SEARCH

SEARCH BY CITATION

Keywords:

  • solid tumours;
  • children;
  • aetiology;
  • environment;
  • space-time clustering;
  • England

Abstract

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

The aetiology of most childhood solid tumours (other than central nervous system [CNS] tumours) is unclear. To investigate whether certain environmental exposures may be involved, we have analysed for space-time clustering using population-based data from North West England for the period 1954–98. Knox tests for space-time interactions between cases were applied with fixed thresholds of close in space, <5 km, and close in time, <1 year apart. Addresses at birth and at diagnosis were used. Tests were repeated replacing geographical distance with distance to the Nth nearest neighbour. N was chosen such that the mean distance was 5 km. Data were also examined by a second order procedure based on K-functions. There was significant evidence of space-time clustering for Wilms' tumours (p = 0.03 and 0.04, using the geographical distance and nearest neighbour versions of the Knox test; and p = 0.07 and 0.03, using the geographical distance and nearest neighbour versions of the K-function method), and soft tissue sarcomas (p = 0.01, using both the geographical distance and nearest neighbour versions of the Knox test; and p = 0.001 and 0.002, using the geographical distance and nearest neighbour versions of the K-function method) based on time and location at birth, but not time and location at diagnosis. There was little or no evidence of space-time clustering amongst other diagnostic groups. These are the first results to demonstrate space-time clustering for childhood Wilms' tumours and soft tissue sarcomas. The results are consistent with environmental exposure hypotheses, relating to locations pre-natally or peri-natally. © 2003 Wiley-Liss, Inc.

The aetiology of solid tumours, other than central nervous system tumours (non-CNS solid tumours) in children is far from clear. There are 2 possible general mechanisms: genetic susceptibility and environmental exposures. The very early age of onset and the embryonal nature of many non-CNS solid tumours of childhood suggest a prenatal origin for most, if not all, of these tumours and genetic factors are likely to be important.1 Mutations in a number of tumour suppressor genes confer increased risk for certain embryonal tumours but inherited single gene defects would account for a small minority of cases only. De novo germline mutations in affected children have been reported,2, 3, 4, 5, 6, 7 however, and the possibility that parental pre-conceptional exposures might give rise to such mutations should be considered. Although it is likely that high penetrance genes play a role in a minority of cases, other low penetrance genes and gene-environment interactions may be important and pre- or post-natal exposure of a genetically susceptible child might also be a factor.

A number of statistically significant associations between certain parental occupations or occupational exposures and childhood non-CNS solid tumour risk have been reported. Associations with parental exposures may indicate parental germ cell mutation or transplacental exposure of the fetus. These parental associations include: exposure to pesticides, metals, chemicals, solvents, petroleum products, paints, pigments, plastic and resin fumes and maternal use of sex hormones before or during the index pregnancy.8 There is inconsistency, however, between studies and relative risks for the noted associations were all small.

Certain environmental agents, including industrial and traffic pollution, pesticides and infections, are likely to occur at localised geographical points at certain times or time periods. If such agents are involved in the aetiology of specific childhood non-CNS solid tumours, pre-conceptionally, pre-natally or post-natally, the distribution of cases may exhibit space-time clustering. Space-time clustering is said to occur when excess numbers of cases are observed within various small geographical locations, but only at limited points in time. Such clustering should occur amongst particular diagnostic types sharing a common aetiology.

There have been a number of studies that have applied formal statistical methods9, 10 to population-based incidence and mortality data on childhood non-CNS solid tumours,8, 11 but these studies were inadequate for the following reasons: 1) lack of reliable diagnostic data leading to inaccurate or inappropriate classification of cases; 2) some of the studies were based on time and location at death and these will suffer from incompleteness and assume (most probably incorrectly) that the aetiological factor is present close to time of death; and 3) limitations of the statistical methods employed.

We have examined incidence data from the Manchester Children's Tumour Registry (MCTR) for presence of space-time clustering using up-to-date and rigorous statistical methods. This registry is population-based and contains high quality verified diagnostic data over the 45-year time span of the study. Ascertainment of cases has been consistently high throughout its existence, even during the early years.12 The aims of our study were to test predictions of space-time clustering that might arise as a result of various possible environmental exposures and to distinguish between exposures around the times of birth and onset by using time/place of birth and diagnosis respectively.

