Childhood leukemia and proximity to nuclear power plants: A systematic review and meta-analysis

Journal of Cancer Policy 6 (2015) 44–56

Contents lists available at ScienceDirect

Journal of Cancer Policy journal homepage: www.elsevier.com/locate/jcpo

Childhood leukemia and proximity to nuclear power plants: A systematic review and meta-analysis William Mueller ∗ , Clare Gilham London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, United Kingdom

a r t i c l e

i n f o

Article history: Received 2 September 2015 Accepted 18 October 2015 Available online 24 October 2015 Keywords: Childhood leukemia Nuclear power Meta-analysis Ionizing radiation Epidemiology

a b s t r a c t Objective: To evaluate the association between childhood leukemia and residential proximity to nuclear power plants (NPP). Methods: We performed a systematic review by searching the MEDLINE database for published studies of childhood leukemia incidence and proximity to NPP. The primary analysis included children <15 years of age living within 25 km of a NPP, and the secondary analysis focused exposure of children <5 years of age living within 5 km of such facilities. Results: A meta-analysis including eight studies (1,665 cases) of childhood leukemia within 25 km of NPPs produced a pooled estimate of 1.00 (95% confidence interval (CI) = 0.95–1.05). A secondary analysis of a subset of three case-control studies (53 cases) examining the risk in children <5 years of age within 5 km of a NPP produced a meta-estimate of 1.45 (95% CI = 0.74–2.86), and an analysis of the same parameters using four studies (76 cases) from ecological/cohort studies generated a significantly elevated pooled estimate of 1.33 (95% CI = 1.05–1.68). Conclusion: Meta-estimates for ecological/cohort and case-control studies did not provide evidence of an increase in leukemia incidence in children <15 years of age living <25 km of a NPP. A subset of studies including children <5 years of age living <5 km from a NPP produced significantly elevated estimates for ecological/cohort studies. Continuing to undertake large-scale studies of populations surrounding nuclear facilities is encouraged to refine potential risks and better understand methodological nuances. © 2015 Elsevier Ltd. All rights reserved.

Contents 1. 2.

3.

4.

5.

Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 2.1. Study identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 2.2. Statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.1. Studies included . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.2. Study characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.3. Primary statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 3.4. Secondary statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.1. Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.2. Potentially influential factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.3. Comparison with other findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4.4. Strengths and limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4.5. Alternative explanations for observed results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

∗ Corresponding author at: 411-308 Palmerston Ave, Toronto, Ontario M6J 3X9, Canada. E-mail address: [email protected] (W. Mueller). http://dx.doi.org/10.1016/j.jcpo.2015.10.003 2213-5383/© 2015 Elsevier Ltd. All rights reserved.

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

45

Funding source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Appendix A. List of excluded studies with rationale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Appendix B. A comparison of the quality of eligible studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

1. Background Childhood leukemia is the most common type of childhood cancer, representing more than a third of all cancer diagnoses in children below 15 years of age [1]. This particular cancer is unique both in its short latency period and average incident peak early in life, around two to four years of age [2]. Genetic links, such as Down’s syndrome, neurofibromatosis, and Fanconi’s anemia, have been identified as risk factors for leukemia, but are believed to be responsible for less than 10% of all leukemia cases [3]. Studies in various populations, including atomic bomb survivors, nuclear power workers, and children exposed to nuclear fallout, have convincingly indicated increased leukemia risks from exposures to low levels of radiation [4]. Recently, Kaatsch [5,6] conducted a case-control study (known as the Kinderkrebs in der Umgebung von Kernkraftwerken or, more commonly as, the KiKK study) in Germany and noted a higher than expected risk of leukemia in children less than five years of age living within five kilometres of nuclear power plants (NPPs). Such findings have renewed concerns and invited research efforts into better understanding the nature of a possible link between childhood leukemia and residential proximity to a NPP. Yet subsequent investigations have been unable to replicate such results: Bithell [7] followed a similar protocol of children five years and under living within five kilometres of a NPP and were not able to identify a positive association. In light of recent conflicting results, numerous researchers have encouraged the pooling of international data on childhood leukemia cases in close proximity to NPPs to increase statistical power [8,9]. Moreover, Muirhead [10] recommended a systematic review to be undertaken to help understand strengths and limitations of each study in hopes to better understand the issue globally. Although a meta-analysis [11] was conducted relatively recently, all of the studies used to generate meta-rates were published pre2000, thus excluding updated findings. Interest in nuclear power may have tempered since the Fukushima Daiichi meltdown in 2011, but there are still 435 nuclear reactors currently generating electricity in 30 countries, with 72 new nuclear plants presently under construction in 15 countries [12]. Gaining a better understanding of a potential association between nuclear power and childhood leukemia would be important to protect the health of populations currently living within close proximity to such facilities and also in the consideration of siting new NPPs. We have conducted a systematic review to identify all analytical studies investigating incidence of childhood leukemia and proximity to NPPs, as well as performed a meta-analysis to assess any such association.

2. Materials and methods

mation for a systematic review. One of us1 (WM) undertook the review by first searching the MEDLINE database for studies published in peer-reviewed journals as of July 31, 2015. Broad search terms were used to maximize search results: (childhood leukemia) AND (nuclear power). Once all search results were compiled and assessed, we examined the reference list of selected studies, as well as a meta-analysis [11] and three reviews [9,17,18] on this topic identified during the search, for any additional publications that satisfied all eligibility criteria. The following inclusion criteria were established and applied to all results from the MEDLINE search: • Study must be original research of an ecological, cohort or casecontrol study design; • Primary outcome must be incidence of childhood leukemia, including at least one group of children <15 years of age; • Exposure must be residential proximity to NPPs (i.e. electricitygenerating facilities), either at diagnosis or birth, and it must be made clear where cases reside; • Study estimates must include Standardized Incidence Ratios (SIRs) for ecological and cohort studies or Odds Ratios (ORs) for case-control studies for either individual NPP sites or grouping of sites; • If study estimates are not provided, necessary information to calculate estimates and confidence intervals (CIs) must be available; and • Published in a peer-reviewed journal in English. First, we examined titles and abstracts, and, only if deemed relevant, undertook a review of the full text. All eligible studies were assigned a unique identification code. Fig. 1 provides the numbers of studies identified at each stage of the search, as well as reasons for excluding studies. Risk estimates were recorded (or calculated) separately, where permitted, for various exposure and age groups. Estimates were calculated for exposed children living at intervals of 5 km up to 25 km from a NPP at either time of diagnosis or birth. To be clear, these are arbitrary intervals [19]: a distance of 5 km was selected to correspond with Kaatsch [5,6], which spurred several studies of similar design in other countries [7,20,21]. Proximity at less than 25 km was used as a surrogate for “exposed”, for comparability across published research on this subject. Tables 1 and 2 present information extracted from each of the eligible studies. Preference was given to defining exposure by residence at birth, since residence at birth would better account for higher sensitivities of the fetus to radiation and provide time for latency periods of leukemia, compared to residence at diagnosis [8]. The primary analysis defined exposure as children <15 years living within 25 km of a NPP and the secondary analysis focused on the most sensitive exposure and age [9]: children <5 years living within 5 km. Lastly, while the focus of the current study is on exposure to NPPs specifically, estimates were included for proximity to nuclear reprocessing sites as a sensitivity analysis; studies of such sites have identified childhood leukemia excesses, so understanding the effect of these sites

2.1. Study identification We followed PRISMA [13,14] and MOOSE [15] guidelines, as well as the Cochrane Handbook [16], for performing and recording infor-

1 Although published guidance on systematic reviews recommends multiple authors to be involved in the review, using one author was necessary to satisfy academic requirements. CG verified the included and excluded studies.

