Why the Hospitalization Rates on Covid19 are so Difficult to Model

Some individuals I highly respect are both convinced that the modeling on future hospitalizations and deaths by highly credentialed institutions and the federal and state government political decisions made as a result of that modeling are highly predictive and good enough for decision making.  I will present a contrary position here: there is so much we not only do not know, but cannot know, that the modeling is inherently imprecise and error-prone.  Moreover, accumulating more data has not materially improved its accuracy and precision.  I will explain why and discuss the policy implications of my conclusions.

Let’s start with a set of basic premises that underly the modeling:

  • Adults who carry the virus can be asymptomatic for 2-14 days and can infect others during that period.  The evidence on the ability of children to infect others is less clear

https://fullfact.org/health/covid-19-in-children/

  • Infections can spread to another person through droplets expelled from the carrier not wearing a mask or other protective covering by sneezing, coughing, loud talking, singing, or spitting, or by having droplets alight on a surface that someone else touches while the virus still is active.  The person to whom the virus is transmitted ingests it through his or her mouth, nose, and eyes.
  • The transmission distance for the virus has been estimated to be as little as 3 feet (WHO) to 6 feet (CDC).  One study has concluded that, under certain conditions, it could be over 20 feet, but Dr. Fauci has considered that study not to be credible.
  • The seriousness of the infection can vary from a person having no symptoms to having so severe an infection that it results in death.
  • The hospitalization, intensive care, and death rates for infected people are unknowable, because we do not know how many people have been infected.
  • However, the evidence is quite clear that each decade of age increases the risk, as do the presence of co-morbidities (other diseases, such as Type 2 diabetes, heart disease, hypertension, lung conditions like emphysema, or metabolic conditions like obesity), vulnerabilities due to cancer therapy, or conditions that require medications that weaken the immune system.
  • The studies from which many of the apparently certain conclusions have been reached are often small studies undertaken over a short period of time that have not been replicated by other studies and have not been “peer-reviewed,” that is, reviewed by an authoritative expert panel, which is the custom for rigorous studies.
  • The use of control groups is often not possible because of the speed with which the studies are conducted.  The government, academic institutions, independent research institutions and healthcare firms have knowingly sacrificed precision to get usable, “good enough” results to others.  Regrettably, the media and the public often fail to understand the limitations on these studies.  How modelers treat them can vary the results from the models.

If we know the facts identified above and were highly confident of the findings from every study, and could accumulate date on social contacts and population health, it would seem, at first glance, that modeling should be able to be done with some degree of precision.

However, here is where the ability of modeling to work breaks down:

  • There are multiple ways to alter behavior, short of issuing stay-at-home orders and shutting down businesses.  What we do not know is whether a tightly enforced mandate to use masks or other face coverings and to practice social distancing while in public would have achieved a significant percentage of the saved hospitalizations.  In other words, if there are four actions in a government directive, we do not know the degree to which each of them has contributed to the outcome.
  • At a detail level, even if we could predict the number of people who are in close proximity to one another by assessing cell phone data (probably the best available source), we do not know whether infected people are engaging in any of the behaviors that would make disease transmission more likely.  We cannot determine whether they were talking, spitting, sneezing, coughing  or expelling their viral loads during an encounter with others.
  • We also do not know whether they were wearing a face covering to protect others or washed their hands prior to touching surfaces.
  • We also do not know whether others in the vicinity were facing toward or away from the infected person, so we cannot assess that factor.

However, let’s assume that we knew how many people could get infected.  Here is what we do not know about the likelihood of progressing to a hospitalization or death:

  • We know that people with particular medical conditions are more vulnerable than others, but our fundamental policy of not allowing personally identifiable health data to be shared without our consent makes detailed model inherently imprecise.
  • Moreover, a health record at the hospital where a Covid19 patient is admitted is highly likely to be incomplete, because we have no way of integrating records with all the different medical events in a person’s life and all the diagnoses that may have been done on that individual.  Because our diagnostic codes determine billing and payment, some diagnoses may not be included in a record, even if relevant to Covid19 vulnerability, because the patient’s health plan will not pay for them.  For example, while some plans will pay for a consultation and a diagnosis of obesity, others will not, so the obesity code will not be included in the record, even if relevant downstream to determining disease susceptibility.
  • We believe that vulnerability is influenced by genetic differences among people, but we do not yet know how, why or to what extent, as this study indicates.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7214856/