MATERIAL AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

The non-central nervous system (non-CNS) solid tumours that are included in our study comprise: retinoblastoma, neuroblastoma, peripheral neuroectodermal tumours, Wilms' tumour, hepatic tumours, sarcomas, germ cell tumours and other rare miscellaneous neoplasms. All such cases, diagnosed between 1 January 1954 and 31 December 1998 and registered by the MCTR, were analysed. Ordnance Survey (OS) 8-digit (i.e., 4-digit Easting and 4-digit Northing) grid references were allocated to each case with respect to addresses at time of birth and diagnosis. These grid references locate each address to within 0.1 km.

The following aetiological hypotheses were tested: 1) a primary factor influencing geographical or temporal heterogeneity of incidence of childhood non-CNS solid tumours is related to exposure to an environmental agent relatively close to disease onset or in-utero or peri-natally; 2) geographical or temporal heterogeneity of incidence of childhood non-CNS tumours is modulated by differences in susceptibility between males and females and patterns of exposure related to level of population density; and 3) the age of onset of a childhood non-CNS solid tumour is related to the timing of exposure.

There are 4 possible space-time interactions: 1) between times and places of diagnosis; 2) between times and places of birth; 3) between time of diagnosis and place of birth; and 4) between time of birth and place of diagnosis. The interpretation of these interactions will depend on the extent of migration between birth and diagnosis among cases.13

Knox space-time clustering tests9 were applied to the data with thresholds fixed, a priori, as: close in space, <5 km, and close in time, <1 year apart. These limits were chosen to be the same as those used in 3 other studies of space-time clustering of the leukaemias and CNS tumours based on MCTR data.13, 14, 15 These limits are somewhat arbitrary, but this problem is overcome by using the K-function method described below.16 Where the observed number (O) of pairs that are both close in time and close in space, is greater than the expected number (E), this indicates a tendency for pairs of cases that are close in space to have similar times and vice versa.

One-sided tests were used to detect a significant interaction. The strength of interactions (S) was indicated by calculating [(O − E)/E] × 100 counts of pairs that are close in space and close in time. The distribution of case addresses is used as a proxy for population distribution. To adjust for different levels of population density (with respect to space and not with respect to time) the tests were repeated replacing geographical distance thresholds by the maximum of the distances to the Nth nearest neighbour of each of the pair, using all locations of all the cases of childhood malignancy in the MCTR population-based data set, 1954–98, (i.e., all leukaemias, CNS tumours and non-CNS solid tumours). For the calculation of the Nth nearest neighbour at diagnosis, all addresses at diagnosis were used. Cases were excluded if the address at diagnosis was not available. For the calculation of the Nth nearest neighbour at birth, all addresses at birth were used, and addresses at diagnosis (only if the address at birth was not available). This approach is similar to the method proposed by Jacquez.17 N was chosen such that the mean distance was 5 km. N was found to be 134 for birth locations and 136 for diagnosis locations.

Two problems are apparent with the Knox test. First, boundary problems may be important because it can be impossible or less probable for some cases to be close in 1 dimension to other cases. The second problem concerns the arbitrariness of the thresholds chosen, which often results in multiple testing. A simplification, avoiding adjustment for boundary conditions, of a second order procedure based on K-functions16 is used in the present analyses to overcome the problem of multiple testing. Nearest neighbour [NN] approaches were used as described above in relation to classical Knox tests.

The reasons for using both the Knox and K-function methods are: 1) the Knox test provides comparability with other studies of space-time clustering (that have used this well known test); 2) the Knox test provides a measure of “strength of clustering”, which is easily interpretable as an excess of cases or a deficit of cases, compared to the expected number of cases; and 3) the K-function method circumvents the methodological problems associated with the Knox test (choice of boundaries and multiple testing problems), but does not have an easily interpretable “strength of clustering” measure. Hence, the Knox test and K-function methods should be regarded as being complementary.

