IJE Advance Access originally published online on August 31, 2006
International Journal of Epidemiology 2006 35(6):1504-1513; doi:10.1093/ije/dyl193
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Article |
Childhood leukaemia and socioeconomic status: fact or artefact? A report from the United Kingdom childhood cancer study (UKCCS)
1 Epidemiology and Genetics Unit, Department of Health Sciences, University of York, York YO10 5DD, UK.
2 Academic Centre for Defence Mental Health, Institute of Psychiatry, King's College London, Weston Education Centre, Cutcombe Road, London, SE5 9RJ, UK.
3 Academic Unit of Paediatric Oncology, Christie Hospital and Central Manchester and Manchester Children's University Hospitals NHS Trusts, Manchester, M20 4BX, UK.
* Corresponding author. E-mail: alex.smith{at}egu.york.ac.uk
 |
Abstract |
|---|
 Top Abstract Materials and methodsResultsDiscussionReferences |
|---|
Background It is widely believed that children of high socioeconomic status (SES) are more likely than those of low SES to develop acute lymphoblastic leukaemia (ALL). Such observations have led to wide-ranging speculations about the potential aetiological role of factors associated with affluence and modernization.
Methods Children (0–14 years) newly diagnosed with cancer in the UK between 1991 and 1996 were ascertained via a rapid hospital-based case finding system (n = 4430, of which 1578 were ALL). Children without cancer (controls) were randomly selected from primary care population registries for comparative purposes (n = 7763). Area-based deprivation scores were assigned as markers of SES at two time points—birth and diagnosis. An individual-based marker of SES—social class—was assigned using father's occupation as recorded on the child's birth certificate.
Results No differences in area-based measures of deprivation were observed between cases and controls at time of diagnosis, either for all cancers combined [n = 4430, odds ratio (OR) = 1.00 (95% confidence intervals (CI) 0.98–1.01)] or for ALL alone (n = 1578 OR = 0.99, 95%CI 0.96–1.01). Findings were similar at time of birth (all cancers, OR = 0.99 95%CI 0.98–1.01, ALL OR = 0.98, 95%CI 0.96–1.00). In addition, no case-control differences were observed when an individual-based measure of SES—social class—based on father's occupation at time of birth was used.
Conclusions The comprehensive nature of the data, coupled with complete case-ascertainment and representative population-based controls suggests that SES in the UK is not a determinant of ALL in children. We believe the small effects reported for SES in some past studies may be artefactual.
Keywords bias, childhood cancer, childhood leukaemia, epidemiology, socioeconomic status
Accepted 27 July 2006
Historical data from resource-rich countries suggest that children of high socioeconomic status (SES) may be more likely than those of low SES to develop acute lymphoblastic leukaemia (ALL) in childhood. Although the evidence is far from conclusive (reviewed in ref. 1,2), such associations have been evoked in support of a wide-range of potential aetiological factors linked with affluence and modernization. This includes increased maternal age at child-bearing, improved levels of hygiene, and increased exposures in the home to power-frequency magnetic fields from electrical appliances.3,4
Positive associations between high SES and ALL in children have been reported in some ecological studies.5–8 Such analyses generally cover comparatively long periods of time, the numerator usually being derived from cancer registrations and the denominator from census estimates—inaccuracies in either potentially producing spurious results. In contrast, most case–control studies have reported either no association1,9–12 or the reverse—the lack of effect being ascribed by some to biased control recruitment.13–15 It has also been suggested that since earlier studies (reviewed in ref. 16) were more likely to produce positive correlations and later ones the inverse, that risk factors have changed over time.8 It is unclear, however, whether this change over time is real or is an artefact of the bias caused by different study designs.
