Estimation of true number of COVID-19 infected people in Japan using LINE questionnaire

Shin-ichiro Tanaka; Shinya Oku

Firstly, LINE questionnaire will be described here. As discussed in the report by MHLW, high fever caught in the questionnaire is considered as including the fever with COVID-19 as its cause. In contrast with the ongoing PCR test, this questionnaire is quasi-completely randomized, 9 exhibiting its large scale of 23,992,701 replies. This means that these data can be served as the fundamental data to deduce the true number of infected. As it should be noted that the data contains the fever with other reasons than COVID-19, the data itself could not be regarded as the number of the infected. The authors underpinned the rate of high fever per prefecture, which were served as the data source for estimating true number of the infected. Figure 1 exhibits the number of high fever/responses per prefecture by LINE questionnaire. Bar charts were provided after sorting data in descending order. Except for the three prefectures which showed the extremely high numbers, the difference among the rest remains within a certain range. As reported in the news, the spread of COVID-19 infection is not homogenous nationwide; i. e., in Tokyo, only seldom infections were found when many were found in Hokkaido in its initial days. Therefore, it was expected that the distribution of high-fever patient caused by COVID-19 shows a similar distribution to the confirmed cases according to the regions visibly. In contrast, the distribution of high-fever patient caused by non-COVID-19 reasons was considered as stable regardless of the regions; that the identical distribution, with some extent of statistical error, could be expected. If the variations according to the regions could be taken into account, the part of the high-fever patients caused by COVID-19 could be extracted. In order to evaluate the appropriateness of this model, the comparison between the national distribution of PCR positive patients at the timing of LINE questionnaire and that of high-fever patients assessed in LINE questionnaire, would be effective. It could not be expected to determine the true absolute patient number with PCR test data, however, the relative distribution reflects the true patient distribution properly, assuming that the philosophy to design PCR test per prefecture remains identical nationwide. With this consideration, the authors could believe that the simple comparison between the distributions of PCR positivity and LINE high-fever suffice. Needless to state, when pursuing the trend of positive patients, not only the regional difference but also the time-dependent change of a patient history from the infection via the latent period to the appearance of initial symptom, should be considered. Simply applying the current statistical data as the data for infected might cause a loss of accuracy; rather the data at the exact timing of LINE questionnaire should be applied. Furthermore, as the fixation of number of the infected happens certain days later from the infection, waiting for the continuous symptom(s) to decide to perform PCR test, not the confirmed number of a date, but the newly symptomatic patient number of the same date should be used. Meanwhile, as the information of confirmed number of patients at a certain date includes only the limited number of initial symptom dates, the exclusive use of these data makes the analysis statistically non-robust.