Insulin-dependent diabetes mellitus (IDDM) is a chronic autoimmune disease of the pancreas. It is an

Transcription

1 International Journal of Epidemiology International Epidemiological Association 1997 Vol. 26, No. 6 Printed in Great Britain Small Area Variation in the Incidence of Childhood Insulin-Dependent Diabetes Mellitus in Yorkshire, UK: Links with Overcrowding and Population Density A STAINES,* H J BODANSKY,** P A MCKINNEY, F E ALEXANDER, R J Q MCNALLY, G R LAW, H E B LILLEY, C STEPHENSON AND R A CARTWRIGHT Staines A (Division of Public Health, Nuffield Institute for Health, Fairbairn House, Clarendon Road, Leeds LS2 9PL, UK), Bodansky H J, McKinney P A, Alexander F E, McNally R J Q, Law G R, Lilley H E B, Stephenson C and Cartwright R A. Small area variation in the incidence of childhood insulin-dependent diabetes mellitus in Yorkshire, UK: Links with overcrowding and population density. International Journal of Epidemiology 1997; 26: Background. The incidence of insulin-dependent diabetes mellitus (IDDM) incidence varies between and within countries. The origins of this variation are disputed, but they involve both genetic and non-genetic influences. To explore the role of environmental factors in the aetiology of IDDM we have examined the incidence in small geographical areas and related it to variables derived from national censuses. Methods. This is an ecological analysis of incidence data from a register of children with IDDM covering the counties of West Yorkshire, North Yorkshire and Humberside in the north of England. All children aged 16, diagnosed with IDDM between 1978 and 1990 were eligible for inclusion. Spatial variation in incidence between electoral wards was investigated using Poisson regression, in relation to socioeconomic status, population density, urban-rural status and measures of geographical isolation. Ward child populations varied in size from 84 to 7197 (mean = 1545). Results. Rates were significantly lower in wards of high population density and with many overcrowded houses. The rate ratio for areas in the upper half of the childhood density distribution was 0.88 (95% confidence interval (CI) : ) and for the two upper tertiles of household overcrowding the rate ratios were 0.84 (95% CI : ) and 0.68 (95% CI : ) respectively. Conclusions. The incidence of childhood IDDM was associated with environmental factors including population density and overcrowded homes. A possible inference from these data is that patterns of infection are involved in the occurrence of IDDM. Analytical epidemiological studies will be needed to investigate these ideas further. Keywords: diabetes mellitus, insulin-dependent, ecological analysis, United Kingdom, deprivation, overcrowding, census, childhood Insulin-dependent diabetes mellitus (IDDM) is a chronic autoimmune disease of the pancreas. It is an * Division of Public Health, Nuffield Institute for Health, Fairbairn House, Clarendon Road, Leeds LS2 9PL, UK. ** The Diabetes Centre, Leeds General Infirmary, Leeds LS1 3EX, UK. Paediatric Epidemiology Group, University of Leeds, Leeds LS2 9LN, UK. Department of Public Health Sciences, University of Edinburgh, Edinburgh, UK. LRF unit, University of Leeds, Leeds LS2 9NG, UK. Reprint requests to: Dr A Staines, Department of Public Health Medicine and Epidemiology, University College Dublin, Earlsfort Terrace, Dublin 2, Ireland increasingly important public health problem in Western Europe. 1,2 The aetiology of this condition is not well understood; neither are the causes of its increasing incidence. A plausible model for the aetiology of this disease is an interaction between several environmental agents, at least some of which are changing over time, and a defined genetic susceptibility. 3 One way of exploring these environmental effects is to examine geographical variation in diabetes incidence. Differences in incidence between countries are well known, 4 although earlier ideas of a simple north to south gradient have not been substantiated by more extensive studies. 2,5 7 The factors influencing this international variation are not known.

