Uncertainties about Japan’s COVID-19 data

In our earlier posts (here and here) we tried to clarify the actual policies being implemented in Japan with regard to COVID-19 testing. We pointed out inconsistencies and other official data issues that make it difficult for the public to get a good grasp of the actual number of tests being done and how many people are probably infected now. Four weeks have passed since our last post on this issue. At the time, 20,340 tests had been reported in Japan, with 1214 positive cases nationwide, 171 in Tokyo. Nationwide there were fewer than 100 new cases reported daily. At the time of writing today, the most recent data available shows 150,692 people tested, 13,448 total positive cases nationwide, and 3,908 cases in Tokyo. The highest daily count was reported on April 11, with 700 cases, compared to the current average of about 400 per day.

These counts, of course, represent a notable increase over February and March, which should have silenced the voices that insisted that Japan had already “beaten” the coronavirus. At the same time, while the increase is large, and the health care system already shows worrying signs of overloading, the country does not yet appear to have entered a phase of devastating exponential growth like that seen already in many places in the world. As we frequently point out, it is very difficult for even symptomatic people in Japan to get tested. We found this to be of such concern that we created an online map where people can report their experiences. Considering the small number of tests being done, as gauged by the percentage of the population which has been tested (roughly 0.008%), how soon would we know that a larger outbreak was happening? We must ask the same questions we asked in March: How many cases are being missed by the current “cluster countermeasures” approach, which intentionally limits testing? How dangerous might the situation become if a much larger proportion of COVID-19 positive people, specifically those who are asymptomatic or pre-symptomatic (and so not likely to qualify for testing under current guidelines), is not identified and quickly isolated to slow down the spread of the virus?

Dr. Kenji Shibuya, Director of the Institute of Population Health at Kings College in London, gave an online presentation to the Tokyo Foreign Correspondents Club (FCCJ) last week in which he laid out the implications he sees in the available data and clarified areas of concern. As we noted previously, he has raised cautions about the risks of relying exclusively on the “cluster countermeasures” approach, and has urged a much larger expansion of testing in Japan. We encourage everyone to watch the video of his presentation. Specifically, he estimated that the actual number of infections in Japan is likely to be at least ten times higher than the number currently being reported through official channels, for a number of reasons. The following day, Hokkaido University professor Hiroshi Nishiura, a key member of the Ministry of Health, Labor, and Welfare’s task force on coronavirus clusters, gave a press conference in Tokyo in which he concurred with Dr. Shibuya’s estimate, adding, in fact, that it may be even higher. At the same time, Nishiura reiterated several times that data now shows a decreasing trend of cases in the nation. The most recent data graphs seem to show that as well.

Nationwide case count for Japan as of April 27, 2020 (source: https://covid19japan.com/)
Tokyo COVID-19 case data, as of April 27, 2020 (source: https://stopcovid19.metro.tokyo.lg.jp/en/)

As Dr. Shibuya pointed out in his presentation, there are many caveats that should be kept in mind when looking at the numbers reported by Japanese government sources. For several reasons, he believes it is difficult to draw firm epidemiological conclusions from the data we are given. The implications of the roughly two-week delay between the onset of a COVID-19 infection and symptoms becoming serious enough that the patient fulfills Japan’s stringent PCR testing criteria are very important and should be understood. The daily numbers for positive cases we are given only indicate the date of compilation or reporting of cases which have already existed for weeks. These numbers are a lagging indicator, a narrow snapshot of the spread of infection two weeks prior. If much more extensive testing were being done, Shibuya says, particularly of asymptomatic and pre-symptomatic cases, we would have a fuller and more current picture of the spread.

Another caveat is that results for tests done at private laboratories are not fully represented, due to delayed and inconsistent reporting. As an example, although the bilingual online dashboard provided by the Tokyo Metropolitan Government is one of the better official communication efforts, data from private labs is only included as a weekly total added every Friday, as opposed to the daily reporting from public institutions. Consequently the case numbers appear to spike mysteriously every Friday. Similar spikes affect the testing total graphs. The explanation for this is not evident anywhere on the city’s webpage.

