-May CDC had expected almost 1 million Americans to die anyway, this translates into about 30,000 extra deaths and since the only really unusual aspect of 2020 is the presence of COVID-19, we can reasonably assume that so far, the extra deaths caused directly by the virus has been somewhat offset by reductions in deaths from other sources (e.g. traffic accidents and maybe even flu deaths). It is also likely that deaths from heart and lung disease are slightly lower this year because some of those who got sick and died from COVID-19 would have died anyway from their underlying medical conditions.

While the above may seem like good news, there is no guarantee that the reductions in deaths from causes other than COVID-19 will continue to dampen the effect of the virus on overall mortality rates. In fact, there is cause to be worried. This is because first, the opening of the economy may cause traffic accidents and other hazards of modern life to go back to their “normal” level of danger and second, because it is possible that other factors (like economic hardship, mental health issues and the impact of deferred medical treatment during the crisis) will cause this offset to disappear and even reverse. As a result, in the coming months, there may be more deaths due to the pandemic and its indirect effects.

At this point, you may be wondering what good actuarial analysis is if we can’t say definitively what will happen in the future, but that is exactly the point. As actuaries, we are trained to observe and discern what is happening today and then to use that knowledge to develop plausible scenarios for what might happen tomorrow. When I look at the data, my conclusion is that so far, at least with respect to deaths, COVID-19 has not been as bad as many had feared. That is not to say that there aren’t many segments of society that have been devastated by the pandemic, in human and economic terms. In fact, Nursing Homes, Hospitals as well as the Transportation, Hospitality and other sectors of the economy have been dramatically impacted. On the other hand, the overwhelming majority of us are still healthy and many industries have been surprisingly unaffected. For example, the Life Insurance industry which is tasked with protecting families and others against economic losses due to mortality has so far come through the crisis in good shape.

But what about the future? How bad could things get and is it possible that many more people will die? This is a critical question for both individuals and organizations (including Life Insurance companies) and while I will not add my prediction to the overwhelming number of forecasts out there, I will talk about how an actuary (or at least this actuary) looks at a couple of key variables that will determine the answer.

IFR, “Herd Immunity” and how to evaluate your own risk of dying

The question in the back and the front of the mind of almost everyone I know is: “What are my chances of dying from this thing?”

As intractable and unknowable as it might be for any individual to determine their exact probability of dying from COVID-19, actuarial analysis can provide a very good estimate of that probability for our population as a whole and for age groups within the population.

Simple math tells us that the answer to the above question is the product of two separate probabilities:

Probability #1 — The probability of getting infected with the virus

        Multiplied by

     Probability #2 — The probability of dying if you get infected

Getting an estimate for probability #1 is very difficult, but we know it can’t be more than 100% and is likely to be somewhat less. One important way we can estimate how much less than 100% is by considering how many people will be infected by the time the pandemic ends. Virologists and epidemiologists speak of the concept of “herd immunity” which says that by the time a certain percentage of the population has been infected, the virus will die out.

  • For different diseases the level of exposure necessary to achieve herd immunity will vary considerably, and for a new virus like COVID-19 there is even greater uncertainty. Here is a basic description of herd immunity.
  • A more comprehensive discussion of herd immunity and COVID-19 (where the author estimates it at 67%) can be read here.

If we assume that herd immunity for COVID-19 is achieved at 60%, then we have a value of .6 for probability #1. Of course, if you are more careful than your neighbor you might be able to make sure you are not one of the 60% who eventually get infected, but for now we are just trying to put an upper bound on your probability of dying, so let’s assume you don’t do anything extraordinary (vs everyone else) to protect yourself and we will use 60% as our estimate.

At the end of this essay, we will talk about other reasons why your probability of getting the disease might be less than 60%, but let’s now move on to probability #2; your chances of dying if you get infected. This is called the Infected Fatality Rate (IFR), and there is a great deal of confusion and misinformation circulating about this probability. Therefore, it is important to carefully review the relevant data. Many media articles have spoken about “fatality rates” of 5% or higher on average along with truly frightening numbers ranging up to 20% for the oldest members of society. Lately more and more articles are recognizing that just taking the ratio of deaths/confirmed cases grossly overstates the probability of dying from COVID-19 by confusing Case Fatality Rate (CFR) with the IFR that we spoke about earlier.

In its Research Brief on the Impact of COVID-19 the Society of Actuaries noted that estimates of the IFR of the virus are between .1% and 1.0%. The entire study can be viewed here.

That the SOA has put such a wide range on this probability reflects the innate conservatism of the profession. In general, that’s a good thing, but I think today we can get a little closer to the actual number. In particular, testing has improved, and new studies are being conducted that include determination of COVID-19 status of large samples of individual pulled from populations that include both suspected cases and those not obviously exposed to the virus.

For example, the SOA research brief says:

“Data from Iceland, where testing has been more broadly implemented than in other countries, indicates that the prevalence may be around 1%. Using the deaths that have occurred in Iceland and its population age mix, this would extrapolate to a global CFR of about 0.6%.”

Additional studies are being conducted on almost a daily basis, and the Society of Actuaries is doing their best to keep up with the flood of new data. In fact, they conduct and continuously update their analysis of exactly the questions I am discussing here.

  • In one of the most relevant studies, SOA digs more deeply into the sources of death data and discusses the problems associated with teasing out information on those dying from COVID-19 deaths from those other causes. View that here.
  • Another SOA  study addresses the question of “how bad will it get?” but as you will see, the SOA is reluctant to speculate on just how many people will die of the virus because of the many many uncertainties we have been discussing. Click here to read it.

I fully support my colleagues’ conservatism in this regard, but I also believe that it is worth trying to try to look down the road a little bit and estimate the implications of what we have discovered so far about COVID-19 mortality.

