Vol.21, No. 1 Printed in Great Britain
International Journal of Epidemiology © International Eptdemiological Association 1992
Seasonality of Preterm Births in Japan SHINYA MATSUDA'f AND HIROAKI KAHYO*
Seasonal variation in birthweight is well known.'"* It is, however, difficult to identify a particular cause of this seasonality, because many factors influence birthweight including period of gestation, maternal constitution, nutrition, general morbidity, smoking, and so on.5'6 Seasonality in these factors, therefore, might explain the seasonal variation in birthweight. Among factors associated with birthweight, period of gestation is known as the most important and definitive factor.3'6 Relatively few studies, however, have been concerned with the seasonality of the gestation period.7 In this investigation, we focused on the seasonal changes in proportion of preterm births in Japan over the period 1979-1983 and found apparent seasonality in proportion of preterm births in Japan.
and Information Department, Ministry's Secretariat, Ministry of Health and Welfare, Japan. The data on length of gestation is described in the vital statistics as a frequency table with one weekintervals (36 weeks and less, 37, 38, 39, 40, 41, 42 weeks and more). From this table the proportion of preterm births (less than 37 weeks gestation) were calculated. Vital registration of live-born infants in Japan is almost 100% complete. The total number of live singleton infants registered in the vital statistics during the five-year period was 7665006. Climate in Japan
The Japanese climate differs greatly from region to region. The country's north-south length of 3000km puts one end in the sub-arctic zone and the other in the sub-tropical zone. The influence of ocean currents are also important. Most of Japan, however, enjoys a moderate, oceanic type of climate with four distinct seasons, i.e. spring (March to May), summer (June to August), autumn (September to November), and winter (December to February). Another important characteristic of the climate is the rainy season which lasts from June to July, and typhoons which often hit the southwestern part of Japan between August and October.
MATERIALS AND METHODS Data Source
The data used for this investigation were derived from computer tapes of the vital statistics of Japan for January 1979 to December 1983 provided by Statistics •Department of Human Ecology, School of Medicine, University of Occupational and Environmental Health, Yahatanishi-ku Kitakyushu, 807, Japan. tDepartment of Hygiene, Faculty of Medicine, Kyoto University, Japan
91
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
Matsuda S (Department of Human Ecology, School of Medicine, University of Occupational and Environmental Health, Yahatanishi-ku, Kitakyushu, 807, Japan) and Kahyo H. Seasonality of preterm births in Japan. International Journal of Epidemiology 1992; 21: 91-100. Seasonal variations in the proportion of preterm births in Japan from January 1979 to December 1983 are analysed using a traditional method of time-series analysis, which divides the variation in a series into trend, seasonal variation, other cyclic change, and remaining irregular fluctuations. It is shown that the proportion of preterm births in Japan have a clear seasonal periodicity with two peaks in summer and winter. Analysis of seasonality by period of gestation shows that interesting differences in kurtosis and skewness exist between summer and winter, i.e. the summer increase in preterm births was characterized by an increase of skewness which means an extension of the lower part of the distribution. On the other hand, the winter increase was characterized by a decrease of kurtosis which corresponds to a flat-topped distribution. This result suggests that causes of preterm births might be different between the two seasons. Theoretical simulations based on actual birth data in Japan over the period, are carried out to examine how season of conception could influence seasonal variations in the proportion of preterm births. Results show that, at least for first births, seasonality in conception rates could be one explanatory factor for the observed seasonal variation in proportions of preterm births. Another analysis reveals that conception in May and June are more likely to result in preterm births in Japan.
92
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Statistical Approach
,
+ k
-m
2
)
N where rk
= autocorrelation coefficient at lag months k, x, = index value at t th month, m, = mean of index values from x, to x ^ N-k
m, = Z x/(N - k) and, m2 = mean of index values from xk + , to xN; N-k
m, = Z x,, t /(N - k) 1 - 1
'•"•*
RESULTS Figure 1 shows plots of the time series of the proportion of preterm, term, and post-term births, for male first births and male subsequent births, respectively, with a significant regression line. While preterm and term births show upward trends, post-term births show downward trends. These findings were the same for female births.