46

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

Studies Idenfied by MEDLINE (n=95) Excluded (n=75): • Reviews of Various Topics (n=35) • Descripve and other Study Types (n=12) • Editorials/Comments/Leers (n=10) • Exposure not Distance to NPP (n=9) • Not Relevant (n=7) • Non-English (n=1) • Outcome other than Childhood Leukemia (n=1)

Selected for Further Review (n=20) Idenfied from Review of References (n=11) Excluded (n=17) • Overlap with More Recent Studies (n=11) • Distance not an Exposure (n=3) • Mortality Data only or >15 years of age (n=3) Included for Analysis (n=14) Fig. 1. Studies included and excluded through the systematic review.

Table 1 Characteristics of the eligible case-control studies. Author

Setting & # of facilities Study period Age group Exposed group Reference group Leukemia type

Heinävaara, 2010 [30] Finland—2 NPPs France—19 NPPs Sermage-Faure, 2012 [20]

1977–2004 2002–2007

Kaatsch, 2008 [6]

Germany—16 NPPs

1980–2003

<15 years <5 years <15 years <15 years <5 years

Bithell, 2013 [7]

UK—13 NPPs

1962–2007

<5 years

5–10 km <5 km <5 km <15 km <5 km <10 km <5 km <10 km <25 km

on the pooled estimate would be useful to test the robustness of any detected association.

>30 km >20 km

>5 km >10 km >25 km

For all ecological and cohort studies, SIRs were recorded for the aforementioned exposure, age, and nuclear power facility categories of individual studies. The SIR in ecological studies compares the incidence of childhood leukemia within a specified distance of a NPP over a given time period with that of the national level or a smaller, specified geographic area over the same period. While ecological studies compare individual data with national rates, true cohort studies require the comparison of individual data for both exposed and unexposed groups. For case-control studies, the OR was recorded separately for the same categories. The OR indicates the ratio of leukemia cases to controls either born or diagnosed within a given proximity to a NPP to the ratio of such groups outside the given proximity; again, this occurs over a given time period. Risk estimates were calculated for all possible combinations as permitted by the data, i.e. location and age of study population, and type of nuclear power facility. The standard error was either recorded for each estimate or calculated from given CIs. All standard errors were calculated for both case-control and ecological/cohort studies by the following formula

1 6 14 58 All leukemia 37 95 All leukemia and NHL 10 56 445

0.71 (0.05–10.25) 1.60 (0.66–3.87) 1.90 (1.05–3.45) 1.03 (0.79–1.35) 2.19 (1.51–3.18) 1.33 (1.06–1.67) 0.72 (0.32–1.61) 1.02 (0.70–1.48 1.06 (0.92–1.22)

included in the Cochrane Handbook [16]. This calculation produces the standard error of the natural logarithm of the SIR and OR. s.e.

2.2. Statistical analysis

Total # of cases OR (95% CI)

All leukemia Acute Leukemia

[ln (upper 95% CI) − ln (lower 95% CI)] 3.92

(1)

If only the expected and observed values were presented with the SIR in ecological studies, the equations below were utilized to generate lower and upper 95% CIs [22]: √ 2 #observed − (1.96 × 0.5) Lower 95% CI = (2) #expected √ 2 #observed + (1.96 × 0.5) Upper 95% CI = (3) #expected To generate 95% CIs for ORs in case-control studies, error factors were calculated based on the following formulae [23]:



Lower 95% CI = exp

ln(OR) − 1.96

 Upper 95% CI = exp





ln(OR) + 1.96

1 1 1 1 + + + a c b d

1 1 1 1 + + + a c b d







(4)



(5)

Meta-analyses were performed for the primary and secondary analyses. As childhood leukemia is a relatively rare disease, the

Table 2 Characteristics of the eligible ecological/cohort studies. Setting & # of facilities

Study period

Age group

Exposed group

Reference group

Leukemia type

Total # of cases

SIR (95% CI)

Lane, 2013 [37]

Canada—3 NPPs

1990–2008

<5 years

<25 km

Ontario incidence rates

All leukemia and NHL

165

0.88 (0.76–1.03)

Bithell, 1994 [41]

England and Wales—8 NPPs

1966–1987

<15 years <15 years

<25 km

UK National Registry of Childhood Cancers

All leukemia and NHL

393 480

1.01 (0.91–1.12) 0.98 (0.89–1.07)

Heinävaara, 2010 [30]

8 NPPs + 1 reprocessor Finland—3 NPPs

1980–2000

<15 years

5–15 km

All leukemia

504 16

0.99 (0.91–1.08) 1.01 (0.60–1.70)

1990–2007 for cohort.

<5 years

<5 km

Finnish national rates French National Registry of Childhood Hematopoietic Malignancies

14

1.37 (0.81–2.32)

131 24 272 4

0.98 (0.78–1.23) 1.14 (0.73–1.78) 1.05 (0.90–1.22) 2.07 (0.71–6.04)

Sermage-Faure, 2012 [19]

France—19 NPPs

Guizard, 2001 [42]

France—La Hague (nuclear reprocessor)

1978–1998

<15 years

<15 km <5 km <15 km <10 km

Kaatsch, 2008 [5]

Germany—16 NPPs

1980–2003

<5 years

<20 km <5 km

Michaelis, 1992 [45]

Germany—18 NPPs + 2 research reactors

<5 years

<10 km <5 km

<15 years

<15 years Black, 1994 [31]

Scotland—1 nuclear reprocessor

1968–1991

<15 years

Sharp, 1996 [43]

Scotland—3 NPPs, 1 nuclear reprocessor

1968–1993

<15 years

Spycher, 2011 [20]

Switzerland—4 NPPs

1985–2009

<5 years

<16 years COMARE, 2011 [18]

Ma, 2011 [44]

UK—13 NPPs

US (Illinois)—7 NPPs

1969–2004

1986–2005

<5 years

<15 years

<15 km <5 km <15 km <12.5 km Enumeration districts with population centroid <25 km <5 km <15 km <5 km <15 km <5 km <10 km <25 km <10 miles (∼16 km)

Acute Leukemia

French national rates

Acute Leukemia only for <5 years, all leukemia for <15 years.