More recent studies suggest that the actual IFR is on the low side of range that the SOA proposes. In fact, the CDC recently updated their analysis of the probability of dying after infection and their best estimate is now slightly under 0.3%. The CDC has even broken this probability out by age category and tried to take into account an additional confounding problem with some of the original IFR studies – that a large number of infected individuals show no symptoms at all. Their full report can be read here.

CDC’s current best estimate of age-based probabilities of dying if infected (discounted for the 35% of infected who are assumed to be “asymptomatic”) is:

  • Age — < 50   Probability of dying = .0003 (approximately 1/3000
  • 50-64 Probability of dying = .0013 (approximately 1/750)
  • 65 +   Probability of dying = .0085 (approximately 1/120)
  • Overall probability of dying if infected = .0024 (approximately 1/400)

Note that if you want to use those numbers to assess your own risk, you need to be aware that those numbers don’t take into account herd immunity (or other ways the pandemic could end before everyone is infected). You also need to recognize that those rates are not only average rates for each age group, but that if you are at the older end of that age cohort your risk could be substantially higher than the average. On the other hand, if you believe that your chances of catching the virus is less than 100% you need to discount those probabilities further.

If we assume that 60% of all Americans eventually get infected that will give rise to a total death toll that can be directly attributed to COVID-19 of 475,000 (.0024 x .6 x 330,000,000). This amount is, of course horrifying and far worse than the Hong Kong flu, but it is nowhere near as bad as several of the mass mortality events (e.g. the Spanish flu and the Civil War) that have befallen this country in the past.

And it might not even be as bad as that. Here is why.

Actuarial Assumptions and some Final Thoughts

One of my former actuarial colleagues was once asked by a corporate client how confident he was in his estimates of the company’s projected pension costs that was in the Actuarial Report. These projections were going to be of critical importance to management’s strategic plans for the coming year and so his question was very reasonable. The actuary told the client that the one thing he was sure of was that his projections would be wrong. He couldn’t even say by how much they would be wrong, but what he could say was that those projections were his best estimate of what might happen based on the most accurate and complete current information he could obtain and actuarial assumptions that he had developed consistent with the purpose of the projection. The rest was arithmetic. That is the nature of what actuaries do – we make our best guess about what the future might look like based on what we can discern about the present and reasonably assume about the unknowns – both the incomplete information we have, and the way things will change over time.

It is up to you to decide whether the assessment of our risks that I have laid out above is hopeful or discouraging. My sole concern is that the numbers I have provided are based on sound actuarial methodology and assumptions that are “reasonably conservative”. The principle of using “conservative” assumptions is considered by many to be an “actuarial best practice” and for this purpose, I think conservatism is entirely appropriate

So how are my assumptions conservative? Well, first I’ve assumed no vaccine is developed before herd immunity is achieved. If one is developed, that could dramatically reduce the number of Americans who are ultimately infected to a percentage lower than the 60% I have assumed. A lower probability of getting infected, translates directly to a lower risk of dying from the disease.

Secondly, I have assumed that we don’t get better at preventing those infected from dying. The race toward an effective therapy continues and it is highly conceivable that the IFR rates that inform the CDC’s estimates could go down further as more people don’t progress from symptomatic to hospitalization to death. We will undoubtedly get better at treating the disease, but I have assumed that we won’t get better fast enough to make a difference.

Finally, I have assumed that there will be no benign mutation in the virus itself. There is a large body of literature among virologists that suggest that the natural tendency of viruses is to evolve toward becoming less virulent over time. The basic rationale is that the most virulent strains kill those that they infect and therefore don’t get to be passed on to others, while the weaker variants continue to infect a greater and greater portion of the population. I have not taken this potential scenario into account even though many virologists believe it is possible, if not likely. Click here for more.

At the end of the day, we all need to make our own personal choices about how we behave socially and interact with others as long as the pandemic persists.  I have tried to do here what actuaries are best at – to detach from any value judgment about how we are behaving now or in the past, but instead to objectively and dispassionately evaluate the risks of COVID-19 to ourselves and those that we care about – to, as Ruskin says, “substitute facts for impressions”. If I have been successful, you won’t know what to do, but you will have just a little more information to make the difficult choices that we have to make every day.

 Note on Reviewers:

I believe in independent qualified peer review for any serious exploration of these kinds of issues, but to my mind the “qualified” part is just as important as the “independent”. In this instance I have asked several actuaries and other qualified experts that I know to read this essay and give me feedback. For those that provided substantive review I am extremely grateful and have incorporated many of their suggestions. A partial list of those who were kind enough to look at this piece before it was published includes:

Specific Comments made by reviewers not reflected in article:

James Kenney suggested that my estimate of COVID-19 deaths might be too low because 100% is not the maximum value for “probability #1” since we still don’t know whether or not someone can get “re-infected” with the virus. It’s possible that someone could survive an initial infection but succumb to a subsequent infection months or years from now.

Eric Baum suggested that my estimate of COVID-19 deaths might be too high because more current projections suggest that the pandemic “appears to be petering out” possibly because of partial immunity that already exists within the population

Tom Herzog suggested I have avoided complex mathematical constructs, but I could have used a “massive computer simulation (an easy task for modern computers) and the results could then be presented in the form of an entire (predictive) probability distribution Instead of a point estimate.”

David Ballard suggested that “The rising case fatality by age is very important as is the increased mortality associated with comorbidity. These issues probably introduce an important component of complexity to this analysis that is very important for actuaries to include in their estimates of excess mortality for different population subgroups.” He also suggested the excess mortality associated with deferred medical care could be significant in the coming months/years.