2 1308 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY Many studies of within country variation have been published. Most, 8 26 though not all, have identified within country variation. Some studies have linked variation in IDDM incidence to characteristics of the areas studied such as climate, 23 latitude, 11,15 deprivation 19 and urban-rural status. 12 For many reasons, principally the choice of overly large units for analysis, but also inconsistency in the variables chosen for analysis, and the difficulty in comparing concepts such as social status between different studies, these are not easy to interpret. In order to explore these issues, an ecological analysis has been done on the Yorkshire Regional Children s Diabetes Register (YRCDR) 33 using electoral wards as the unit of analysis, and variables derived from national censuses. METHODS The YRCDR identifies children with IDDM, diagnosed while aged under 17 years and normally resident in the counties of West Yorkshire, North Yorkshire and Humberside, the former Yorkshire Regional Health Authority. Cases are identified through direct contact with hospital clinics, surveys of general practitioners, and from computerized hospital discharge records. Ascertainment is estimated to be over 97% complete. 33 This report analyses data on 1490 children diagnosed with IDDM, between 1978 and The unit of analysis was the electoral ward. There were 536 of these in 1981 and 532 in Cases diagnosed from 1978 to 1984 were allocated to 1981 wards, and those diagnosed from 1985 to 1990 were allocated to 1991 wards. Populations were estimated for each single year of age, from zero to 16, and for each sex, using census data. This was done by using the total population figures in 5-year age groups, allocated to single years of age in proportion to the population figures for people resident in households. Cases were allocated to wards using the Central Postcode Directory. This is a list of postal delivery points in the UK, linked to administrative and political boundaries. Ward characteristics were derived from the small area statistics of the 1981 and 1991 censuses held on the Midas service at the University of Manchester. 34,35 The following ward characteristics were investigated: population density, socioeconomic status, isolation and urban-rural status. Childhood population density was calculated by dividing the total ward population aged zero to 16 by the area in hectares. The total population density was calculated similarly, but using the entire population. Socioeconomic status was measured using the Townsend score. 36 This is a combination of four census measures, unemployment, household access to a car, home ownership and household overcrowding. The total score, expressed as standard deviations from the mean of the original score, and each of the four components, untransformed, were used. Isolation was measured using a score developed by Alexander et al., 37 based on the distance of the population weighted ward centre from built-up areas, as defined on Ordnance Survey maps. Wards were categorized as urban or rural using the Office of Population Censuses and Surveys (OPCS) classification. 38 This is based on the predominant land use among the electoral districts making up the ward. In analysis the six levels provided were combined to form a dichotomous variable: Urban for wholly and predominantly urban wards (levels one to three), and Rural for all others. Statistical analyses were performed using Poisson regression in Genstat and SAS. 39,40 The number of cases observed in each ward was the dependent variable, and the log of the expected number of cases was used as the offset. The expected number of cases was derived using the iterative method of Breslow and Day. 41 This method is used because a full model with terms for every combination of age, sex and area, poses computational difficulties with convergence, and the deviance is hard to interpret because the expected value in each unit is very small. 42 Extra-Poisson variation, that is greater variability than that predicted by the Poisson model, is not uncommon in spatial analysis. 43 This is probably due to intrinsic differences between the areas compared. Adjustment for extra-poisson variation used the methods of Breslow 43 as adapted by Alexander et al. 37 Continuous variables were cut into strata so that each stratum contained an equal proportion of the childhood population; exact equality was not possible because each stratum had to be composed of complete wards. For model selection each variable was cut into two, three and five strata. The choice of whether to include variables as continuous variates, or one of the sets of strata produced was made by considering likelihood ratio tests, and examination of the estimates for each set of strata. Models were selected and tested using standard methods. 44 The final model selected was further tested in a model with separate terms for age, sex, and area. RESULTS There is substantial variation in the size of electoral wards in the three counties. The area of the wards ranged from 17 to hectares (mean = 2595); the childhood population, aged 16 years, ranged from 84 to 7197 (mean = 1545); the total population ranged

3 CHILDHOOD INSULIN-DEPENDENT DIABETES MELLITUS I N YORKSHIRE, UK 1309 TABLE 1 Univariate analysis of the effect of census derived variables on diabetes incidence. Population density and childhood population density are measured in people per hectare Variable Value Rate ratio 95% confidence interval Urban-rural Urban Isolation Built-up Population density Childhood population density Townsend score Components of the Townsend score Unemployment 9.2% Housing tenure 33.5% Overcrowding 2.3% Access to a car 40.5% from 498 to (mean = 6641). The mean number of cases observed in each ward during the period of the study was 1.4 (range 0 10). All of the variables, except urban-rural status, showed significant associations with the incidence of diabetes on univariate analysis (Table 1). In this Table, for ease of presentation, results are shown for the higher level of each variable when cut into two strata. The univariate associations are qualitatively similar for the continuous variables, and variables cut into two, three or five strata (data available on request). The variables tested were highly intercorrelated. Multivariate analysis suggested that the best fitting model contained two variables, childhood population density, as a two-level variable, and the level of overcrowding, as a three-level variable (Tables 2 and 3). These variables made a greater contribution to the model than the Townsend score itself. With these two variables in the model the addition of other variables gave no improvement in fit (data available upon request). There was no evidence for any interaction between these two (likelihood ratio statistic = 0.43, P 0.5). Detailed analysis of spatial variation in the residuals from this model suggested that very little remained. There was evidence for a modest amount of extra- Poisson variation, with the dispersion parameter estimated at 1.1 (P ). As adjustment for this did not affect the results in any significant way, the unadjusted results are presented. The final testing, using a full model with degrees of freedom, did not substantively alter the results (data available on request). As an exploratory analysis the final model chosen for the principal analysis was fitted separately for males and females, and for children aged 8 and 8 years (Table 4). This showed that the effect of household overcrowding was similar for males and females, and for children 8 and 8 years. The effect of childhood population density was much weaker for males, and for younger children. DISCUSSION These results provide strong evidence that, in Yorkshire, children living in areas with many overcrowded houses, and with a high population density are at significantly reduced risk of developing IDDM. Given the apparently contradictory results of earlier studies, our results need to be put in context. While it is unlikely that the observed patterns are due to chance, there are some important confounding variables which could not be measured for this study. Many studies have shown that breastfeeding is associated with a lower risk of IDDM, and increased maternal age with a higher risk. 45,47 However no small area data on breastfeeding is available, and while maternal age is recorded on UK birth certificates, it is confidential information. These and other confounders cannot be TABLE 2 Analysis of deviance for the best fitting model, that containing childhood population density and overcrowding Variable d.f. Deviance Mean P-value deviance Childhood density (people/hectare) Overcrowding (% of houses with more than one person per room) Residual deviance Total deviance

4 1310 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY TABLE 3 Effect estimates for the best fitting univariate model, that containing childhood population density and overcrowding Variable Value Rate ratio 95% confidence interval Childhood population density (people per hectare) Overcrowding (% of 1.5% 1.00 houses with more than 3.2% person per room) 3.2% excluded as explanations for the patterns which we observed. Only replications of this type of study elsewhere can really address such issues. Many studies of the variation of IDDM within countries have been carried out. Some simply showed variation, and did not try to explain it, for example studies from Scotland, England, Brazil, and Italy. 14,17,21,22 Uniform rates of IDDM within countries have been reported in an early Swedish study, and in papers from France, Finland and Japan. 25,28 30 Some studies have explicitly compared different types of area, and found little difference. Thus, Hamman et al. 27 in Colorado found little evidence of variation in incidence between five areas of the state. Verge et al., 31 in New South Wales, found similar rates when comparing urban and rural areas of the state, as did Grzwya and Sobel 32 in the province of Rzeszow in Poland. Other studies have found differences. Bruno et al. 20 found an excess of IDDM in Turin compared with the surrounding rural areas. Rewers et al. 13 in Poland found higher rates in the city of Poznan than in the surrounding rural areas. Dacou-Voutetakis et al. 24 found higher rates in metropolitan and semi-rural areas than in rural or urban areas. Crossley and Upsdell 9 found higher rates in the South Island of New Zealand than in the more urbanized North Island. Christau et al. 8 found higher rates in the south of Copenhagen, which is relatively deprived, than in the north of the city. In Norway, Joner and Søvik 15 found a lower incidence in the north of the country than in the south. Dahlquist et al. 11 observed a similar pattern of geographical variation in Sweden, but Finland does not display such a pattern. 29 The causes of these conflicting results are unclear, but may be related to variations in ascertainment, in population estimation, in definition of areal units, and random variation, as well as genuine differences. The studies discussed so far have a common weakness, which is that each has used large areas for comparison, making it less likely that characteristics of the areas represent the environment of individuals. The rationale for the choice of areas for comparison is unclear, and the designation of those areas as, for example, affluent or urban can seem arbitrary, and does not facilitate replication elsewhere. TABLE 4 Results for the best fitting model (childhood population density and overcrowding) by age and sex Variable Group Value Rate 95% ratio confidence interval Childhood population Baseline density (children aged Male ( ) 16 years per hectare) Female ( ) ( ) ( ) All ( ) Overcrowding (% of houses Baseline 1.5% 1.00 with more than one person Male 3.2% 0.82 ( ) per room 3.2% ( ) Female 3.2% 0.85 ( ) 3.2% ( ) % 0.67 ( ) 3.2% ( ) % 0.82 ( ) 3.2% ( ) All 3.2% 0.84 ( ) 3.2% ( )

5 CHILDHOOD INSULIN-DEPENDENT DIABETES MELLITUS I N YORKSHIRE, UK 1311 Two studies from Sweden have analysed IDDM incidence at a smaller scale, that of county. Nyström et al. 18 analysed diabetes incidence in the 24 counties of Sweden, and found significant variation, associated with latitude, but not with population density nor with being a coastal rather than an interior county. Dahlquist and Mustonen 23 analysed the relationship between IDDM incidence and climate at county level in Sweden and found that incidence was higher in counties with less sunshine and colder climates. Only five studies have used truly small area comparisons, i.e. using areas most of which have populations of less than people, namely the present study, and studies from Montreal, Scotland, Newcastle in England and Northern Ireland. 10,12,16,19,26 Colle et al. 10 found a lower risk in the bottom three-fifths of areas in Montreal, as classified by the average income of the households in each census tract. Waugh 12 identified a lower risk in urban areas of Tayside, Scotland. Patterson and Waugh 19 in a study covering the whole of Scotland, and including all of the cases from the earlier Scottish study 12 found a lower risk for children in deprived postcode sectors, an effect that was stronger in urban areas. However Crow et al. 16 in a study conducted in the north-east of England, which lies between Scotland and Yorkshire, found an increased incidence in poorer areas. Patterson et al. 26 in Northern Ireland, found lower risk in deprived areas, in areas of high population density, and those with a high proportion of overcrowded homes. These four UK studies used very similar measures of deprivation, but the