Data collected from a national case–control study are used in this report to investigate the potential impact of various sources of bias on the association between SES and ALL. The comprehensive nature of the data, coupled with complete case-ascertainment and representative population-based controls, permits a more thorough investigation of this topic than has been possible in the past.
|
Materials and methods |
|---|
|
TopAbstractMaterials and methodsResultsDiscussionReferences |
|---|
The United Kingdom Childhood Cancer Study (UKCCS) is a national population-based case–control study, and information about its conduct and ethical approvals are described in detail elsewhere.3 Briefly, for the purposes of study management, 10 UKCCS administrative areas were created; the conduct of the study within each area being the responsibility of a single epidemiological centre (Fig. 1). Proactive notification systems were set up in all hospitals treating children with cancer throughout England, Wales, and Scotland. Children 0–14 years diagnosed in 1991–96 were eligible, although the specific malignancy groups targeted varied over the years and between UKCCS administrative areas. In Scotland, this was from January 1991 to December 1994 and in England & Wales between September 1992 and December 1994 for all cancers including lymphomas and leukaemias, lymphomas and leukaemias only in 1995, and leukaemias only in 1996.
|
For comparative purposes, children without cancer (two per case) matched on sex, month and year of birth and UKCCS area of residence at diagnosis were randomly selected and recruited from population registries held by the (former) Family Health Service Authorities (FHSA) (England & Wales) and Health Boards (HB) (Scotland). At the time of the study, there were 98 FHSAs and 15 HBs responsible for primary national health care organization and at least 98% of the population was registered on these lists.17 On average, each FHSA register contained details of 650 311 people, the smallest being the Scottish HB Orkney (19 644) and the largest the English FHSA of Hampshire (1 539 831); the average number of children (0–14 years) per FHSA was 122 278, ranging from 3959 to 291 218. The first two control families whose GP gave permission to approach them are used as the comparison group here; this is regardless of whether or not their families agreed to be interviewed.
Each FHSA/HB comprised, on average, 2206 census enumeration areas (range 147–7055). Employing methods described elsewhere, data from the 1991 UK census were used to assign a deprivation score to each enumeration area—151 719 in total.3 Briefly, for each census area—which contains aggregated data on about 360 people (200 households) of which 70 are between the ages of 0 and 14—the proportion of households without a car, the number of overcrowded households, and the number of persons unemployed was calculated. Deprivation categories were produced by dividing the continuous deprivation score (range –6.15 to +7.75) for all enumeration areas into five equally sized groups, where group one represented the most affluent and group five the most deprived.
Birth certificates for all eligible cases (4430) and controls (7663) were obtained from the Office for National Statistics (ONS). The National Health Service Central Register (NHSCR) provided copies of birth certificates for births occurring in England and Wales (87%) and the General Register Office (GRO) provided details in an electronic format for those registered in Scotland (13%). Birth certificates not only provide details about the child, i.e. name, sex, date of birth, and birth weight, but also mother's usual address, and father's employment status and occupation at the time of the birth. Occupation was coded using the UK 1990 Standard Occupational Classification Scheme,18 and a measure of approximate social class derived using the Registrar General's social class classification scheme. These SES proxies, together with the deprivation score assigned to the address at the time of diagnosis (cases)/pseudo-diagnosis (controls),3 form the basis of the report.
In order to aid interpretation of any patterns that might emerge, four disease groups are considered—ALL alone, all leukaemias combined, all non-leukaemias, and all cancers. In the analyses presented, all cases are included regardless of whether their parents' were interviewed (3835) or not (595). While 3835 interviewed cases had two ‘first-choice’ controls selected, seven had only one, yielding 7663 in total—controls were not selected for the 595 non-interviewed cases. The 7663 nationally representative randomly selected controls form the basis of the comparisons made here—5526 had parents who were interviewed and 2137 had parents who were not. As in previous UKCCS publications19–21 in order to increase precision and statistical power all available controls were used as the comparison group in the main analyses. Accordingly, odds ratios and 95% confidence intervals were estimated using unconditional logistic regression22 with adjustment for UKCCS study area (n = 10), sex, and age at diagnosis. To check for consistency, risk estimates were also calculated using conditional logistic regression. All analyses were performed in Stata.23
|
Results |
|---|
|
TopAbstractMaterials and methodsResultsDiscussionReferences |
|---|
Table 1 shows that at the time of diagnosis, the numbers of control families in each deprivation category was roughly equal (
20%) suggesting that, in terms of deprivation, the sample was broadly representative of Britain as a whole. Importantly, no differences between cases and controls are evident, either for all cancers combined [n = 4430, odds ratio (OR) = 1.00 (95% confidence intervals (CI) 0.98–1.01)] or for ALL in particular (n = 1578, OR = 0.99, 95%CI 0.96–1.01).
