Age-Dependent Variations in Kawasaki Disease Incidence in Japan

This cross-sectional study investigates rates of Kawasaki disease among pediatric patients in Japan by age, including analyses of seasonal cycles, interannual variations, and correlations across regions.


eAppendix 1. Time Series of KD Incidence Rate
In the main body of the paper Figure 1 shows the �me series of incidence rate normalized by the rate in 1987-1992.Here, using the same data, we show the �me series that has not been normalized to the 1987-1992 period (eFigure 1).eFigure 1: Time Series of KD Incidence Rate by Age Group eFigure 1: Time series of KD incidence rate per 100,000 for each age group, based on daily smoothed data.The incidence rate is computed rela�ve to the popula�on of 0-to 14-year-olds for each year to account for the falling birth rate in Japan.

eAppendix 2. Significance of differences between seasonal cycles of KD incidence over �me
To test for significant differences between seasonal cycles (Figure 2 e-h, main body), a random sampling approach was used to determine baseline distribu�ons.A smoothed, normalized seasonal cycle, averaged from 1988 to 2019, was computed with a 31-day running mean.
Residual values for each day of each year were found by subtrac�ng the long-term seasonal cycle from the normalized values for each year.To construct a baseline sample ensemble, against which to compare the true seasonal cycles, sample daily series averaged over a 4-year period were constructed by averaging four residual values for each day of the year randomly selected from the 32 year observed residuals.These residuals were added to the long term smoothed seasonal cycle.This process was repeated 1000 �mes, yielding an envelope of 1000 sample 4-year aggregate seasonal cycles.For each month, the value from the observed seasonal cycle was ranked and compared to the distribu�on of the random seasonal cycles.When the observed seasonal cycle fell in the highest or lowest 2.5% of the ranking, it was considered significant at the 95% level. in the spring, while in the later years there was lower incidence in the winter and much higher incidence in the fall.

eAppendix 3. Correla�ons of seasonal cycles between prefectures
To assess whether the seasonal cycle varied spa�ally across Japan, the KD historical record was disaggregated into prefectures.The seasonal cycle of each of the 47 prefectures was correlated with that of every other prefecture (eFigure 3). Figure 3 of the main body shows the averages of these correla�ons as well as the loca�on of each of the 47 prefectures.In the prefectures near Tokyo (prefectures 12-14) both toddlers and children 3 and older have the highest correla�ons between prefectures.In the southern prefectures (41-46) the toddlers do not show any significant correla�on with other prefectures, while the children 3 and older have several significant correla�ons with prefectures across the country.Also of note is Hokkaido (prefecture #1), where the toddlers correlate more broadly with the rest of the country then the children 3 and older.between prefectures for toddlers and children 3 and older, with values less than 0.59 (95% significance) colored dark blue.Prefectures are arranged in columns and rows from north (Hokkaido, 1) to south (Kagoshima,46 and Okinawa,47).See Figure 3 in the main body for maps with numbered prefectures.Note these heatmaps are symmetric.

eAppendix 4. Changing correla�on of seasonal cycles between prefectures
The coherence of the seasonal cycle across prefectures changed over �me for most age groups.
To illustrate this, the KD incidence within the 47 prefectures were condensed into five regional subsets (eFigure 4, right).For each region and for each age group, the �me series of the number of KD pa�ents per day was correlated between like-�meseries from the other four regions for each year.For each age group, correla�ons for each prefecture were averaged across the five regions for each year to obtain a single series of correla�on traversing the en�re 1988-2019 period.These �me varying all-Japan aggregate correla�ons are shown in eFigure 4.
Data for the figure was smoothed with a 3-year running mean.
Similar to results shown in the main body, eFigure 4 shows that infants had no sta�s�cally significant correla�on between regions, and that a low level of associa�on persisted across the

eAppendix 1 .
Time Series of Kawasaki Disease Incidence Rate eFigure 1.Time Series of Kawasaki Disease Incidence Rate by Age Group eAppendix 2. Significance of Differences Between Seasonal Cycles of Kawasaki Disease Incidence Over Time eFigure 2. Statistical Tests of Differences Between Seasonal Cycles eAppendix 3. Correlations of Seasonal Cycles Between Prefectures eFigure 3. Seasonal Cycle Correlations Between Prefectures (1988-2019) eAppendix 4. Changing Correlation of Seasonal Cycles Between PrefectureseFigure 4. Correlation of Daily Incidence Between Regions by Year eAppendix 5. Changing Attendance in Childcare Institutions eFigure 5. Number and Percentage of Children Attending Japanese Childcare Institutions This supplemental material has been provided by the authors to give readers additional information about their work.© 2024 DeHaan LL et al.JAMA Network Open.

eFigure 2
provides a ledger of significant changes in seasonal cycle throughout the period of record.The observed seasonal cycle (colored lines, eFigure 2 a-d) frequently falls outside the random sample (grey lines, eFigure 2 a-d).This is most notable for the toddlers in 2016-2019, but occurs in many other instances as well, demonstra�ng the varia�on in seasonal cycle over the years.Consistent with other findings presented here, the infants have very few significant changes in seasonal cycle (eFigure 2a & 2e).While the 2-year-olds (eFigure 2c & 2g) and children 3 years and older (eFigure 2d & 2h) have more significant changes than the infants, the largest number of significant changes happen with the toddlers (eFigure 2b & 2f).In the early years the toddlers had significantly lower incidence in the fall and significantly higher incidence © 2024 DeHaan LL et al.JAMA Network Open.
Panels a-d: Monthly normalized seasonal cycles of 4-year bins.Light grey lines are random sampling; colored lines are observed.Panels e-h: Significance of differences between seasonal cycles of 4-year bins (shown in panels a-d) and the 32-year average for each month of the year.Blue colors indicate significantly lower incidence; red colors indicate significantly higher incidence.Dark shades indicate that the �me period exhibits a sta�s�cally different seasonal cycle for the given month at the 99% confidence level; lighter shades indicate a 95% confidence level.

eFigure 3 :
Seasonal Cycle Correla�ons between Prefectures (1988-2019) eFigure 3. Correla�ons of KD seasonal cycles en�re record from 1988 to 2019.In sharp contrast, two-year-olds and children 3 years and older experienced increasing correla�ons in the past two decades, rising to an average of 0.6 or greater by 2019.Toddlers exhibited a more complicated patern, wherein correla�on between regions also increased in the early 2000s, but then dropped star�ng around 2010 and began to rebound a�er 2015.The reduc�on in correla�on in 2010 coincides with the large shi� in the seasonal cycle of countrywide KD incidence shown in Figure 2f of the main body.Increased correla�on of regional toddler incidence a�er 2015 is somehow associated with a more widespread shi� in the seasonal cycle to have a reduced peak in KD incidence in winter and © 2024 DeHaan LL et al.JAMA Network Open.increased incidence in autumn KD incidence.To test the robustness of the features shown in eFigure 4, this computa�on was repeated (not shown) using differing numbers of regions and different smoothing applied to the data.While the exact values differ with the number of regions and the level of smoothing, the age-related correla�on structure obtained in eFigure 4 here are common to all varia�ons of the calcula�on.eFigure 4: Correla�on of Daily Incidence between Regions by Year eFigure 4. Correla�on of daily incidence between each of five regions of Japan (shown right) and every other region.Correla�on is computed for each year individually and averaged over the five regions.Thick lines indicate at least one pair of regions had a correla�on significant at 95% (above 0.59).eAppendix 5. Changing atendance in childcare ins�tu�ons The number of children in daycare ins�tu�ons in Japan has steadily increased for the last two decades, with a significant increase in ins�tu�ons in 2015, as shown in eFigure 5. eFigure 5: Number and Percent of Children Atending Japanese Childcare Ins�tu�ons eFigure 5: Number of childcare ins�tu�ons of various types since 1999 (top).Percent of children in childcare ins�tu�ons (botom).Local discre�onary type: 0-2-year-old, Kindergarten type: 3-5-year-old, Unified type: 0-5-year-old, Day nursery type: 0-5-year-old.Data from the