unit of analysis used by Crow et al, 16 the enumeration district, is much smaller than those used in this study, and those of Patterson et al. 19,26 The results of Crow et al. 16 have not been confirmed in any other study of small area variation of IDDM incidence. There is no apparent explanation for this discrepancy in results. Our results are based on a diabetes register believed to be over 97% complete, with uniform completeness of ascertainment over time. 33 Our conclusion, that diabetes incidence is more strongly related to measures of population density and overcrowding than to a global deprivation score or measures of unemployment and home ownership is robust to alternative models and routine checking. It is also consistent with four 10,12,19,26 of the existing comparable studies, though not with the fifth. 15 No previous studies appear to have analysed these associations in subgroups of the overall study population, and so our finding in a subgroup analysis that population density appears to be less important for males, and for younger children is hard to assess. Pending replication or refutation elsewhere the subgroup findings should be regarded as particularly tentative. The results presented here are not consistent with studies from Italy, 20 Poland 13 and Greece 24 which have used larger units for analysis. They are consistent with results from studies in Copenhagen, 8 and in New Zealand. 9 As discussed before, these discrepancies may be due to differences in definitions, the size of the areas analysed, or to difficulties with ascertainment. Why might these small area patterns be associated with a lower incidence of IDDM? One correlate of household crowding is poverty. However, other measures of poverty, like unemployment levels and home ownership, are less strongly linked with IDDM, once population density and overcrowding are considered. Exposure to pollution is another possible factor but this cannot be measured in a study like ours. However there is little prior evidence linking IDDM with pollution. Overcrowded conditions lead to a high prevalence of many diseases, 48 but are especially strongly linked to the spread of infectious diseases. 48,49 It is not plausible that this pattern could arise from genetic variation in the population. There has been much interest in the links between IDDM and infection. 50 Many viruses have been suggested as causes of IDDM, but so far virological studies have been inconclusive. Animal models, such as NOD mice and BB rats may provide some insight The incidence of diabetes in NOD mice is variable, and is affected by many things including hormones, diet, various immunological manipulations, genetic background and sex In particular, exposure to infectious disease early in life can greatly reduce the risk of IDDM both in NOD mice 53,54 and in the BB rat. 51 Most theories linking IDDM and infection have suggested a model where a virus initiates an autoimmune process, perhaps by attacking the pancreas directly, so exposing pancreatic antigens, or by molecular mimicry. 3,50,60 Early exposure to infection may abrogate a very early stage of this process, 60 or may mature the immune system so that any transient anti-beta cell immune response that develops later is controlled. Given the wrong genetic background, already pre-set for diabetes, and either a lack of early exposure to infection, and/or late exposure to certain infections, the process of islet-cell destruction may escape from control, and diabetes will ensue. This remains entirely speculative as a mechanism for IDDM, and extrapolation of results from experiments in rats and mice to disease in humans must be tentative, 59 although there is some evidence favouring a related mechanism as part of the aetiology of childhood acute

6 1312 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY lymphoblastic leukaemia. 61 This hypothesis could be tested in a case-control study where residential histories, histories of social contact, and high quality infectious disease histories are obtained. At least one such study is already underway in Leeds in England, and the results are awaited with interest. ACKNOWLEDGEMENTS This study was supported by a grant from the former Yorkshire RHA. We thank the paediatricians, diabetologists, diabetes health visitors, specialist diabetes nurses and clinic staff for their generous co-operation. Dr A Staines, R McNally and R Cartwright are supported by the Leukaemia Research Fund. The Children s Research Fund financially supports the register. Census data are copyright Her Majesty s Government, and are made available by the ESRC. We would like to acknowledge the helpful and constructive comments of the referees, which have improved this paper. REFERENCES 1 Karvonen M, Tuomilehto J, Libman I, LaPorte R. A review of the recent epidemiological data on the worldwide incidence of Type 1 (insulin-dependent) diabetes mellitus. Diabetologia 1993; 36: Gortmaker S L, Sappenfield W. Chronic childhood disorders: Prevalence and impact. In: Haggerty R J (ed.). Symposium on Chronic Disease in Children. Pdr Clin N A 1984; 31: Drash A L, Lipton R B, Dorman J S et al. The interface between epidemiology and molecular biology in the search for the causes of insulin-dependent diabetes mellitus. Ann Med 1991; 23: Diabetes Epidemiology Research International. Geographic patterns of childhood insulin-dependent diabetes mellitus. Diabetes 1988; 37: Tuomilehto J, Podar T, Brigis G et al. Comparison of the incidence of insulin-dependent diabetes mellitus in childhood among five Baltic populations during Int J Epidemiol 1992; 21: Songini M, Loche M, Muntoni S et al. Increasing prevalence of juvenile onset type 1 (insulin-dependent) diabetes mellitus in Sardinia: the military service approach. Diabetologia 1993; 36: Mazzella M, Caratozzolo A, Cotelessa M et al. Incidence of Type 1 Diabetes in the Liguria Region, Italy. Diabetes Care 1994; 17: Christau B, Kromann H, Andersen O O et al. Incidence, seasonal and geographical patterns of juvenile-onset insulindependent diabetes mellitus in Denmark. Diabetologia 1977; 13: Crossley J R, Upsdell M. The incidence of juvenile diabetes mellitus in New Zealand. Diabetologia 1980; 18: Colle E, Siemiatycki J, West R et al. Incidence of juvenile onset diabetes in Montreal-demonstration of ethnic differences and socio-economic class differences. J Chron Dis 1981; 34: Dahlquist G, Blom L, Holmgren G et al. The epidemiology of diabetes in Swedish children 0 14 years a six-year prospective study. Diabetologia 1985; 28: Waugh N R. Insulin-dependent diabetes in a Scottish region: incidence and urban/rural differences. J Epidemiol Community Health 1986; 40: Rewers M, LaPorte R E, Walczak M, Dmochowski K, Bogaczynska E. Apparent epidemic of insulin-dependent diabetes mellitus in Midwestern Poland. Diabetes 1987; 36: Barclay R P C, Craig J O, Galloway C A S, Richardson J E, Shepherd R C, Smail P J. The incidence of childhood diabetes in certain parts of Scotland. Scot Med J 1988; 33: Joner G, Søvik O. Increasing incidence of diabetes-mellitus in Norwegian children 0 14 years of age Diabetologia 1989; 32: Crow Y J, Alberti K G, Parkin J M. Insulin dependent diabetes in childhood and material deprivation in northern England, Br Med J 1991; 303: Metcalfe M A, Baum J D. Incidence of insulin dependent diabetes in children aged under 15 years in the British Isles during Br Med J 1991; 302: Nyström L, Dahlquist G, Östman J et al. Risk of developing insulin-dependent diabetes mellitus (IDDM) before 35 years of age: Indications of climatological determinants for age at onset. Int J Epidemiol 1992; 21: Patterson C C, Waugh N R. Urban/rural and deprivational differences in incidence and clustering of childhood diabetes in Scotland. Int J Epidemiol 1992; 21: Bruno G, Pisu E, Merletti F, Ruggiorio M, Vuolo A, Pagano G. Sex differences in incidence of IDDM in age-groups years. Diabetes Care 1993; 16: Ferreira S R, Franco L J, Vivolo M A, Negrato C A, Simoes A C, Ventureli C R. Population-based incidence of IDDM in the state of Sao Paulo, Brazil. Diabetes Care 1993; 16: Cherubini V, Ravaglia E, Cantarini M, Bartolotta E. Incidence of IDDM in the Marche region, Italy. Diabetes Care 1994; 17: Dahlquist G, Mustonen L. Childhood onset diabetes-time trends and climatological factors. Int J Epidemiol 1994; 23: Dacou-Voutetakis C, Karavanaki K, Tsoka-Genatas H. National data on the epidemiology of IDDM in Greece. Diabetes Care 1995; 18: Sterky G, Holmgren G, Gustavson K H et al. The incidence of diabetes mellitus in Swedish children Acta Paed Scand 1978; 67: Patterson C C, Carson D J, Hadden D R. Epidemiology of childhood IDDM in Northern Ireland : Low incidence in areas with highest population density and most household crowding. Diabetologia 1996; 39: Hamman R F, Gay E C, Cruickshanks K J et al. Colorado IDDM Registry. Incidence and validation of IDDM in children aged 0 17 years. Diabetes Care 1990; 13: Levy-Marchal C, Papoz L, de Beaufort C et al. Incidence of juvenile type 1 (insulin-dependent) diabetes mellitus in France. Diabetologia 1990; 33: Tuomilehto J, Lounamaa R, Tuomilehto-Wolfe E et al. Epidemiology of childhood diabetes mellitus in Finland background of a nationwide study of type 1 (insulin-dependent) diabetes mellitus. Diabetologia 1992; 35:

