SEARCH

SEARCH BY CITATION

Keywords:

  • etiology;
  • cancer;
  • children;
  • infections;
  • space-time clustering

Abstract

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Previously, we identified space-time clustering in certain childhood cancers around diagnosis residence. These findings provided support for the involvement of environmental agents in etiological processes occurring close to diagnosis. We have reanalyzed the same British population-based dataset. The aim of the study was to determine whether there was space-time clustering around the residence at birth in relation to time of birth and separately from time of diagnosis. A total of 29,553 cases, diagnosed during the period 1969–1993, were examined by a second-order procedure based on K-functions. Locations were birth addresses, but separately, both dates of birth and diagnosis were analyzed. There was statistically significant space-time clustering for Hodgkin lymphoma (HL) and central nervous system (CNS) tumors (p = 0.047 and 0.01, respectively, based on birth date) and for total leukemia at ages 1–4 years only, Non-Hodgkin lymphoma (NHL) and Wilms tumor (p = 0.01, 0.02 and 0.006, respectively, based on diagnosis date). These results, interpreted together with other epidemiological evidence, suggest an etiological role for environmental factors focused around birth address for certain childhood cancers. For HL and CNS tumors, findings suggest that etiological exposures occurred at similar ages or in utero. For leukemia, NHL and Wilms tumor there is support for exposures occurring at similar times before diagnosis. For leukemia, HL, NHL and CNS tumors, but not Wilms tumor, the findings are consistent with infectious hypotheses. © 2008 Wiley-Liss, Inc.

Space-time clustering is said to occur when an excess of cases is seen within small geographical areas over limited temporal periods, and this pattern cannot be attributed to general excesses in those areas or at those periods. We have previously reported statistically significant space-time clustering for certain childhood cancers diagnosed in Great Britain during the period 1969–1993. The analyses were based on place of diagnosis and time of diagnosis. Space-time clustering was evident for acute lymphoblastic leukemia (ALL), soft tissue sarcomas and osteosarcomas. These findings supported the involvement of environmental factors (especially infections) in etiological processes acting around residence at diagnosis.1, 2

One study has estimated that ∼50% of children diagnosed with cancer moved residential address between birth and diagnosis. However, most of the moves were local.3 As residential migration is so extensive, space-time clustering based on residence at birth will have a different interpretation from space-time clustering based on residence at diagnosis. Space-time clustering based on date and place of birth only would suggest a widespread, possibly epidemic process affecting fetuses or newborns. It would be consistent with an etiological exposure that occurs around the place of birth either at similar ages after the time of birth or prenatally. Furthermore (in the absence of clustering based on date or place of diagnosis), such clustering indicates that the disease might have a highly variable latent period. In contrast, space-time clustering based on place of birth and date of diagnosis (without space-time clustering based on date and place of birth) might be explained by an etiological exposure at similar times before diagnosis, with a fairly constant latent period. Most space-time clustering studies have analyzed place and date of diagnosis only.1, 2, 4–6 Only a few previous studies have found evidence of space-time clustering acting around birth amongst cases of childhood leukemia, CNS tumors, soft tissue sarcomas and Wilms tumors.7–11 One earlier study from northwest England found space-time clustering based on date of diagnosis and place of birth for childhood leukemia.3

We reanalyzed a national population-based childhood cancer data set from Great Britain (GB), of cases diagnosed during the period 1969–1993, for space-time clustering.1, 2 The aim of this study was to determine whether there was space-time clustering around the residence at birth in relation to date of birth and separately in relation to date of diagnosis.

Material and methods

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

All cases diagnosed with childhood cancer during the period 1969–1993 who had a residential address recorded for the time of birth were included in the study. Anonymous details were obtained from the population-based National Registry of Childhood Tumours, which now covers the entire geographical area of the UK (England, Wales, Scotland and Northern Ireland).12 Birth addresses were available for almost 92% of cases registered. This analysis was restricted to cases from Great Britain (England, Wales and Scotland).