Figure 2 shows seasonal indices of proportion of preterm births. All of them have a similar appearance with two peaks in winter and summer (or rainy season), and with two troughs in spring and autumn. There is a little difference in the summer peak among them, i.e. male first births have a peak in June, female first births in July, and subsequent births of both sexes in August. In the comparison between first births and subsequent births, the latter have a lower peak in summer than the former. Figure 3 shows correlograms of the proportion of preterm births, which reveal strikingly regular patterns both for first and subsequent births. The former show high autocorrelation coefficients with a peak correlation of about 0.80 for male and about 0.50 for female at six-month intervals. And the latter also show high coefficients with about 0.80 for male and about 0.50 for female at every 12-month lag time. DISCUSSION Figure 1 shows an upward trend in the proportion of preterm and term births in Japan over the period. Ishizuka et al10>" compared neonatal mortality rates of low birthweight infants between 1980 and 1985 in Japan using the data from main hospitals (1980: about 25 000 cases in 528 hospitals; 1985: about 30000 cases in 648 hospitals), and reported that neonatal mortality rates had decreased from 8.8% in 1980 to 4.9% in 1985. They also compared the mortality rates between hospitals equipped with NICU (Neonatal Intensive Care Units) and hospitals without NICU and observed that neonatal mortality rates were much higher in hospitals without NICU. Their result suggests that one of the contributing factors to the upward trend in the proportion of preterm births may be recent developments in neonatal care for premature infants. Another interesting result shown in Figure 1 is that the proportion of post-term births show a downward trend, while term births show an upward trend. Factors associated with this phenomenon are unknown. A greater use of induced labour could be one of the factors for this phenomenon. Further studies using birth records from obstetric facilities which have more detailed information, especially on conditions at onset of labour, are necessary to clarify factors associated with this phenomenon. There is no doubt that a seasonal rhythm in the proportion of preterm births exists as shown by the high autocorrelation coefficient of monthly preterm birth data in Japan. The seasonality was observed for both sexes and both parities. The appearance of their seasonal indices were similar but there was a little difference in timing of the summer peak. The male peak was
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
Traditional methods of time-series analysis which divide the variation in a series into trend, seasonal variation, other cyclic change, and remaining irregular fluctuations, were adopted for the investigation.8 This approach is considered appropriate for a particular variable when the variation is dominated by trend and/or seasonality.8 The procedure is as follows. 1) Firstly, time series for the monthly data of male first births (primiparae), male subsequent births (multiparae), female first births, and female subsequent births, were plotted and then examined for linear trends by least squares regression. 2) An observed value (y) for each month (t) was divided by the corresponding yearly average (y' t ), (the latter was calculated by 12-month moving average) in order to calculate an index value for each month (yt/y',*100). The average of index values was calculated in each calendar month for five years, and then accommodated in the way where the summed value of 12 months became 1200% using a weighted average in order to calculate seasonal index for each calendar month. 3) Autocorrelation coefficients were calculated up to 36 lag months for the data sets using index values and then correlograms were plotted to examine the seasonal rhythm. The formula is as follows;
93
SEASONALITY OF PRETERM BIRTHS IN JAPAN
Multiparae
Primiparae Preterm
5.0-
Preterm A ft
5.0-
9
4.54.03.5-
4.5-
1
vv PWft
3.0G
v
V
VJ r
y = 3.876 + 0.005X 1 2
24
36
48 60 MONTH
3J-
w
3.00
V
WLrVfrM
y = 4.065 + 0.006x 12
94.0 n
24
36
48 60 MONTH
Term y = 91.649 + 0.018x
91.0-
93.0-
90.0-
92.0 • y = 90.511 + 0.018x 91.0
89.0 12
24
36
Postterm
7.0-
12
48 60 MONTH
6.0-
6.0 i * 5.0-
5.0-
4.0-
4.0-
y = 5.585 - 0.023X
3.0
24
36
48 60 MONTH
Postterm y = 4.285 - 0.024x
3.02.0
12
24
36
48 60 MONTH
12
24
36
48
60
MONTH
FIGURE 1. Seasonal fluctuation and secular trends in the proportion of preterm, term and post-term births (male, all Japan, 1979-1983).
earlier than female peak in first births and first births were earlier than subsequent births. As we reported elsewhere," mean birthweight in Japan also shows a clear seasonal periodicity, having two peaks in spring and autumn and two troughs in summer and winter. This pattern contrasts with the seasonal change in the proportion of preterm births. This fact suggests that a part of the seasonality in the mean birthweight could be explained by the seasonality of preterm births. Strictly speaking, the peaks and troughs in seasonality occur at slightly different times. This disagreement could be explained by multifactorial characteristics of mean birthweight. Further studies focusing on other factors related to birthweight are necessary to clarify seasonality in birthweight as a whole.
There are two possible explanations for the seasonality of preterm births, i.e. the seasonality of the gestation period itself and the seasonal change in time between menstruation periods. Because weeks of gestation is estimated from the date of the last menstrual period (LMP), if seasonal change in the period of menstruation exists, then the estimated gestation period could be influenced by season. A previous report12 has revealed seasonal change in gonadotropic hormones in animal experiments. However, there has been no investigation of the existence of seasonality in time between menstruation periods in man. If the seasonal rhythm of preterm births observed in this study reflects a change in the period after conception, we have to clarify whether the seasonal change is
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
Term
92.0 n
4.0-
J
I
94
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY Male,
115 -
Female,
Prlmlparae
Prlmlparae
1 1 5 -i
SI
SI 11010510095-
90 2
6
Male,
115-
8
10 12 MONTH
2
Multlparae
4
6
Female,
115-
8
10
12
Multlparae
SI
110-
110-
105-
105-
100-
100-
95-
95-
90
90 0
2
4
6
8
10
0
12
2
4
6
8
FIGURE 2.
12
Seasonal index of the proportion of preterm births (all Japan, 1979-1983).
Primiparae
1.0
10
MONTH
MONTH
Multlparae
1.0
-0.5-
-0.5-
-1.0 4-
-i.o 412
2 4
0
36
TIME LAG (MONTH)
12
2 4
36
TIME LAG (MONTH) MALE • —
FEMALE
FIGURE 3. Correlogram of the proportion of preterm births (all Japan, 1979-1983).