7 CHILDHOOD INSULIN-DEPENDENT DIABETES MELLITUS I N YORKSHIRE, UK Japan IDDM Epidemiology Study Group. Lack of regional variation in IDDM risk in Japan. Diabetes Care 1993; 16: Verge C F, Silink M, Howard N J. The incidence of childhood IDDM in New South Wales, Australia. Diabetes Care 1994; 17: Grzywa M A, Sobel A K. Incidence of IDDM in the province of Rzeszow, Poland, 0 29-year-old age-group, Diabetes Care 1995; 18: Staines A, Bodansky H J, Lilley H E B, Stephenson C, McNally R J Q, Cartwright R A. The epidemiology of diabetes mellitus in the United Kingdom: The Yorkshire Regional Childhood Diabetes Register. Diabetologia 1993; 36: Rhind D (ed.). A Census User s Handbook. London: Routledge Kegan & Paul, Openshaw S (ed.). Census User s Handbook. Chichester: John Wiley & Sons, Townsend P, Phillimore P, Beattie A. Health and Deprivation. Inequality and the North. London: Croom Helm, Alexander F E, Ricketts T J, McKinney P A, Cartwright R A. Community lifestyle characteristics and risk of acute lymphoblastic leukaemia in children. Lancet 1990; 336: Craig J. An Urban-Rural Categorisation for Wards and Local Authorities. London: HMSO, Genstat 5 Committee. Genstat 5, Release 3, Reference Manual. Oxford: Clarendon Press, SAS Institute. SAS Statistics User Guide for Version Cary, IL: SAS Institute, Breslow N E, Day N E. Indirect standardization and multiplicative models for rates, with reference to the age adjustment of cancer incidence and relative frequency data. J Chron Dis 1975; 28: Bithell J F, Dutton S J, Neary N M, Vincent T J. Controlling for socioeconomic confounding using regression methods. J Epidemiol Community Health 1995; 49 (Suppl. 2): S Breslow N E. Extra-Poisson variation in log-linear models. Appl Stat 1984; 33: Atkinson A C. Plots, Transformations and Regressions. Oxford: Oxford University Press, Verge C F, Howard N J, Irwig L, Simpson J M, Mackerras D, Silink M. Environmental factors in childhood IDDM. A population-based, case-control study. Diabetes Care 1994; 17: Patterson C C, Carson D J, Hadden D R, Waugh N R, Cole S K. A case-control investigation of perinatal risk factors for childhood IDDM in Northern Ireland and Scotland. Diabetes Care 1994; 17: Dahlquist G, Blom L, Lonnberg G. The Swedish Childhood Diabetes Study a multivariate analysis of risk determinants for diabetes in different age groups. Diabetologia 1991; 34: Davey Smith G, Bartley M, Blane D. The Black Report on socioeconomic inequalities in health 10 years on. Br Med J 1990; 301: Anderson R M, May E R. Dynamics of Infectious Disease in Humans. Oxford: Oxford University Press, Szopa T M, Titchener P A, Portwood N D, Taylor K W. Diabetes mellitus due to viruses, some recent developments. Diabetologia 1993; 36: Schwimmbeck P L, Dyrberg T, Oldstone M B A. Abrogation of diabetes in BB rats by acute virus infection. J Immunol 1988; 140: Baxter A G, Koulmanda M, Mandel T E. High and low diabetes incidence nonobese diabetic (NOD) mice: origins and characterisation. Autoimmunity 1991; 9: Wilberz S, Partke H J, Dagnaes-Hansen F, Herberg L. Persistent MHV (mouse hepatitis virus) infection reduces the incidence of diabetes mellitus in non-obese diabetic mice. Diabetologia 1991; 34: Takei I, Asaba Y, Kasatani T et al. Suppression of development of diabetes in NOD mice by lactate dehydrogenase virus infection. J Autoimmunity 1992; 5: Hawkins T, Gala R R, Dunbar J C. The effect of neonatal sex hormone manipulation on the incidence of diabetes in nonobese diabetic mice. Proc Soc Exp Biol Med 1993; 202: Nomura Y, Stein E, Mullen Y. Prevention of overt diabetes and insulitis by intrathymic injection of syngeneic islets in newborn nonobese diabetic (NOD) mice. Transplantation 1993; 56: Beales P E, Williams A J, Albertini M C, Pozzilli P. Vitamin E delays diabetes onset in the non-obese diabetic mouse. Horm Metab Res 1994; 26: Hanafusa T, Miyagawa J, Nakajima H et al. The NOD mouse. Diabetes Res Clin Pract 1994; 24 (Suppl.): S Rossini A A, Handler E S, Mordes J P, Greiner D L. Human autoimmune diabetes mellitus: lessons from BB rats and NOD mice Caveat emptor. Clin Immunol Immunopathol 1995; 74: Atkinson M A, Maclaren N K. The pathogenesis of insulindependent diabetes mellitus. New Engl J Med 1994; 331: Greaves M F, Alexander F E. An infectious aetiology for common acute lymphoblastic leukaemia in childhood? Leukemia 1993; 7: (Revised version received February 1997)