Seasonally in the Clinical Onset of Insulin-dependent Diabetes Mellitus in Finnish Children

Transcription

1 merican Journal of pidemiology opynght 6 by The Johns Hopkins University School of Hygiene and Public Health ll rights reserved Vol., o. Printed in U.S.. Seasonally in the linical Onset of nsulin-dependent Diabetes Mellitus in Finnish hildren Marjatta Karvonen, Jaakko Tuomilehto, sa Virtala, Janne Pitkaniemi, ntti Reunanen, va Tuomilehto-Wolf, and Hans K. kerblom for the hildhood Diabetes in Finland (DiMe) Study Group Seasonal patterns in the incidence of insulin-dependent diabetes based on,6 cases diagnosed at age years or under in Finland are described for the years 87. Seasonal patterns were estimated presenting the data as short Fourier series up to three harmonics together with a possible linear trend. This method allows an arbitrary shape for the seasonal effect. Likelihood ratio tests and kaike's information criterion were used to determine the number of harmonics necessary to model the seasonal pattern and to test differences among age- and sex-specific subgroups in the population. Seasonal patterns in incidence were compared between sexes and between the three 5-year age groups with each controlling for the other's effect. significant seasonal pattern in the incidence of insulin-dependent diabetes was found for the sexes combined and for two age groups (- and - years). statistically significant seasonal pattern could be confirmed for males, but not for females. During a calendar year, one cycle with a decreased incidence of insulin-dependent diabetes in June was found among younger boys. mong older boys, there were two distinct cycles with a decreased incidence, the first in June and the second during ovember-december. The most visible seasonal pattern was a lower number of cases diagnosed in June, while during the rest of the year the incidence remained relatively stable and high. The average annual incidence was 5.6 per, persons without any upward peaks. m J pidemiol 6;:7-76. diabetes mellitus, insulin-dependent; incidence data; seasons Seasonal variation in incidence of insulin-dependent diabetes mellitus in children was reported in the s when higher rates of "acute diabetes" were found during the late autumn, winter, and early spring (). Peaks in incidence, with one peak in winter months and the other during the late summer, were detected in northern Sweden among children aged - years registered from 8 to 77 (). Several other epidemiologic studies have described seasonal patterns in the onset (or, better, at diagnosis) of new cases of insulin-dependent diabetes in children. Most studies have reported higher occurrence of insulin-dependent diabetes during the cold autumn and winter months than during the wanner spring and summer months (), but these findings are difficult Received for publication December, 4, and in final form October, 5. bbreviations:, kaike's information criterion; DDM, insulindependent diabetes mellitus. ational Public Health nstitute, Department of pidemiology and Health Promotion, Helsinki, Finland. hildren's Hospital, Department of Pediatrics, Stenbackinkatu, F Helsinki, Finland. Reprint requests to Dr. Marjatta Karvonen, ational Public Health nstitute, Department of pidemiology and Health Promotion, Diabetes and Genetic pidemiology Unit, Mannerheimintie 6, F- Helsinki, Finland. to compare because of differences in methodology. Seasonal variation in the diagnosis of insulin-dependent diabetes has been considered as an indirect evidence for environmental exposure in the development of insulin-dependent diabetes. Recent studies have provided more indirect evidence for an association between viral infections and the pathogenesis of insulin-dependent diabetes, but the final evidence for viruses causing insulin-dependent diabetes is still missing (). Some communicable diseases occur more frequently during the cold winter months in areas where the climate changes during the year. Therefore, infectious diseases could play a role, at least as a triggering factor in the onset of clinical symptoms of insulin-dependent diabetes. Some limited data about the seasonality of the incidence of insulin-dependent diabetes exist in Finland, where the incidence of insulin-dependent diabetes in children is the highest in the world and has risen steeply over the last 4 decades (4). During the 5s, the onset of insulin-dependent diabetes seemed to be distributed rather evenly in time, although the number of incident cases was highest in September (5). Over the period from 68 to 7, a seasonal variation in incidence of insulin-dependent diabetes (defined as 7 Downloaded from on pril