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
SI
4
SEASONAUTY OF PRETERM BIRTHS IN JAPAN
On the other hand, the available literature reports many factors associated with preterm births.5 These include low pre-pregnancy weight, general morbidity, genital tract infection, environmental toxins, prior preterm births, cigarette smoking, etc. If seasonal changes in these factors exist, then part of the seasonal periodicity in preterm births could be explained through changes in these factors. For example, it is well known that appetite and basal metabolic rate change according to season,15 and this might cause seasonal change in maternal pre-pregnancy weight. So far as general morbidity is concerned, many upper respiratory infections such as influenza are more
prevalent in the winter season. According to De Vries et al.l6 frequency of detection of aflatoxin, which has harmful effects on birth outcome, was significantly higher in maternal and cord bloods during the wet months in a developing country. In order to examine the increase of preterm births in summer and winter in detail, we calculated the mean, standard deviation, skewness and kurtosis of gestation periods for each month over the period, stratified by sex and parity and then carried out time-series analyses on these statistical measures to obtain seasonal indices. According to the results, standard deviations of gestation period were larger both in summer and winter, but interesting differences in kurtosis and skewness existed between the two seasons, i.e. the summer increase in preterm births was characterized by a decrease of skewness which meant the extension of the lower part of the distribution, while the kurtosis in summer was almost the same as the yearly mean. On the other hand, the winter increase was characterized by a decrease of kurtosis which meant a flat-topped distribution while skewness was around the yearly mean. This result suggests that decreases in gestation period occur in the lower part of the distribution in summer, but that decrease occurs all over the distribution in winter, and therefore that the cause of preterm births might be different between the two seasons. Figure 4 shows differences between the two seasons schematically. Another important consideration in the cause of seasonality in preterm births is derived from the recently published WHO Report on the social and biological effects on perinatal mortality.17 This suggested that a wide variation in conception combined with constant risk of preterm delivery could produce the seasonal variation in preterm births. Table 1 shows actual numbers of preterm, term, and post-term births of male first births in each month over the period. Figure 5 shows seasonal indices of the average daily number of births calculated from this data using the time-series analysis. While first births show two peaks in January and September and two troughs in May and November, subsequent births show one peak in May to July. For first births, peaks in the seasonal index of total births is preceded by peaks in the proportion of preterm births. This result suggests that there are seasonal variations in conception rates in Japan. Based on birth data in Table 1, it is possible to estimate the number of conceptions in each month. For convenience of calculation, we presumed preterm, term, and postterm births corresponded to births in the eighth, the ninth, and tenth month of gestation, respectively. Using this presumption, for male first births, the hypothetical number of conceptions in each month and cor-
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
indirect acting through variation in factors associated with gestation periods, or direct, constituting an immediate physiological response in the onset of labour to environmental changes, e.g. day length, temperature, moisture, barometric pressure, and so on. The onset of labour results from the gradual convergence of a number of elements, including stretching of the uterus by the fetus, oestrogen, progesterone and oxytocin, etc. each of which has a complex relationship with the others.13 If climatic factors have direct effects on these factors, part of the seasonality in preterm births could be explained by season itself. Animal experiments in the field of chronobiology indicate that the length of gestation is managed by the interaction between an 'internal' timer and the circadian timer.M The internal timer which is genetic, counts absolute time after conception and once gestational maturity has reached a certain point, parturition is governed by the circadian timer. The SCN (suprachiasmatic nuclei) is the site of a known circadian pacemaker in the mammalian brain which generates and regulates the rhythmic expression of numerous biological processes, e.g. circadian, seasonal, and 'circannual' rhythms. The daily light-dark cycle entrains this endogenous pacemaker to a 24-hour period via a direct retinohypothalamic projection to the SCN. Reppert et al.u showed that artificial change of the light-dark cycle caused a change in the timing of the onset of labour in the rat, which usually occurs between the 21st or the 22nd subjective day, and that destruction of the SCN eliminated the circadian gating. This fact is of much interest when discussing the cause of seasonal periodicity in perterm births, i.e. seasonal change in day length might be partly responsible for seasonality in preterm births. It is not appropriate, of course, to extrapolate results from animal experiments to human physiology without careful consideration. Further studies are necessary to clarify the relationship between reproductive events in man and the seasonal change in environment.
95
96
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY [Winter] Kurtosls: decreased Skewness: around the yearly mean
Yearly mean r
a
Qestattonal Weeks
Gettatlonal Weeks
Yearly mean
Qestatlonal Weeks FIGURE 4. Seasonal changes in the distribution of gestational periods. The summer increase of the preterm births is characterized by the increase of skewness which means the extension of the lower part of the distribution. On the other hand, the winter increase ofpreterm births is characterized by the decrease ofkurtosis which corresponds to the flat-topped distribution.
Male
110 < SI
100-
90-
800
V"A/ \\ J
SI
100-
90-
—o- -
Primlparae
Primlparae
—-•-- -
Muttlparae
Hultlparae 80
2
FIGURE 5.
4
6
8
10 12 MONTH
0
2
4
6
8
10
12
MONTH
Seasonal index of average daily number of births (all Japan, 1979-1983).