In the UK, there are around 1.7 million postcodes. These are unique alphanumeric geographical identifiers for delivery of mail and may include a number of residential addresses (often 15 to 20 houses), a smaller number of multiple occupancy dwellings or a single commercial address.13

Ordnance Survey grid references were allocated to the centroid of the postcode of the address at the time of birth. Hence, the Easting and Northing co-ordinates of the birth address postcode were geo-referenced to within 0.1 km.

The International Classification of Diseases for Oncology (ICD-O) was used to define the following diagnostic groups for analysis: (i) leukemia; (ii) ALL, acute lymphoblastic leukemia; (iii) ANLL, acute nonlymphocytic leukemia; (iv) lymphomas; (v) HL, Hodgkin lymphoma; (vi) NHL, non-Hodgkin lymphoma; (vii) CNS, central nervous system tumors; (viii) astrocytoma; (ix) PNET, primitive neuroectodermal tumors; (x) soft tissue sarcomas; (xi) bone tumors; (xii) osteosarcoma; (xiii) sympathetic nervous system tumors; (xiv) Wilms and other renal tumors; (xv) germ cell and related tumors; (xvi) retinoblastoma; (xvii) hepatic tumors; (xviii) carcinoma; (xix) all cancers except leukemia and lymphoma and (xx) all cancers.14

There are 2 possible space-time interactions (based on residential address at birth) between: (a) date and place of birth; and (b) date of diagnosis and place of birth. We tested the following prior hypotheses that were suggested from previous studies.3–5, 7–11 (i) Space-time clustering based on date and place of birth will be found specifically in cases of CNS tumor, soft tissue sarcoma and Wilms tumour; and (ii) space-time clustering based on date of diagnosis and place of birth will be found especially in cases of leukemia (particularly ALL aged 1 to 4 years).

Many space-time clustering analyses have used the Knox test15 as the method of choice.4, 5 However, one problem with the Knox test is that thresholds are arbitrarily chosen. Use of many thresholds would result in multiple testing. A simplification of a second order procedure based on K-functions16 was used to partially circumvent this arbitrary choice of thresholds which is implicit in the Knox test and also to avoid multiple testing. This procedure involved a set of 225 Knox-type calculations, where the thresholds changed over a prespecified set of values.

Previous analyses have used a set of fixed geographical distances to define 2 cases as being close in space (taken as 0.5, 1, 1.5, …, 7.5 km) in addition to variable nearest neighbor (NN) distances.1, 3, 8–10 However, analysis based on the NN approach is likely to be more appropriate when both urban and rural areas are included. Thus, in this study, we have only considered NN (and not fixed geographical) distances. Inspection of the data showed that the mean distance between the 25th NNs was ∼5 km (although the distribution of these distances had high variability). Instead of applying a set of fixed distance thresholds (0.5, 1, 1.5, …, 7.5 km), we used a set of variable NN thresholds (defined by distances between the 18th, 19th, …, 32nd NNs). For close times, we took the set of values: 0.1, 0.2, …, 1.5 years.

The observed value of the K-function was calculated. The (otherwise unknown) distribution of the K-function was simulated. At each simulation, the dates of birth were randomly re-allocated to each of the cases in the analysis and a realization of the K-function was obtained from these simulated data. This was repeated for a total of 999 random permutations of time. Statistical significance in all analyses was taken as p < 0.05 and was assessed by comparing the observed value with the simulated distribution.

The K-function method does not provide a measure of the magnitude of the clustering. For statistically significant (p < 0.05) or marginally significant (0.05 ≤ p < 0.10) results we derived a value from the Knox test (with critical values for closeness in space and time taken as distances between 25th NNs and 1 year, respectively). The excess was estimated as S = [(OE) / E] X 100, where O and E were the observed and expected numbers of close space-time pairs. It should be noted that the variability of S is dependent on E and that there are no theoretical boundaries to the range of values that S may take.

Gender specific effects were studied by analyzing “male: any” and “female: any” clustering pairs (where “any” means that the corresponding partner case in a close pair may be either male or female).

