Blog

Methodological Problems in Epidemiology

As much of the world looks to slowly ramp down COVID-19 isolation measures, it remains unclear whether this global social experiment should be considered wise or foolish. The prevalence of infections is < 1% in every country in the world except the microstate San Marino. This is better than projected by most models, and could be interpreted as a success for isolation, an overestimation of the virus's infectiousness, or a natural seasonal effect. This question is not resolvable insofar as it depends on the counterfactual of what would have happened if isolation was not imposed. As mentioned in the last post, spread to 60% of the population with millions of deaths was never realistic. That alarmist scenario relied on a naive application of epidemiological models that have poor predictive ability. Using an SEIR model with the estimated parameters for COVID-19, one indeed gets a grim picture. Yet if one were to insert the parameters for seasonal influenza (R0 = 1.3, avg. incubation period = 2 days, avg. duration of infectiousness = 5 days, mortality rate = 0.1%) into the same model, you would have over 40% infected and 150,000 fatalities in the first year, far more than what occurs in reality. The reproduction rate of a disease depends not only on the duration of contagiousness, but also the likelihood of infection per contact (secondary attack rate) and contact rate. These last two are highly variable by region, social structure, and perhaps even individual physical susceptibility.

Conventional compartmentalized models have poor predictive ability for seasonal influenza, as they do not account for other factors besides herd immunity and isolation that could slow the spread of disease. A Los Alamos study was able to create a model with parameters that fit to past seasonal data and should hopefully have predictive power for future seasons. Such an approach, however, is useless for novel pandemics. As the authors note, these models are all highly sensitive to choice of prior parameters, but we cannot know these until after the epidemic has run its course.

The problem of predictive modeling is exacerbated by the poor quality of public health data, which is often woefully incomplete or inconsistent, with categorizations often driven by policies or other unscientific criteria. Public health systems do a better job of recording the number of infected than they do for those exposed or recovered. Even here they are limited to those who seek medical treatment, and often diagnoses are made by symptoms rather than definitive tests. Cause of death on death certificates is driven by bureaucratically imposed standards. Even in scientific studies, researchers classify subjects according to one or another cause of death, and treat comorbidities as risk factors increasing the chance of death by the primary cause. It would be more rigorous to acknowledge that there is not always a single cause of death, and instead to treat comorbidities as contributing causes by factor analysis. This would let us know the mortality contribution of each disease to the population, but it would remain generally impossible to give a single “cause of death” for each individual.

Some parameters of COVID-19 are fairly well known at this point. The infected are contagious from 48 hours before showing symptoms to 3 days afterward. The secondary attack rate is surprisingly low, only 0.45% (compared to 5%-15% for seasonal flu). Thus the relatively high R0 is attributable not so much to high contagiousness, but to the longer duration of contagiousness, especially while presymptomatic, so that infected people have more contacts while contagious than seasonal flu victims would. The 2009-10 H1N1 pandemic, by contrast, had a secondary attack rate of 14.5%, yet it infected 61 million out of 307 million in the US, just under 20%. It is implausible that COVID-19, with its much lower attack rate, could ever attain a comparable prevalence level.

Why, then, are the death statistics so much higher than would be suggested by the low infectiousness and low prevalence? On the one hand, many jurisdictions, notably New York, have decided to include so-called “probable” COVID-19 related deaths, and most public health data includes no serious attempt to account for comorbidities as causal factors, though they occur in well over 90% of fatal cases. On the other hand, the increase in deaths versus last year in many areas greatly exceeds even this high count, so it could be argued we are undercounting COVID-19 fatalities. The problem here is that many of the excess deaths could be caused not by COVID-19 per se, but by the overloading of medical facilities, resulting in less than immediate critical care. Some of these excess deaths might even be caused by the quarantine measures, as diagnostic and non-emergency medical visits have been cancelled.

It would not be uncommon for the number of deaths to be revised upward or downward by a large factor retrospectively. A year after the H1N1 pandemic, a study suggested that the deaths attributed to H1N1 ought to be revised 15 times higher. Whether H1N1 deaths were undercounted or COVID-19 deaths are overcounted remains to be seen, and is unlikely to be resolved, given the problems of data and methodology we have touched upon.

The truly frightening thing is that major public health policy decisions are made on woefully inadequate data and modeling, which will likely be radically revised after each pandemic passes, and the moment for decision-making is past. Public health officials will always err on the side of caution, but as we have noted in the previous post, this is not practicable for an indefinite period of time. At some point we must be willing to poke our heads out of our caves and assume the risk of living.