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
[Summer] Kurtosis : around the yearly mean Skewness: Increased
97
SEASONALITY OF PRETERM BIRTHS IN JAPAN
TABLE 1. Number of preterm, term, and post-term births and theoretical simulation of seasonal pattern of preterm births (male, first births, all Japan, 1979-1983). Actual data Preterm
1244 1032 1024 942 962 1114 1197 1220 1033 1109 1110 1335 1338 1030 1067 1026 1083 1143 1293 1268 1202 1080 1117 1314 1248 1024 1023 975 1003 1140 1210 1176 1095 1056 1033 1287 1237 952 966 1005 1020 1131 1175 1233 1054 1035 1117 1301 1345 1107
29413 26211 27051 23490 23353 22978 26699 28221 27 281 26549 25028 27419 28215 25973 26243 23436 23 198 22564 26146 27415 27159 26192 24199 26589 27339 25154 25481 24043 23 341 22266 25560 27097 26788 26072 23666 25970 26780 24177 24610 22488 22554 22947 26280 27 767 26879 25364 23681 26473 26859 24167
Estimated numbers Post-term
Conception
Preterm
1585 1541 1647 1616 1635 1426 1661 1809 1658 1834 1526 1479 1619 1397 1377 1300 1322 1069 1333 1495 1491 1444 1332 1297 1338 1281 1353 1391 1343 1261 1350 1466 1551 1523 1347 1284 1271 1181 1333 1248 1231 1152 1337 1504 1288 1336 1225 1167 1142 1041
29102 29699 26149 25721 25601 29622 31076 30335 29108 27616 30148 30947 28688 28573 25 825 25293 24980 28784 30199 29871 28726 26576 29044 29934 27 755 27896 26409 25 577 24619 28166 29858 29487 28514 26006 28274 29248 26747 26810 24685 24711 25304 28915 30230 29448 27643 25883 28732 29201 26707 28001 25994 25198 24629 28600 29506 28733 27222
1174 1198 1055 1038 1033 1195 1254 1224 1174 1114 1216 1248 1157 1153 1042 1020 1008 1161 1218 1205 1159 1072 1172 1208 1120 1125 1065 1032 993 1136 1204 1190 1150 1049 1141 1180 1079 1082 996 997 1021 1166 1219 1188 1115 1044 1159 1178 1077 1130
Estimated
Term
Post-term
Sum
26506 27050 23817 23427 23317 26980 28304 27 629 26512 25153 27459 28187 26129 26024 23521 23037 22752 26216 27 505 27207 26164 24205 26453 27264 25279 25408 24053 23296 22423 25654 27195 26857 25971 23686 25 752 26639 24361 24419 22483 22507 23047 26336 27533 26821 25177 23574 26169 26596 24325
1420 1450 1276 1255 1250 1446 1517 1481 1421 1348 1472 1511 1400 1395 1261 1235 1219 1405 1474 1458 1402 1297 1418 1461 1355 1362 1289 1248 1202 1375 1457 1439 1392 1269 1380 1428 1306 1309 1205 1206 1235 1411 1476 1437 1349 1263 1402 1425
29525 26305 25 736 25767 29484 30974 30320 29107 27790 30055 30816 28793 28466 25936 25306 25148 28653 30115 29840 28694 26779 28958 29802 27 865 27828 26447 25578 24807 28060 29760 29464 28459 26219 28201 29098 26871 26721 24789 24733 25419 28790 30132 29412 27658 26082 28610 29075 26880
"^ preterm
3.573 3.946 4.014 4.638 4.253 3.952 3.872 3.827 4.376 4.152 3.755 4.004 3.661 3.933 3.983 4.617 4.251 4.001 3.8S4 3.736 4.377 4.172 3.758 4.037 3.827 3.902 3.8S2 4.579 4.291 3.999 3.903 3.686 4.352 4.184 3.708 4.027 3.727 4.022 4.128 4.587 4.234 3.943 3.791 3.775 4.444 4.117 3.704 4.204
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
Jun-78 Jul-78 Aug-78 Sep-78 Oct-78 Nov-78 Dec-78 Jan-79 Feb-79 Mar-79 Apr-79 May-79 Jun-79 Jul-79 Aug-79 Sep-79 Oct-79 Nov-79 Dec-79 Jan-80 Feb-80 Mar-80 Apr-80 May-80 Jun-80 Jul-80 Aug-80 Sep-80 Oct-80 Nov-80 Dec-80 Jan-81 Feb-81 Mar-81 Apr-81 May-81 Jun-81 Jul-81 Aug-81 Sep-81 Oct-81 Nov-8I Dec-81 Jan-82 Feb-82 Mar-82 Apr-82 May-82 Jun-82 Jul-82 Aug-82 Sep-82 Oct-82 Nov-82 Dec-82 Jan-83 Feb-83
Term
98 TABLE 1.
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY Continued (male, first births, all Japan, 1979-1983) Actual data
Estim!ited numbers
Estimated
Term
Post-term
Conception
Preterm
Term
Post-term
Sum
•fa preterm
Mar-83 Apr-83 May-83 Jun-83 Jul-83 Aug-83 Sep-83 Oct-83 Nov-83 Dec-83
1094 1060 1038 1140 1200 1261 1085 1050 1146 1224
25706 23716 23079 22442 26215 27063 26181 24970 24070 26400
1195 1188 1184 1059 1149 1245 1243 1291 1167 1151
26271
1049 1016 994 1154 1190 1159 1098 1060
25 503 23675 22950 22432 26049 26874 26170 24794 23928
1304 1367 1269 1230 1202 1396 1440 1402 1329 1282
27 856 26058 25213 24816 28441 29429 28708 27256
3.766 3.899 3.942 4.65 4.184 3.938 3.825 3.889
Total Proportion
67529 4.034
1 524667 91.084
81709 4.881
responding numbers of preterm, term and post-term births could be estimated as shown below. For example, the hypothetical number of conceptions in June 1978 is estimated as the sum of the number of preterm births in January 1979, term births in February 1979, and post-term births in March 1979, which adds up to 29102. Suppose the probability of preterm, term, and post-term births are 4.034%, 91.084%, and 4.881
International Journal of Epidemiology © International Eptdemiological Association 1992
Seasonality of Preterm Births in Japan SHINYA MATSUDA'f AND HIROAKI KAHYO*
Seasonal variation in birthweight is well known.'"* It is, however, difficult to identify a particular cause of this seasonality, because many factors influence birthweight including period of gestation, maternal constitution, nutrition, general morbidity, smoking, and so on.5'6 Seasonality in these factors, therefore, might explain the seasonal variation in birthweight. Among factors associated with birthweight, period of gestation is known as the most important and definitive factor.3'6 Relatively few studies, however, have been concerned with the seasonality of the gestation period.7 In this investigation, we focused on the seasonal changes in proportion of preterm births in Japan over the period 1979-1983 and found apparent seasonality in proportion of preterm births in Japan.