The primary analysis was restricted to the main diagnostic groups. Only those diagnostic groups that exhibited space-time clustering with a significance level of p < 0.05 for the total cases in the specified diagnostic group, using at least 2 of the 4 methods (the geographical or NN versions of the Knox test, or the geographical or NN versions of the K-function method) and including a NN threshold version, were considered further. Such groups were analysed within age and gender subgroups, and by level of population density.

Two age-groups were chosen: 1) ages less than the median age at the time of diagnosis; and 2) ages greater than (or equal to) the median age at the time of diagnosis. Median ages were calculated separately for each respective diagnostic group or sub-group. This analysis was done to differentiate between early or late onset cancers being involved in the clustering and hence to provide clues with respect to the timing of the putative environmental exposure. For gender, however, if there is a difference between males and females, then clustering involving ‘male:male’ and ‘male:female’ pairs would point to susceptibility among males, whereas clustering involving ‘female:female’ and ‘male:female’ pairs would point to susceptibility among females. Thus, analysis by gender proceeded by examining pairs of cases including at least 1 male (‘male:any’ clustering pairs) and pairs of cases including at least 1 female (‘female:any’ clustering pairs).

Addresses were classified as being located in a more densely populated area, or being located in a less densely populated area. For birth addresses the median distance for the 134th nearest neighbour was found to be 3.4 km. Birth locations, whose 134th nearest neighbour was ≤3.4 km were classified in the ‘more densely populated’ category, and birth locations, whose 134th nearest neighbour was >3.4 km were classified in the ‘less densely populated’ category. For addresses at time of diagnosis the median distance for the 136th nearest neighbour was found to be 3.5 km. Diagnosis locations, whose 136th nearest neighbour was ≤3.5 km were classified in the ‘more densely populated’ category, and diagnosis locations, whose 136th nearest neighbour was >3.5 km were classified in the ‘less densely populated’ category. Analysis by population density was undertaken by considering pairs of cases including at least one from the ‘more densely populated’ category (‘more densely populated:any’ clustering pairs) and pairs of cases including at least 1 from the ‘less densely populated’ category (‘less densely populated:any’ clustering pairs). It should be noted that analyses by population density (especially the analyses of ‘less densely populated:any’ clustering pairs) are potentially subject to a strong diluting influence from edge effects because neither the ‘more densely populated’ areas nor the ‘less densely populated’ areas form a single spatially contiguous zone.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Our study included 128 cases of retinoblastoma, 287 cases of neuroblastoma, 92 cases of peripheral neuroectodermal tumours (76 of bone and 16 extra-skeletal), 229 cases of Wilms' tumour, 33 cases of hepatic tumour (31 hepatoblastomas and 2 hepatocellular carcinomas), 370 cases of sarcoma (comprising 105 osteosarcomas, 12 chondrosarcomas,159 rhabdomyosarcomas, 9 other bone tumours, and 85 other soft tissue sarcomas), 140 cases of germ cell tumour (comprising 57 gonadal germ cell tumours, 78 non-gonadal germ cell tumours, and 5 other gonadal tumours), and 110 rare miscellaneous neoplasms (comprising 12 adrenal cortical carcinomas, 21 nasopharyngeal carcinomas, 6 thyroid carcinomas, 36 melanomas and other malignant neoplasms of the skin, 31 carcinomas of other primary sites and 4 pleuropulmonary blastomas). There were 4 sib-pairs of retinoblastoma, but only one of these pairs was diagnosed <1 year apart, and none were born <1 year apart. There were no other like-diagnosis sib-pairs of cases.

Most of the evidence of space-time clustering occurred using place and time of birth (Tables I–IV. Table I shows that 2 diagnostic groups gave p < 0.05 using at least 2 of the 4 analysis methods and including a NN threshold version, based on time of birth and place of birth (Wilms' tumour; sarcomas). NN threshold analyses show that the clustering is not just a result of varying population density. There was no evidence of space-time clustering for other groups (p > 0.1, using all 4 methods).

