Key words seasonality - childhood - recognition - diabetes type 1
Introduction
Genetic as well as environmental factors affecting the immune system have been implicated
in the pathogenesis of type 1 diabetes (T1D). An interplay between genetic and environmental
triggers results in a varying frequency of incidence rates between various populations.
It has been suggested that such environmental factors as infections and vitamin D
levels may influence the seasonality of T1D [1 ].
Among several viruses, enteroviral infections, especially those caused by coxackievirus
B, have been linked to the development of T1D. Krogvold et al. provide evidence for
the presence of enterovirus in pancreatic islets of newly recognized T1D patients
[2 ]. The authors conclude that there is a possibility that a low-grade enteroviral infection
in the pancreatic islets contributes to T1D progression. Further studies show that
antibodies against the group B 1 coxackieviruses are more frequent among diabetic
children as compared to a control [3 ]
[4 ]. Coxsackievirus B was the most frequently isolated serotype (30%) among 116 isolations
of enteroviruses performed in Polish population [5 ].
In the temperate climate regions of Poland the greatest number of enteroviral infections
occur during the summer and early autumn months. The tropical and subtropical areas
of Poland maintain a constant level of enteroviral infections throughout the year
[6 ]. Since viral infections and vitamin D level [7 ] demonstrate seasonal fluctuations, some studies have tested the hypothesis that
the diagnosis of diabetes may also display seasonality patterns. The seasonality pattern
seems to be dependent on geographical position [8 ]. However, the results are conflicted as only 42 of 105 centers collected by the
World Health Organization Diabetes Mondiale Project over the period 1990–1999 reported
seasonality in the incidence of T1D. A different pattern was reported by two out of
the four centres with significant seasonality in the southern hemisphere [8 ]. Seasonality at the clinical diagnosis of T1D was noted in European countries with
the highest percentage of incident cases in January and the lowest in June. Similar
seasonality patterns were noted in different age-groups, females and males. The pattern
differed by the year of manifestation [9 ].
The aim of the current study was to investigate the seasonality of T1D recognition
in different age groups of children living in the eastern and central regions of Poland.
Material and methods
We included data of all children and adolescents with newly recognised T1D admitted
between January 2010 and December 2014 to seven diabetes/endocrinology departments
of hospitals distributed across the following Polish regions: Podkarpackie, Warmińsko-Mazurskie,
Lubelskie, Świętokrzyskie, Podlaskie and Mazowieckie. These regions are situated in
eastern and central Poland. Depending on the hospital, data regarding birth and diabetes
recognition were collected retrospectively from paper or electronic documentation
or prospectively from electronic databases. Type 1 diabetes in children was diagnosed
according to International Society for Paediatric and Adolescent Diabetes criteria
[10 ].
We assessed in all children seasonality in the incidence of T1D. Additional comparison
between autumn to winter months (from September to February) and spring to summer
months (from March to August) was conducted, and OR was calculated in reference to
months with lower incidence. Moreover, the difference in diabetes incidence between
the warmest and the coldest months was evaluable. Meteorological data regarding temperature
was provided by the Polish Institute of Meteorology and Water Management.
Due to the higher incidence of diabetes among younger children in Poland, and possibly
other seasonal patterns in different ages, patients were divided into two age groups:
0–4 and 5–17 years. Additional analysis was performed based on place of residence
(rural/urban area) and depending on sex.
The study protocol was approved by the Ethics Committee of the Medical University
of Warsaw.
Statistical analysis
Analysis of time series data was performed using a statistical R package. We used
R project for Satistical Computing version 3.3.1, with package forecast version 7.1
and package TTR version 0.23-1. To estimate trend and seasonal components of a seasonal
time series classical seasonal decomposition (Census Method 1) was used, as the best
fit to the data. In the model a seasonal time series consists of a trend component,
a seasonal component and an irregular component. The seasonal component is computed
as the average for each point in the season. The medial average of a set of values
is the mean after the smallest and largest values are excluded. The resulting values
represent the (average) seasonal component of the series. The original series were
adjusted by subtracting from it, the seasonal component. Trend component is different
from the seasonal component in that it is usually longer than one season, and different
cycles can be of different lengths. The random or irregular (error) component were
isolated by subtracting from the seasonally adjusted series the trend component. Comparisons
of groups were performed by chi-square test and analyses of trends were tested using
Pearson correlation in Statistica 6.0. Pearson correlation coefficient analysis was
done for comparing the two time series for groups of boys and girls, or rural and
urban place of residency. In this analysis the pattern of curves, not the group data,
were compared. A high correlation between two patterns from different groups indicates
that these patterns are likely similar, but maybe on a different level. The Odds Ratio
(OR) was calculated with 95% confidence intervals (CI). P-values less than 0.05 were
considered as significant.