and Information Department, Ministry's Secretariat, Ministry of Health and Welfare, Japan. The data on length of gestation is described in the vital statistics as a frequency table with one weekintervals (36 weeks and less, 37, 38, 39, 40, 41, 42 weeks and more). From this table the proportion of preterm births (less than 37 weeks gestation) were calculated. Vital registration of live-born infants in Japan is almost 100% complete. The total number of live singleton infants registered in the vital statistics during the five-year period was 7665006. Climate in Japan
The Japanese climate differs greatly from region to region. The country's north-south length of 3000km puts one end in the sub-arctic zone and the other in the sub-tropical zone. The influence of ocean currents are also important. Most of Japan, however, enjoys a moderate, oceanic type of climate with four distinct seasons, i.e. spring (March to May), summer (June to August), autumn (September to November), and winter (December to February). Another important characteristic of the climate is the rainy season which lasts from June to July, and typhoons which often hit the southwestern part of Japan between August and October.
MATERIALS AND METHODS Data Source
The data used for this investigation were derived from computer tapes of the vital statistics of Japan for January 1979 to December 1983 provided by Statistics •Department of Human Ecology, School of Medicine, University of Occupational and Environmental Health, Yahatanishi-ku Kitakyushu, 807, Japan. tDepartment of Hygiene, Faculty of Medicine, Kyoto University, Japan
91
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
Matsuda S (Department of Human Ecology, School of Medicine, University of Occupational and Environmental Health, Yahatanishi-ku, Kitakyushu, 807, Japan) and Kahyo H. Seasonality of preterm births in Japan. International Journal of Epidemiology 1992; 21: 91-100. Seasonal variations in the proportion of preterm births in Japan from January 1979 to December 1983 are analysed using a traditional method of time-series analysis, which divides the variation in a series into trend, seasonal variation, other cyclic change, and remaining irregular fluctuations. It is shown that the proportion of preterm births in Japan have a clear seasonal periodicity with two peaks in summer and winter. Analysis of seasonality by period of gestation shows that interesting differences in kurtosis and skewness exist between summer and winter, i.e. the summer increase in preterm births was characterized by an increase of skewness which means an extension of the lower part of the distribution. On the other hand, the winter increase was characterized by a decrease of kurtosis which corresponds to a flat-topped distribution. This result suggests that causes of preterm births might be different between the two seasons. Theoretical simulations based on actual birth data in Japan over the period, are carried out to examine how season of conception could influence seasonal variations in the proportion of preterm births. Results show that, at least for first births, seasonality in conception rates could be one explanatory factor for the observed seasonal variation in proportions of preterm births. Another analysis reveals that conception in May and June are more likely to result in preterm births in Japan.
92
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Statistical Approach
,
+ k
-m
2
)
N where rk
= autocorrelation coefficient at lag months k, x, = index value at t th month, m, = mean of index values from x, to x ^ N-k
m, = Z x/(N - k) and, m2 = mean of index values from xk + , to xN; N-k
m, = Z x,, t /(N - k) 1 - 1
'•"•*
RESULTS Figure 1 shows plots of the time series of the proportion of preterm, term, and post-term births, for male first births and male subsequent births, respectively, with a significant regression line. While preterm and term births show upward trends, post-term births show downward trends. These findings were the same for female births.