Table I. SPACE-TIME CLUSTERING TESTS FOR THE MAIN DIAGNOSTIC GROUPINGS1
Disease groupKnox testK-function analysis5
Geographical distance2NN threshold4Geographical distance6NN threshold7
OESp3OESp3Ip8Ip8
  • 1

    Using time and place of birth for children ages 0–14 years from North West England and diagnosed during the period 1954–98. O, observed space-time pairs (cases are close in time if dates of birth differ by <1 year); E, expected space-time pairs; S, strength = {(O− E)/E} × 100 counts of pairs that are close in time and space; I, observed integral = ∫ R(s,t) ds dt, where R(s,t) = [K(s,t) − K1(s)K2(t)]/√[K1(s) K2(t)], K(s,t) = proportion of pairs whose distance apart is ≤t in time and ≤s in space, K1(s) = proportion of pairs whose distance apart is ≤s, and K2(t) = proportion of pairs whose distance apart is ≤t.

  • 2

    Cases are close in space if their locations are <5 km apart.

  • 3

    1-sided p-value derived from the Poisson distribution.

  • 4

    Cases are close in space if the locations of 1 (or both) was nearer than the other's 134th NN in the total data set.

  • 5

    Cases are close in time if dates differ by <t where t is in the range 1–15 months.

  • 6

    Cases are close in space if distances between their locations differ by <s where s is in the range 0.5–7.5 km.

  • 7

    Cases are close in space if either is within the distance to the Nth nearest neighbour of the other (in the total data set) where N is in the range 127–141.

  • 8

    p-value obtained by simulation (999 runs) with dates of birth randomly re-allocated to the cases in the analysis.

Retinoblastoma811.2−28.5%0.7854.84.5%0.52−17.050.94−11.400.73
Neuroblastoma4746.02.2%0.462927.45.9%0.40−1.260.5013.690.22
Peripheral neuroectodermal tumours33.8−21.5%0.5353.069.3%0.18n/an/a13.830.17
Wilms' tumour4835.834.2%0.033323.938.2%0.0419.170.0733.110.03
Hepatic tumours11.1−12.6%0.3210.738.4%0.51n/an/a−7.230.68
Sarcomas8262.630.9%0.015038.330.6%0.0443.330.00545.990.01
Germ cell tumours97.520.6%0.3385.836.9%0.23−2.020.5419.000.14
Rare miscellaneous neoplasms35.1−41.2%0.7532.522.7%0.44−3.400.5811.340.22

The 2 diagnostic groups that showed most evidence for space-time clustering based on time of birth and place of birth, i.e., Wilms' tumour and sarcomas, were studied further.

The group of Wilms' tumours (Table II) was divided into 2 age-groups: 1) ages less than the median age at the time of diagnosis (32 months or less); and 2) ages greater than (or equal to) the median age at diagnosis (33 months or more). This analysis showed that clustering was confined to those cases ages 33–179 months at diagnosis. Separate analyses were carried out for ‘male:any’ and ‘female:any’ pairs. More significant evidence of space-time clustering was observed for ‘male:any’ pairs (p < 0.05, for all methods) than for ‘female:any’ pairs. The ‘strength of clustering’ was markedly greater for ‘male:any’ pairs than for ‘female:any’ pairs. Finally, separate analyses by population density were carried out. For Wilms' tumour, significant space-time clustering was only present for ‘less densely populated:any’ pairs.

Table II. SPACE-TIME CLUSTERING TESTS FOR WILMS' TUMOUR1
Disease groupKnox testK-function analysis5
Geographical distance2NN threshold4Geographical distance6NN threshold7
OESp3OESp3Ip8Ip8
  • 1

    Using time of birth and place of birth, for: 1) children ages <median age at diagnosis; 2) children ages ≥median age at diagnosis; 3) ‘male:any’ pairs; 4) ‘female:any’ pairs; 5) ‘more densely populated:any’ pairs; and 6) ‘less densely populated:any pairs, O, observed space-time pairs (cases are close in time if dates of birth differ by <1 year); E, expected space-time pairs; S, strength = {(O− E)/E} × 100 counts of pairs that are close in time and space; I, observed integral = ∫ R(s,t) ds dt, where R(s,t) = [K(s,t) − K1(s) K2(t)]/√[K1(s) K2(t)], K(s,t) = proportion of pairs whose distance apart is ≤t in time and ≤s in space, K1(s) = proportion of pairs whose distance apart is ≤s, and K2(t) = proportion of pairs whose distance apart is ≤t.