Results
The study group consisted of 2174 children (1007 girls) with a mean age of 9.3±4.5
years. The characteristic of the study population is shown in [Table 1 ]. We noted significant seasonality in the incidence of T1D. The estimated seasonal
factors are given for the months January-December in [Table 2 ]. The largest seasonal factor is for January (about 15) and the lowest is for June
(about -14), indicating that there seems to be a peak in diagnosis of diabetes in
January and a minimum rate in June. In the model a seasonal time series consists of
a trend component, a seasonal component, and an irregular component (random). [Fig. 1 ] shows the original time series (top), the estimated trend component (second from
top), the estimated seasonal component (third from top), and the random component
(bottom). The estimated trend component shows a very intensive increase at the end
of 2012 and then some constancy between 2013 and 2014.
Fig. 1 Decomposition of seasonal diabetes diagnosis data in an additive model.
Table 1 Characteristics of the study population.
0–17 years
0–4 years
5–9 years
10–17 years
Number
2174
476
689
1009
Female/Male
1007/1167
241/235
338/351
428/581
Age (years)
9.3±4.5
3.1±1.2
7.4±1.5
13.4±2.2
Urban area
Number
1313
287
419
607
Female/Male
608/705
145/142
198/221
265/342
Age (years)
9.2±4.5
3.1±1.1
7.5±1.4
13.3±2.2
Rural area
Number
861
189
270
402
Female/Male
399/462
96/93
140/130
163/239
Age (years)
9.3±4.5
3.2±1.2
7.4±1.5
13.4±2.2
Table 2 The estimated seasonal factors given for the months January–December.
Month
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Factor
15.03
3.80
0.63
−1.59
−8.95
−14.01
−7.46
−6,57
3.21
13.15
2.32
0.43
The estimated seasonal factor is the result of decomposition (after subtract the trend
and irregular components). The largest seasonal factor is noted in January what is
mean that in January is peak in diabetes diagnosis in each year.
A total of 423 (19%) children were diagnosed in the warmest months (June to August
with a mean temperature of 16.8°C) compared to 636 (29%) diagnosed with T1D in the
coldest months (December to February with a mean temperature of −1.6°C), OR 0.57 95%CI
[0.51-0.67], p<0.0001. T1D onset was noted more frequently in autumn-winter than in
spring-summer: 1270 (58%) vs. 904 (42%) cases, respectively, OR 1.97 95%CI [1.75-2.22],
p<0.0001 ([Fig. 2 ]).
Fig. 2 Relationship between incidence of type 1 diabetes recognition and the mean temperature
in different months at diabetes diagnosis.
There were 476 children aged 0–4 years (241 girls) with a mean age of 3.1±1.2 years
and 1698 children aged 5–17 years (766 girls) with a mean age of 10.9±3.6 years. We
noted a more flat seasonal pattern in children 0–4 years of age compared with subjects
5–17 years of age with a weak correlation trend comparison between both groups, r=0.69,
p=0.001 ([Fig. 3 ]).
Fig. 3 Seasonal pattern at diabetes onset in different age groups.
In the trend comparison in the T1D incidence between girls and boys the correlation
was not statistically significant, which means that this trend is different (r=0.5,
p<0.09). However we noted that these curves pattern seemed to be “shifted one month”
(in the girls group peaks are observed earlier), after adjusting to this observation
these curves were similar (strong correlation r=0.7192, p<0.008) ([Fig. 4 ]).
Fig. 4 Seasonal pattern of type 1 diabetes recognition in female and male.
A total of 1313 children (608 girls) with a mean age of 9.2±4.5 lived in an urban
area and 861 children (399 girls) with a mean age of 9.3±4.5 lived in a rural area.
The curve of T1D incidence in the sequenced months of urban children had a very strong
correlation in trend comparison, when compared to the rural group (r=0.8172, p<0.001),
which means that no difference was observed.
Discussion
Our study showed a sinusoidal seasonal pattern in the incidence of T1D with a lower
number of children diagnosed in spring-summer and a higher frequency in autumn-winter.
Similar to our results other authors confirmed the seasonality in the incidence of
T1D [7 ]
[11 ]
[12 ]
[13 ]
[14 ]
[15 ]
[16 ]
[17 ]. However, no significant seasonality was observed in some studies [7 ]
[18 ]
[19 ]. The seasonal pattern in the diagnosis of T1D in our study showed the highest incidence
in January and the lowest in June, which is in accordance with previous reports [9 ]
[20 ]
[21 ]. In their study, Samuelsson et al. evaluated the seasonal pattern of diagnosis of
T1D between 1977–2001 in children in the south-east of Sweden. Due to an increased
rate of T1D incidence during the 25 years of follow-up the authors divided the study
period into periods of 5 years and it was only during the last two periods that significant
seasonal variation occurred [9 ]. There were no differences in seasonal pattern between four 5-year periods in European
countries during 1989–2008 [12 ]. In our study it is predicted that the diagnoses of T1D in the coming years will
be the highest in January and the lowest in June. Health forecasting seems to be a
useful tool for health service in diabetes care planning.