Figure 2 shows seasonal indices of proportion of preterm births. All of them have a similar appearance with two peaks in winter and summer (or rainy season), and with two troughs in spring and autumn. There is a little difference in the summer peak among them, i.e. male first births have a peak in June, female first births in July, and subsequent births of both sexes in August. In the comparison between first births and subsequent births, the latter have a lower peak in summer than the former. Figure 3 shows correlograms of the proportion of preterm births, which reveal strikingly regular patterns both for first and subsequent births. The former show high autocorrelation coefficients with a peak correlation of about 0.80 for male and about 0.50 for female at six-month intervals. And the latter also show high coefficients with about 0.80 for male and about 0.50 for female at every 12-month lag time. DISCUSSION Figure 1 shows an upward trend in the proportion of preterm and term births in Japan over the period. Ishizuka et al10>" compared neonatal mortality rates of low birthweight infants between 1980 and 1985 in Japan using the data from main hospitals (1980: about 25 000 cases in 528 hospitals; 1985: about 30000 cases in 648 hospitals), and reported that neonatal mortality rates had decreased from 8.8% in 1980 to 4.9% in 1985. They also compared the mortality rates between hospitals equipped with NICU (Neonatal Intensive Care Units) and hospitals without NICU and observed that neonatal mortality rates were much higher in hospitals without NICU. Their result suggests that one of the contributing factors to the upward trend in the proportion of preterm births may be recent developments in neonatal care for premature infants. Another interesting result shown in Figure 1 is that the proportion of post-term births show a downward trend, while term births show an upward trend. Factors associated with this phenomenon are unknown. A greater use of induced labour could be one of the factors for this phenomenon. Further studies using birth records from obstetric facilities which have more detailed information, especially on conditions at onset of labour, are necessary to clarify factors associated with this phenomenon. There is no doubt that a seasonal rhythm in the proportion of preterm births exists as shown by the high autocorrelation coefficient of monthly preterm birth data in Japan. The seasonality was observed for both sexes and both parities. The appearance of their seasonal indices were similar but there was a little difference in timing of the summer peak. The male peak was
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
Traditional methods of time-series analysis which divide the variation in a series into trend, seasonal variation, other cyclic change, and remaining irregular fluctuations, were adopted for the investigation.8 This approach is considered appropriate for a particular variable when the variation is dominated by trend and/or seasonality.8 The procedure is as follows. 1) Firstly, time series for the monthly data of male first births (primiparae), male subsequent births (multiparae), female first births, and female subsequent births, were plotted and then examined for linear trends by least squares regression. 2) An observed value (y) for each month (t) was divided by the corresponding yearly average (y' t ), (the latter was calculated by 12-month moving average) in order to calculate an index value for each month (yt/y',*100). The average of index values was calculated in each calendar month for five years, and then accommodated in the way where the summed value of 12 months became 1200% using a weighted average in order to calculate seasonal index for each calendar month. 3) Autocorrelation coefficients were calculated up to 36 lag months for the data sets using index values and then correlograms were plotted to examine the seasonal rhythm. The formula is as follows;
93
SEASONALITY OF PRETERM BIRTHS IN JAPAN
Multiparae
Primiparae Preterm
5.0-
Preterm A ft
5.0-
9
4.54.03.5-
4.5-
1
vv PWft
3.0G
v
V
VJ r
y = 3.876 + 0.005X 1 2
24
36
48 60 MONTH
3J-
w
3.00
V
WLrVfrM
y = 4.065 + 0.006x 12
94.0 n
24
36
48 60 MONTH
Term y = 91.649 + 0.018x
91.0-
93.0-
90.0-
92.0 • y = 90.511 + 0.018x 91.0
89.0 12
24
36
Postterm
7.0-
12
48 60 MONTH
6.0-
6.0 i * 5.0-
5.0-
4.0-
4.0-
y = 5.585 - 0.023X
3.0
24
36
48 60 MONTH
Postterm y = 4.285 - 0.024x
3.02.0
12
24
36
48 60 MONTH
12
24
36
48
60
MONTH
FIGURE 1. Seasonal fluctuation and secular trends in the proportion of preterm, term and post-term births (male, all Japan, 1979-1983).
earlier than female peak in first births and first births were earlier than subsequent births. As we reported elsewhere," mean birthweight in Japan also shows a clear seasonal periodicity, having two peaks in spring and autumn and two troughs in summer and winter. This pattern contrasts with the seasonal change in the proportion of preterm births. This fact suggests that a part of the seasonality in the mean birthweight could be explained by the seasonality of preterm births. Strictly speaking, the peaks and troughs in seasonality occur at slightly different times. This disagreement could be explained by multifactorial characteristics of mean birthweight. Further studies focusing on other factors related to birthweight are necessary to clarify seasonality in birthweight as a whole.
There are two possible explanations for the seasonality of preterm births, i.e. the seasonality of the gestation period itself and the seasonal change in time between menstruation periods. Because weeks of gestation is estimated from the date of the last menstrual period (LMP), if seasonal change in the period of menstruation exists, then the estimated gestation period could be influenced by season. A previous report12 has revealed seasonal change in gonadotropic hormones in animal experiments. However, there has been no investigation of the existence of seasonality in time between menstruation periods in man. If the seasonal rhythm of preterm births observed in this study reflects a change in the period after conception, we have to clarify whether the seasonal change is
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
Term
92.0 n
4.0-
J
I
94
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY Male,
115 -
Female,
Prlmlparae
Prlmlparae
1 1 5 -i
SI
SI 11010510095-
90 2
6
Male,
115-
8
10 12 MONTH
2
Multlparae
4
6
Female,
115-
8
10
12
Multlparae
SI
110-
110-
105-
105-
100-
100-
95-
95-
90
90 0
2
4
6
8
10
0
12
2
4
6
8
FIGURE 2.
12
Seasonal index of the proportion of preterm births (all Japan, 1979-1983).
Primiparae
1.0
10
MONTH
MONTH
Multlparae
1.0
-0.5-
-0.5-
-1.0 4-
-i.o 412
2 4
0
36
TIME LAG (MONTH)
12
2 4
36
TIME LAG (MONTH) MALE • —
FEMALE
FIGURE 3. Correlogram of the proportion of preterm births (all Japan, 1979-1983).