  • Epigenetics (whether particular genes are turned on or off) can alter the susceptibility to a disease, such as cancer or a potentially lethal virus.  The epigenetic factors can be environmental (e.g. exposure to air pollutants), behavioral (changed diet or alterations in physical activity levels or even attitudinal (having a sense of purpose in life alters genetic expression to increase the strength of an immune system).  No modeling can get enough data to assess epigenetic factors.
  • By their nature, epigenetic factors not only result in different individuals being affected by contracting the virus at the same time, but the same individual being affected differently at different points in time.  In effect, I might react differently to an infection in February than I would in May, based on epigenetic factors that change for me in the intervening months.
  • There are clear and unexplained differences in individual, ethnic gender, and racial susceptibility to the serious effects of the virus.  Hypotheses about Vitamin D levels in individual systems and blood types have been proposed as explanations, but there is insufficient data to validate these hypotheses.  If they were valid, modeling does not enable us to know who has what blood types or exposures to Vitamin D.
  • We also know that the virus continues to mutate.  We do not know whether what we call the Covid19 virus that infected someone in February is materially different from the virus that infects a person in May. In fact, it is clearly not the same.  What we do not know is how much it has drifted or shifted during that time.

Predictions usually improve as more data is collected, but the data that is being collected is so incomplete and is based on changing population susceptibility and a mutating virus that the usual patterns may not apply.  The modeling is not only not precise or sufficiently accurate, but it cannot get better, given our unwillingness to have detailed health data shared with the modelers or public health authorities.

Public Policy Implications of this Continually Shifting Public Health Landscape

Faced with this level of inherent and continuing uncertainty, what should lawmakers do?

They have to make difficult trade-offs between the negative impacts of shutting down businesses and forcing people to stay at home and the potential benefit of “flattening the curve” in reducing the incidence of hospitalizations.

The best way of thinking about this kind of situation was set out in a 1997 article in the Harvard Business Review entitled “Strategy Under Uncertainty.” Hugh Courtney, the Dean of the Northeastern University School of Business and co-authors Jane Kirkland and Patrick Viguerie identified four types of uncertainty. They draw on research and thinking done by McKinsey & Co.

“This article is based on research sponsored by McKinsey’s ongoing Strategy Theory Initiative (STI).”

What most describes the Covid19 situation is Level 4 uncertainty, described by the authors as follows:

“At level 4, multiple dimensions of uncertainty interact to create an environment that is virtually impossible to predict. Unlike in level 3 situations, the range of potential outcomes cannot be identified, let alone scenarios within that range. It might not even be possible to identify, much less predict, all the relevant variables that will define the future.”

https://hbr.org/1997/11/strategy-under-uncertainty

Later in the article, they provide guidance for decision makers in a Level 4 situation.  These are the key messages from their guidance:

  • A “one-size-fits-all” response is inadequate.  Decision makers must be flexible.
  • Decision makers need to preserve as many options as possible and move from one to another as rapidly as data is available. Big bets are highly risky in this kind of environment.
  • They should always devise strategies and tactics that might work in the widest range of possible futures, and possess little or no downside risk.

These are some of the executive or regulatory options:

Reducing population risk

We know that frail people of advanced age, such as those in assisted living facilities and nursing homes or those in hospice care, are particularly vulnerable to the Covid19 virus or any future virus.  We need to take special care with them.

We also need remote healthcare interaction with individuals who are vulnerable because of pre-existing illnesses, diseases, or care conditions.

Vulnerable populations need special attention and need to be separately managed from the rest of the population.  These conclusions are incontrovertible, even if everything else is uncertain.

Reducing contact risk

  • Mandating the wearing of face masks or other coverings when close to others in public spaces.  Although there is disagreement about the effectiveness of wearing masks to prevent others from getting infected, it is a relatively low risk, low cost requirement.  This would likely reduce disease transmission.
  • Reducing the frequency with which people need to travel to work, school, shopping and social activity.  Many organizations are voluntarily adopting “work from home” practices.  Others rotate workers to reduce the in-office workforce by 50% or more.  Still others spread their workers out to reduce in-person contact.  Some combine these practices.
  • For essential workers who must be at a particular work site, reducing the number of workdays from 5 to 3 a week reduces exposures by 40%.
  • Reducing the contact with surfaces inside a facility by plexiglass shields, cashless transactional activity, reducing the number of locations inside a facility for takeout order pickup or other transactional work, and reducing the amount of handling of items that others will touch can all be part of a risk reduction redesign.