  • 2

    Cases are close in space if their locations are < 5 km apart.

  • 3

    1-sided p-value derived from the Poisson distribution.

  • 4

    Cases are close in space if the locations of 1 (or both) was nearer than the other’s 134th NN in the total data set.

  • 5

    Cases are close in time if dates differ by <t where t is in the range 1–15 months.

  • 6

    Cases are close in space if distances between their locations differ by <s where s is in the range 0.5–7.5 km.

  • 7

    Cases are close in space if either is within the distance to the Nth nearest neighbour of the other (in the total data set) where N is in the range 127–141.

  • 8

    p-value obtained by simulation (999 runs) with dates of birth randomly re-allocated to the cases in the analysis.

Wilms' tumour(ages <33 months)118.234.2%0.2075.039.8%0.2410.510.184.010.36
Wilms' tumour(ages ≥33 months)218.8139.9%0.0003136.696.1%0.0240.350.00342.150.009
Wilms' tumour(male: any)3726.539.7%0.032818.254.1%0.0221.040.0434.330.03
Wilms' tumour(female: any)3527.527.2%0.102518.237.6%0.0714.700.1634.110.05
Wilms' tumour(more densely populated: any)3629.223.5%0.121613.122.5%0.249.180.1614.050.17
Wilms' tumour(less densely populated: any)1914.134.7%0.122415.456.3%0.0213.620.1639.170.03

The sarcomas group (Table III) was further examined by sub-type (osteosarcoma, chondrosarcoma, rhabdomyosarcoma, other bone sarcomas, other soft tissue sarcomas). There was no evidence for clustering amongst the bone tumours (all bone tumours or osteosarcoma by itself). The strongest clustering occurred amongst all soft tissue sarcomas (p < 0.05 using all 4 methods). Rhabdomyosarcoma by itself, however, was also statistically significant (I = 25.16, p = 0.03, using the geographical distance version of the K-function method; and I = 39.99, p = 0.03, using the NN threshold version of the K-function method). There was no indication of any cross-clustering between the soft tissue sarcomas and the bone tumours because the results for combined groups (Table I) were weaker than for all soft tissue sarcomas. Therefore, we studied the group that showed significant evidence of space-time clustering in more detail: all soft tissue sarcomas.

Table III. SPACE-TIME CLUSTERING TESTS FOR THE MAIN DIAGNOSTIC GROUPS OF THE SARCOMAS1
Disease groupKnox testK-function analysis5
Geographical distance2NN threshold4Geographical distance6NN threshold7
OESp3OESp3Ip8Ip8
  • 1

    Using time of birth and place of birth. O, observed space-time pairs (cases are close in time if dates of birth differ by <1 year); E, expected space-time pairs; S, strength = {(O − E)/E} × 100 counts of pairs that are close in time and space; I, observed integral = ∫R(s,t) ds dt, where R(s,t) = [K(s,t) − K1(s)K2(t)]/√[K1(s) K2(t)], K(s,t) = proportion of pairs whose distance apart is ≤t in time and ≤s in space, K1(s) = proportion of pairs whose distance apart is ≤s, and K2(t) = proportion of pairs whose distance apart is ≤t.

  • 2

    Cases are close in space if their locations are <5 km apart.

  • 3

    1-sided p-value derived from the Poisson distribution.

  • 4

    Cases are close in space if the locations of 1 (or both) was nearer than the other's 134th NN in the total data set.

  • 5

    Cases are close in time if dates differ by <t where t is in the range 1–15 months.

  • 6

    Cases are close in space if distances between their locations differ by <s where s is in the range 0.5–7.5 km.