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
SI
4
SEASONAUTY OF PRETERM BIRTHS IN JAPAN
On the other hand, the available literature reports many factors associated with preterm births.5 These include low pre-pregnancy weight, general morbidity, genital tract infection, environmental toxins, prior preterm births, cigarette smoking, etc. If seasonal changes in these factors exist, then part of the seasonal periodicity in preterm births could be explained through changes in these factors. For example, it is well known that appetite and basal metabolic rate change according to season,15 and this might cause seasonal change in maternal pre-pregnancy weight. So far as general morbidity is concerned, many upper respiratory infections such as influenza are more
prevalent in the winter season. According to De Vries et al.l6 frequency of detection of aflatoxin, which has harmful effects on birth outcome, was significantly higher in maternal and cord bloods during the wet months in a developing country. In order to examine the increase of preterm births in summer and winter in detail, we calculated the mean, standard deviation, skewness and kurtosis of gestation periods for each month over the period, stratified by sex and parity and then carried out time-series analyses on these statistical measures to obtain seasonal indices. According to the results, standard deviations of gestation period were larger both in summer and winter, but interesting differences in kurtosis and skewness existed between the two seasons, i.e. the summer increase in preterm births was characterized by a decrease of skewness which meant the extension of the lower part of the distribution, while the kurtosis in summer was almost the same as the yearly mean. On the other hand, the winter increase was characterized by a decrease of kurtosis which meant a flat-topped distribution while skewness was around the yearly mean. This result suggests that decreases in gestation period occur in the lower part of the distribution in summer, but that decrease occurs all over the distribution in winter, and therefore that the cause of preterm births might be different between the two seasons. Figure 4 shows differences between the two seasons schematically. Another important consideration in the cause of seasonality in preterm births is derived from the recently published WHO Report on the social and biological effects on perinatal mortality.17 This suggested that a wide variation in conception combined with constant risk of preterm delivery could produce the seasonal variation in preterm births. Table 1 shows actual numbers of preterm, term, and post-term births of male first births in each month over the period. Figure 5 shows seasonal indices of the average daily number of births calculated from this data using the time-series analysis. While first births show two peaks in January and September and two troughs in May and November, subsequent births show one peak in May to July. For first births, peaks in the seasonal index of total births is preceded by peaks in the proportion of preterm births. This result suggests that there are seasonal variations in conception rates in Japan. Based on birth data in Table 1, it is possible to estimate the number of conceptions in each month. For convenience of calculation, we presumed preterm, term, and postterm births corresponded to births in the eighth, the ninth, and tenth month of gestation, respectively. Using this presumption, for male first births, the hypothetical number of conceptions in each month and cor-
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
indirect acting through variation in factors associated with gestation periods, or direct, constituting an immediate physiological response in the onset of labour to environmental changes, e.g. day length, temperature, moisture, barometric pressure, and so on. The onset of labour results from the gradual convergence of a number of elements, including stretching of the uterus by the fetus, oestrogen, progesterone and oxytocin, etc. each of which has a complex relationship with the others.13 If climatic factors have direct effects on these factors, part of the seasonality in preterm births could be explained by season itself. Animal experiments in the field of chronobiology indicate that the length of gestation is managed by the interaction between an 'internal' timer and the circadian timer.M The internal timer which is genetic, counts absolute time after conception and once gestational maturity has reached a certain point, parturition is governed by the circadian timer. The SCN (suprachiasmatic nuclei) is the site of a known circadian pacemaker in the mammalian brain which generates and regulates the rhythmic expression of numerous biological processes, e.g. circadian, seasonal, and 'circannual' rhythms. The daily light-dark cycle entrains this endogenous pacemaker to a 24-hour period via a direct retinohypothalamic projection to the SCN. Reppert et al.u showed that artificial change of the light-dark cycle caused a change in the timing of the onset of labour in the rat, which usually occurs between the 21st or the 22nd subjective day, and that destruction of the SCN eliminated the circadian gating. This fact is of much interest when discussing the cause of seasonal periodicity in perterm births, i.e. seasonal change in day length might be partly responsible for seasonality in preterm births. It is not appropriate, of course, to extrapolate results from animal experiments to human physiology without careful consideration. Further studies are necessary to clarify the relationship between reproductive events in man and the seasonal change in environment.
95
96
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY [Winter] Kurtosls: decreased Skewness: around the yearly mean
Yearly mean r
a
Qestattonal Weeks
Gettatlonal Weeks
Yearly mean
Qestatlonal Weeks FIGURE 4. Seasonal changes in the distribution of gestational periods. The summer increase of the preterm births is characterized by the increase of skewness which means the extension of the lower part of the distribution. On the other hand, the winter increase ofpreterm births is characterized by the decrease ofkurtosis which corresponds to the flat-topped distribution.
Male
110 < SI
100-
90-
800
V"A/ \\ J
SI
100-
90-
—o- -
Primlparae
Primlparae
—-•-- -
Muttlparae
Hultlparae 80
2
FIGURE 5.
4
6
8
10 12 MONTH
0
2
4
6
8
10
12
MONTH
Seasonal index of average daily number of births (all Japan, 1979-1983).