Whilst the mean distance between the 25th NNs was 5 km, the distribution of these distances was highly skewed. The median distance between the 25th NNs was 2.8 km. To test whether population density was associated with space-time clustering, cases were split into 2 groups: 50% were classified as belonging to a “more densely populated” group and 50% were classified as belonging to a “less densely populated” group on the basis of whether the 25th NN was nearer or further away than the median distance (2.8 km) of the 25th NN. Analysis by population density was then done by considering clustering pairs that included at least 1 case from the “more densely populated” category and clustering pairs that included at least 1 case from the “less densely populated” category. It should be stressed that these analyses of population density (especially analyses of “less densely populated: any clustering pairs”) may be prone to a dilution from edge effects because “less densely populated” areas are not always contiguous.

Results

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

The study included 29,553 cases of childhood cancer that had complete birth address. Details of numbers of cases by sub-group are given in Table I. Twins have been excluded from all analyses.

Table I. Numbers of Cases for Analyses of Space-Time Clustering of Childhood Cancer Around the Residence at Birth (Great Britain, Diagnosed 1969–1993)
Diagnostic group and age (years)TotalMalesFemalesIn more densely populated areasIn less densely populated areas
  1. ALL, acute lymphoblastic leukemia; ANLL, acute nonlymphocytic leukemia; HL, Hodgkin lymphoma; NHL, non-Hodgkin lymphoma; CNS, central nervous system; PNET, primitive neuroectodermal tumors.

Leukemia, ages 0–149,7635,5004,2634,9414,822
Leukemia, ages 1–44,8502,7492,1012,4112,439
Leukemia, ages 5–144,3652,4801,8852,2472,118
ALL, ages 0–147,7594,4413,3183,9193,840
ALL, ages 1–44,1402,3611,7792,0592,081
ALL, ages 5–143,3351,9391,3961,7111,624
ANLL, ages 0–141,528792736779749
Lymphomas, ages 0–142,9392,0558841,5671,372
HL, ages 0–141,226851375649577
HL, ages 0–9444342102244200
NHL, ages 0–141,4851,050435789696
NHL, ages 0–9937674263494443
CNS tumors, ages 0–146,8633,7393,1243,4753,388
Astrocytoma, ages 0–142,6141,3061,3081,3051,309
PNET, ages 0–141,410889521693717
Soft tissue sarcomas, ages 0–141,9211,083838965956
Bone tumors, ages 0–141,262638624645617
Osteosarcoma, ages 0–14669334335343326
Sympathetic nervous system tumors, ages 0–142,0021,1098931,0021,000
Wilms and other renal tumors, ages 0–141,820920900896924
Germ cell and related tumors, ages 0–14915456459460455
Retinoblastoma, ages 0–14911469442461450
Hepatic tumors, ages 0–14245140105143102
Carcinoma, ages 0–14789327462386403
All cancers, except leukemia and lymphoma, ages 0–1416,8518,9507,9018,4958,356
All cancers, ages 0–1429,55316,50513,04815,00314,550

Table II presents the main results of the analyses from the NN threshold method. Based on date and place of birth, there was statistically significant space-time clustering for cases of HL (p = 0.047, S = 24.3%) and CNS tumor (p = 0.01, S = 3.9%). The clustering of cases of CNS tumor was not attributable to a particular sub-type (for astrocytoma, p = 0.60 and for PNET, p = 0.11). Based on date of diagnosis and place of birth, there was statistically significant space-time clustering for cases of leukemia, aged 1–4 years (p = 0.01, S = 4.0%). Clustering for ALL, aged 1 to 4 years, was less significant (p = 0.06; S = 3.7%). There was also statistically significant clustering for cases of lymphoma (p = 0.03, S = 10.6%), which was attributable to clustering amongst cases of NHL (p = 0.02, S = 14.1%), and for Wilms tumor (p = 0.006, S = 21.0%).

Table II. Main Results of Analyses of Space-Time Clustering of Childhood Cancer Around the Residence at Birth (Great Britain, Diagnosed 1969–1993), Obtained Using the NN Threshold Method
Diagnostic group and age (years)Place of birth, date of birth: p-value (S)Place of birth, date of diagnosis: p-value (S)
  • NN, nearest neighbor; S, strength of clustering; ALL, acute lymphoblastic leukemia; ANLL, acute nonlymphocytic leukemia; HL, Hodgkin lymphoma; NHL, non-Hodgkin lymphoma; CNS, central nervous system; PNET, primitive neuroectodermal tumors.