  • 7

    Cases are close in space if either is within the distance to the Nth nearest neighbour of the other (in the total data set) where N is in the range 127–141.

  • 8

    p-value obtained by simulation (999 runs) with dates of birth randomly re-allocated to the cases in the analysis.

Osteosarcoma, chondrosarcoma and other bone tumours109.64.7%0.4834.2−28.0%0.602.230.415.750.34
Osteosarcoma97.225.8%0.2932.79.4%0.5210.450.1717.700.13
Rhabdomyosarcoma and other soft tissue tumours3623.652.5%0.012817.263.0%0.0148.000.00161.940.002
Rhabdomyosarcoma1713.031.2%0.16128.738.4%0.1725.160.0339.990.03

The group of all the soft tissue sarcomas (Table IV) was also divided into 2 age-groups: 1) ages less than the median age at the time of diagnosis (48 months for rhabdomyosarcoma; and 92 months for other soft tissue sarcomas); and 2) ages greater than (or equal to) the median age at diagnosis (for the sub-diagnostic group).

Table IV. SPACE-TIME CLUSTERING TESTS FOR RHABDOMYOSARCOMA AND OTHER SOFT TISSUE TUMOURS1
Disease groupKnox testK-function analysis5
Geographical distance2NN threshold4Geographical distance6NN threshold7
OESp3OESp3Ip8Ip8
  • 1

    Using time of birth and place of birth, for: 1) children ages <median age at diagnosis; 2) children ages ≥median age at diagnosis; 3) ‘male:any’ pairs; 4) ‘female:any’ pairs; 5) ‘more densely populated:any’ pairs; and 6) ‘less densely populated:any pairs’ O, observed space-time pairs (cases are close in time if dates of birth differ by <1 year); E, expected space-time pairs; S, strength = {(O − E)/E} × 100 counts of pairs that are close in time and space; I, observed integral = ∫ R(s,t) ds dt, where R(s,t) = [K(s,t) − K1(s) K2(t)]/√[K1(s) K2(t)], K(s,t) = proportion of pairs whose distance apart is ≤t in time and ≤s in space, K1(s) = proportion of pairs whose distance apart is ≤s, and K2(t) = proportion of pairs whose distance apart is ≤t.

  • 2

    Cases are close in space if their locations are < 5 km apart.

  • 3

    1-sided p-value derived from the Poisson distribution.

  • 4

    Cases are close in space if the locations of 1 (or both) was nearer than the other's 134th NN in the total data set.

  • 5

    Cases are close in time if dates differ by <t where t is in the range 1–15 months.– 6Cases are close in space if distances between their locations differ by <s where s is in the range 0.5–7.5 km.

  • 7

    Cases are close in space if either is within the distance to the Nth nearest neighbour of the other (in the total data set) where N is in the range 127–141.

  • 8

    p-value obtained by simulation (999 runs) with dates of birth randomly re-allocated to the cases in the analysis.

Aged < median for specific sub-group96.244.4%0.1865.215.4%0.4213.890.1211.220.25
Aged ≥ median for specific sub-group86.132.2%0.2643.418.3%0.4413.300.1413.560.23
Male: any2819.742.2%0.052515.165.3%0.0140.390.00563.490.001
Female: any2216.136.9%0.091710.660.5%0.0436.250.0144.000.02
More densely populated: any3217.979.3%0.002168.393.2%0.0148.480.00165.550.003
Less densely populated: any1210.613.2%0.371912.552.6%0.0523.370.0336.930.03

This analysis did not show statistically significant clustering in either age group, due to the reduction in the numbers of cases. Second, separate analyses were carried out for ‘male:any’ and ‘female:any’ pairs. Significant evidence of space-time clustering for both ‘male:any’ and ‘female:any’ pairs was observed. The ‘strength of clustering’ was slightly more marked for males than for females. Finally, separate analyses by population density were done. Space-time clustering was more significant and the ‘strength of clustering’ was somewhat greater for ‘more densely populated:any’ pairs.