German national incidence rates

All leukemia

12 34

0.85 (0.51–1.41) 1.41 (0.99–2.00)

German national incidence rates

Acute leukemia

95 19

1.09 (0.89–1.33) 3.01 (1.25–10.31)

152 30 274 6

1.28(0.99–1.69) 1.44 (0.81–2.79 1.06 (0.88–1.28) 3.29 (1.41–7.67)

75

0.96 (0.76–1.21)

8

1.20a (0.60–2.41)

51 12 84 20

0.92a (0.69–1.23) 1.05a (0.59–1.86) 0.89a (0.71–1.11) 1.22 (0.77–1.94)

61 430 71

0.86 (0.32-2.34) 0.93 (0.84-1.03) 1.08 (0.86-1.36)

Scottish national rates Scottish national rates

All leukemia and NHL All leukemia and NHL

All cases >15 km

All leukemia

UK national average

Illinois state rates

All leukemia and NHL

All leukemia

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

Author

47

48

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

% Weight Authors

ES (95% CI)

Spycher 2011

0.89 (0.71, 1.11) 5.95

Sharp et al 1996

0.90 (0.70, 1.15) 4.87

(I-V)

Bithell et al 1994

0.98 (0.89, 1.07) 35.00

Heinavaara et al 2010

1.01 (0.60, 1.70) 1.10

Lane et al 2013

1.01 (0.91, 1.12) 26.52

Sermage-Faure et al 2012

1.05 (0.90, 1.22) 12.71

Ma et al 2011

1.08 (0.86, 1.36) 5.46

Michaelis et al 1992

1.06 (0.88, 1.28) 8.39

I-V Overall (I-squared = 0.0%, p = 0.872)

1.00 (0.95, 1.05) 100.00

D+L Overall

1.00 (0.95, 1.05)

.1

1

10

Fig. 2. Forest plot of leukemia cases <15 years of age and address within 25 km of NPP, displaying both fixed effects and random effects outputs on a log scale.

OR approximates relative risks [24]; therefore, ecological/cohort and case-control study estimates were combined in meta-analyses where there was no duplication of data. The set of studies selected for each analysis was such that the most complete datasets were included, while adhering most closely to the distance and age parameters. Both fixed effects and random effects methods were used to weight studies accordingly when calculating summary rates. Random effects methods are generally used where there is significant heterogeneity among individual study estimates, though meta-estimates and associated CIs generated are not necessarily more conservative [25]. Nevertheless, some researchers have reasoned that even in the absence of quantified or visual heterogeneity in forest plots among studies of the same age groups and geographical exposures, the multiple types of nuclear power generators operating with differing capacities justify random effects as the most appropriate method for calculating summary measures in this context [11]. To measure publication bias in each analysis, funnel plots were developed and trim and fill tests were completed, which suppress studies with extreme effect sizes [16,26]. Published guidance on the best metrics to use for axes was followed: standard errors were used as measures of study size for the vertical axis and either the SIR or OR was used for the horizontal axis [27,28]. All statistical analyses were performed in Stata, version 12 (Stata Statistical Software; Stata Corporation, College Station, TX, USA). 3. Results 3.1. Studies included From the systematic review of the MEDLINE database, 20 relevant studies were identified, and the manual search of references from relevant publications further identified 11 papers. Ultimately, a total of 14 studies were selected for analysis (Fig. 1). Of the 31 studies that were initially identified as relevant, most of the 17

were excluded because more recent studies were available that contained more complete datasets; Appendix A includes the list of excluded studies with rationale. One study [29] identified in the search was excluded, as it was available in French only; however, an abstract was available in English and it was determined the study results were less complete than Sermage-Faure [20], so, ultimately, it would not have been included in the analysis. 3.2. Study characteristics Of the 14 studies that were selected, ten were ecological studies, two were case-controls studies, and two other studies provided both cohort and case-control analyses. The study period for all studies combined began in 1962 and terminated in 2009. Studies selected represented nine countries with a total of 63 NPPs, plus three nuclear reprocessing facilities. Although the primary outcome included in each study was childhood leukemia, at least one study [19] also included similar diseases, for example, non-Hodgkin’s lymphoma. Most research included all types of childhood leukemia, with Sermage-Faure [20] identifying only acute leukemia cases. The number of cases included within each study ranged from one to 504. Tables 1 and 2 present the characteristics of the case-control and cohort studies, respectively. The quality of studies varied considerably, with a general trend of recent research using more sophisticated methods and improved control for confounders. Exposure was predominantly assessed as address at diagnosis; however, several studies provided estimates based on either address at birth or a reconstruction of residential histories [7,20,30]. Studies also differed in the measurement from a NPP: some measured the actual distance from residence to NPP, while others defined distance as proximity to a population centroid; that is, the weighted geographical centre of a given area’s population. For example, outputs from the secondary analysis in the Committee on Medical Aspects of Radiation in the Environ-

49

.25

.2

s.e. of logSIR .1 .15

.05

0

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

.6

.8

1 logSIR

1.2

1.4

1.6

Fig. 3. Funnel plot of leukemia cases <15 years of age and address within 25 km of NPP on a log scale.

%

Weight

Authors

ES (95% CI)

(I-V)

Bithell et al 2013

0.72 (0.32, 1.61)

15.30

Sermage-Faure et al 2012

1.60 (0.66, 3.87)

12.77

Kaatsch et al 2008b

2.19 (1.51, 3.18)

71.93

I-V Overall (I-squared = 67.1%, p = 0.048)

1.77 (1.29, 2.43)

100.00

D+L Overall

1.45 (0.74, 2.86)

.1

1

10

Fig. 4. Forest plot of case-control studies of leukemia cases <5 years of age and address within 5 km of NPP, displaying both fixed effects and random effects outputs on a log scale.

ment (COMARE) study [19] were used, as the study methodology to measure distance was more comparable to other eligible studies identified in the systematic review. While the ecological studies were mostly consistent with the use of national level data for the reference group, distances designated for controls in the casecontrol studies varied from 5 km to 30 km. Also, as one might

expect with increased follow-up time, recent studies with more data involved greater numbers of cases. Appendix B highlights quality indicators for each of the included studies. Four studies produced ORs, 11 provided SIRs and one presented an IRR. Reference groups for ecological studies were based mainly on national incidence rates of childhood leukemia. Two

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

.5

.4

s.e. of logOR .3 .2

.1

0

50

1

2 logOR

3

4

5

Fig. 5. Funnel plot of case-control studies of leukemia cases <5 years of age and address within 5 km of NPP on a log scale.