These kinds of changes in micro-environments are best implemented at a local community level.  In fact, not only is federal, state, and county government unit too far away from being able to implement these changes, even a large municipality may be too far away.  The relevant decision unit may be a ZIP code or even a census tract within a city or town, because virus transmission risks may vary that much.

Prohibiting indoor gatherings

Prohibiting large indoor gatherings in which individuals cannot be spaced several feet apart has serious economic consequences, but it is highly likely to reduce the spread of the virus.  There are ways of reducing economic impact, but no ways of eliminating them.  Because of the certainty of benefit, this is a regulatory directive that clearly tilts toward being a good decision in a time of uncertainty.

Outdoor gatherings

Prohibiting outdoor gatherings is more complicated, because there is significant data which suggests that the virus does not survive outdoors.

Moreover, there are more ways to increase space among people at outdoor gatherings and not adversely affect the economics or the wellbeing of those participating in them.  A sporting event can be managed relative to spectator contact with other spectators.

The interesting question is what to do about beaches.  Although it is cumbersome to do so, public beach access can be controlled in multiple ways.  Communities can severely constrain parking near beaches or they can require access permits that are limited in number.  Ironically, there have been many court cases in which communities have been sued for limiting beach access to residents.  The communities have lost in most cases, but courts may be willing to let them temporarily employ restrictive practices that were previously prohibited.

Shutting down businesses and stay at home orders

The decision to shut down businesses is far more draconian and has serious economic and health consequences for both the business owners and the employees.  Unlike the regulatory options listed above, this is costly and creates the risk of irreversible damage to the individuals and communities involved.

Stay-at-home orders have other serious negative consequences, especially when combined with the shutdown of businesses.

These two kinds of directives have serious downsides, but uncertain preventive outcomes beyond the directives listed above.

Constitutional considerations

America has a First Amendment that guarantees freedom of assembly, freedom of religion and freedom of speech.  Political and religious gatherings have to be addressed with attention to these constitutional rights.

Outdoor political rallies and protests are manageable by lawmakers because local governments have always had to issue permits for political advocacy in public spaces.  Indoor rallies and protests can be prohibited or limited because of the availability of technology tools, such as Zoom, in place of indoor gatherings.

Religious services are more challenging because they often require congregants to interact in very personal ways with priests, ministers, rabbis or imams. Religious rituals generally require in-person interactions.  However, even these kinds of services can be managed by limiting attendance at each service and doubling the number of services on a day of observance.

Preventing travel across state lines by someone who has been in a community with a higher incidence of infections is arguably constitutionally deficient unless the federal government mandates it or endorses a state’s decision to implement it.

Conclusions

  • Models are imprecise and can be subject to significant error.  The risk of big error is not likely to decrease in the near term in any situation like this one.  There is too much posturing about “science” and “evidence” and “public health” when the support for mandated actions is inadequate.
  • The tools lawmakers should use need to be those that are likely to be helpful, low risk, and easy to implement, with more limited downside consequences.
  • The best place to regulate public behavior is at the most local level.  Federal and state governments can authorize limitations or prohibitions on migrations into communities, but the exact form of population and contact control is best implemented locally.
  • Limiting the size of indoor gatherings, although costly to those who depend on them for revenue and profit, delivers high benefit.  Despite the high cost, these limitations are defensible.
  • There are more flexible ways of limiting outdoor gatherings.  Those with lower economic impact should be used first.  For example, limiting the number of people with beach access is preferable to closing a public beach.  Communities do this all the time by parking regulations, beach patrols, and making lengthy beach visits unattractive options.
  • Closing businesses and stay-at-home orders are drastic courses of action and their incremental benefit over less restrictive options is unknowable and highly questionable.  The science and the evidence is unlikely to demonstrate that the balance of benefits exceeds the obvious harms from these regulatory orders.
  • Constitutional rights can and should be respected.  Doing so can be done within a public health framework.

There is no perfect solution.  A few people will be exposed to the virus and die, even if everything recommended here is followed, but a better balance can be struck over time.

Lawmakers and public health authorities should strike that balance.  Modeling is unlikely to get them there.  Working closely with a wide range of stakeholders to craft workable solutions for each locality is a far more productive and less disruptive path.