It is speculated that seasonality in diabetes incidence may be caused by some environmental
factors such as viral infection. The seasonal pattern in diabetes onset in Japanese
children found in 1987 was linked to an epidemic of Coxsackie B3 in 1987 [16 ]. With the exception of 1987, no seasonal variation in the month of onset was observed
in this cohort. It was confirmed that school-aged children are of particular importance
for viral infection transmission and temporary school closures are effective intervention
to reduce influenza transmission [22 ]. Although we do not have any data regarding viral epidemics in our regions during
the study period, the increase in diabetes onset from September may be linked with
the beginning of the school year. As compared to during holidays, children during
the school year are more prone to infections, have less activity, higher stress levels
and spend less time in the sun, which results in lower levels of vitamin D3. These
environmental factors may influence the immune system, decrease insulin sensitivity
and accelerate diabetes onset.
The different seasonal pattern in children 0–4 years of age compared to older subjects
may be partly explained by exposure to other environmental factors. In Poland great
number of enteroviral infections occur during the summer. Considering that seasonal
infections are more common in younger children than in adolescents, we may speculate
that some children attend kindergarten also during summer when they are exposed to
enteroviral infections. Furthermore, according to TEDDY study, seroconversion occurs
in children mainly between 27.7 (15.2–48.3) months of age[23 ]. One of the possible explanation of the more flat seasonal pattern in children 0–4
years of age may be obligatory vitamin D supplementation in lactating women and their
infants. Additionally, in this age group high aggressive destruction of beta cells
is noted which may be less dependent on environmental triggering factors [24 ]. Some authors demonstrate no seasonal pattern in the 0–4 years age group [25 ]. Patterson et al. claim the least seasonal variation in those under 5 years of age
as compared to older children registered in 23 European counties during 1989–2008
[12 ]. In a Greek population, children less than 3 years old had a peak incidence occurring
during the warmer months and the older group had peak values occurring during the
cold months [10 ]. Similarly, Rosenbauer et al. report significant seasonal variation in diabetes
incidence in German children under 5 years of age with the highest rate during summer
(June to August) and the lowest rate in spring (March to May) [26 ].
The current study showed that in the girls group T1D peaks were observed one month
earlier than in boys. The possible explanation of this phenomenon may be more expressed
symptoms of yeast vulvitis in girls which may influence more frequent urine analysis
compared to boys and in consequence earlier recognition of diabetes. A significant
seasonality in the incidence of T1D diagnosis in both genders has been confirmed in
other studies [9 ]
[10 ]
[21 ]
[24 ]. On the other hand, some authors suggestthat a statistically significant seasonal
pattern could be confirmed for males, but not for females [27 ]
[28 ]. Interestingly, Samuelsson et al. report not only seasonal variation of T1D incidence
among Swedish children but also seasonal variation of C-peptide at diagnosis [20 ]. This seasonal variation of C-peptide was significantly correlated to the seasonal
variation of diagnosis and was more pronounced in boys than in girls.
Recent studies confirm the role of vitamin D in the incidence of T1D possibly by suppressing
immune responses. High-dose vitamin D supplementation early in life protected against
T1D [29 ]. Moreover, vitamin D deficiency in pregnancy may increase the incidence of type
1 diabetes in genetically predisposed subjects [30 ]. In Poland, vitamin D deficiency affects both children and pregnant women. The main
reason for vitamin D deficiency in Poland are climate conditions (with adequate vitamin
D synthesis only between April – September) and limitations of skin synthesis due
to the use of sun screens, living indoors, or air pollution [31 ]. We hypothesised that there might be differences in limitations of skin synthesis
between children living in urban and rural places that may cause differences in seasonal
patterns. Although in our population more children with newly recognised T1D lived
in urban than in rural areas we noted similar seasonal variation at diabetes onset
in both groups.
The limitation of our study is the short period of observation. Future data is needed
to observe if the trend of seasonality will be in the increase or stabilize in the
current position.
In conclusion, seasonal variation in incidence of T1D in children with newly recognised
diabetes supports the role of different environmental factors in diabetes onset. Most
children were diagnosed with diabetes in autumn and winter. The youngest children
aged 0–4 years showed less marked variation of the incidence of diabetes as compared
to older subjects. Future analyses of environmental factors are needed to explain
the causes of seasonality.