After all, as recently as the early twentieth century, people went about their business even while living under the threats of smallpox, polio, and measles, any one of which had higher infectiousness and fatality rates than the current pandemic. By objective criteria, there is nothing exceptional about COVID-19 as an infectious disease. What is exceptional is the post-WWII belief that life should be free from deadly risk, enabled by technological means to perform many service economy jobs remotely.

Overreaction vs. Sober Risk Assessment of COVID-19

COVID-19 was at first believed to be a public health threat on par with SARS, with a mortality rate around 10%. Since then, better data has shown that it has much lower case mortality, comparable to the case mortality of ordinary pneumonia (which is about 1.4%, see here and here). It is a threat only to the elderly and those with pre-existing health problems, again like ordinary pneumonia. Bizarrely, the world’s politicians, public health officials, journalists, and other opinion leaders have instead decided to escalate their reaction, as though unaware of the change in factual reality, or unwilling to admit error.

The most striking thing about the cycle of one-upmanship in overreaction is that the solution is always to curtail freedom. If people are willing to renounce their freedoms over small risks, how easily will governments be able to curtail freedom when there is a more serious threat. As with the exploitation of 9/11, this objective is attained by promoting excessive fear, which short-circuits reasoning even among the educated.

There are two types of factual distortions when making these faulty risk assessments. First, the risk of the new threat is overestimated. Second, already existing risks are underestimated or ignored altogether. These errors combined to create a gross overestimate of marginal risk.

According to a study of 1099 Chinese patients, published in the New England Journal of Medicine, the mortality of COVID-19 is 1.4% of those who test positive. Since at least as many others are asymptomatic and never tested, true mortality is likely 0.5% to 0.8%.

The increased risk of death is mortality times prevalence. In China, prevalence is 1 in 15,000. In Italy it’s 1 in 5000. In the USA it’s 1 in 200,000. In all these nations, the risk of death is less than or equal to dying in a car accident. So driving a car instead of taking public transit to avoid COVID-19 may actually increase your risk of death. In any event the marginal risk, positive or negative, is miniscule. Someone genuinely worried about this level of risk should avoid driving or riding in an automobile.

Suppose that containment fails, as seems likely, and further that this new strain becomes as prevalent as other forms of flu, so that COVID-19 should have about 2% prevalence, i.e. 1/5 of flu cases (10% prevalence). The increased risk of death, compared with average flu mortality of 0.1%, would be 1/50 * 1/200 = 1/10,000. Here I assume mortality of 0.6% for COVID-19 vs 0.1% for average flu. This is to compare apples to apples, since the flu figures include (estimated) unreported cases. Most sites get this wrong, and compare the flu figures for all cases against the COVID-19 figures for positively-tested patients only.

This figure of 1 in 10,000 is likely overstated, since it excludes consideration that many of the “excess deaths” are in people with preexisting conditions who would have died of something else shortly. This pessimist scenario, in a nation of 300 million, would result in 30,000 excess deaths.

Preventing such a scenario is certainly worthy of strenuous measures, but not without limit. One must also consider the effectiveness of such measures, and the cost in terms of public health. Sinking the economy and depriving people of months of income may cause comparable excess deaths, especially if people are prevented from getting cancer screenings as some health systems are recommending. 30,000 excess deaths represents a 50%-60% increase in annual flu deaths, but there are other bigger killers, both those existing, and those we may create by excessive reaction to this new public health risk. A simplistic attitude that “no measure is too big” fails to be a rational form of risk management.

At some point, we may have to grapple with the possibility that containment does not work. The USA may not have the same legal means at its disposal to compel quarantine that may exist in the more centralized authority of Italy or China. Also working against containment is the low mortality rate, the possibility of carrying the virus in mildly symptomatic and asymptomatic individuals, and the unusually long incubation period. Indeed, once the virus proliferates beyond a certain threshold, containment of COVID-19 would seem to be as impracticable as containing the common cold or the flu. While we may not have reached that point yet, we must recognize the possibility that at some point continued efforts at containment are not worth the cost, simply because of their low probability of success.