With respect to other space-time combinations, for germ cell tumours, there was one significant result, based on time of diagnosis and place of birth (I = 39.94, p = 0.02, using the NN threshold version of the K-function method), which is likely to be due to chance, because this result was not replicated by any other method. All other diagnostic groups gave p > 0.05 using all 4 methods.

There were significant results in 2 diagnostic groups, based on time of birth and place of diagnosis for: Wilms' tumour (S = 40.2%, p = 0.04, using the NN threshold version of the Knox test); sarcomas (I = 26.39, p = 0.03, using the geographical distance version of the K-function method) and rare miscellaneous neoplasms (I = 26.70, p = 0.04, using the NN threshold version of the K-function method). All other diagnostic groups gave p > 0.05 using all 4 methods.

The single results for Wilms' tumour and the sarcomas are likely to be due to cases who have not changed address between birth and diagnosis. The single result for the rare miscellaneous neoplasms is due to a single clustering pair for nasopharyngeal carcinoma (S = 1411.1%, p = 0.06, using the geographical distance version of the Knox test, and S = 2166.7%, p = 0.04, using the NN threshold version of the Knox test).

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

The analyses presented here have been carried out using rigorous statistical methods on high quality incidence data. The MCTR retains pathological material and can therefore review diagnoses in light of developments in histopathological techniques and changes in diagnostic classification. Periodic diagnostic review of cases ensures accuracy and consistency of diagnoses over time. Our study is entirely novel and indeed it would not have been possible for other datasets. Space-time clustering has been identified for two distinct diagnostic groups among the non-CNS solid tumours. Specifically, clustering was present amongst cases of Wilms' tumour and soft tissue sarcoma. During the period 1954–98, there were 287 cases of neuroblastoma, 255 cases of soft tissue sarcoma and 232 cases of Wilms' tumour in the MCTR area.18 There was, however, no evidence for space-time clustering amongst cases of neuroblastoma. Thus, the findings for the Wilms' tumours and the soft tissue sarcomas can not be attributed solely to the numbers of cases analysed.

The most statistically significant space-time clustering was restricted to space-time pairs involving time of birth and place of birth. There was little evidence for space-time clustering based on time of diagnosis and place of diagnosis. A large number of children (48%) moved residence between birth and diagnosis. For cases of Wilms' tumour and soft tissue sarcoma, this means that cases who lived close to one another around the time of birth had dates of birth close in time and, hence, shared the same spatial-temporal environment at birth, and, most probably, peri-conceptionally and during gestation. They did not necessarily share the same spatial-temporal environment at diagnosis. For both Wilms' tumours and the soft tissue sarcomas, these results provide support for a putative aetiological mechanism involving a pre-natal or a peri-natal exposure to an environmental agent that contributes to the onset of the disease. Because there was little evidence for space-time clustering based on time of diagnosis and place of birth, this indicates that the latent periods, for the development of both Wilms' tumours and soft tissue sarcomas, are variable (i.e., not of constant length in time).

For Wilms' tumour, the analyses have found 3 interesting results. First, the space-time clustering was confined to the older group of cases (the 50% of ages 33 months or more). Second, the space-time clustering was much stronger for space-time pairs involving at least 1 male case than for space-time pairs involving at least 1 female case. Third, space-time clustering was only present for the ‘less densely populated:any’ pairs of cases. The first finding is consistent with an environmental factor acting during gestation rather than affecting parental germ cells. Observations of morphological patterns of Wilms' tumour precursor lesions suggest that sub-groups of Wilms' tumours arise at earlier or later time points during gestation with correspondingly earlier or later ages of clinical onset.19 Our data would be consistent with the clustering being related to precursor lesions arising later in gestation. In this context it would be of interest to re-examine the morphological features of the tumours among the clustering pairs of cases. The second finding is consistent with differences in susceptibility to an environmental agent (or agents) between males and females. The third finding is consistent with an agent that is present in more ‘rural’ areas, for limited time periods.