|
At the time of the child's birth, however, with comparatively more families living in less affluent areas, both control and case distributions appear less typical of the population as a whole. This reflects the fact that around half of the families moved after the birth of their child—generally to a more affluent area. Indeed, children living in areas classified as most deprived (category five) were twice as likely to move as those living in the most affluent areas (category one) (Table 2). Once more, however, no case–control differences with respect to deprivation are evident.
|
At both time-points (birth and diagnosis), although the 595 non-interviewed cases tended to live in more deprived areas, the results were broadly similar when risk estimates were calculated using conditional logistic regression and the analysis was restricted to interviewed cases and their individually matched controls (Table 3). Furthermore, no impact on the findings at the time of diagnosis was observed when children whose GP refused contact (601 of 7663) were included instead of their replacements (Table 4). Unfortunately, it was not possible to repeat the time at birth analysis including children whose GP refused permission to contact, as no information on residence at birth was available.
|
|
Importantly, as expected for both cases (4430) and controls (7663), a positive correlation between the area-based measure of SES (deprivation score) and the individual-based measure of SES (social class assigned on the basis of father's occupation as recorded at birth certification) was observed (Spearman rank correlation +0.3, P < 0.0001) (Fig. 2). Again, use of this individual-based measure revealed no case–control differences (
2 = 1.6, P = 0.4 for all cancers combined, and 2 = 0.3, P = 0.8 for ALL) (Table 5).
|
|
With the exception of Table 3, which excludes non-interviewed cases, the results described so far have been based on all cases and the first two randomly selected ‘first-choice’ controls—regardless of whether or not their families agreed to participate. As in many case–control studies, when control parents refused a further family was selected and so on until two control families were recruited. In total 3835 case families (87% of all diagnoses) and 7621 control families (72% of the first two selected) were interviewed. When the analysis was restricted to participating (interviewed) families a positive association with SES emerged (Table 6).
|
|
Discussion |
|---|
|
TopAbstractMaterials and methodsResultsDiscussionReferences |
|---|
We found no evidence to support the suggestion that SES, and by implication the factors that underpin its assignment, is a determinant of childhood leukaemia. The completeness and representativeness of the UKCCS data is a major strength of our analyses.3 With respect to cases, unlike many studies, we were not reliant on secondary sources for ascertainment. Indeed, in the years 1993–1994, when all childhood malignancies diagnosed in England, Wales, and Scotland were actively targeted, the number identified (all cancers = 2650, of which 722 were ALL) was
10% higher than predicted on the basis of cancer registration.24 There are two possible areas relating to control data omission that could potentially have influenced the findings presented here. The first relates to the lack of controls for cases who were not interviewed (595 of the 4430), and the second relates to the omission of control families whose GPs did not give permission for them to be approached (601 of the 7663). With respect to the former, non-interviewed cases were scattered across all 10 UKCCS administrative areas. Although within these areas non-interviewed cases tended to live in marginally more deprived neighbourhoods than interviewed cases, there is no reason to suspect that the SES distribution of their controls—had they been selected—would have differed from the overall control distribution. This is because, the population-based sampling frames for control selection were large, each being around 122 278 children (650 300 people), ensuring that overmatching on SES was unlikely. With respect to the latter, the reasons why GPs declined are complex and were not necessarily related to the SES of the family. Accordingly, when the analyses were repeated including these children in the comparison group rather than the children who replaced them, there was no impact on the findings at the time of diagnosis (Table 4), and it seems improbable that they would have affected the findings at birth.
Incomplete case ascertainment is also important, although it is generally considered less often. In our data, broadly similar results were seen for the unmatched analyses which included all cases and first choice controls (Table 1), and the matched analyses that was restricted to interviewed cases and their individually matched first choice controls (Table 3). Nonetheless, the exclusion of the 595 non-interviewed cases lowered the estimates in the most deprived categories—across all diagnostic groups, not just ALL. Indeed, although the effect is small, it is possible that systematic SES variations in case ascertainment could contribute to the pattern of results seen in some ecological studies,5–8 leading to the erroneous conclusion that childhood leukaemia is less common among the poor.