2 8 Karvonen et al. onset of first symptoms) was reported in children in northern Finland aged - years, with peaks in incidence in pril and September (6). The aim of this paper was to analyze seasonal patterns in incidence of insulin-dependent diabetes in children in Finland during the most recent years.,(r) = a, a, h cos TTht\ + / a sin MTRLS D MTHODS Since 87, all hospitals in Finland treating diabetic children have participated in the prospective nationwide registration of childhood insulin-dependent diabetes. Details of the procedures in case ascertainment have previously been described elsewhere (7). urrently, the collection of incidence data continues as a part of the World Health Organization DMOD project (Dbetes MODiale) (8). n Finland, all children with insulin-dependent diabetes are treated in a hospital at the time of diagnosis (7). The diabetes nurses in the pediatric wards of the hospitals record necessary information on standardized forms and send them to the ational Public Health nstitute in Helsinki. t has been shown that the case ascertainment is virtually percent complete (7). During 87,,6 children aged years or under were diagnosed with insulin-dependent diabetes. Population data were obtained from the ational Population Registry, which is updated continuously. The Finnish population aged years or under varied from 5,4 to 68,8 during 87. Statistical methods The average annual incidence rates were calculated per, population per year. The midyear populations aged years or under were used as the denominator. The 5 percent confidence intervals were estimated assuming the Poisson distribution of the cases. ge adjustment of the rates was performed using year intervals (-4, 5-, and - years) with the proportions /, /, and /, respectively, as the standard according to the previous approach by the Diabetes pidemiology Research nternational Group (). The average annual incidence of insulin-dependent diabetes was 5.6 per, population per year (5 percent confidence interval 4.-7.) in Finland during 87. Seasonal patterns in incidence of insulin-dependent diabetes were evaluated by using the method described by Jones et al. (). The method fits sine waves to the incidence data with a fundamental period of one cycle per year, and when necessary higher harmonics can be used. t also allows for different lengths of time intervals and different sizes of populations at risk.the incidence model for group i at time t can be written as where a, = a parameter that allows the size of each group to be arbitrary; «i* = c ii,cos((j) ih ) and j / = -c, 7l sin(^, 7l ); a, = a parameter that allows the size of each group to be arbitrary; b; = (t - 77), the time trend; c lh = the amplitude of the cyclic curve; $ ih = the phase angle; P = the fundamental period of the seasonal model; h = the harmonic of the fundamental frequency; p = the number of terms in the sum. n this case, the 6-year data starting on January, 87, have been divided into 7 intervals, at the time points f, =, t = + 8, t = + 8 +, etc., with a day as the unit of time and the fundamental period of seasonal model P = 65. This method allows an arbitrary shape for the seasonal effect by presenting the data as a short Fourier series with up to three harmonics together with a possible linear trend. Likelihood ratio tests and kaike's information criterion () () were used to determine the number of harmonics necessary to model the seasonal pattern adequately and to test differences among subgroups in the population. Maximum likelihood estimation based on Poisson distribution was used to estimate the parameters of the model. Seasonal patterns in incidence were also compared between males and females and between the three 5-year age groups, each controlling for the other's effect. RSULTS The observed numbers of cases each month are shown in table and the -month moving averages of the insulin-dependent diabetes mellitus (DDM) incidence are shown in figure. The month-to-month variation was more pronounced in males than in females, and it did not follow any obvious pattern in either sex. The values of (log-likelihood), the number of estimated parameters, and from the analysis of seasonal variation in incidence fitting different coefficients to each sex (two) and age groups (three) with and without time trend are shown in tables -6. The m J pidemiol Vol., o., 6 Downloaded from on pril

3 5 O O O D TBL. ases Year Males ll Females ll Total ll of insulin-dependent diabetes meliitus January February March in Finland pril per month May from January Month June ,87 to July December, (7 months) ugust September, October ovember December Total luual 7 7 5, ,6 Downloaded from on pril