Very little of Japan’s testing and case reporting system is automated or fully online. Unbelievably, all of the data is still sent by fax, and needs to be manually transcribed, entered, and compiled. This introduces unnecessary delays and makes the likelihood of data entry errors higher. Considering the inefficient data compilation process, the difference between what the data categories actually represent as opposed to what they appear to represent, and ongoing inconsistencies regarding the number of tests reported, it is hard to evaluate what is actually going on based on these numbers alone. Most concerningly, none of this data is publicly available, so outside researchers have no way to independently validate, replicate, or scrutinize it. Shibuya related his experience trying to obtain COVID-19 data from Japan’s National Institute for Infectious Diseases (NIID) to analyze, which should be a very straightforward process, particularly for a researcher of his status and reputation. He was told that he would need to fill out a number of specific forms, and that the process could take months. Without access to the background dataset held by the NIID we can tell very little from the publicly available data.

A recent study based on a large Italian data sample from Lombardy indicated that 43% of the COVID-19 transmissions seen there arose from asymptomatic infections. Also, most new infections occurred before the lockdown, from asymptomatic infections in the same household. Other studies reinforce similar conclusions regarding the predominance of asymptomatic infections in the population. Despite the reassuring recent Japan case graph curves, there remains a strong possibility that Japan has entered an explosive phase of COVID-19 transmission which the current national disease surveillance system has not yet detected because asymptomatic and pre-symptomatic cases, which form the majority, are almost entirely excluded from the testing criteria. This is bolstered by recent reports from Keio University Hospital that of 67 patients admitted to the hospital for non-respiratory reasons who were screened for COVID-19, four of them (6%) were found to be positive. Shibuya and others note that this is neither a large nor representative sample, but the prevalence is unexpectedly high and should not be ignored.

Based on known asymptomatic-to-symptomatic case ratios from other countries, both Shibuya and Nishiura consider ten times the reported Japanese cases to be a low-end estimate of actual infections, while noting that in other countries it has been as high as a factor of 20 to 50. Assuming that the Keio findings of 6% prevalence represent an overestimate, but that an infection rate of 3-4% of the Tokyo population is plausible based on other evidence, Shibuya noted that, “A simple calculation will give you a figure how many people are infected right now.” Taking 13 million as the population of the Tokyo Prefecture, 3% is 390,000 people. The operative term here is “plausible.” Results of modeling done by Dr. Nishiura reported two weeks ago estimated that without strong countermeasures the number of serious COVID-19 cases in Japan could reach 850,000, and the death toll as high as 400,000.  These two sets of estimates are in general agreement on the possible magnitude of the outbreak here. The effectiveness of the voluntary social distancing and other countermeasures that have been implemented remain a relatively uncertain variable.

Large numbers of deaths, however, still do not appear to be evident. We previously noted concern that due to the lack of testing, many observers have suspected that deaths recorded simply as pneumonia may in fact be due to COVID-19, in effect “hiding” them from being properly counted. Influenza mortality, however, is difficult to quantify even under normal conditions, Shibuya points out, with many variations, co-morbidities, and other contributing factors. For various reasons, none of them duplicitous, deaths from contagious diseases are often underestimated. Because in the case of influenza it is impossible to test everyone, we must often rely on an excess mortality estimation process instead. A recent paper from Tokyo University researchers shows that seasonal influenza activity in Japan was markedly lower in 2020 than in previous years, and suggests that measures taken to constrain COVID-19 may be one reason.

2019-2020 Japan influenza mortality. The blue points indicate increased mortality in March-April 2020 (source : NIID)

Recent mortality rate data from NIID indeed shows a statistical increase in influenza mortality in Tokyo in March and early April, 2020. Shibuya and others conclude that this may include some undiagnosed coronavirus deaths. Again, this does not mean deaths are being intentionally hidden, but that the reporting system sometimes makes it difficult to determine with full confidence actual causes of death.

The success South Korea has had so far in minimizing the spread of COVID-19 within its borders, according to Shibuya, is because they focused on basics: to test and isolate. Because of their unfortunate experiences with MERS and SARS in earlier years, they were well prepared for this outbreak. We can say that the same is true for Taiwan. Japan’s focus on border control and identifying clusters of major symptoms — “cluster countermeasures” which will miss asymptomatic or pre-symptomatic mild cases being transmitted within the community, and has already led to dangerous hospital transmission — has left the country vulnerable. The country needs to be conducting 100,000 tests per day, and Shibuya believes this goal is achievable. Nishiura is more sanguine. His personal stated opinion is that even if testing is increased to 20,000 per day, response will still be limited by the capacity of health care facilities. Because of personnel issues, logistics, and other issues, he noted, “The reality is we do not have the capacity to do what Korea has been doing with intensive testing and tracing.”

A key point to keep in mind, Shibuya explained, is that the perspectives of physicians and public health specialists are different, and sometimes opposite. For physicians, the primary objective is to reduce disability and save patient lives. Therefore it is natural to want to focus on cases with major symptoms, and not on asymptomatic ones, in order to prioritize resources for severe cases. If only limited testing is available, consequently, this thinking suggests that it is best to focus on those who have symptoms. From a clinical standpoint this seems reasonable. “But from a public health perspective,” Shibuya says, “unless you tackle quite a few asymptomatic and pre-symptomatic cases, we cannot contain the epidemic.” The apparently unshakeable focus on the “cluster countermeasures” policy in Japan, despite the clear looming risks, is an indication of how strongly the physician viewpoint dominates the typical Japanese approach to epidemics..

What should we expect in coming weeks? Nobody knows. Perhaps social distancing and the cooperative discipline of the population is truly helping, and the apparent bending of the curve we’ve seen over the past week will continue. It’s clear, however, that actual case numbers in Japan are vastly underreported. The ramifications of the underreporting point to a continuing risk of a much larger outbreak. At this point we believe it is clear that any relaxation of social distancing measures in Japan should be delayed as long as possible. We also think greatly expanded testing is called for, and support Dr. Shibuya’s target of 100,000 per day, while recognizing that this may arrive too late if it is achievable at all. More importantly, based on what we’ve seen so far we wonder if increased testing in Japan would actually lead to meaningful changes in the government’s overall response or the ability of the health care system to deal with COVID-19 patients. What would they do differently? We definitely think that effective mechanisms to sustain social distancing need to be found, however. Without adequate social and financial support and better targeted guidance, it is far too difficult for many people to stay at home as much as is required. We hope that good testing at some point in the near future will indicate a clear minimization of the spread of the virus over the course of several consecutive weeks, and allow the country to move into the next phase of recovery. But none of us will be able to relax until an effective vaccination program has been developed and completed.



To repeat some relevant background information we’ve touched on earlier, Japan does not have an epidemic emergency agency like the CDC in the US or similar agencies in South Korea, Taiwan, and many other countries. Instead, Japan’s National Institute for Infectious Diseases (NIID), which is essentially a research organization, is tasked with collating and analyzing epidemic and pandemic data. It uses a system called the National Epidemiological Surveillance of Infectious Diseases (NESID), which was established in 1981 as part of the Infectious Disease Surveillance Center (IDSC). Due to the complexity of this system, and the difficulty of modifying how it operates, it moves cautiously and slowly, and is not very adaptable under the stress of new outbreaks like we are seeing with COVID-19. In practical terms, this means that data sources which have been part of the system for a long time, particularly public health institutions, are prioritized over others, regardless of their actual relevance to emergency response. Reporting criteria are fairly rigid. A lot of the issues we continue to observe regarding COVID-19 response in Japan appear to be the result of the system is working as it was designed to. It is neither agile nor flexible.