  • 1

    Statistically significant (p < 0.05).

Leukemia, ages 0–140.220.21
Leukemia, ages 1–40.360.011 (4.0%)
Leukemia, ages 5–140.440.33
ALL, ages 0–140.310.09 (1.7%)
ALL, ages 1–40.410.06 (3.7%)
ALL, ages 5–140.440.43
ANLL, ages 0–140.590.20
Lymphomas, ages 0–140.120.031 (10.6%)
HL, ages 0–140.0471 (24.3%)0.47
HL, ages 0–90.08 (32.8%)0.15
NHL, ages 0–140.110.021 (14.1%)
NHL, ages 0–90.720.041 (24.0%)
CNS tumors, ages 0–140.011 (3.9%)0.06 (2.0%)
Astrocytoma, ages 0–140.600.06 (16.2%)
PNET, ages 0–140.110.72
Soft tissue sarcomas, ages 0–140.180.35
Bone tumors, ages 0–140.550.07 (20.6%)
Osteosarcoma, ages 0–140.08 (26.4%)0.10
Sympathetic nervous system tumors, ages 0–140.200.97
Wilms and other renal tumors, ages 0–140.140.0061 (21.0%)
Germ cell and related tumors, ages 0–140.590.06 (18.7%)
Retinoblastoma, ages 0–140.230.30
Hepatic tumors, ages 0–140.810.06 (113.8%)
Carcinoma, ages 0–140.310.16
All cancers, except leukemia and lymphoma, ages 0–140.130.05 (1.5%)
All cancers, ages 0–140.0011 (72.3%)0.021 (105.4%)

Results of the analyses by gender (“male: any” and “female: any” clustering pairs) are shown in Table IIIa and IIIb. Based on date and place of birth, there was statistically significant space-time clustering of “male: any” clustering pairs for HL (p = 0.02, S = 32.8%), CNS tumors (p = 0.01, S = 3.8%) and PNET (p = 0.047, S = 23.2%) and of “female: any” pairs for lymphomas (p = 0.03, S = 16.7%), HL (p = 0.02, S = 32.6%) and CNS tumors (p = 0.03, S = 4.8%). Based on date of diagnosis and place of birth, there was statistically significant space-time clustering of “male: any” clustering pairs for leukemia, aged 1–4 years (p = 0.02, S = 3.6%), ANLL (p = 0.04, S = 20.6%), lymphomas (p = 0.046, S = 11.3%), NHL (p = 0.02, S = 14.9%), CNS tumors (p = 0.02, S = 2.8%), astrocytoma (p = 0.03, S = 18.9%), Wilms tumor (p = 0.002, S = 34.3%), germ cell tumors (p = 0.02, S = 35.2%) and hepatic tumors (p = 0.04, S = 125.4%) and of “female: any” pairs for leukemia, aged 1–4 years (p = 0.03, S = 4.7%), ALL (p = 0.04, S = 2.2%), lymphomas (p = 0.03, S = 15.7%), NHL (p = 0.003, S = 33.9%) and Wilms tumor (p = 0.04, S = 14.6%).

Table III. Results of Analyses of Space-Time Clustering of Childhood Cancer Around the Residence at Birth (Great Britain, Diagnosed 1969–1993), Obtained Using the NN Threshold Method
Diagnostic group and age (years)Place of birth, date of birth: p-value (S)Place of birth, date of diagnosis: p-value (S)
  • NN, nearest neighbor; S, strength of clustering; ALL, acute lymphoblastic leukemia; ANLL, acute nonlymphocytic leukemia; HL, Hodgkin lymphoma; NHL, non-Hodgkin lymphoma; CNS, central nervous system; PNET, primitive neuroectodermal tumors.