Although not a major factor in the findings presented here, the effects of missing information can be considerable in epidemiological studies examining SES proxies. In general, the lower the SES the more likely it is that data will be missed. This may be more apparent in case–control studies, where participation is often differential—participation rates among controls falling more steeply with SES than participation rates among cases.25,26 In our data, for example, when the analysis was restricted to subjects who agreed to be interviewed odds ratios appeared to be higher among the less affluent. This is because less affluent control families were more likely to refuse to be interviewed. This bias, which can in turn result in artefactual correlations between disease and the wide range of exposures linked to SES, bedevils the interpretation of data from many case–control studies.
Our study is unusual in that we were also able to look across the country as a whole at SES markers at two time points in the child's life—birth and diagnosis. In our data, families from more deprived areas were more likely to have moved between birth and diagnosis—generally to more affluent areas. This change not only illustrates how mobile families are, but also illustrates the transient nature of area-based measures of SES, emphasizing the importance of measuring area-based scores of SES at the same time points for cases and the comparison group.
Most studies that have investigated SES and childhood leukaemia have only used one indicator of SES—either based on the individual12,13(occupation, education, etc.) or on the area in which they live10 (deprivation urban/rural etc.). We had both measures available at birth where, reassuringly, the correlation between paternal occupation-based social class and area-based deprivation at birth was good. It is important to note that since occupation was recorded at birth registration, the data are relatively complete and are not subject to errors introduced by recall bias. Unfortunately, these analyses could not be repeated at diagnosis since occupational information was not available for non-responders.
In summary, we believe that the small effects reported for SES in some past studies of childhood leukaemia may be artefactual. Our findings, which are based on the analyses of data with complete UK coverage, suggest that SES in the UK is not a determinant of childhood ALL.
KEY MESSAGES
|
|
Acknowledgments |
|---|
The United Kingdom Childhood Cancer Study is sponsored and administered by the Leukaemia Research Fund. This study was conducted by 12 teams of investigators (10 clinical and epidemiological and 2 biological) based in university departments, research institutes, and the National Health Service in Scotland. The work was coordinated by a Management Committee. It was supported by the UK Children's Cancer Study Group of paediatric oncologists and by the National Radiological Protection Board. We thank the members of the UK Childhood Cancer Study Group for their support and staff of local hospitals, general practitioners, general practice staff, and UKCCS interviewers and technicians. We especially thank the families of the children included in the study, without whom this investigation would not have been possible.
|
References |
|---|
|
TopAbstractMaterials and methodsResultsDiscussionReferences |
|---|
1 Raaschou-Nielsen O, Obel J, Dalton S, Tjonneland A, Hansen J. Socioeconomic status and risk of childhood leukaemia in Denmark. Scand J Public Health 2004;32:279–86.[CrossRef][ISI][Medline]
2 Poole C, Greenland S, Luetters C, Kelsey JL, Mezei G. Socioeconomic status and childhood leukaemia: a review. Int J Epidemiol 2006;35:370–84.
3 UK Childhood Cancer Study Investigators. The United Kingdom Childhood Cancer Study: objectives, materials and methods. UK Childhood Cancer Study Investigators. Br J Cancer 2000;82:1073–102.[CrossRef][ISI][Medline]
4 Little J. Epidemiology of Childhood Cancer. IARC Scientific Publications, 1999.
5 McWhirter WR. The relationship of incidence of childhood lymphoblastic leukaemia to social class. Br J Cancer 1982;46:640–45.[ISI][Medline]
6 Alexander FE, Ricketts TJ, McKinney PA, Cartwright RA. Community lifestyle characteristics and risk of acute lymphoblastic leukaemia in children. Lancet 1990;336:1461–65.[CrossRef][ISI][Medline]
7 Draper GJ, Vincent TJ, O'Connor CM, Stiller CA. Socio-economic factors and variations in incidence rates between County Districts. The geographical epidemiology of childhood leukaemia and non-Hodgkin lymphomas in Great Britain, 1966–83, 1991, pp. 37–45.
8 Borugian MJ, Spinelli JJ, Mezei G, Wilkins R, Abanto Z, McBride ML. Childhood leukemia and socioeconomic status in Canada. Epidemiology 2005;16:526–31.[CrossRef][ISI][Medline]
9 Stewart A, Webb J, Hewitt D. A survey of childhood malignancies. Br Med J 1958;30:1495–507.
10 Dockerty JD, Draper G, Vincent T, Rowan SD, Bunch KJ. Case-control study of parental age, parity and socioeconomic level in relation to childhood cancers. Int J Epidemiol 2001;30:1428–37.