4 Karvonen et al. FGUR. The number of cases of insulin-dependent diabetes diagnosed each month in Finland, 87-. Data are presented as -month moving averages for males ( ), females (), and both sexes ( ). MOTH TBL. Summary of model selection for three age groups* and for both sexes, Finland, January,87 to December, o. of harmonics o. of estimated parameters Without trend With trend * ge groups: -4, 5-, and - years, t, kaike's information criterion. lt best models can be conveniently selected based on, with the minimum being the best. The change in (log-likelihood) for models can be tested as chi-square with degrees of freedom equal to the difference in the number of parameters. n the first analysis (table ), different coefficients were fit to each of the three age groups (-4, 5-, and - years) and both sexes for up to three harmonics with and without trend. The best fit in this test seemed to be the model with one harmonic without trend and with minimum (.). However, the difference in the values between models with null, one, and two harmonics were small, and changes in the (log likelihood) terms tested by chi-square test showed that these three models had an equal fit. The next analysis was done by assuming that the two youngest age groups (-4 and 5- years) have similar coefficients. Therefore, the two youngest age groups were pooled, and the analysis was performed for males and females in the age groups - and - years. The likelihood ratio test for this pooling was (-44.) =., and the chi-square test with 8 df was nonsignificant, indicating that pooling was acceptable. The best fit was the model without trend, with two harmonics and the minimum TBL. Summary of model selection for two age groups* and both sexes, Finland, January,87 to December, o. of harmonics o. of estimated parameters Without trend With trend * ge groups: - and - years, t, kaike's information criterion lt m J pidemiol Vol., o., 6 Downloaded from on pril

5 Seasonally in the Diagnosis of DDM TBL 4. Summary for fitting the model (without trend) for age groups - and - years for males and females, Finland, January,87 to December, o. of harmonics *, kaike's information criterion. ge group - years Males Females Males * ge group - years Females (table ). For younger males (age - years), the model with one harmonic and the minimum. was the best fit, and for older males (- years), the model with two harmonics and the 7. was the best fit, indicating a significant monthly variation in incidence among males in both age groups (table 4). mong younger females, the minimum was. for the model with one harmonic, showing a small seasonal variation. The observed and estimated seasonal pattern for males in age groups - and - years with an overall 5 percent confidence interval for the curve is shown in figure. Of several poolings, the model with two harmonics without trend for males and females together and for two age groups (- years and - years) was the best fit with the minimum of 4.6 (table 5). The likelihood ratio test (.86, 8 df) showed that pooling was acceptable. The stepwise fitting of this model for - harmonics is shown in table 6. The younger age group (- years) had a significant seasonal pattern with one harmonic (minimum, -.), whereas in the older age group (- years), we found the pattern with two harmonics with minimum of.. The observed and estimated seasonal pattern for sexes pooled in the two age groups (- and - years) with an overall 5 percent confidence interval for the curve is shown in figure. n both age groups, with pooled sexes and in males alone (but not in females alone), a seasonal pattern in incidence was seen. mong all study subjects and also in males alone, the number of cases diagnosed in June decreased {p < ). n the older age group, we found also another low-incidence period during the cold season, in ovember-december. However, June was the only month for which the incidence was significantly different (lower) from the overall incidence level. There was no single month with a significantly higher observed incidence than expected. DSUSSO Our results showed that in Finland the incidence of insulin-dependent diabetes mellitus in children fluctuates to some extent within a calendar year and also within a longer period of time. n this analysis, we applied a recently developed statistical method previously used to analyze seasonal patterns in incidence in the Denver insulin-dependent diabetes register (). This method allows us to test whether a seasonal pattern exists for sex and age groups, and it can also be used to test whether two or more groups have a same seasonal pattern in incidence. lthough there was a significant seasonal pattern in the incidence of insulin-dependent diabetes for sexes pooled and for two age groups (- and - years), we could confirm a statistically significant seasonal pattern only for males. mong younger boys, we found one distinct cycle during a calendar year with a decreased incidence in June. mong older boys, there were two distinct cycles with a decreased number of cases in June and a smaller one during ovember- December. The lower number of cases diagnosed in June was the most visible seasonal pattern, while during the rest of the year the incidence remained relatively stable and high. seasonal variation in the onset of insulin-dependent diabetes with higher rates during cooler seasons, autumn and winter, has been described previously by several investigators (). However, our results show that there were no high peaks in incidence during the cool months that would have significantly differed from the overall incidence. With the same method to determine the seasonal pattern in the incidence of insulin-dependent diabetes in Denver a seasonal pattern was found only for children aged - years (). n several other studies, a seasonal variation in incidence of insulin-dependent diabetes has been proposed for both sexes only among children aged 5 years or older (-6, 8). The only exception is a report from anada (7) in which a high incidence seasonal peak was found among children m J pidemiol Vol., o., 6 Downloaded from on pril