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
[Summer] Kurtosis : around the yearly mean Skewness: Increased
97
SEASONALITY OF PRETERM BIRTHS IN JAPAN
TABLE 1. Number of preterm, term, and post-term births and theoretical simulation of seasonal pattern of preterm births (male, first births, all Japan, 1979-1983). Actual data Preterm
1244 1032 1024 942 962 1114 1197 1220 1033 1109 1110 1335 1338 1030 1067 1026 1083 1143 1293 1268 1202 1080 1117 1314 1248 1024 1023 975 1003 1140 1210 1176 1095 1056 1033 1287 1237 952 966 1005 1020 1131 1175 1233 1054 1035 1117 1301 1345 1107
29413 26211 27051 23490 23353 22978 26699 28221 27 281 26549 25028 27419 28215 25973 26243 23436 23 198 22564 26146 27415 27159 26192 24199 26589 27339 25154 25481 24043 23 341 22266 25560 27097 26788 26072 23666 25970 26780 24177 24610 22488 22554 22947 26280 27 767 26879 25364 23681 26473 26859 24167
Estimated numbers Post-term
Conception
Preterm
1585 1541 1647 1616 1635 1426 1661 1809 1658 1834 1526 1479 1619 1397 1377 1300 1322 1069 1333 1495 1491 1444 1332 1297 1338 1281 1353 1391 1343 1261 1350 1466 1551 1523 1347 1284 1271 1181 1333 1248 1231 1152 1337 1504 1288 1336 1225 1167 1142 1041
29102 29699 26149 25721 25601 29622 31076 30335 29108 27616 30148 30947 28688 28573 25 825 25293 24980 28784 30199 29871 28726 26576 29044 29934 27 755 27896 26409 25 577 24619 28166 29858 29487 28514 26006 28274 29248 26747 26810 24685 24711 25304 28915 30230 29448 27643 25883 28732 29201 26707 28001 25994 25198 24629 28600 29506 28733 27222
1174 1198 1055 1038 1033 1195 1254 1224 1174 1114 1216 1248 1157 1153 1042 1020 1008 1161 1218 1205 1159 1072 1172 1208 1120 1125 1065 1032 993 1136 1204 1190 1150 1049 1141 1180 1079 1082 996 997 1021 1166 1219 1188 1115 1044 1159 1178 1077 1130
Estimated
Term
Post-term
Sum
26506 27050 23817 23427 23317 26980 28304 27 629 26512 25153 27459 28187 26129 26024 23521 23037 22752 26216 27 505 27207 26164 24205 26453 27264 25279 25408 24053 23296 22423 25654 27195 26857 25971 23686 25 752 26639 24361 24419 22483 22507 23047 26336 27533 26821 25177 23574 26169 26596 24325
1420 1450 1276 1255 1250 1446 1517 1481 1421 1348 1472 1511 1400 1395 1261 1235 1219 1405 1474 1458 1402 1297 1418 1461 1355 1362 1289 1248 1202 1375 1457 1439 1392 1269 1380 1428 1306 1309 1205 1206 1235 1411 1476 1437 1349 1263 1402 1425
29525 26305 25 736 25767 29484 30974 30320 29107 27790 30055 30816 28793 28466 25936 25306 25148 28653 30115 29840 28694 26779 28958 29802 27 865 27828 26447 25578 24807 28060 29760 29464 28459 26219 28201 29098 26871 26721 24789 24733 25419 28790 30132 29412 27658 26082 28610 29075 26880
"^ preterm
3.573 3.946 4.014 4.638 4.253 3.952 3.872 3.827 4.376 4.152 3.755 4.004 3.661 3.933 3.983 4.617 4.251 4.001 3.8S4 3.736 4.377 4.172 3.758 4.037 3.827 3.902 3.8S2 4.579 4.291 3.999 3.903 3.686 4.352 4.184 3.708 4.027 3.727 4.022 4.128 4.587 4.234 3.943 3.791 3.775 4.444 4.117 3.704 4.204
Downloaded from http://ije.oxfordjournals.org/ at University of Sussex on August 27, 2015
Jun-78 Jul-78 Aug-78 Sep-78 Oct-78 Nov-78 Dec-78 Jan-79 Feb-79 Mar-79 Apr-79 May-79 Jun-79 Jul-79 Aug-79 Sep-79 Oct-79 Nov-79 Dec-79 Jan-80 Feb-80 Mar-80 Apr-80 May-80 Jun-80 Jul-80 Aug-80 Sep-80 Oct-80 Nov-80 Dec-80 Jan-81 Feb-81 Mar-81 Apr-81 May-81 Jun-81 Jul-81 Aug-81 Sep-81 Oct-81 Nov-8I Dec-81 Jan-82 Feb-82 Mar-82 Apr-82 May-82 Jun-82 Jul-82 Aug-82 Sep-82 Oct-82 Nov-82 Dec-82 Jan-83 Feb-83
Term
98 TABLE 1.
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY Continued (male, first births, all Japan, 1979-1983) Actual data
Estim!ited numbers
Estimated
Term
Post-term
Conception
Preterm
Term
Post-term
Sum
•fa preterm
Mar-83 Apr-83 May-83 Jun-83 Jul-83 Aug-83 Sep-83 Oct-83 Nov-83 Dec-83
1094 1060 1038 1140 1200 1261 1085 1050 1146 1224
25706 23716 23079 22442 26215 27063 26181 24970 24070 26400
1195 1188 1184 1059 1149 1245 1243 1291 1167 1151
26271
1049 1016 994 1154 1190 1159 1098 1060
25 503 23675 22950 22432 26049 26874 26170 24794 23928
1304 1367 1269 1230 1202 1396 1440 1402 1329 1282
27 856 26058 25213 24816 28441 29429 28708 27256
3.766 3.899 3.942 4.65 4.184 3.938 3.825 3.889
Total Proportion
67529 4.034
1 524667 91.084
81709 4.881
responding numbers of preterm, term and post-term births could be estimated as shown below. For example, the hypothetical number of conceptions in June 1978 is estimated as the sum of the number of preterm births in January 1979, term births in February 1979, and post-term births in March 1979, which adds up to 29102. Suppose the probability of preterm, term, and post-term births are 4.034%, 91.084%, and 4.881
