Spatiotemporal association of low birth weight with Cs-137 deposition after Fukushima at the prefecture level in Japan: an ecological study

Background: Perinatal mortality increased in contaminated prefectures after the Fukushima Daichi Nuclear Power Plant (FDNPP) accidents in Japan in 2011. Elevated numbers of surgeries for cryptorchidism and congenital heart malformations were observed throughout Japan from 2012 onward. The thyroid cancer detection rate (2011 to 2016) was associated with the dose-rate at the municipality level in the Fukushima prefecture. Since the birth weight is a simple and objective indicator for gestational development and pregnancy outcome, the question arises whether the annual birth weight distribution was distorted in a dose-rate-dependent manner across Japan after Fukushima. Data and Methods: The Japanese Ministry of Health, Labour, and Welfare provides prefecture-specific annual counts for 26.158 million live births from 1995 to 2018, of which 2.366 million births (9.04%) with weights < 2500g. Prefecture-specific spatiotemporal trends of the low birth weight proportions were analyzed. Logistic regression allowing for level-shifts from 2012 onward was employed to test whether those level-shifts were proportional to the prefecture-specific dose-rates derived from Cs-137 deposition in the 47 Japanese prefectures. Results: The overall trend of the low birth weight proportion (LBWp) in Japan discloses a jump in 2012 with a jump odds ratio (OR) 1.020, 95%-confidence interval (1.003,1.037), p-value 0.0225. A logistic regression of LBWp on the additional dose-rate after Fukushima adjusted for prefecture-specific spatiotemporal base-line trends yields an OR per µSv/h of 1.098 (1.058, 1.139), p-value < 0.0001. Further adjusting the logistic regression for the annual population size and physician density of the prefectures, as well as for the counts of the dead, the missing, and the evacuees due to earthquake and tsunami (as surrogate measures for medical infrastructure and stress) yields an OR per µSv/h of 1.109 (1.032, 1.191), p-value 0.0046. Conclusion: This study shows increased low birth weight prevalence related to Cs-137 deposition and the corresponding additional dose-rate in Japan from 2012 onward. Previous evidence suggesting compromised gestational development and pregnancy outcome under elevated environmental ionizing radiation exposure is corroborated.


Introduction
Low birth weight (LBW) is defined as having a birth weight of < 2500 g. It is an objective and reliable indicator used as comprehensive demographic reporting measure of fetal development and pregnancy outcome [1][2][3][4]. Environmental pollutants are consistently linked to general untoward pregnancy outcome and specific reductions in birth weight [5][6][7][8][9][10][11][12]. LBW has been suggested as an indicator of genetic detriment caused by mutation in humans exposed to ionizing radiation [13]. Analyses of birth weight and duration of pregnancies in relation to maternal age, parity, and infant survival indicated that nonsurvivors were significantly lighter at birth than survivors [14]. LBW is closely linked to fetal and perinatal mortality and morbidity [15]. It has been reported to be associated with disorders in perinatal periods, in childhood, and in adulthood [16,17]. Studies in Great Britain showed that people who had low birth weight were at increased risk of coronary heart disease and the disorders related to it [18]. Animal and human studies have shown that the LBW proportions (LBWp) increase with toxic exposure and with radiation exposure [19][20][21]. Smoking increases the LBWp in a dose-dependent manner [22], possibly due to elevated radionuclides in tabaco [23]. Females subject to pelvic radiotherapy experience an increased risk of pre-term delivery and LBW among their offspring [24]. Treatment of female childhood cancer patients may entail restricted fetal growth and pre-term births [25]. LBW was reported after dental radiography during pregnancy [26]. A cohort study in China identified multiple risk factors of LBW including radiation exposure of fathers [27]. A natural experiment in Taiwan revealed that prenatal exposure to a continuous low-level irradiation reduced the gestational length and increased the LBW proportion [28]. In Belarus, increased LBW prevalence was reported from the highly Chernobyl-contaminated regions Gomel and Mogilev [29]. Temporarily elevated LBWp was seen in Sweden after Chernobyl [30].
Since several radiation inducible genetic and carcinogenic effects were observed in Japan after Fukushima [31][32][33][34][35][36][37], an increase in the LBW proportion was also conceivable. Among the investigations after the Fukushima nuclear accidents, there are reports that LBW is increasing and reports that deny the increase. In the following we shortly address two reports that are questionnaire-based surveys with a response rate in the 50% range and one survey of a small number of births in one clinic in Fukushima [38][39][40]. Questionnairebased studies are prone to selection bias and studies with small populations (mostly in clinical settings) may likely entail type-2 errors [41]. A questionnaire-based pregnancy and birth survey was conducted by the Radiation Medical Science Center for the Fukushima Health Management Survey [38]. In this study, an increase of the LBW proportion is documented in the combined contaminated Iwaki and Soso regions compared to the remainder of the Fukushima prefecture: OR 1.163, p-value 0.0723. This observation is supported by a corresponding increase of the stillbirth proportion in Soso and Iwaki with OR 1.923, p-value 0.1321. Since this study [38] had a participation rate of below 60%, it is likely that significant effects would be obtained with lager populations considered during longer periods. Maternal and perinatal data (2008 to 2015) were retrospectively collected for singleton live births at a hospital located 23 km from the Fukushima nuclear power plant [39]. In 1101 births, LBWp was compared pre-and post-disaster. There was no increased LBWp in any year from 2011 onward. However, with 4 years before/after the accident, i.e., 140 births per year, which means about 10 LBW-births per year, it was unlikely to receive a meaningful result, i.e., there is a large type-2 error probability in this study [39]. A more recent investigation considered 12,804 maternal outcomes during 2011-2014 in the Fukushima Prefecture [40]. However, this study neither analyzed perinatal outcomes with distance from the nuclear accident nor chronological factors. Therefore, it is unclear whether increases of LBW are due to a temporary cause of the earthquake/tsunami or due to radiation exposure. These surveys cover the Fukushima Prefecture only incompletely over short periods. In the Miyagi Prefecture, the overall rate of LBW infants was reported to be 8.7%, which tended to be lower than LBWp in 2012 of 9.3% and 9.8% in 2013 [42]. The spatiotemporal trends of the LBW rate in most of the surrounding prefectures have not yet been scrutinized, although data is publicly available.
In this study, we analyzed data of the Japanese governments 'Demographical Survey', which accounts for almost all live births and LBW children registered in Japan excluding births to parents living abroad. Therefore, not only Fukushima Prefecture but also the whole country with differently contaminated prefectures was targeted, and statistical accuracy is guaranteed by using official nearly complete long-term data from 1995 to 2018, i.e., 16 years (1995 to 2010) before and 7 years (2012 to 2018) after the nuclear power plant accidents in Fukushima in March 2011.

Vital statistics and auxiliary information
The Japanese Statistics Bureau publishes demographical information compiled by the Ministry of Health, Labor, and Welfare. Statistics include the annual numbers of live births and the annual counts of children with a low birth weight of < 2500 g (LBW), see Table 1 or Vital Statistics of Japan: https://www.mhlw.go.jp/english/database/db-hw/vs01.html. We investigated the spatiotemporal distribution across 47 Japanese prefectures from 1995 to 2018 of 26.158 million live births, of which 2.366 million births (9.04%) with weights < 6 2500 g. Since medical supply may impact the prevalence of LBW, physician density by prefecture in Japan (https://stats-japan.com/t/kiji/10343) was deployed as an appropriate surrogate confounder variable in the spatiotemporal logistic regression models. The counts of earthquake related deaths, the counts of the dead and missing after earthquake and tsunami, as well as the number of evacuees to and within any prefecture were obtained from official sources [43] and served as additional explicit potential ecological confounding variables, see Table 2. Table 1 Annual live births, live births with low birth weight (LBW: birth weight < 2500 g), and LBW proportions (LBWp) in Japan stratified by exposure status of prefectures; see Table 3 and Fig. 2 Table 3), whereas the UNSCEAR data set 2013/2014 documents a mean value of 153.957 kBq/m 2 Cs-137, which amounts to a factor 6 underestimation of the deposition in Fukushima by Yasunari et al. [44], see the data file https://www.unscear.org/docs/ publications/2013/UNSCEAR_2013_Annex-A_Attach_C-2.xls [45].
Fukushima is implausibly lesser contaminated than Miyagi, and Ibaraki, see Table 3.
The Yasunari et al. data decay with r -3.27 at distance r form the FDNPP, see Figure 1, whereas a theoretical decay law of r -1.42 is expected according to UNSCEAR [46], and has empirically confirmed for the Fukushima prefecture [31].
Since strong underestimation of radiation exposure would exaggerate any dose-specific radiation risk estimates, we suggest and propagate a correction and a rescaling of the  Fig. 1 and listed in Table 3.   [ 44], rescaled Cs-137 deposition according to [44][45][46], and dose-rate [µSv/h] derived from the rescaled deposition.

Statistical methods
A powerful method to assess data that contains spatial as well as temporal information is spatiotemporal (logistic) regression [47][48][49][50][51][52]. The basic idea is to adjust a regression model for region-specific trend functions and to allow for local or global level-shifts at certain points in time, or, preferably, to allow for the heights of local drops or jumps to be proportional to the contamination level of the regional strata. The advantage of this spatiotemporal method is optimization of adjustment and minimization of confounding by considering partial trends of regional units. Values of the outcome variable (here LBWp) in those partial trends are modeled and compared within the same regional stratum, as the target variable describing the interesting characteristic (LBW) varies from year to year.
Information on several regional units is  Table 1 for the corresponding absolute counts and the LBW proportions (LBWp). As a first step, we fit to this overall LBWp a  Table 2), as well as additionally taking into account the prefecture-specific population , and this would likely be an overestimation of the radiation effect due to an obvious underestimation of the overall Cs-137 deposition across Japan by Yasunari et al. [44].
To more directly assess and display the relative impacts of the earthquake and the tsunami versus the effects of the Cs-137 deposition and the associated dose-rate on LBW, we compared the three contaminated prefectures Fukushima, Iwate, and Miyagi, where the dead and missing persons due to earthquake and tsunami were numerous (n = 18,359), to the somewhat weaker contaminated Ibaraki, Tochigi, and Yamagata where only relatively few immediate deaths occurred and few persons were missing (n = 31), see Table 2. Figures 6A and 6B show that for the less versus strongly earthquake and tsunami impacted groups of prefectures, the LBWp jump heights are similar with largely overlapping 95%-CIs and with similar p-values. Therefore, the long-term increasing LBWp is essentially independent of the direct or protracted impact of earthquake and tsunami.

Discussion
This study strengthens the evidence provided by previous investigations [19,[25][26][27][28][29][30] that elevated exposure to medical or environmental ionizing radiation increases the prevalence of low birth weight (LBW) children. The proportion of low birth weight babies in Japan (LBW < 2500 g) was increasing continuously from 1995 to a peak value in 2006, see  Table 2 we compiled these potential ecological confounders. However, the adjustment for the additional confounders (population counts, physician density, and the triple-disaster-related stress indicators from 2012 onward) does not change our effect estimate of approximately 10% per 1 µSv/h. The reason for this may be that all major prefecture-specific information is already captured by the prefecture-specific spatiotemporal base-line trends accounted for in the adopted spatiotemporal regression approach [52].
Our investigation disclosed a positive association between the Cs-137 deposition across Japan after Fukushima with the prevalence of low birth weight (< 2500 g). Therefore, previously reported epidemiological health detriment after Fukushima [31][32][33][34][35][36][37]53] can be generalized and corroborated. Nevertheless, the question whether ionizing radiation exposure of future parents or peri-conceptional and embryonic radiation exposure impair fetal and post-natal development remains a controversial issue. There are articles in favor of and against this hypothesis, e.g., [54,55]. One problem with statistically negative studies in the clinical setting is sample size -typically in the range of only a few thousand [55] or below. Small sample sizes generally entail low statistical power implying large type-2 error probabilities. It may be rather improbable to detect relevant changes in low birth weight proportions, say in the order of 10%, with population sizes ranging only in the thousands. For example, a two-sided one-sample binomial test for testing a hypothetically increased LBWp of 0.11 against a typical null-LBWp of 0.10 (i.e. 10% increase) requires a sample size of 7,248 to achieve a statistical power of at least 80%. For more realistic twosample scenarios, additional independent LBW-determinants, and higher biological variability, the required sample sizes for obtaining meaningful results would be even larger. Therefore, it is of no surprise that no unequivocal evidence has been obtained yet.
Under the headline "Radiation-induced mutation rates in man", UNSCEAR [13] emphasized already in the year 1958 "All the results obtained are subject to an inevitable sampling error which necessitates the collection of a very large amount of data. A number of quantitative characters, such as birth weight, size and various anthropometric measurements, as well as statistical data, such as neo-natal mortality, have been suggested and examined. Unfortunately, the precise genetic component in these variables is not known; on the contrary, they are known to be dependent upon factors which are economic (standard of living), demographic (age of parents, order of birth, etc.) and sociological (medical care)." The sample size issue may be resolved when instead of at most thousands of births in clinical settings many millions of births in ecological studies can be considered: After the nuclear accidents of Chernobyl and Fukushima, the populations of whole countries have been exposed to significant additional ionizing radiation [51,[56][57][58]. Moreover, in a large-scale ecological design, as the one presented here, the socio-demographic and environmental determinants of the low birth weight prevalence can be considered similar in and comparable between the regional units (prefectures). The differences within and between the regional trends from 2012 onward can be assessed by spatiotemporal logistic regression adjusted for appropriately chosen base-line trend parameters and further LBW determinants and ecological confounders [52].

Conclusion
This study shows increased low birth weight prevalence across Japan related to the prefecture-specific dose-rate derived from Cs-137 deposition after Fukushima. One (1) µSv/h (equivalent to 8.8 mSv/a) increases the odds of observing low birth weight events by 10% in the order of magnitude. Therefore, previous investigations suggesting compromised gestational development and impaired pregnancy outcome under elevated ionizing radiation levels have been corroborated by the present study. These findings, in the overall view, call for intensifying bio-physical research in exposure mechanisms and exposure pathways of natural or artificial ionizing radiation. Biological, epidemiological, and medical research should aim at clarifying the genetic and carcinogenic consequences of enhanced radiation in the environment. Radiation-induced genetic effects may occur without immediately obvious link to spectacular incidents or accidents [58,59]. Therefore, the legislator, the nuclear industry, and the nuclear and radio-pharmaceutical medicine must impose and exert even greater care when processing, employing, and disposing radioactive materials.

Consent for publication
Not applicable. Only anonymous data is being used.

Availability of supporting data
The employed data has exclusively been published previously and/or it is contained in the Tables and in the Figures included in this paper. Fukushima corrected and remaining depositions rescaled to a decay of r-2 with distance r, see Table 3.

Figure 2
Geographic region value plot of the decadic logarithm for the rescaled Cs-137 deposition in 47 Japanese prefectures after the Fukushima nuclear power plant accidents as of March 2011 [44], see Table 3; indication of the positions of the earthquake epi-center, the FDNPP, and a 300 km geo-circle around FDNPP; for the prefecture codes see Table 2 or Table 3.  [45] and Figure 5A in [31].

Figure 4
Low birth weight (LBW) proportion in Japan 1995 to 2018; 4th degree polynomial logistic regression trends allowing for jumps from 2012 onward; A: Japan; B: Japan excluding 10 exposed prefectures; C: 5 moderately exposed prefectures; D: 5 highly exposed prefectures.

Figure 5
Odds ratios for the jumps in the low birth weight proportion trends (LBWp) from 2012 onward by prefecture-specific dose-rates derived from the rescaled Cs-137 deposition in the Japanese prefectures from March 20th to April 19th 2011; restricted linear regression yields trend p-value < 0.0001; the left data point summarizes and represents 37 slightly radiologically impacted prefectures, see Table 1.