6 Karvonen et al. i c i o V R < ML / - YRS O TRD/ HRMO V R T ML/ - YRS O TRD / HRMOS MOTH ML / - YRS O TRD/ HRMOS D V R T O FGUR. stimated and observed (solid line) seasonal pattern in the diagnosis of insulin-dependent diabetes in males in Finland, 87-. The value on y-axis denotes the mean of the incidence during the entire study period for the particular age group. The inner interval of two confidence intervals shown is the pointwise 5 percent confidence interval, and the outer interval is a 5 percent confidence interval for the entire curve. The equations for the seasonal pattern (where tis in days) are: males age - years X(t) = +.7 cos(7rf/65) -.66 sin(7rt/65); males age - years \(t) = - 8 cos(irf/65) + 48 sin(irt/65) cos(47rt/65) +.6 sin(47rt/65); males age - years \(t) = cos(7rt/65) -.4 sin(trt/65) -.64 cos(47rt/65) +.57 sin(47rt/65). m J pidemiol Vol., o., 6 Downloaded from on pril

7 Seasonally in the Diagnosis of DDM TBL 5. Summary of model selection for two age groups* and pooled sexes, Finland, January,87 to December, o. Of harmonics o. of estimated parameters Without trend With trend * ge groups: - and - years, t, kaike's information criterion lt TBL 6. Summary for fitting the model for age groups - and - years with pooled males and females, Finland, January,87 to December, o. of harmonics ge group - years * *, kaike's information criterion. ge group - years aged - years, but in the anadian study the majority of children in this age group were age 5 years or older. We also found a seasonal pattern in incidence for males in the younger age group. Since the s, there has been increasing interest in the possibility that a virus infection may play a role in the etiology of insulin-dependent diabetes (,, ). However, virologic studies on patients have not yet produced convincing evidence to prove this hypothesis, although the possibility of numerous viruses in the etiology has been suggested. Our results, in Finland, the country with the highest incidence of insulindependent diabetes in the world, showed that there was a different seasonal pattern for females and males with one nadir among younger males and two nadirs among older males. This does not fit the assumption that the seasonal variation in incidence were caused by a virus infection. The occurrence of diseases caused by viruses, such as mumps, measles, and oxsackie B, that have been proposed to trigger insulin-dependent diabetes show an epidemic pattern, and the duration of the high-incidence period is often short. However, the very high and relatively stable incidence of insulindependent diabetes in Finland supports the assumption that if a virus infection were involved in the etiology of diabetes, such a virus should be endemic in the country. The incidence of common virus infections in children is at the highest during the cold seasons late autumn, winter, and early spring. Therefore, the seasonal pattern in incidence of insulin-dependent diabetes and the nadir around June can support the assumption of a high endemic level of infective agent. The incidence of insulin-dependent diabetes in children increases with age. n our study, nearly 6 percent of the children had been cared outside the home before the school age of 7 years and thus were probably exposed to various infections before school age. Therefore, if an infective agent were be a pathogen causing insulin-dependent diabetes, it must have had a low degree of infectivity. On the other hand, the high and stable incidence of insulin-dependent diabetes suggests chronic or recurrent infections and thus a high endemic level of infections in the country. lmost all Finnish children aged years or under have been vaccinated against several common viruses that previously caused occasional epidemics. However, in spite of such a vaccination program, the incidence of insulin-dependent diabetes is still high and is increasing in Finland, especially among the youngest age groups. Our results suggest that for a virus infection to cause the beta cell damage, the behavior of such a pathogen would have to be very atypical for viruses that cause usually short epidemics in a susceptible population. t is very likely that the seasonality in diagnosis of insulin-dependent diabetes, if found, is not related to causal factors that trigger insulin-dependent diabetes, but it may be associated with the expression of symptoms in subjects who already are at an advanced stage in developing the disease that may take months or years. Many physiologic parameters (for instance, blood glucose, blood lipids, blood pressure, and body weight) and health habits (diet and physical activity) have a seasonal pattern. These may be sufficient to explain the seasonal effect that we observed in the diagnosis of insulin-dependent diabetes. n addition, the beginning of the school holidays in June may play a role in the low rate of newly diagnosed children in Finland. nterestingly, Somersalo et al. (5), who reviewed a series of 5 diabetic children in Finland diagnosed during the 4s and early 5s, found a peak in onset in September (.5 percent of cases diagnosed in that month). They pointed out that the school term began on September and that a peak in newly diagnosed cases could mean that either infec- m J pidemiol Vol., o., 6 Downloaded from on pril

8 4 Karvonen et al. MLS FML/- YRS O THM) M HRMO MOTH ML & FML / - YRS O TRD / HRMOS MOTH i D.75.5 ML & FML / - YRS O TRD/ HRMOS V R T O FGUR. stimated and observed (solid line) seasonal pattern in the diagnosis of insulin-dependent diabetes in both sexes in Finland, 87-. The value on y-axis denotes the mean of the incidence during the entire study period for the particular age group. The inner interval of two confidence intervals shown is the pointwise 5 percent confidence interval, and the outer interval is a 5 percent confidence interval for the entire curve. The equations for the seasonal pattern are: males and females age - years \(t) = cos(7rt/65) -. sin(irt/65); males and females age - years (() = + 84 cos(irt/65) sin(7rf/65) -.88 cos(4irt/65) sin(47rf/65); males and females age - years (f) = +.84 cos(7rt/65) sin(trf/65) cos(4-rt/65) +.45 sin(47rt/65). m J pidemiol Vol., o., 6 Downloaded from on pril

9 Seasonally in the Diagnosis of DDM 5 tious or psychologic factors are related to the onset of childhood diabetes. Many earlier papers have been based on data in which a considerable number of children newly diagnosed with DDM had ketoacidosis at diagnosis. n other words, it means that the time between the onset of clinical manifestation to the actual date bf diagnosis may have been rather long. n our present series, only about one fifth of newly diagnosed cases of DDM had signs of ketoacidosis, which is one of the lowest rates reported. This indicates that the delay prior to diagnosis has been short. Most of the previous studies of seasonality in the risk of insulin-dependent diabetes have been based on relatively few cases. s we have demonstrated in this study, to analyze the month-to-month variation with a sufficient power, it is necessary to have several hundred or maybe over, cases of insulin-dependent diabetes. KOWLDGMTS The hildhood Diabetes in Finland (DiMe) Study Group: Principal nvestigators H. K. kerblom and J. Tuomilehto; oordinators R. Lounamaa and L. Toivanen; Local nvestigators. Fagerlund, M. Flittner, B. Gustafsson,. Haggqvist,. Hakulinen, L. Herva, P. Hiltunen, T. Huhtamaki, -P. Huttunen, T. Huupponen, M. Hyttinen, T. Joki, R. Jokisalo, M-L. Kaar, S. Kallio,.. Kaprio, U. Kaski, M. Knip, L. Laine, J. Lappalainen, J. Maenpaa, -L. Makela, K. iemi,. iiranen,. uuja, P. Ojajarvi, T. Otonkoski, K. Pihlajamaki, S. Pontynen, J. Rajantie, J. Sankala, J. Schumacher, M. Sillanpaa, M-R Stahlberg, -H. Strahlmann, T. Uotila, M. Vare, P. Varimo, and G. Wetterstrand. RFRS. dams F. The seasonal variation in the onset of acute diabetes. rch ntern Med 6;7: Hagglof B, Holmgren G, Wall S. ncidence of insulin-dependent diabetes mellitus among children in a orth-swedish population Hum Hered 8;:48-.. hristau B, Kromann H, Ortved-ndersen O, et al. ncidence, seasonal and geographical patterns of juvenile-onset insulindependent diabetes mellitus in Denmark. Diabetologia 77; : Durruty P, Ruiz F, arcia de los Rios M. ge at diagnosis and seasonal variation in the onset of insulin-dependent diabetes in hile (southern hemisphere). Diabetologia 7; : Dahlquist G, Blom L, Holmgren G, et al. The epidemiology of diabetes in Swedish children - years a six-year prospective study. Diabetologia 85;8: Tajima, LaPorte R, Hibi, et al. comparison of the epidemiology of youth-onset insulin-dependent diabetes mellitus between Japan and the United States (llegheny ounty, Pennsylvania). Diabetes are 85;8 (Suppl. ):. 7. llen, Palta M, lessio DJD. ncidence and differences in urban-rural seasonal variation of type (insulin-dependent) diabetes in Wisconsin. Diabetologia 86;: Siemiatycki J, olle, ubert D, et al. The distribution of type (insulin-dependent) diabetes mellitus by age, sex, secular trend, seasonality, time clusters, and space-time clusters: evidence from Montreal, 7-8. m J pidemiol 86; 4: Joner G, Sovik O. ncreasing incidence of diabetes mellitus in orwegian children - years of age 7-8. Diabetologia 8;:7-8.. Ludvigsson J, foke O. Seasonality of type (insulindependent) diabetes mellitus: values of -peptide, insulin antibodies and haemoglobin lc show evidence of a more rapid loss of insulin secretion in epidemic patients. Diabetologia 8;:84-.. Bingley PJ, Gale M. ncidence of insulin dependent diabetes in ngland: a study in the Oxford region, BMJ 8;8: Schranz G, Prikatsky V. Type diabetes in the Maltese slands. Diabet Med 8;6:8-.. Hamman R, Gay, ruickshanks K, et al. olorado DDM Registry. ncidence and validation of DDM in children aged - yr. Diabetes are ;: Serrano Rios M, Moy S, Martin Serrano R, et al. ncidence of type (insulin-dependent) diabetes mellitus in subjects - years of age in the omunidad of Madrid, Spain. Diabetologia ;:4-4.. Metcalfe M, Baum JD. ncidence of insulin-dependent diabetes in children aged under years in the British sles during 88. BMJ ;: Songini M. ncidence of type diabetes mellitus in Sardinia. Medicographia ; (Suppl. l):-.. Green, ndersen PK, Svendsen J, et al. ncreasing incidence of early onset type (insulin-dependent) diabetes mellitus: a study of Danish male birth cohorts. Diabetologia ;5:8-8.. lamin, Omar M, Zein K, et al. pidemiology of childhood type diabetes in Sudan 87-. Diabetes are,:56-.. Goday, astell, Tresserras R, et al. Sex differences in seasonality at onset of type (insulin-dependent) diabetes mellitus in atalonia. Diabetologia ;5 (Suppl. ):.. Kostraba J, Gay, ai Y. ncidence of insulin-dependent diabetes mellitus in olorado. pidemiology ;:-8.. Podar T, Tuomilehto-Wolf, Tuomilehto J, et al. nsulindependent diabetes mellitus in native stonians and immigrants to stonia. m J pidemiol ;5:-6.. Lipman TH. The epidemiology of type diabetes in children - yr of age in Philadelphia. Diabetes are ; :-5.. Szopa TM, Titchener P, Portwood D, et al. Diabetes mellitus due to viruses some recent developments. Diabetologia ;6: Tuomilehto J, Rewers M, Reunanen, et al. ncreasing trend in type (insulin-dependent) diabetes mellitus in childhood in Finland. nalysis of age, calendar time and birth cohort effects during 65 to 84. Diabetologia ;4: Somersalo O, Hiekkala H, Rantakallio P, et al. Studies in childhood diabetes. Review of 5 cases.. Onset and physical development. nn Paediatr Fenn 6;6: Reunanen, kerblom HK, Kaar ML. Prevalence and tenyear (7-7) incidence of insulin-dependent diabetes in children and adolescents in Finland. cta Paediatr Scand 8;7: Tuomilehto J, Lounamaa R, Tuomilehto-Wolf, et al. pidemiology of childhood diabetes mellitus in Finland background of a nationwide study of type (insulin-dependent) diabetes mellitus. Diabetologia ;5: WHO Multinational Project for hildhood Diabetes, WHO DMOD Project Group. Diabetes are ;:-8.. Diabetes pidemiology Research nternational Group. Geographic patterns of childhood insulin-dependent diabetes mellitus. Diabetes 88;7:-. m J pidemiol Vol., o., 6 Downloaded from on pril

10 6 Karvonen et al.. Jones R, Ford P, Hamman R. Seasonality comparisons among. Gundersen. s diabetes infectious origin? J nfect Dis 7; groups using incidence data. Biometrics 88;44:-44. 4:7.. kaike H. nformation theory and an extension of the maxi-. Gamble DR. possible virus etiology for juvenile diabetes, mum likelihood principle. n: Petrov B, saki F, eds. Second n: reutzfeld W, Kobberling J, eel V, eds. The genetics of international symposium on information theory. Budapest, diabetes mellitus. ew York, Y: Springer-Verlag, 76: Hungary: katemia Kaido, 7: m J pidemiol Vol., o., 6 Downloaded from on pril