11 Shaw G, Lavey R, Jackson R, Austin D. Association of childhood leukemia with maternal age, birth order, and paternal occupation. A case-control study. Am J Epidemiol 1984;119:788–95.
12 Steensel-Moll HA, Valkenburg HA, Vandenbroucke JP, van Zanen GE. Are maternal fertility problems related to childhood leukaemia? Int J Epidemiol 1985;14:555–59.
13 Dockerty JD, Skegg DCG, Elwood JM, Herbison GP, Becroft DMO, Lewis ME. Infections, vaccinations, and the risk of childhood leukaemia. Br J Cancer 1999;80:1483–89.[CrossRef][ISI][Medline]
14 Ross JA, Potter JD, Shu XO, Reaman GH, Lampkin B, Robison LL. Evaluating the relationships among maternal reproductive history, birth characteristics, and infant leukemia: a report from the Children's Cancer Group. Ann Epidemiol 1997;7:172–79.[CrossRef][ISI][Medline]
15 Schuz J, Grigat JP, Brinkmann K, Michaelis J. Residential magnetic fields as a risk factor for childhood acute leukaemia: results from a German population-based case-control study. Int J Cancer 2001;91:728–35.[CrossRef][ISI][Medline]
16 Greenberg RS, Shuster JL, Jr. Epidemiology of cancer in children. Epidemiol Rev 1985;7:22–48.
17 RCGP. Profile of UK Practices, Information Sheet No 2. London: Royal College of General Practitioners, 1999.
18 Office of Population Censuses and Surveys. Standard Occupational Classification Structure and Definition of Major, Minor and Unit Groups. London: HMSO, 1990.
19 Beral V, Fear NT, Alexander F, Appleby P. UK Childhood Cancer Study Investigators. Breastfeeding and childhood cancer. Br J Cancer 2001;85:1685–94.[CrossRef][ISI][Medline]
20 Gilham C, Peto J, Simpson J et al. Day care in infancy and risk of childhood acute lymphoblastic leukaemia: findings from UK case-control study. Br Med J 2005;330:1294–99.
21 Roman E, Simpson J, Ansell P, Lightfoot T, Mitchell C, Eden TO. Perinatal and reproductive factors: a report on haematological malignancies from the UKCCS. Eur J Cancer 2005;41:749–59.[CrossRef][ISI][Medline]
22 Breslow NE, Day NE. Statistical Methods in Cancer Research. Volume 1—The Analysis of Case-Control Studies. Lyon: International Agency for Research on Cancer, 1980.
23 StataCorp. Stata Statistical Software: Release 8.2. College Station, Texas: Stata Corporation, 2003.
24 Parkin DM, Kramarova E, Draper GJ et al. International Incidence of Childhood Cancer, Vol. II. IARC Scientific Publications No. 144, 1998.
25 Hartge P. Raising response rates: getting to yes. Epidemiology 1999;10:105–07.[ISI][Medline]
26 Law GR, Smith AG, Roman E. The importance of full participation: lessons from a national case-control study. Br J Cancer 2002;86:350–55.[CrossRef][ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  A. M. Hughes, S. Crouch, T. Lightfoot, P. Ansell, J. Simpson, and E. Roman Eczema, Birth Order, and Infection Am. J. Epidemiol., May 15, 2008; 167(10): 1182 - 1187. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. J Borugian, J. J Spinelli, G. Mezei, R. Wilkins, and M. McBride Childhood leukaemia and socioeconomic status. Int. J. Epidemiol., October 1, 2007; 36(5): 1155 - 1155. [Full Text] [PDF]
|
|
|
|
|
|
A. Smith, E. Roman, and J. Simpson Childhood leukaemia and socioeconomic status Int. J. Epidemiol., October 1, 2007; 36(5): 1156 - 1156. [Full Text] [PDF]
|
|
|
|
|
|
M. J Borugian, J. J Spinelli, G. Mezei, R. Wilkins, and M. McBride Childhood leukaemia and socioeconomic status Int. J. Epidemiol., August 1, 2007; 36(4): 929 - 929. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||