% Weight Authors

ES (95% CI)

(I-V)

Spycher, 2011

1.20 (0.60, 2.41)

11.08

COMARE, 2011

1.22 (0.77, 1.94)

25.21

Sermage-Faure et al., 2012

1.37 (0.81, 2.32)

19.40

Kaatsch et al., 2008a

1.41 (0.99, 2.00)

44.31

I-V Overall (I-squared = 0.0%, p = 0.953)

1.33 (1.05, 1.68)

100.00

D+L Overall

1.33 (1.05, 1.68)

.1

1

10

Fig. 6. Forest plot of ecological/cohort studies of leukemia cases <5 years of age and address within 5 km of NPP, displaying both fixed effects and random effects outputs on a log scale.

studies [21,30], included individual data of the unexposed group. Four studies [5,6,20,31] produced eligible estimates of childhood leukemia with lower 95% CI limits above the null and none included upper 95% CI limits below the null (see Tables 1 and 2).

3.3. Primary statistical analysis A meta-analysis of the eight studies (1665 cases) of exposed children <15 years of age residing at time of diagnosis (or born)

51

.4

.3

s.e. of logSIR .2

.1

0

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

1

1.5

2

2.5

logSIR

Fig. 7. Funnel plot of ecological/cohort studies of leukemia cases <5 years of age and address within 5 km of NPP on a log scale.

within 25 km of a NPP compared to (predominantly) national incidence rates generated a pooled estimate of 1.00 (95% = 0.95–1.05) using a random effects model (heterogeneity <0.0%) (Fig. 2). Five of the eight studies compared incidence within 15 km of a nuclear installation. A funnel plot consisting of the log of study estimates presented little asymmetry (Fig. 3) and a trim and fill test presented no evidence of heterogeneity (p > 0.8), leaving the meta-estimate unchanged. Conducting a sensitivity analysis on these findings by integrating also those studies [31–33] that included populations living within 25 km of a nuclear reprocessing facility minimally increased the pooled estimate of cohort studies to 1.01 (95% = 0.94–1.07), with heterogeneity remaining unsubstantial at 15.5% (p > 0.8). 3.4. Secondary statistical analysis One study [20] included both ecological and case-control results of the same data and four other publications [5–7,19] also contained both cohort/ecological and case-control analyses of the same populations. Thus, meta-analyses were performed separately for each study type to provide a more robust analysis, though with relatively small sample sizes. A meta-analysis of the three casecontrol studies (53 cases) examining the risk of children <5 years of age living within 5 km of a NPP produced a pooled OR of 1.45 (95% CI = 0.74–2.86) (Fig. 4). A funnel plot was produced and a trim and fill test identified evidence of heterogeneity (p < 0.05), though meta-estimates were not adjusted (Fig. 5). An analysis of the same parameters using results from four cohort studies (76 cases) generated a significantly elevated pooled SIR of 1.33 (95% CI = 1.05–1.68) Table 3 Summary of meta-analysis results. Variables

Fixed effects

Random effects

<5 years old & <5 km (NPPs) (case-control) <5 years old & <5 km (NPPs) (cohort) <15 years old & <25 km (NPPs) <15 years old & <25 km (NPPs + reprocessing)

1.77 (1.29–2.43)* 1.33 (1.05–1.68)* 1.00 (0.95–1.05) 1.00 (0.95–1.06)

1.45 (0.74–2.86) 1.33 (1.05–1.68)* 1.00 (0.95–1.05) 1.01 (0.94–1.07)

*

p-value < 0.05.

(Fig. 6). The CIs of the estimates used from each of these studies included the null. A funnel plot was also produced, and the estimate remained the same after a trim and fill test, which did not identify heterogeneity (p > 0.9) (Fig. 7). Table 3 presents a summary of all meta-analysis results for both fixed effects and random effects models. 4. Discussion 4.1. Main results The calculation of meta-estimates provided no evidence of an increase in leukemia incidence in children <15 years of age residing within 25 km of a NPP. Pooling studies focusing on children <5 years of age living within 5 km of a NPP produced mixed findings: significantly elevated estimates of leukemia incidence using cohort studies, but not with case-control research.

4.2. Potentially influential factors There are at least two opposing factors involved when comparing populations in close and distant proximities to NPPs. Larger distances include a broader radius, associated with greater populations and thus more cases; therefore, the larger population ought to produce a more precise estimate. Conversely, closer vicinities equate to fewer people overall, yet those residents are conceivably subject to higher exposures from NPP emissions than those living farther. Therefore, studies at closer proximities might tend to produce a less diluted, albeit less precise (smaller sample sizes), effect estimate. This scenario is demonstrated in the present study, as 1665 cases were included in the primary analysis (<25 km), yet only 53 cases and 76 cases respectively in the case-control and cohort secondary analyses (<5 km). Although residential addresses provide more accurate information on location, some researchers assert that population centroids might be more useful, since residents spend much of their time in the general vicinity of their home; nevertheless, applying this method in districts with more than one population centroid, i.e., a

52

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

clustered population distribution, could exacerbate exposure misclassification [34]. The potential for such misclassification also arose from the collation of studies in the primary analysis, in which the proximity of the exposed group ranged from living 15 to 25 km from a NPP. While this varying definition does not present issues within a single study, pooling data for a meta-analysis with higher exposed groups would tend to bias risk estimates away from the null. Despite this tendency, estimates within the 15–25 km consistently did not indicate increased leukemia risk for children residing within such perimeters to nuclear installations. While less consistent than the primary analysis, the case-control and ecological/cohort results for the <5 km data were not incompatible: the 95% CIs of the ecological/cohort summary estimate (1.05–1.68) fit entirely within those of the case-control dataset (0.74–2.86). Interestingly, the heterogeneity in the latter dataset was attributed to one study [7], which used a case-control design, but whose population overlapped with one of the cohort studies [19]. Exposure was defined differently in the two studies: COMARE [19] used the population centroid of electoral wards to measure distance, while Bithell [7] used measured distances of each address to the nearest NPP. Interestingly, although the study period for COMARE was actually shorter (1969–2004) compared to Bithell (1962–2007), exposure parameters resulted in double the cases identified in the former study compared to the latter (20 versus ten). This disparity is incongruous with the KiKK study, which demonstrated more consistent findings when using either specific residential addresses (37 cases) or population centroids (34 cases) to measure exposure [35]. While consistent, methodological biases of this study necessitate caution when interpreting results. Kaatsch [6] acknowledged that certain communities with a NPP were less likely to provide control information, potentially overestimating resulting risk estimates, but an exclusion of cases and controls from applicable areas did not lead to an appreciable difference. Still, Kinlen [35] raised further potential biases that might weaken KiKK study results. In addition to instances of control selection prior to diagnosis year, at the initiation of the German Childhood Cancer Registry, paediatric oncologists were the sole source of case ascertainment; it is possible that with an incomplete registry in earlier years, cases near NPPs might be more likely to be included, considering concerns over radiation. Though valid, it is difficult to quantify the effects of these issues on calculated KiKK study risk estimates, and even more so for results of the current meta-analysis.