For the soft tissue sarcomas, the analyses have found 3 interesting results. First, the space-time clustering was only apparent in cases of soft tissue sarcoma and not among cases of bone tumour. Second, the space-time clustering was stronger for space-time pairs involving at least 1 male case than for space-time pairs involving at least 1 female case. Third, space-time clustering was stronger for ‘more densely populated:any’ pairs than for ‘less densely populated:any’ pairs. The first finding is consistent with differences in the aetiology of soft tissue sarcomas and bone tumours, with evidence of shared environmental factors among rhabdomyosarcoma and other soft tissue sarcomas. The second finding is consistent with differences in susceptibility to an environmental agent (or agents) between males and females. The third finding is consistent with an environmental agent that is more evident in ‘more densely populated’ areas, for limited periods, than in ‘less densely populated’ areas.

Agents that may cause localised variations in incidence include airborne infections, electrical power lines, traffic pollution, agrochemicals and factory emissions. The patterns of incidence, as exhibited by the findings of space-time clustering, are not consistent with a constant sustained exposure either geographically or over time because the statistical tests are designed to detect marked peaks in incidence in time and space. Thus putative causative agents such as power lines and constant factory emissions are not supported by this analysis.

The agents responsible for the observed clustering are much more likely to exhibit patterns of temporary occurrence at many points in time and space. Potential agents include: traffic pollution, modulated by changing weather patterns; certain agrochemical exposures, which have a seasonal and variable application, such as pesticides; diet including maternal consumption of fresh fruit and vegetables that may be a factor in itself or that may be seasonally contaminated by pesticides or other chemicals; and infectious agents.

For Wilms' tumours, the findings, in particular of space-time clustering only for less densely populated areas, are consistent with an environmental exposure such as pesticides. Three recent studies from Brazil,20 Norway21 and the UK22 have all linked parental exposure to pesticides with statistically significantly increased risk of Wilms' tumour in the child. For the soft tissue sarcomas, the finding that space-time clustering was confined to pairs of cases involving those from more densely populated areas would point to environmental factors that are more likely to be present in such places. Obvious candidate agents would include traffic pollution and infections. With respect to the latter the risk of person to person transmission would be greater than in more sparsely populated places.

The finding of one space-time clustering pair for nasopharyngeal carcinoma, which is an extremely rare disease in children in the UK, is of interest. One putative environmental exposure is infection with Epstein-Barr virus.23

It is possible for apparent space-time clustering to arise due to population shifts. It is not possible in our present study to adjust for such shifts, using the method of Kulldorff and Hjalmars,24 because this would need post-code denominator counts, which are not available. There is a danger, however, that adjustment for population shifts may produce p-values that are too conservative. Furthermore, we believe that the significant space-time clustering results, for Wilms' tumour and soft tissue sarcomas cannot be explained away as being artefactual. The lack of space-time clustering for other diagnostic groups, the specificity of the significant results (in terms of age, diagnostic group and centering on time of birth, place of birth), and the differences found between males and females point to the involvement of specific effects and would argue against the space-time clustering being artefactual.

Previous studies of clustering amongst non-CNS solid tumours have suffered from methodological limitations.8, 11 A study, from the UK, covering the period 1953–6425 did not find clustering amongst solid tumours, but that study did not consider diagnostic sub-type.

In summary, we have uniquely found evidence of space-time clustering cases of Wilms' tumour and soft tissue sarcoma. Clustering was most apparent amongst space-time pairs involving time of birth, place of birth. The results are consistent with a role for environmental agents in at least a proportion of cases in these particular diagnostic groups, with mechanisms acting pre-natally, or around the time of birth. There is also a possibility, however, that these results represent chance findings. Our study has only been made possible by the availability of high quality and consistent population-based diagnostic and residential address data. The associations need to be confirmed by other studies, using indepen dent data. Future studies should also consider candidate environmental agents.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

The Manchester Children's Tumour Registry is supported by Cancer Research UK. J.M. Birch is Cancer Research UK Professorial Fellow and O.B. Eden is Cancer Research UK Professor of Paediatric Oncology at the University of Manchester. We thank Mrs. E.A. Dale, Mrs. D.A. Elliott, Mrs. J.F. Hogg and Mr. C. Nikolaisen for all their hard work on data processing and verification.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES