C International Epidemlologial Assoiation 1998 Printed in Great Britain International Journal of Epidemiology 199827:587-591 Urbanization and hildhood leukaemia in Taiwan Chung-Yi Li, a Ruey S Iin b and Chuan-Hsiung Iin Bakground In the 1980s soioeonomi development was dramatially rapid in the urbanized muniipalities of Taiwan due to a prospering eonomy. This study addressed the question: Could differenes in the inidene of hildhood leukaemia (age <15) be demonstrated between urban and rural ommunities in Taiwan between 1981 and 1990? Methods The log-linear regression model was used to assess the effets of age, level of urbanization, and alendar year on the variation of hildhood leukaemia inidene rates between 1981 and 1990. Results Between 1981 and 1990, the overall inidene rate of hildhood leukaemia inreased by 20% (rate ratio () =, 95% : -1.5). As ompared to rural areas, metropolitan regions showed a signifiandy higher inidene rate during the study period ( =, 95% : -1.6). This urban-rural differene was partiularly notable among hildren <5 years old ( = 1.5, 95% : -1.9). Dose-response analysis further indiated that risk of hildhood leukaemia was monotonially assoiated with levels of urbanization. The signifiant gradient in the risk of hildhood leukaemia with urbanization was ontributed solely by hildren in the years age group. Conlusions We notied a relationship between urbanization and risk of leukaemia in hildren. Beause of a dramati influx of people into metropolitan areas during the 1980s, our findings may have provided support for the putative assoiation between 'population mixing' or 'population density' and risk of hildhood leukaemia. Whether suh assoiation an be attributable to virus infetion or other aetiologially related leukemogens warrants further investigations. Keywords Childhood leukaemia, population density, population mixing, infetion, relative risk, urbanization Aepted 21 ovember 1997 Taiwan has experiened rapid soioeonomi development resulting in hanges in many aspets of lifestyle and disease patterns in the past 30 years. In the 1950s infetious diseases were the predominant ause of deaths in hildren. Today, aidents and malignanies are the leading auses of death for hildren below the age of 15. Among hildhood aners, leukaemia is the most ommon one. Childhood leukaemia inidene rate has shown a global variation in whih Afrian ountries generally experiene a lower risk.' Competing auses of death and underdiagnosis of aner " Department of Publi Health, College of Mediine, Fu Jen Catholi University, 510 Chung Cheng Rd., Hsinhuang, Taipei Hsiea 24205, Taiwan. b Institute of Epidemiology, College of Publi Health, ational Taiwan University, Taipei. Taiwan. Department of Hospital and Health Care Administration, Chung Tai Junior College, Taihung, Taiwan. have been suggested to aount for suh variation. onetheless, fators assoiated with modernization ould have ontributed to a higher hildhood leukaemia inidene rate in the developed world as well. Epidemiologial studies have shown that risk of hildhood leukaemia is assoiated with soioeonomi status as measured by residential area and paternal oupation. 2 ' 3 Beause modernization, urbanization, and soioeonomi status are known to be highly interrelated, these epidemiologial findings may suggest that fators assoiated with urbanization have played a role in the ausation of hildhood leukaemia. Between 1980 and 1990 Taiwan has experiened a remarkably dramati advanement in eonomi development resulting in rapid urbanization and enormous lifestyle hanges. The purpose of the present study is to investigate whether suh a rapid eonomi development in Taiwan, along with urbanization and lifestyle hanges, has affeted the ourrene of hildhood leukaemia. We used the data of inident hildhood leukaemia ases reported 587
588 ITERATIOAL JOURAL OF EPIDEMIOLOGY Table 1 -speifi distribution of types of hildhood leukaemia in Taiwan, 1981-1990 Aute lymphoblasti leukaemia o. (%) rate (10~ 5 ) 417 (6) 2.3 246 (64.6) 199 (45.8) 862 (57.6) 1.6 on-lymphoblastl leukaemia o. (%) 142 (2) 108 (28.3) 134 (30.9) 384 (25.6) rate (10~ 5 ) 0.8 0.6 Chroni leukaemia fr unlassified o. (%) rate (10~ 5 ) 124 (18.2) 27 (7.1) 0.1 101 (23.3) 0.5 252 (16.8) 0.5 All types o. (%) 683 (100.0) 381 (100.0) 434 (100.0) 1498 (100.0) rate (10~ 5 ) 3.8 2.0 2.3 2.8 to Taiwan's Caner Registry Center between 1981 and 1990 to explore time trends and geographial variations of hildhood leukaemia inidene rates. Materials and Methods Study subjets and base population All hildhood leukaemia ases (International Classifiation of Diseases for Onology (ICD-O) odes 980-994) diagnosed between 1979 and 1990 in Taiwan were retrieved from the national aner registry. For eah ase, sex, age at diagnosis, type of leukaemia, and distrit of residene were olleted for analysis. The auray and ompleteness of aner registry were evaluated elsewhere. 4 It suggested that the registration for the first 2 years is less omplete. To avoid a false sense of rising seular trend in hildhood leukaemia, the 1979 and 1980 data were exluded from the analyses. The annual distrit-age-speifi (,, years) population sizes between 1981 and 1990 were derived from the national annual ensus statistis published by Ministry of the Interior of Taiwan. With the above information, we were able to alulate annual distrit-age speifi inidene rates of hildhood leukaemia. Classifiation of urban/rural distrits Eah distrit in Taiwan (a total of 361 distrits) is given an index of one to eight aording to its degree of urbanization. The level of urbanization was determined using the urban-rural lassifiation of Tzeng and Wu, in whih population density, age omposition, mobility, eonomi ativity and family inome, eduational level, and health and sanitation failities were integrated into a single index. 5 The distrit with the highest degree of urbanization, suh as the Taipei metropolitan area, was given an index of one, while the mountainous areas with the least degree of urbanization were indexed as eight. For the purpose of study, we redassified the distrits into four levels of urbanization with urbanization indies of one or two ombined into the ategory 'metropolitan', three or four into the ategory 'ity', five or six into the ategory 'town', and seven or eight into the ategory 'rural area'. Statistial analysis Childhood leukaemia inidene rates were alulated for five 2-year periods i.e. 1981-1982, 1983-1984, 1985-1986, 1987-1988 and 1989-1990, for the age groups, and years, respetively. The log-linear model, with the natural logarithm of inidene rates as the dependent variable, was used to fit grouped data from three age groups (,, ), five 2-year periods (1981-1982 to 1989-1990), and four levels of urbanization (metropolitan, ity, town, and rural areas). The log-linear model was fitted with standard Poisson distribution assumption and was used to assess the effets of age, period, and urbanization on the variation of hildhood leukaemia inidene rates. 6 The ratio of two inidene rates was alulated using the formula e m, where m is the regression oeffiient. The statistial signifiane for the effet of hildhood leukaemia inidene rates by age, period, and level of urbanization individually or ombined was tested using analyses of deviane for the loglinear model. The analysis of deviane was also used to test for seular trend of inidene rates and for dose-response relationship between risk of hildhood leukaemia and levels of urbanization by fitting a model with a term for period oded 1 to 5 and a term for levels of urbanization oded 1 to 4, respetively. We used GLIM omputer pakage to fit the log-linear regression model. 7 Results Desriptive epidemiology A total of 1498 ases of hildhood leukaemia (<15 years old) diagnosed between 1981 and 1990 were reported to the Caner Registry Center of Taiwan. Among them, 57.6% were aute lymphoblasti leukaemia, 25.6% were non-lymphoblasti leukaemia, and the other 16.8% were hroni leukaemia and those unlassified. The distributions of types of hildhood leukaemia were not similar aross age groups. Aute lymphoblasti leukaemia was the ommonest subtype in all age groups with a pronouned peak at age, whereas the inidene rate of non-lymphoblasti leukaemia was rather similar in all age groups (Table 1). The annual variation in age-speifi inidene rates showed a signifiant inrease of 20% (rate ratio () =, 95% : -) from 2.3 per 10 5 in 1981-1982 to 2.7 per 10 5 in 1989-1990 (Table 2). The inidene rate in hildren aged was onsistently higher than those of older ages over the study period (data not shown). Although variable, the inidene rates among hildren of different ages all showed a moderate but non-signifiantly rising trend in the 1980s. Predition of inidene rates by age, period and urbanization We fitted separate models to assess the rude and adjusted effets of age, year of diagnosis (alendar year or period), and level of urbanization on the variation of hildhood leukaemia inidene rates (Table 3). The model fitted with age or level of urbanization was signifiantly better than that fitted with the onstant term alone, indiating that the variation in hildhood leukaemia inidene rates an be signifiantly explained by age
URBAIZATIO AD CHILDHOOD LEUKAEMIA I TAIWA 589 Table 2 -speifi relative risk estimates and 95% onfidene interval () of hildhood leukaemia, using '1981-1982' as referene, In Taiwan, 1981-1990 " a Calendar year 1981-1982 1983-1984 142 59 64 265 147 0.8-86 -1.9 98 1.6-2.1 331-1985-1986 130 0.8-72 0.8-1.6 87-1.9 289 0.9-1987-198S 124 0.8-80 0.9-1.7 112 1.7-2.4 316-1989-1990 140-1.5 84-1.9 73 297 - Test for trend in the rate of hildhood leukaemia inidene Deviane differene DF" differene 2.09 1.68 0.46 3.32 * DF = degree of freedom; = rate ratio. b Statistial signifiane based on f} Jest. umber of ases. Table 3 Analysis of deviane for log-linear model, assessing the effet of age, alendar year (period), and urbanization level on the hildhood leukaemia inidene in Taiwan, 1981-1990 Fitting Deviane DF 1 Effet Deviane differene DF differene onstant term onstant + age onstant + period onstant + urbanization onstant + age + period onstant + age + urbanization onstant + period + urbanization onstant + age + period + urbanization ' DF = degree of freedom. b Signifiane based on x 2 test. Crude effet. d Adjusted eflet. 253.6 128.6 244.4 242.0 117.6 114.3 233.2 103.9 59 57 55 56 53 54 52 50 age 0 period 0 urbanization 0 age" period d urbanization 125.0 9.2 11.6 129.3 10.4 13.7 2 4 3 2 4 3 >0.05 >0.05 and level of urbanization individually. On the other hand, the rude effet of period on the variation in hildhood leukaemia inidene rates was ompared to null. In the assessment of adjusted effets, the addition of age to the period-urbanization model gave the most signifiant improvement. The adjusted effet of level of urbanization was statistially signifiant as well. Although the addition of period to the age-urbanization model gave some further improvement, suh improvement was not statistially signifiant. The analysis of deviane indiated that both age and level of urbanization were able to signifiantly predit the variation in hildhood leukaemia inidene rates. Time trend of Inidene rates by ages Table 2 shows the seular trend of age-spedfi relative risk estimates, using the inidene rate of 1981-1982 as referene, of hildhood leukaemia between 1981 and 1990. Over the 10-year period, the overall inidene rate of hildhood leukaemia inreased by 20% ( =, 95% : -). The inrease was partiularly evident in hildren between 5 and 9 ( =, 95% : -1.9). onetheless, the rising trends for the overall and for the three age-speifi inidene rates were all ompared to null statistially, suggesting again that the seular hange of risks of hildhood leukaemia was not signifiant. -speifi inidene rates by level of urbanization Table 4 shows the overall and the age-speifi relative risk estimates, using the inidene rate of rural areas as referene, of hildhood leukaemia with inreasing levels of urbanization. The metropolitan hildren experiened a fold ( =, 95% : -1.6) higher inidene rate than that of hildren in rural areas (2.9 per 10 5 versus 2.2 per 10 5 ). Further age-speifi metropolitan/rural omparison showed that the rate ratio was
590 ITERATIOAL JOURAL OF EPIDEMIOLOGY Table 4 Trend of relative risk estimates of hildhood leukaemia with inreasing level of urbanization, using 'rural area' as referene, by age in Taiwan, 1981-1990 * «Levels of urbanization Rural area Town City Metropolitan 78 40 54 172 152 0.9-1.5 86 0.8-1.7 98 336-158 -1.7 87 0.8-1.7 110 0.9-1.7 355 - " DF = degree of freedom; «. rate rauo; = 95% onfidene Interval. b Statistial signifiane based on % 2 test. umber of ases. 295 1.5-1.9 168 0.9-1.9 172 635-1.6 Test for trend In the rate of hildhood leukaemia Inidene Deviane differene 13.65 2.70 0.48 12.43 DF" differene 1 >0.1 1 >0.1 1 mostly noted in hildren of 0 4 years old (4.4 per 10 5 versus 2.9 per 10 5, = 1.5, 95% : -1.9), and that the rate ratios for the other ages were only moderately and non-signifiantly elevated (Table 4). The trend analysis also suggested that the overall inidene rates showed a statistially signifiant urbanization gradient (P for trend test < 0.01), and the urbanization gradient was solely evident (P for trend test < 0.01) in hildren below 5 years old. Disussion Between 1981 and 1990 the inidene rate of leukaemia signifiantly inreased by 20% in Taiwan's hildren under 15 years old. Although a slightly rising trend in the inidene rates was observed, this did not reah statistial signifiane. In addition, the overall inidene rate was signifiantly higher in metropolitan than in rural areas. Suh metropolitan/rural differene was mostly notable in hildren aged <5 years. The dose-response analysis further showed a signifiant gradient in the inidene rates with levels of urbanization, in partiular among hildren under ages of 5. The main results from the study may imply that fators related to modernization or urbanization ould be assoiated with the ourrene of hildhood leukaemia. These fators may inlude lifestyle hanges, sodoeonoml improvement, and growing diversity and the amount of environmental and industrial hazards. One may reasonably postulate that improvements in aner diagnosis, aess to medial are, and registration ompleteness may aount for a higher inidene rate of hildhood leukaemia in Taiwan's urban regions. Beause the signifiant urbanizationhildhood leukaemia assoiation was observed only in hildren less than 5years old, the assoiation found between level of urbanization and inidene rates would not be merely an artefeat. Moreover, assessment of Taiwan's aner registry also showed no urban/rural differene in registration ompleteness. 4 Although improvements in aner diagnosis, aess to medial are, and registration ompleteness made little ontribution to the observed urbanization-hildhood leukaemia assoiation, they were likely to be responsible for the moderate seular inrease in the inidene rates. Similar findings have been reported in some previous studies. Two British studies onluded that there was evidene of a general tendeny for lymphoid leukaemia to luster in areas of higher soioeonomi status, in partiular for those aged under 5. 8 ' 9 A review of other studies in the US, UK, and Australia also indiated higher risks of hildhood leukaemia assoiated with higher soioeonomi status. 1 Beause some indiators used in our study to determine levels of urbanization, suh as family inome, eduational level, and health and sanitation failities were highly related to soioeonomi status, our results may also suggest a possible relation of hildhood leukaemia to higher soioeonomi status. Moreover, the similarity of age ranges (i.e. <5 years old) at whih soioeonomi status most signifiantly assoiated with hildhood leukaemia as reported in the British study and in this study was of partiular interest. This may have an impliation that a ommon aetiologi fator manifested in different ways. Childhood leukaemia has frequently been proposed as assoiated with exposure to infetion as a result of population mixing. 8 A reent study in the US showed a 40% higher risk in areas of high population density than in low population-density areas. 10 One latest study of Chinese hild population in Hong Kong also indiated signifiant evidene of lustering of aute lymphoblasti leukaemia in areas with extreme population mixing. 11 Taiwan's eonomy developed very rapidly in the 1980s. During this period there was a dramati inrease in population mixing arising from influxes of the rural population into metropolitan areas for better employment opportunities, and hildhood leukaemia might have inreased as a onsequene of population mixing. As shown in Table 5, the population of all ages inreased by 24.2% over the study period in metropolitan regions, and the metropolitan hildhood population (<15 years)