The reactions have been so rapid, and so outpace the actual facts on the ground, even when the number of cases is statistically negligible, that we cannot consider them to represent the result of careful deliberation. Rather, as in the closure of multiple universities on the same day, it is more like the imitative behavior of a panicked and stampeding herd. In such a climate, it may take more courage to do less than to do more. It is very easy to say that money is no object and leave the private sector to pay for government largesse. Those of us who have to make budgets and do not have the power to print money may have a different perspective. This is not a mere economic problem, for it can swiftly transform into a humanitarian catastrophe at least as great as the one ostensibly being prevented.

Update: 28 March 2020

Misinformation continues to spread. First, there is the oft-repeated claim that, absent our draconian containment measures, the virus would spread to 60% of the population, resulting in millions of deaths in the US. This is a cumulative figure over two or three seasons, ignoring the near-certain fact that pharmaceutical measures and natural antibodies will reduce the virulence of the disease by next season. It is effectively an impossible scenario, and again is not comparing apples to apples, as the seasonal flu death figure is annual.

Second, the mortality rate continues to be overstated. As testing becomes limited only to those who are hospitalized, the “mortality rate” of tested positives will increase, since you are actually measuring only the most severely affected subset of cases. Worse, in Italy, anyone who dies with coronavirus is counted as a death due to COVID-19, although 99% of fatal cases had comorbid conditions. The best data from South Korea, which has far more aggressive testing, currently points to a mortality rate of 0.7%. Using this figure as an upper bound and applying the more exact population of 327 million for the US yields a “pessimist” scenario of 39,000 excess deaths this season. We may get there anyway as outright containment has proven ineffective, and we are now hoping only for mitigation, i.e., slowing the spread.

The Collusion Delusion

Whether you are more concerned about the content of the Pentagon Papers or the fact that they were leaked illegally is determined by your stance on the Vietnam War. Likewise with the 2016 leaks of e-mails of the Democratic National Committee. Are you more concerned that the DNC was in the tank for Hillary Clinton and that her populist stances were just cynical hypocrisy, or more concerned that this was leaked illegally? Again, this depends entirely on your politics. Among the e-mails, we found evidence of a close relationship between the DNC and journalists, with the former asking the latter to run favorable stories when needed. We should not be surprised, then, that the media, having its last veneer of impartiality thoroughly shredded, should seek strident retribution against those who committed the leaks, and the Trump campaign which benefited from them. This has elided into a brazen attempt to delegitimize the outcome of the election, and reverse it if possible through impeachment.

This farce began with a joke made in response to another e-mail scandal, that of Hillary Clinton’s private server while she was Secretary of State. In response to a subpoena, she provided the e-mails on this server only after deleting tens of thousands of messages that were supposedly personal and unrelated to work. Candidate Trump called out this blatantly illegal non-compliance, humorously imploring anyone who had the e-mails, even Russia, to release them as a public service. Naturally, the media went ballistic, accusing Trump of encouraging espionage by a foreign power. They evidently could not keep their lies consistent, for there could be no damaging espionage if the e-mails merely pertained to wedding and yoga appointments, as the credulous press would have us believe.

In July 2016, the first set of DNC e-mails was released by Wikileaks. These revealed that DNC chair Debbie Wasserman Schulz was firmly behind the Clinton candidacy, long before the primaries had been decided. This surprised virtually no one, as bias against the Sanders campaign had long been evident, and the clearing of the field by other Democrats for Hillary to run virtually without opposition in her party was transparent to all but the willfully naive. Only specifics, such as the leaking of debate questions to the candidate, and the enlisting of DNC resources and press allies, added information about the depth of the establishment collusion to force Clinton upon the public, as later confirmed by Donna Brazile.

The second set of e-mails was leaked on the same weekend when Trump’s Access Hollywood tape was leaked as an October surprise to derail his candidacy. The press generally ignored the fact that the behind-the-scenes tape, being proprietary and confidential, must certainly have been stolen or leaked illegally at some point, no less than the DNC e-mails. Again, whether you care more about the content or the mode of release depends on your politics.

Democrats naturally railed against the “Russian hackers” committing espionage against our national institutions, ignoring that the DNC is a private, non-state institution, and that the e-mails were obtained by phishing, not hacking. That is, someone was dumb enough to give away their password to an e-mail scam. John Podesta did this twice (having been misled by his IT person who omitted the word “not”), and we were treated to a host of e-mails showing the cynicism of Democratic party strategy and Hillary’s two-facedness regarding Wall Street. This should have surprised absolutely no one, and indeed this second wave of e-mails made no measurable impact on the polls, having been drowned in the Access Hollywood scandal.