  • 1

    Statistically significant (p < 0.05).

a. For “male: any” clustering pairs  
 Leukemia, ages 0–140.210.16
 Leukemia, ages 1–40.470.021 (3.6%)
 Leukemia, ages 5–140.460.17
 ALL, ages 0–140.520.10
 ALL, ages 1–40.610.08 (3.1%)
 ALL, ages 5–140.440.46
 ANLL, ages 0–140.440.041 (20.6%)
 Lymphomas, ages 0–140.08 (10.1%)0.0461 (11.3%)
 HL, ages 0–140.021 (32.8%)0.31
 HL, ages 0–90.120.10
 NHL, ages 0–140.06 (15.1%)0.021 (14.9%)
 NHL, ages 0–90.640.05 (21.1%)
 CNS tumors, ages 0–140.011 (3.8%)0.021 (2.8%)
 Astrocytoma, ages 0–140.790.031 (18.9%)
 PNET, ages 0–140.0471 (23.2%)0.79
 Soft tissue sarcomas, ages 0–140.09 (15.5%)0.33
 Bone tumors, ages 0–140.690.08 (24.9%)
 Osteosarcoma, ages 0–140.200.18
 Sympathetic nervous system tumors, ages 0–140.230.89
 Wilms and other renal tumors, ages 0–140.270.0021 (34.3%)
 Germ cell and related tumors, ages 0–140.270.021 (35.2%)
 Retinoblastoma, ages 0–140.110.31
 Hepatic tumors, ages 0–140.780.041 (125.4%)
 Carcinoma, ages 0–140.490.18
 All cancers, except leukemia and lymphoma, ages 0–140.08 (1.1%)0.05 (1.7%)
 All cancers, ages 0–14<0.0011 (38.7%)0.07 (65.3%)
b. For “female: any” clustering pairs  
 Leukemia, ages 0–140.540.20
 Leukemia, ages 1–40.360.031 (4.7%)
 Leukemia, ages 5–140.650.62
 ALL, ages 0–140.390.041 (2.2%)
 ALL, ages 1–40.280.09 (3.6%)
 ALL, ages 5–140.580.24
 ANLL, ages 0–140.870.74
 Lymphomas, ages 0–140.031 (16.7%)0.031 (15.7%)
 HL, ages 0–140.021 (32.6%)0.77
 HL, ages 0–90.0481 (75.2%)0.38
 NHL, ages 0–140.06 (26.6%)0.0031 (33.9%)
 NHL, ages 0–90.300.0011 (64.7%)
 CNS tumors, ages 0–140.031 (4.8%)0.18
 Astrocytoma, ages 0–140.800.40
 PNET, ages 0–140.220.72
 Soft tissue sarcomas, ages 0–140.610.63
 Bone tumors, ages 0–140.580.13
 Osteosarcoma, ages 0–140.270.27
 Sympathetic nervous system tumors, ages 0–140.500.99
 Wilms and other renal tumors, ages 0–140.210.041 (14.6%)
 Germ cell and related tumors, ages 0–140.950.10
 Retinoblastoma, ages 0–140.290.42
 Hepatic tumors, ages 0–140.490.58
 Carcinoma, ages 0–140.220.33
 All cancers, except leukemia and lymphoma, ages 0–140.370.09 (1.3%)
 All cancers, ages 0–140.0031 (18.4%)0.011 (41.8%)

Table IVa and IVb show the results of analyses by level of population density (“more densely populated: any” and “less densely populated: any” clustering pairs). Based on date and place of birth, there was statistically significant space-time clustering of “more densely populated: any” clustering pairs for osteosarcoma (p = 0.04, S = 33.8%) and of “less densely populated: any” clustering pairs for CNS tumors (p = 0.01, S = 4.1%) and soft tissue sarcomas (p = 0.009, S = 27.3%). Based on date of diagnosis and place of birth, there was statistically significant space-time clustering of “more densely populated: any” clustering pairs for leukemia, aged 1 to 4 years (p = 0.003, S = 7.4%), lymphomas (p = 0.04, S = 9.5%) and NHL (p = 0.007, S = 13.7%) and of “less densely populated: any” clustering pairs for CNS tumors (p = 0.006, S = 6.0%), astrocytoma (p = 0.02, S = 21.1%), bone tumors (p = 0.045, S = 29.2%), Wilms tumors (p = 0.02, S = 21.6%) and germ cell tumors (p = 0.03, S = 31.6%).

Table IV. Results of Analyses of Space-Time Clustering of Childhood Cancer Around the Residence at Birth (Great Britain, Diagnosed 1969–1993), Obtained Using the NN Threshold Method
Diagnostic group and age (years)Place of birth, date of birth: p-value (S)Place of birth, date of diagnosis: p-value (S)
  • NN, nearest neighbor; S, strength of clustering; ALL, acute lymphoblastic leukemia; ANLL, acute nonlymphocytic leukemia; HL, Hodgkin lymphoma; NHL, non-Hodgkin lymphoma; CNS, central nervous system; PNET, primitive neuroectodermal tumors.

  • 1

    Statistically significant (p < 0.05).

a. For “more densely populated: any” clustering pairs  
 Leukemia, ages 0–140.350.22
 Leukemia, ages 1–40.430.0031 (7.4%)
 Leukemia, ages 5–140.390.69
 ALL, ages 0–140.300.32
 ALL, ages 1–40.470.05 (5.0%)
 ALL, ages 5–140.510.74
 ANLL, ages 0–140.700.27
 Lymphomas, ages 0–140.08 (12.2%)0.041 (9.5%)
 HL, ages 0–140.06 (37.8%)0.70
 HL, ages 0–90.230.42
 NHL, ages 0–140.180.0071 (13.7%)
 NHL, ages 0–90.600.10
 CNS tumors, ages 0–140.09 (4.2%)0.35
 Astrocytoma, ages 0–140.520.28
 PNET, ages 0–140.07 (19.5%)0.79
 Soft tissue sarcomas, ages 0–140.310.28
 Bone tumors, ages 0–140.560.13
 Osteosarcoma, ages 0–140.041 (33.8%)0.31
 Sympathetic nervous system tumors, ages 0–140.160.81
 Wilms and other renal tumors, ages 0–140.100.18
 Germ cell and related tumors, ages 0–140.430.33
 Retinoblastoma, ages 0–140.470.15
 Hepatic tumors, ages 0–140.880.11
 Carcinoma, ages 0–140.420.40
 All cancers, except leukemia and lymphoma, ages 0–140.300.11
 All cancers, ages 0–140.041 (30.0%)0.07 (55.3%)
b. For “less densely populated: any” clustering pairs  
 Leukemia, ages 0–140.250.27
 Leukemia, ages 1–40.260.24
 Leukemia, ages 5–140.540.22
 ALL, ages 0–140.510.08 (2.2%)
 ALL, ages 1–40.370.31
 ALL, ages 5–140.450.38
 ANLL, ages 0–140.790.27
 Lymphomas, ages 0–140.390.23
 HL, ages 0–140.180.26
 HL, ages 0–90.150.20
 NHL, ages 0–140.230.13
 NHL, ages 0–90.750.10
 CNS tumors, ages 0–140.011 (4.1%)0.0061 (6.0%)
 Astrocytoma, ages 0–140.380.021 (21.1%)
 PNET, ages 0–140.05 (28.7%)0.55
 Soft tissue sarcomas, ages 0–140.0091 (27.3%)0.36
 Bone tumors, ages 0–140.540.0451 (29.2%)
 Osteosarcoma, ages 0–140.470.11
 Sympathetic nervous system tumors, ages 0–140.480.99
 Wilms and other renal tumors, ages 0–140.100.021 (21.6%)
 Germ cell and related tumors, ages 0–140.760.031 (31.6%)
 Retinoblastoma, ages 0–140.210.39
 Hepatic tumors, ages 0–140.330.12
 Carcinoma, ages 0–140.390.17
 All cancers, except leukemia and lymphoma, ages 0–140.110.13
 All cancers, ages 0–140.0031 (27.6%)0.13

Discussion

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

This systematic space-time clustering study has used the same statistical methods and high-quality data as before.1, 2 Furthermore, it is also the largest analysis of such clustering around residence at birth, including 29,553 cases. On the basis of residence and date of birth, we found space-time clustering for cases of HL and CNS tumor. On the basis of residence at birth and date of diagnosis, we found space-time clustering for cases of leukemia (aged 1 to 4 years), NHL and Wilms tumor. The small values of strength of clustering (S) that were found for a number of specific diagnostic groups (e.g. CNS tumors based on place and date of birth) indicate that only a small proportion of cases contributed to the space-time clustering at the specified critical values of closeness in space (<distance to the 25th NN) and closeness in time (<1 year). In contrast, the large values of S found for all cancers (based on place of birth and both dates of birth and diagnosis) indicate that there was space-time clustering both within specific diagnostic groups and between different diagnostic groups.

The results for leukemia are noteworthy. The finding (based on place of birth and date of diagnosis) was restricted to cases aged 1 to 4 years (the childhood incidence peak).4, 5, 12 However, clustering included cases of both ALL and ANLL. This contrasts with our previously published study that reported similar clustering of place and date of diagnosis that was specific to cases of ALL in the childhood peak.1, 2 There was no evidence of space-time clustering based on place and date of birth. Taken together, these findings are consistent with a role for infections in etiology. The precipitating exposure would occur at similar times before diagnosis. We recall that there are 3 hypotheses concerning an infectious origin for childhood leukemia.17–20 Greaves' hypothesis specifically concerns precursor B-cell ALL and suggests that increased risk is associated with delay in exposure to common infections.17 Kinlen suggested that greatly increased incidence of childhood leukemia is associated with situations of highly unusual population mixing in previously isolated areas where a new infection is introduced to residents that have low herd immunity.18, 19 Smith suggested that higher incidence rates are linked with an infectious exposure that occurred in utero.20 Our results are consistent with Kinlen's hypothesis, but not Smith's hypothesis, because there was no evidence of space-time clustering based on place and time of birth. Another recent study suggested an association between incidence of childhood leukemia and influenza epidemics. This finding was interpreted as providing support for the role of an infection, occurring close to time of diagnosis, in triggering childhood leukemia in individuals that had low herd immunity.21 The data used in this study overlapped with the data analyzed here (they were from the same study area and were diagnosed during the time period 1974–2000). In this study there was only weak support for Greaves' hypothesis, because space-time clustering based on time of diagnosis and place of birth involved cases of both ALL and ANLL. In contrast to our earlier study (based on place and time of diagnosis), space-time clustering of childhood ALL was apparent for females, but not males.1, 2 Also, we found that space-time clustering of childhood leukemia was only present for cases from “more densely populated areas.” In the previous study we had found that clustering was confined to cases from “less densely populated areas.” An additional analysis of the same dataset considered in this study found increased incidence of leukemia associated with lower residential population density at time of diagnosis.2 However, another regional study from northwest England (which overlapped with the present national dataset) found higher incidence of childhood leukemia for cases resident at time of diagnosis in areas of greater population density.22 Also, more generally 3 other recent studies (from the USA, Taiwan and Sweden) have found higher incidence of childhood leukemia for cases resident at time of diagnosis in areas of greater population density,23–25 one multinational study (from Europe and Australia) found higher incidence for cases resident at time of diagnosis in areas of intermediate population density,26 whilst another study (from Great Britain analyzing data from the period 1953–1980) found higher mortality associated with cases born in areas of greater population density.27

For the lymphomas, there were 2 marked and contrasting findings. For HL, space-time clustering was present based on place and date of birth only. This suggests that the relevant etiological exposure occurred at similar ages after birth or in utero and that there was a heterogeneous latent period from exposure to diagnosis. Conversely, NHL space-time clustering was found based on place of birth and date of diagnosis only. This indicates that the exposure occurred at similar times before diagnosis. However, the findings for HL and NHL are consistent with different infectious etiologies, rather than a single one. Our earlier study (based on place and time of diagnosis) did not identify any space-time clustering amongst the lymphomas.1, 2 For both HL and NHL clustering was more marked for females than for males. This is consistent with a gender-specific effect and may be linked with differential susceptibility to a relevant etiological agent. It should also be noted that space-time clustering for both HL and NHL was limited to cases from more densely populated areas. This is consistent with a role for a directly transforming agent. A number of infectious exposures have been implicated in the etiology of HL and NHL including Epstein-Barr virus (EBV), hepatitis C, human herpes virus 8, HIV and helicobacter pylori.28–30