URBAIZATIO AD CHILDHOOD LEUKAEMIA I TAIWA 591 Table 5 Changes in average population size between '1981-1982' and 1989-1990' by level of urbanization Level of urbanization Metropolitan City Town Rural areas Average annual population size 8- perentage of hange Total population Child population 1981-1982 1989-1990 hange (%) 1981-1982 6 383 593 7 930 500 T24.2 2 131 911 3 815 107 4 765 222 T24.9 1 280 081 4 483 625 4 708 158 T5.0 1 425 181 2 909 168 2 699 361 4-7.2 881 476 1989-1990 2 216 919 1 332 460 1 267 888 659 699 hange (%) T4.0 T4.1 4.H.0 4-25.2 also inreased by 4.0% within the same period. On the other hand, the total and hildhood populations in rural areas dereased by 7.2% and 25.2%, respetively. Given the sustained deline in birth rates sine the early 1960s in Taiwan, the inrease in population in urban regions was largely due to the migration of people from rural parts of the island. From the above information, our study seems able to provide support for the putative assoiation between population mixing and risk of hildhood leukaemia. Whether the possible assoiation between population mixing and risk of hildhood leukaemia has been due to infetion, however, annot be substantiated in our study. In summary, there was a general tendeny for hildhood leukaemia inidene to be higher in urban regions of Taiwan between 1981 and 1990. The assoiation found between urbanization and risk of hildhood leukaemia an be interpreted in two ways. Firstly, beause population density was an integral element of urbanization and there was a notable population mixing in the 1980s, our result may provide empirial support for the hypothesis that population mixing may aount for higher risks of hildhood leukaemia possibly due to inreased suseptibility to infeaion in hildren. Seondly, although our data are supportive of the 'infeaion' hypothesis for hildhood leukaemia, our findings would not prelude the possibility that ertain environmental hazards and/or urbanization-related lifestyle hanges may have played important roles in ausing the urbanizationleukaemia assoiation in hildren. We therefore suggest that further researh is desirable to examine whether the assoiation between urbanization and hildhood leukaemia an be attributable to an individual's exposure to virus infeaion or other urbanization-related hazards. Referenes 1 Greenberg RS, Shuster JL. Epidemiology of aner in hildren. Epidemiol Rev 1985:7:22-48. 2 Browning D, Gross S. Epidemiologial studies of aute hildhood leukemia. A survey of Guyahoga, Ohio. Am J Dis Child 1968,116:576-85. 3 Mwhirter WR. The relationship of inidene of hildhood lymphoblasti leukaemia to sodal lass. Br J Caner 1982;46:645. 4 Yauo RH. Quality Assessment of Caner Registry in Taiwan. MS Dissertation, ational Taiwan University Taipei, Taiwan, 1992 (in Chinese with English Abstrat). 5 Tseng KH, Wu TY. Charateristis of urbanization level in Taiwan Distrits. Geographial Res 1986,12:287-323. 6 Holford TR. The estimation of age, period and ohort effets for vital rates. Biometris 1983:39:311-24. 7 Baker RG, elder JA (eds). The GLIM System. Release 3. Oxford, umerial Algorithms Group, 1978. 8 Kinlen LJ, Clarke K, Hudson C. Evidene from population mixing in British new towns 1946-85 of an infetive basis for hildhood leukaemia. Lanet 1990:336:577-32. 'Draper GJ (ed.). The geographial epidemiology of hildhood leukaemia and non-hodgkin lymphomas in Great Britain, 1966-83. Studies on Medial and Population Subjets. o. 53 London, UK, HMSO, 1991. l0 Muirhead CR. Childhood leukemia in metropolitan regions in the United States: a possible relation to population density. Caner Cause Control 1995:6:383-88. 1 ' Alexander FE, Chan LC, Lam TH el al. Clustering of hildhood leukaemia in Hong Kong: assoiation with the hildhood peak and ommon aute lymphoblasti leukaemia and with population mixing. Br J Caner 1997:75:457-63.