4.3. Comparison with other findings A previous meta-analysis [11] was undertaken on both mortality and incidence data of childhood leukemia and vicinity to NPPs. Mortality data was omitted from the present analysis for several reasons: survival rates for childhood cancers, including leukemia, are on the rise, leading to possible risk underestimates [36]; families tend to move to seek treatment upon diagnosis of a child, potentially skewing mortality data in the vicinity of a NPP [37]; and death certificates may suggest an outcome other than leukemia for underlying cause of death, further complicating case ascertainment [38]. We included multi-site studies, whereas only individual NPPs were assessed in the previous meta-analysis. Another methodological difference was that only cohort studies were included in the previous meta-analysis, thereby excluding potentially useful case-control results merely due to study design. Further, other researchers [39] have criticized the methods used to perform that analysis, including the pooling of different age groups, types of nuclear facilities, and exposure zones; failing to provide justification of studies selected for analysis; and incorporating only studies that present data for individual nuclear sites, even though pooling

data for multiple sites is good practice and typically provides more precise risk estimates. Such criticisms aside, the highest risks for incidence in that study were also younger children (aged 0–9) living in close proximity (<16 km) of NPPs; risk estimates ranged from 1.22 to 1.25 with lower 95% CIs in excess of the null. By contrast, the current study presents no evidence of an increased risk in all children living at farther distances, whereas the previous meta-analysis demonstrated modest, but statistically significant, excesses (1.10–1.12 with lower 95% CIs at 1.00 or higher). Although there was overlap in both studies of the populations surrounding specific NPPs, none of the same studies was used in both meta-analyses, as the current work benefitted from the publication of updated research on those populations.

4.4. Strengths and limitations There are two principal areas in which bias could have affected the results of the present study: (1) the method used for the systematic review and (2) publication bias. Only one medical journal database (MEDLINE) was used during the review to identify potential studies that satisfied the stated eligibility criteria, of which there were 20. From the reference lists of these papers, an additional 11 publications were selected for potential inclusion, implying that MEDLINE failed to retrieve over a third of the total included studies. However, the reference lists of three reviews and an earlier meta-analysis of childhood leukemia and NPP studies were also examined for potential papers beyond the eligible studies identified in the systematic review. It is therefore unlikely that any consequential studies would have been missed through the application of these search methods. Nevertheless, when conducting a meta-analysis there is always the underlying possibility of publication bias, where studies published and referenced are more likely to include significant results [40]. Visual interpretation of funnel plots produced in the present study did not indicate evidence of clear publication bias, albeit with limited study numbers. It is also noteworthy that meta-estimates were predominantly based on European populations, with limited representation from the US, the country with the highest number of NPPs. American research is underway, however, including a focus on populations that are within 5 miles (∼8 km) of a NPP, of which there are over 1 million people, as well as reconstructing doses of children to time and place of birth [41]. A significant limitation of the findings is the reliance on distance as a surrogate exposure to radiation. As discussed above, the decision to use population centroids or individual residential distances can affect study outcomes. Ideally, as demonstrated in the KiKK study [5,6], both methods would be used in an epidemiologic study; any identified disparities in outcomes would warrant further analysis. Even more, although there is a causal association between ionizing radiation and childhood leukemia, radioactive emissions of a NPP are magnitudes of order below levels believed to cause leukemia. Studies that have identified an association with distance and leukemia have explicitly stated that the cause for the excess is unknown, based on current understandings of radiobiology [5,6]. Necessary doses believed to induce leukemia notwithstanding, certain studies have measured actual radiation levels at varying distances from a NPP and have demonstrated that radiation levels do not decrease uniformly with distance; in fact, radiation levels have been found to be greater at farther distances [41]. It is more likely that exposure to radiation is reliant on individual behaviours, including food consumption and water use [20]. Though research examining the association between estimated dose and childhood leukemia have not suggested an association [20,42], other researchers have encouraged, where feasible, the

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

direct measurement of radionuclides in the body to improve exposure assessment [9].

53

One prominent alternative theory to radiation-induced leukemia is that of population mixing [43,44]: it is possible that a subclinical virus might heighten susceptibility to leukemia where marked population mixing occurs. An example of this scenario is population influx into relatively isolated rural areas, such that occurs with the construction and operation of a NPP. This alternative theory would be more compatible in situations where leukemia cases were isolated at the stages of reactor start-up, rather than continual during ongoing operations [45]. Nevertheless, a potential infectious agent has not yet been identified, and it is unclear how long leukemia excesses might continue in a population with prolonged exposure to such an agent.

case-controls studies, and two other studies providing both cohort and case-control designs. The main findings of this study presented no evidence of an increased risk of leukemia in children residing within 25 km of a NPP. Nevertheless, an analysis of a subset of these results, including four ecological/cohort studies, indicated a weak, but statistically significant, excess of leukemia in children <5 years of age living <5 km from a NPP. Despite some evidence from this study of increased leukemia incidence based on a limited number of cases, the use of distance as a surrogate exposure for radiation is not without its challenges, and other hypotheses should continue to be explored. Nevertheless, with the ongoing expansion of nuclear power, health of populations surrounding NPPs should be monitored, larger studies with superior exposure assessment methods should be initiated that pool incidence data for childhood leukemia in areas not previously investigated, and, meanwhile, siting of new NPPs near dense population centres should be appropriately justified.

5. Conclusion

Funding source

4.5. Alternative explanations for observed results

A systematic review of published studies examining potential associations between childhood leukemia incidence and residential proximity to NPPs was undertaken and resulted in 14 studies selected for further analysis, including ten ecological studies, two

No funding was provided for this research.

Appendix A. List of excluded studies with rationale.

#

Author, year

Study design

Setting & # of facilities

Study period

Rationale for exclusion

1

Pobel, 1997 [49]

Case-control

1978–1993

Distance is not an exposure.

2

Viel, 1993 [50]

Cohort

1978–1990

Guizard [32] provides more complete data.

3 4

Cohort Cohort

1990–1998 1990–1998

Sermage-Faurer [20] updates incidence study. Sermage-Faurer [20] updates cohort study.

Cohort Cohort

1978–1992 1990–2005

Guizard [32] provides more complete data. Kaatsch [5] provides more complete data.

Cohort

Germany—1 NPP

1990–1995

Kaatsch [5] provides more complete data.

8 9

Laurier, 2008 [51] White-Koning, 2004 [52] Viel, 1995 [53], Hoffmann, 2007 [54] Hoffmann, 1997 [55] Korblein, 1999 [56] Spix, 2008 [57]

France—1 nuclear reprocessor France—1 nuclear reprocessor France—19 NPPs France—all 29 nuclear sites France—La Hague Germany—1 NPP

Cohort Case-control

Germany—15 NPPs Germany—16 NPPs

1980–1995 1980–2003

10

Kaatsch, 1998 [58]

Cohort

1991–1995

11

Meinert, 1999 [59]

Case-control

1980–1994

Distance is not an exposure.

12

Urquhart, 1991 [60] Ewings, 1989 [61] Gardner, 1990 [62]

Case-control

1968–1986

Cases and controls are matched on distance.

1959–1986 1950–1985

Study does not include an age group of <15 years. Study does not include an age group of <15 years.

Roman, 1987 [63] Goldsmith, 1992 [64] Enstrom, 1983 [65]

Cohort Cohort

Germany—20 nuclear installations Germany—does not specify Scotland—1 nuclear reprocessor UK—1 NPP UK—1 nuclear reprocessor UK—3 NPPs UK—8 NPPs

Kaatsch [5] provides more complete data. Same study as Kaatsch [5] but does not include leukemia-specific data. Kaatsch [5] provides more complete data.

1972–1985 1971–1980

Includes only non-NPP or reprocessing facilities. COMARE [19] provides more complete data.

Cohort

US—1 NPP

1960–1978

Includes leukemia mortality only.

5 6 7

13 14 15 16 17

Cohort Case-control

54

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

Appendix B. A comparison of the quality of eligible studies. Author, year

Ascertainment of cases

Adjustment

Residence at birth known?

Bithell, 2013 [7]

Uses National Registry of Childhood Cancer Cases (NRCT). For birth cohort, controls were cancer free as of age of matched case. Controls selected either from birth register or NRCT. For address cohort, used other cancer controls, excluding leukemia and non-Hodgkin’s Lymphoma (non-LNHL), as reference group. Used NRCT and estimated ascertainment of cases to be 95%.

Cases matched on sex and approximate date of birth. Included variables for social class (father’s occupation at birth), urban/rural status, population density and Carstairs index.

Yes. Produced different Coordinates of address estimates based on <5 km from NPP. both location of address at birth and diagnosis.

Adjusted by 5 year age groups and socioeconomic variables.

No. Uses only address at diagnosis, not birth.

Cases ascertained from Scottish National Cancer Registry. Estimated ascertainment of cases to be over 90%. Used NRCT data. Included cases of NHL.

No adjustments for socio-economic status (SES) or rural-urban status.

No, residence at diagnosis only.

Adjusted expected cases by SES by Carstairs index, rural/urban status, population density.

No, residence at diagnosis only.

Expected calculations based on age and sex.

No, residence at diagnosis only.

Distance to electoral ward. Unclear if centroid or address actually used.

Ecological data adjusted for age group, sex, and time since production. Collected info on socioeconomic status. Four controls were matched to cases on sex, age, and municipality of residence at diagnosis. Measured whether parents had history of radiation work. Controls matched for date of birth, age, sex, NPP area at diagnosis. Accounted for urban and rural setting. Social status and other risk factors were excluded.

Residential histories reconstructed compared to point in time.

Distances weighted by Some moving dates duration from start-up missing for residential to diagnosis. histories for 8% of cases and 4% of controls. Small number of cases overall. No cases or controls within 5 km of plant.

No, only address at diagnosis was included, not residential histories.

Geo-coded, so used exact distances.

No, only address at diagnosis was included, not residential histories.

Geo-coded, so used exact distances.

Bithell, 1994 [46]

Black, 1994 [31]

COMARE, 2011 [19]

Guizard, 2001 [32]

Heinävaara, 2010 [30]

Cases were identified from surveys of doctors, not a registry. Lower response rate in distances further from the NPP; potential under-ascertainment in these areas. Cases were validated by healthcare files, but not exhaustive for early in study period. Cancer cases obtained from Finnish Cancer Registry.

Kaatsch, 2008 [5]

Cases ascertained from German Childhood Cancer Registry (GCCR).

Kaatsch, 2008 [6]

Cases ascertained from German Childhood Cancer Registry (GCCR).

Controls matched for date of birth, age, sex, NPP area at diagnosis. Accounted for urban and rural setting. Social status and other risk factors were excluded.

Measure of distance

Distance defined as population centroid of ward within 25 km of NPP. Addresses of cases checked from case notes and postcoded. Used centroid of electoral wards for exposure status, not precise distance.

Other comments Excludes cases from Sellafield in primary analysis. Assumes non-LNHL cancers unrelated to proximity.

Start date of operations is not mentioned. Sellafield has been excluded, as it was the hypothesis generating observation and also is a nuclear reprocessing site. Also excludes Dounreay. Several cases had Down’s syndrome prior to leukemia, which is a significant risk factor for leukemia.

Communities provided information on eligible controls. Communities with a NPP were less likely to supply controls, which might have overestimated study effect, but sensitivity analysis suggests otherwise. Independent statistician replicated analyses.

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

55

Lane, 2013 [42]

Incidence data collected by Ontario Cancer Registry for 1990-1991 and Canadian Cancer Registry for 1992-2008.

SIRs calculated based on age and sex-specific rates of Ontario population for 1990-2008. No SES adjustment.

No, residence at diagnosis only.

Used addresses linking Cancer Care Ontario data to Ontario property assessment files.

Ma, 2011 [48]

Cases obtained from Illinois State Cancer Registry. Diagnosis using SEER and International Classification of Childhood Cancer. Cases ascertained from German Childhood Cancer Registry (GCCR).

Age-adjusted SIRs calculated for each geo-zone using Illinois state age-specific cancer incidence. Age, sex, and race used as covariates. Control counties matched by regional age structure and a large proportion of pediatric cancer patients having been treated at the same facilities as the installation county.

No, residence of diagnosis only. History of residence not known. No, residence of diagnosis only. History of residence not known.

Distance to zip code centroid of cancer cases used as exposure.

Sermage-Faure, 2012 [20]

Cases ascertained from the French National Registry of Childhood Hematopoietic Malignancies.

No, residence of diagnosis only. History of residence not known.

Actual addresses of the cases and controls were geocoded using the geographic information system MAPINFO.

Sharp, 1996 [33]

Cases ascertained from Central Scottish Cancer Registry.

Controlled for confounders by urban status, median household income, proportion of blue collar workers, proportion of baccalaureate holders, excluded cases and controls living within 200 m of high-voltage power line. Controls highly illustrative of source population. Could not adjust for individual risk factors, e.g. breast-feeding, but not likely to differ within and outside of 5 km threshold. Controlled for age, sex, deprivation, urban-rural specific rates.

No, residence at diagnosis only.

Spycher, 2011 [21]

Used Swiss Childhood Cancer Registry.

Controlled for other nuclear facilities, background ionizing radiation, electromagnetic radiation, traffic exhaust, agricultural pesticides, socioeconomic status based on job, income, education, population mixing,

Yes, analyzed both residence at birth and at diagnosis. Birth cohort did not include cases that had emigrated before diagnosis.

Included enumeration districts with a population centroid within 25 km. Actual distances to residence were computed using ArcGIS. Extensive sensitivity analyses included distance as a continuous variable.

Michaelis, 1992 [47]

Circles around nuclear installations within which least one third of community resides.

Also looks at doses to populations, show that does not decrease uniformly with distance, and can even be higher with greater distance. Provided a 5-year latency period. Selected zones to ensure enough cases and consistency with other studies. For a subset of cases, researchers investigated additional potential environmental risk factors, e.g. father smoking.

Latency period >5 years.

References [1] R. Siegel, D. Naishadham, A. Jemal, Cancer statistics, CA Cancer J. Clin. 62 (2012) 10–29. [2] J.W.W. Coebergh, A.M.J. Reedjik, E. de Vrie, et al., Leukaemia incidence and survival in children and adolescents in Europe during 1978–1997. Report from the Automated Childhood Cancer Information System project, Eur. J. Cancer 42 (2006) 2019–2036. [3] C.H. Pui, Childhood leukemias, NEJM 332 (24) (1995) 1618–1630. [4] D.J. Brenner, R. Doll, D.T. Goodhead, et al., Cancer risks attributable to low doses of ionizing radiation: assessing what we really know, Proc. Nat. Acad. Sci. U. S. A. 100 (24) (2003) 13761–13766. [5] P. Kaatsch, C. Spix, I. Jung, et al., Childhood leukemia in the vicinity of nuclear power plants in Germany, Dtsch. Ärzteblatt Int. 105 (42) (2008) 725–732. [6] P. Kaatsch, C. Spix, R. Schultz-Rath, et al., Leukemia in young children living in the vicinity of German nuclear power plants, Int. J. Cancer 1220 (2008) 721–726. [7] J.F. Bithell, M.F.G. Murphy, C.A. Stiller, et al., Leukaemia in young children in the vicinity of British NPPs: a case-control study, BJC 109 (2013) 2880–2885. [8] C. Kuehni, B.D. Spycher, Nuclear power plants and childhood leukaemia: lessons from the past and future directions, Swiss Med. Wkly. 143 (2014) w13912. [9] D. Laurier, B. Grosche, A. Auvinen, et al., Childhood leukemia risks: from unexplained findings near nuclear installations to recommendations for future research, J. Radiol. Prot. 34 (2014) R53–R68. [10] C. Muirhead, Childhood leukaemia near NPPs, BJC 109 (2013) 2763–2764.

[11] P.J. Baker, D.G. Hoel, Meta-analysis of standardized incidence and mortality rates of childhood leukemia in proximity to nuclear facilities, Eur. J. Cancer Care 16 (2007) 355–363. [12] Nuclear Energy Institute (NEI), World statistics: nuclear energy around the world, 2014. http://www.nei.org/Knowledge-Center/Nuclear-Statistics/ World-Statistics (accessed: 20.08.14). [13] A. Liberati, D.G. Altman, J. Tetzlaff, et al., The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: explanation and elaboration, Ann. Intern. Med. 151 (4) (2009) W-65–W-94. [14] D. Moher, A. Liberati, J. Tetzlaff, et al., Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement, BMJ 339 (2009) 332–336. [15] D.F. Stroup, J.A. Berlin, S.A. Morton, et al., Meta-analysis of observational studies in epidemiology: a proposal for reporting, JAMA 283 (15) (2008) 2008–2012. [16] Cochrane Handbook for Systematic Reviews of Interventions, in: J.P.T. Higgins, S. Green (Eds.), Wiley & Sons, Chichester, 2008. [17] D. Laurier, S. Jacob, O. Bernier, et al., Epidemiological studies of leukaemia in children and young adults around nuclear facilities: a critical review, Radiat. Prot. Dosim. 132 (2) (2008) 182–190. [18] M.K. Janiak, Epidemiological evidence of childhood leukemia around nuclear power plants, Dose-Response 12 (2014) 349–364. [19] COMARE, Further Consideration of the Incidence of Childhood Leukaemia Around Nuclear Power Plants in Great Britain. Fourteenth Report, Health Protection Agency, London, 2011.

56

W. Mueller, C. Gilham / Journal of Cancer Policy 6 (2015) 44–56

[20] C. Sermage-Faure, D. Laurier, S. Goujon-Bellec, et al., Childhood leukemia around French nuclear power plants—the Geocap study, 2002–2007, Int. J. Cancer 131 (5) (2012) E769–E780. [21] B.D. Spycher, M. Feller, M. Zwahlen, et al., Childhood cancer and nuclear power plants in Switzerland: a census-based cohort study, Int. J. Epidemiol. 40 (5) (2011) 1247–1260. [22] P. Boyle, D.M. Parkin, Statistical methods for registries, in: O.M. Jensen, D.M. Parkin, R. MacLennan, et al. (Eds.), Cancer Registration: Principles and Methods, International Agency for Research on Cancer, Lyon, 1991. [23] M. Szumilas, Explaining odds ratios, J. Can. Acad. Child Adolesc. Psychiatry 19 (3) (2010) 227–229. [24] J. Zhang, K.F. Yu, What is the relative risk? A method of correcting the odds ratio in cohort studies of common outcomes, JAMA 280 (19) (1998) 1690–1691. [25] C. Poole, S. Greenland, Random-effects meta-analyses are not always conservative, Am. J. Epidemiol. 150 (5) (1999) 469–475. [26] J.L. Peters, A.J. Sutton, D.R. Jones, et al., Performance of the trim and fill method in the presence of publication bias and between-study heterogeneity, Stat. Med. 26 (2007) 4544–4562. [27] J.A.C. Sterne, M. Egger, Funnel plots for detecting bias in meta-analysis: guidelines on choice of axis, J. Clin. Epidemiol. 54 (10) (2001) 1046–1055. [28] S.P. Megdal, C.H. Kroenke, F. Laden, Night work and breast cancer risk: a systematic review and meta-analysis, Eur. J. Cancer 41 (2005) 2023–2032. [29] S. Bouges, J.P. Daure, M. Hébrard, Incidence des leucémies aiguës, lymphomes et cancers thyroïdiens chez les enfants de moins de 15ans vivant autour du site nucléaire de Marcoule de 1985 à 1995, Rev. Epidemiol. Sante Publ. 47 (1999) 205–217. [30] S. Heinävaara, S. Toikkanen, K. Pasanen, et al., Cancer incidence in the vicinity of finnish nuclear power plants: an emphasis on childhood leukemia, Cancer Causes Control 21 (2010) 587–595. [31] R.J. Black, L. Sharp, E.F. Harkness, et al., Leukaemia and non-Hodgkin’s lymphoma: incidence in children and young adults resident in the Dounreay area of Caithness, Scotland in 1968–1991, J. Epidemiol. Commun. Health 48 (1968) 232–236. [32] A.V. Guizard, O. Boutou, D. Pottier, et al., 2001) The incidence of childhood leukemia around the La Hague nuclear waste reprocessing plant (France): a survey for the years 1978–1998, J. Epidemiol. Commun. Health 55 (2001) 469–474. [33] L. Sharp, R.J. Black, E.F. Harkness, et al., Incidence of childhood leikemia and non-Hodgkin’s lymphoma in the vicinity of nuclear sites in Scotland, 1968–1993, Occup. Environ. Med. 53 (1996) 823–831. [34] C.E. Dunn, R.S. Bhopal, S. Cockings, et al., Advancing insights into methods for studying environment-health relationships: a multidisciplinary approach to understanding legionnaires’ disease, Health Place 13 (2007) 677–690. [35] L.J. Kinlen, A German storm affecting Britain: childhood leukemia and nuclear power plants, J. Radiol. Prot. 31 (2011) 279–284. [36] C.H. Pui, D. Pei, J.T. Sandlund, et al., Long-term results of St. Jude total therapy studies 11, 12, 13A, 13B and 14 for childhood acute lymphoblastic leukemia, Leukemia 24 (2010) 371–382. [37] H. Tunstall, K. Pickett, S. Johnsen, Residential mobility in the UK during pregnancy and infancy: are pregnant women, new mothers and infants ‘unhealthy migrants’? Soc. Sci. Med. 71 (2010) 786–798. [38] R.R. German, A.K. Fink, M. Heron, et al., The accuracy of cancer mortality statistics based on death certificates in the United States, Cancer Epidemiol. 35 (2011) 126–131. [39] C. Spix, M. Blettner, Meta-analysis of standardized incidence and mortality rates of childhood leukaemia in proximity to nuclear facilities, Re: Baker P.J., Hoel D.G., 2007. European Journal of Cancer Care, 16, 355–363, Eur. J. Cancer Care 18 (2009) 429–430. [40] S. Duval, R. Tweedie, Trim and fill: a simple funnel-plot—based method of testing and adjusting for publication bias in meta-analysis, Biometrics 56 (2000) 455–463. [41] D.O. Stram, Analysis of cancer risks in populations near nuclear facilities: phase I. A report by the National Academies of Nuclear and Radiation Studies Board, Health Phys. 106 (2) (2014) 305–306. [42] R. Lane, E. Dagher, J. Burtt, et al., 2013) Radiation exposure and cancer incidence (1990–2008) around nuclear power plants in Ontario, Canada, J. Environ. Prot. 4 (2008) 888–913.

[43] L.J. Kinlen, R. Doll, Population mixing and childhood leukaemia: Fallon and other US clusters, BJC 91 (2004) 1–3. [44] K. Petridou, K. Revinthi, F.R. Alexander, et al., Space—time clustering of childhood leukaemia in Greece: evidence supporting a viral aetiology, Br. J. Cancer 73 (1996) 1278–1283. [45] M. Goodman, J.S. LaKind, J.A. Fagliano, et al., Cancer cluster investigations: review of the past and proposals for the future, Int. J. Environ. Res. Public Health 11 (2014) 1479–1499. [46] J.F. Bithell, S.J. Dutton, G.J. Draper, et al., Distribution of childhood leukemia and non-Hodgkin’s lymphomas near nuclear installations in England and Wales, BMJ 309 (6953) (1994) 501–505. [47] J. Michaelis, B. Keler, G. Haaf, et al., Incidence of childhood malignancies in the vicinity of West German nuclear power plants, Cancer Causes Control 1992 (3) (1992) 255–263. [48] F. Ma, M. Lehnherr, J. Fornoff, et al., Childhood cancer incidence in proximity to nuclear power plants in Illinois, Arch. Environ. Occup. Health 66 (2) (2011) 87–94. [49] D. Pobel, J.F. Viel, Case-control study of leukemia among young people near La Hague nuclear reprocessing plant: the environmental hypothesis revisited, BMJ 314 (7074) (1997) 101–106. [50] J.F. Viel, S. Richardson, P. Danel, et al., Childhood leukemia incidence in the vicinity of La Hague nuclear-waste reprocessing facility (France), Cancer Causes Control 4 (4) (1993) 341–343. [51] D. Laurier, D. Hémon, J. Clavel, Childhood leukemia below the age of 5 years near French nuclear power plants, J. Radiol. Prot. 28 (3) (2008) 401–403. [52] M.L. White-Koning, D. Hémon, D. Laurier, et al., Incidence of childhood leukemia in the vicinity of nuclear sites in France, 1990–1998, Br. J. Cancer 91 (2004) 916–922. [53] J.F. Viel, D. Pobel, A. Carré, Incidence of leukemia in young people around the La Hague nuclear waste reprocessing plant: a sensitivity analysis, Stat.Med. 14 (1995) 2459–2472. [54] W. Hoffman, C. Terschueren, D.B. Richardson, Childhood leukemia in the vicinity of the Geesthacht nuclear establishments near Hamburg, Germany, Environ. Health Perspect. 115 (6) (2007) 947–952. [55] W. Hoffman, H. Dieckmann, H. Dieckmann, et al., A cluster of childhood leukemia near a nuclear power plant in Northern Germany, Arch. Environ. Health 52 (4) (1997) 275–280. [56] A. Körblein, W. Hoffman, Childhood cancer in the vicinity of German nuclear power plants, Med. Glob. Surv. 6 (1) (1999) 18–23. [57] C. Spix, S. Schmiedel, P. Kaatsch, et al., Case-control study on childhood cancer in the vicinity of nuclear power plants in Germany 1980–2003, Eur. J. Cancer 44 (2008) 275–284. [58] P. Kaatsch, U. Kaletsch, R. Meinert, et al., An extended study on childhood malignancies in the vicinity of German nuclear power plants, Cancer Causes Control 9 (1998) 529–533. [59] R. Meinert, U. Kaletsch, P. Kaatsch, et al., Associations between childhood cancer and ionizing radiation: results of a population-based case-control study in Germany, Cancer Epidemiol. Biomark. Prev. 8 (1999) 793–799. [60] J.D. Urquhart, R.J. Black, M.J. Muirhead, et al., Case-control study of leukemia and non-Hodgkin’s lymphoma in children in Caithess near the Dounreay nuclear installation, BMJ 302 (6778) (1991) 687–692. [61] P.D. Ewings, C. Bowie, M.J. Phillips, et al., Incidence of leukemia in young people in the vicinity of Hinkley Point nuclear power station, 1959–1986, BMJ 299 (6694) (1989) 289–293. [62] M.J. Gardner, M.P. Snee, A.J. Hall, et al., Results of case-control study of leukaemia and lymphoma among young people near Sellafield nuclear plant in West Cumbria, BMJ 300 (6722) (1990) 423–429. [63] E. Roman, V. Beral, L. Carpenter, et al., Childhood leukemia in the West Berkshire and Basingstoke and North Hampshire district health authorities in relation to nuclear establishments in the vicinity, BMJ 294 (6572) (1987) 597–602. [64] J.R. Goldsmith, Nuclear installations and childhood cancer in the UK: mortality and incidence for 0–9 year old children, 1971–1980, Sci. Total Environ. 127 (1980) 13–35. [65] J.E. Enstrom, Cancer mortality patterns around the San Onofre nuclear power plant, 1960–1978, Am. J. Public Health 73 (1) (1978) 83–91.