There was space-time clustering for CNS tumors, but not for astrocytoma nor for PNET based on place and time of birth. A previous study from northwest England had identified space-time clustering specifically amongst cases of astrocytoma based on place and time of birth.9 It is possible that the lack of clustering for astrocytoma in this study may be attributable to differences in sub-diagnostic classification, or that the earlier result was due to chance. There is certainly tentative evidence from a number of studies that supports the possible involvement of infections in the etiology of CNS tumors.31–34

For Wilms tumors, space-time clustering was only evident for place of birth and time of diagnosis. This finding also differed from a study from northwest England that showed space-time clustering based on place and time of birth.10 The presence of space-time clustering is consistent with the involvement of a spatio-temporally dispersed environmental agent. However, there is very limited evidence linking Wilms tumor with infections. One case-control study found a significant association with vaginal infection during pregnancy,35 and another case-control study found a protective association with breast-feeding.36 However, another study did not find any association with maternal infection.37 Some studies have linked parental occupational exposure to pesticides with increased risk of Wilms tumor.38–40 Conversely, a recent case-control study found no association with household pesticide exposure during pregnancy or early childhood.41

Overall, only one prior hypothesis (concerning CNS tumors) was confirmed from the present analyses. The results for leukemia and Wilms tumor differed somewhat from the prior hypotheses. There was little or no evidence of space-time clustering for soft tissue sarcomas. The possibility of chance findings cannot be excluded. It should be noted that the Knox test has low power to detect space-time clustering.42 The power of the K-function method has not been studied. Hence, low power might account for some of the inconsistencies in the findings from earlier studies that were based on relatively small numbers of cases. However, the major strength of this study is that to our knowledge it is the largest such analysis conducted, based on nearly 30,000 cases.

There was some overlap between the present data set and several data sets that generated the prior hypotheses. First, cases from northwest England, that were included in the Manchester Children's Tumour Registry (MCTR) during the period 1969–1993, were also analyzed in several earlier studies that covered most or all of the time period from 1954 to 2001.3, 8–10 However, this consisted of only around 1,500 cases. Secondly, there was also overlap with one other study (of around 10,000 cases who died from childhood cancer during the period 1968–1980), but this used a more limited method for statistical analysis.11

It is possible that small-area population shifts over short time intervals may have contributed to overall clustering, especially as this was found to be statistically significant (p = 0.001 based on date and place of birth and p = 0.02 based on date of diagnosis and place of birth). Although Kulldorff and Hjalmars have suggested a method for allowing for population shifts,43 we are not able to apply this to our data. This would need small-area population denominators for short time intervals and these are not available in the UK. However, it should be noted that statistically significant space-time clustering is confined to particular diagnostic groups for which we have some a priori evidence for an environmental (especially infectious) etiology.4, 5

In conclusion, we have found space-time clustering around residence of birth for specific diagnostic groups (HL, CNS tumors, leukemia, NHL and Wilms tumors). For HL and CNS tumors, the clustering suggests the involvement of environmental agents occurring at similar ages after birth or in utero. For leukemia, NHL and Wilms tumors the clustering pattern is consistent with exposures occurring at similar times before diagnosis. Furthermore, for HL, CNS tumors, leukemia and NHL, but not Wilms tumors, the findings are consistent with other evidence suggesting an infectious component to etiology. Whilst the possibility of artifact cannot be excluded, it is important to note that significant results are limited to specific diagnostic groups. Further research needs to be done to clarify the nature and timing of etiologically relevant exposures.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

The authors thank the anonymous referees for their most helpful and constructive comments on an earlier version of this article. They are also grateful to the North of England Children's Cancer Research (NECCR) fund for providing financial support for childhood cancer epidemiology research at Newcastle University. The Childhood Cancer Research Group (University of Oxford) receives Programme Grant support for its core functions from the Department of Health and the Scottish Ministers.

References

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References