All of this would have come to little had not Donald Trump, ever so improbably, won the presidential election. Hillary Clinton soon had the consolation prize of “winning” the popular vote, though in fact she had less than 50%, so she would have lost in a House vote even had there been no electoral college. Ironically, before the election, the electoral college was thought to give Trump an impossibly narrow path to victory, allowing for no mistakes in major battleground states, and on top requiring him to flip some Democratic strongholds. He did precisely that, in part by keeping a grueling travel schedule to the Midwest in the final weeks. Mrs. Clinton, by contrast, feared that Trump would win the popular vote while losing the electoral vote, and devoted resources to padding turnout in non-competitive states such as Illinois. Rather than come to terms with their candidate’s blunders, many Democrats soon took up the “blame Russia” angle.

This, of course, is historical revisionism of the first order. Neither of the Wikileaks releases decided the election in any measurable way. The second batch had no impact on polls, while the first only confirmed an already prevalent sense that the DNC favored Clinton over Sanders. The big needle mover in the final weeks was James Comey re-opening the e-mail server investigation, having found that Anthony Weiner (aka Carlos Danger) had been printing Hillary’s e-mails for her on behalf of his wife Huma Abedin, so his computer might have some of the undisclosed e-mails. This, like Hillary’s e-mail server itself, was simply a workaround for a technologically inept executive. Polls moved appreciably after this re-opening of the investigation, closing the gap between candidates to the margin of error.

Although Obama had known for over a year about the phishing of the DNC, he maintained public silence about possible Russian government involvement, to avoid appearing unduly partisan during the election. It was only after the election that he decided it was significant enough to disclose publicly. While disclosing only weak, equivocal evidence of Russian government involvement, he acted as if the fact were certain, and tried to make this opinion a reality by imposing punitive sanctions against Russian diplomats.

The Americans are shocked – shocked! – that a government should interfere in the political process of another country. Yes, this is the same U.S.A. that regularly bombs countries, foments coups, plots assassinations, and even bribes legislatures to change political outcomes. The U.S. is by far the biggest political meddler in the world, and the biggest practitioner of global espionage, even on allies and the UN. Most notably, it interfered in elections in Ukraine, favoring an anti-Russian party. As always, the U.S., without irony, fails to recognize blowback of its own imperialist actions.

Whether feigning outrage or genuinely shameless, Obama imposed sanctions as a lame duck president trying to force a major foreign policy stance on his successor. The center-left imperialist media did not remark on the inappropriateness of such action, but on the contrary acted as though the president-elect had no right to let other nations know what his intended policy toward them would be. If Trump failed to be duly hostile toward Russia, or even hinted that he intended to reverse Obama’s petty vindictiveness, he would be not so subtly accused of treason. This is an ironic charge from a gang of globalists who have consistently sold out their working class countrymen.

At any rate, the evidence of Russian conspiracy is astonishingly weak in proportion to the political mileage that’s been extracted from it. As Jeffrey Carr points out, it’s doubtful if the “hackers” even spoke Russian. Yes, the hackers were likely based in Russian time zones, but the vast majority of illegal Internet activity comes from Russia, as anyone who runs a website knows. This is hardly proof of Russian government involvement. More significantly, the data extraction tool is one used by a former group believed to have been with the Russian government, but this identifcation is not definitive. Even if the tool were a Russian government creation, that is not proof of involvement, since government hacking tools do get leaked. Such was the case with this year’s Wannacry ransomware attack, which used a leaked NSA exploit. There is nothing technologically sophisticated enough in the DNC spearphishing that could not be done by any reasonably computer savvy individual. Even a leaked NSA document acknowledged there was no direct evidence of Russian government connection, but this was only an inference made by analysts.

Once the “Russian hacking” is made a fact by Obama’s lame duck meddling, the accusation of collusion between the Trump administration and Russia can be made self-fulfilling. Any post-election attempts at detente are portrayed as evidence of such collusion. Most accusations are made only by innuendo. Ironically, Trump’s greatest error, from an optics standpoint, was his firing of James Comey, the same man who did more than any hacker, Russian or otherwise, to cost Hillary the election. We should not expect logical consistency, of course, in politically motivated accusations. Though their errors are comical, the leftist elites are not to be smiled at. They are making unfounded accusations for criminal offenses that can put people in prison. They have no compunction about ruining people’s lives, even non-politicians like the president’s son, in order to score points for the next election. This is, after all, an elite that can kill ten thousand Libyans for a marginal political advantage.