Childhood leukaemia and population movements in France, 1990–2003

In a national study, we investigated the incidence of childhood leukaemia (CL) over a 14-year period in France in relation to several measures based on the proportion of individuals who changed address between the last two national censuses. A positive association was found with the proportion of migrants who came from a distant place. The further the migrants came, the higher was the incidence of leukaemia, particularly among children aged 0–4 years in ‘isolated’ communes at the time of diagnosis (RR=1.4, 95% CI: 1.1,1.8 in the highest category of migration distance). Although the role of the population density was less obvious, a more marked association was found above a certain threshold. No association with the proportion of commuters was observed.

About 20 years ago, a cluster of cases of childhood leukaemia (CL) in the village of Seascale (England) revived interest in an infective basis by leading to the population mixing hypothesis (Kinlen, 1988). This proposed that large population influxes into rural or isolated areas are conducive to an epidemic of the underlying infection(s) of which CL is postulated to be a rare consequence. Childhood leukaemia mortality or incidence has been examined in rural areas after receiving such influxes (Kinlen, 1988(Kinlen, , 2006Kinlen et al, 1990Kinlen et al, , 1993Kinlen et al, , 2002Langford, 1991;Kinlen and John, 1994;Kinlen and Petridou, 1995;Koushik et al, 2001;Boutou et al, 2002). These studies, mainly focusing on historically documented and specific rural population increases (with consequent increases in population density), showed significant increased relative risks, in the range of 1.5 -4.7, in the places with the highest population increases, compared to the reference group.
Elsewhere, population mixing has mainly been defined not as population increases but as increases in the proportion of residents who changed address over a specified period, or in the year before a census, and not specifically in rural areas (Stiller and Boyle, 1996;Dickinson et al, 2002;Parslow et al, 2002;Law et al, 2003;Labar et al, 2004;Nyari et al, 2006;Rudant et al, 2006). Overall, the studies seem to favour a positive association between the incidence of, or mortality associated with, CL and the highest proportions of migrants, although a few have reported a negative association (Parslow et al, 2002;Law et al, 2003). A few of these also found that the further the migrants came, the higher was the incidence of CL (Stiller and Boyle, 1996;Dickinson et al, 2002;Rudant et al, 2006). In the past, a positive association was found with childhood acute lymphoblastic leukaemia (ALL), particularly in isolated areas with a population density greater than a given threshold at the time of birth.
In the present study, we investigated CL incidence on a national scale in relation to the proportion of individuals who changed address over a 14-year period, focusing on the isolation status of residence at diagnosis.

MATERIALS AND METHODS
In 1999, mainland France consisted of 36 565 communes, the smallest administrative unit, 95 départements and 22 régions. Owing to commune merging or splitting before 1999, information was sometimes available only for merged communes. The whole of the country was finally divided into 36 347 communes or combinations thereof, which, for simplicity, are still referred to as communes. This ecological study was conducted on the national scale in France. All cases of CL registered in the French National Registry of childhood haematopoietic malignancies (Clavel et al, 2004) and diagnosed between 1990 and 2003 were included. Each case was associated with its commune of residence at the time of diagnosis.
To define the status of each commune in 1990, two classifications by the French National Institute for Statistics and Economic Studies (INSEE) were considered, the Urban Zoning Classification and the Urban Unit classification (see Rudant et al, 2006). From the former, based on the influence and dependence in terms of employment, 'attracting' communes were those that attracted a substantial number of commuters from other communes for jobs. On the basis of this classification, 'dependent' communes were then defined as those with at least 40% of their economically active population working outside, in 'attracting' communes. Overall, 'attracting' and dependent communes accounted for about 9 and 29% of communes, respectively.
The Urban Unit classification classified the French communes by the population size of the Urban Unit, defined as a group of communes in which the distance between dwellings was not more than 200 m.
On the basis of those criteria, a commune was then considered a priori 'isolated' if it belonged to an Urban Unit with a population of less than 5000 or to a rural commune, and was neither an 'attracting' nor 'dependent' commune in 1990. A few communes called 'urban' by INSEE were therefore included in the 'isolated' commune group, but they are small, not 'attracting', not 'dependent' and situated in remote areas. The 'isolated' communes accounted for 61% of communes and 18.5% of French population in 1990.
The population movements between the last two national population censuses were used as proxy measures of population mixing. The number of individuals who moved from a commune to another between 1990 and 1999 was the initial focus. To take account of the distance covered by the migrants, the numbers of people in a commune outside the département, outside the région or in a distant commune in 1990 were considered. A commune was considered distant from another, defined a priori, if at a greater distance than the median distance covered by the migrants between 1990 and 1999 (100 km overall, 60 km for 'isolated' and 120 km for 'non-isolated' communes). All these measures were considered as proportions of the 1999 population.
We also considered the weighted average migration distance d i defined by in which m ki is the number of migrants who moved from a commune k to given commune i, m i the total number of immigrants in commune i and d ki the distance between the communes i and k. This measure was introduced in regression models with adjustment for the overall proportion of migrants. Lastly, the proportion of regular commuters was evaluated, from the 1999 census data, as the sum of economically active people in a given commune but working outside it and those living outside it but working in it, divided by the 1999 population of the commune.

Statistical analyses
The INSEE provided population estimates by commune for the two census years 1990 and 1999, and population estimates by département from 1990 to 2003 (a French département consists of about 385 communes on average). The populations between the two censuses and after the year 1999, on the commune scale, were then estimated as follows: 1999 pôp depðiÞ;1999 Ã pôp depðiÞ;j in which i and j are the index numbers for a given commune and a given year, respectively, pôp i,j is the population estimate for commune i in year j and pôp dep(i),j is the population estimate for the relevant département for the year j (source INSEE). The numbers of CL cases expected in each French commune over the study period 1990 -2003 were based on the population estimates and the national 5-year age-specific incidence rates provided by the RNHE.
The associations between the various measures of population movements and CL incidence were investigated using Poisson regression models. Each variable was incorporated in the statistical models as a three-category variable, the breakpoints being chosen a priori on the basis of the quintile distribution of the expected numbers, to isolate the highest quintile. The first two groups were then defined so that they each included 40% of the total number of expected cases. The Poisson regression models estimated the SIRR, the ratio of the SIR (standardised incidence ratio) for a category and that of the baseline category. Cut points were defined separately for all, 'isolated' and 'non-isolated' communes. However, when variables varied only slightly with isolation status, in terms of distribution of the expected cases, common cut points for the various groups were adopted. Additional analyses were stratified by age (0 -4, 5 -9 and 10 -14 years), isolation status of the communes ('isolated', 'non-isolated') and population density. The latter was transformed into a threecategory variable, each group containing similar expected numbers at ages 0 -14 years, whereas the 'isolated' group was only split at the median because of the lower expected numbers. The breakpoints were then 180 and 1729 inhabitants per square kilometre (inh km À2 ) in the overall analyses, 394 and 2607 inh km À2 in the 'non-isolated' group and 45 inh km À2 in the 'isolated' group. The analyses of weighted average distance of migration were systematically adjusted for the total proportion of migrants. Lastly, some analyses were performed for common B-cell lymphoblastic leukaemia (common B-cell ALL).
All analyses were performed with the SAS s and R software.

RESULTS
Mainland France was divided into 36 347 communes, 22 252 of which were 'isolated' (60%) and inhabited by only 18.5% of the  1990 (58.5 million in 1999), with about 20% aged up to 14 years. Commune populations range from 1 to more than 2 000 000 inh, with median and mean of about 350 and 1500 inh, respectively. Similarly, population density ranged from 1 to over 22 000 inh km À2 (median 33). As expected, the 'isolated' communes were characterised by a smaller population and a lower density. Overall, 19 657 175 people changed address between the two censuses, 18 336 140 coming from a commune in mainland France of known isolation status. About 18% of the migrants moved to an 'isolated' commune, the median distance covered being about 100 km overall, 60 and 120 km for 'isolated' and 'non-isolated' communes of destination, respectively (Table 1). Overall, half the communes were subject to a total population influx of between 26 and 38% of their 1999 population (Table 1). 'Non-isolated' communes had a slightly greater proportion of migrants than 'isolated' communes, the median proportions from another département or région being 11.4 and 7.8%, respectively, with slightly higher proportions in 'isolated' than in 'non-isolated' communes. The proportions of migrants in a distant commune in 1990 were mainly below 10% and were slightly higher in 'isolated' communes. These showed a higher weighted average migration distance than the 'non-isolated' communes (Table 1). Lastly, half the communes had between 30.5 and 45% regular commuters, and the proportions were markedly greater in the 'non-isolated' group.
Of the 6288 children aged 0 -14 years with CL in the National Registry of Childhood Haematopoietic malignancies between 1990 and 2003, commune of residence at diagnosis was missing for 14 (0.2%) leaving 6274 in the present analyses, half aged less than 5 years at diagnosis. A total of 4090 cases of common B-cell ALL were registered over 1990 -2003, 65% of the total. Overall, 5220 cases of CL (83% of the total) were diagnosed in a 'non-isolated' commune.
Overall, no association with the proportions of all migrants, of migrants from another département or of commuters was found (Table 2). However, an increased incidence was observed with the proportion of migrants who came from another région, particularly in the 'isolated' commune group, a proportion greater than 13% being associated with a significant 18% increase in CL incidence. Similarly, there was a slight increase in incidence in the group with the highest proportions of migrants from distant communes (SIRR ¼ 1.08 [1.02; 1.16]). This was seen in both isolation groups, greater in 'isolated' communes although of borderline significance (SIRR ¼ 1.14 [0.97; 1.33]). Adjusting for the total proportion of incomers, the weighted average migration distance was weakly associated with CL incidence, more marked for the 'isolated' communes (SIRR ¼ 1.17 [1.02; 1.34] for a distance in the range 82 -185 km and 1.24 [1.02; 1.5] for a distance greater than 185 km).
Overall, most of the above results were found, mostly reinforced, in children aged 0 -4 years ( Table 3). The highest proportion of migrants from another région was associated with a 20% increase in CL incidence. In the 'isolated' commune group, more than 12% of migrants from a distant commune showed a 25% increase in CL (SIRR ¼ 1.25 [1.01; 1.56]). Similarly, allowing for the total proportion of migrants, the 'isolated' communes with a weighted average migration distance of more than 185 km were at a significant increased risk (SIRR ¼ 1.41 [1.09; 1.83]). Some of the associations were still observed in the 5-to 9-year age group, but none above that age group (results not shown). When stratifying by the population density, there was no association with the total proportion of migrants (results not shown) but a nonsignificant increase was found in the 'isolated' communes with a population density below 45 inh km À2 (SIRR ¼ 1.28 [0.91; 1.79] for proportions of migrants above 38%).
Considering all communes, whatever the proxy measure, based on the migration distance -from another région, or a distant commune or weighted average distance -the positive association with CL incidence at ages 0 -4 years seemed restricted to lowdensity communes (p180 inh km À2 ; In the 'isolated' commune group, the influence of population density was less clear-cut. However, a striking association with the weighted average migration distance was observed with a population density greater than 445 inh km À2 (SIRR ¼ 1.25 [0.97; 1.61] and SIRR ¼ 1.72 [1.23; 2.39] for a distance of 82 -185 km and over 185 km, respectively). The results observed in the 'nonisolated' group were quite similar to those observed overall, but of slightly smaller magnitude and at only borderline statistical significance.
When the analyses were restricted to common B-cell ALL, the results were quite similar but the role of the population density in 'isolated' communes became less clear (not shown). Although there were slight inconsistencies, the analyses for the two subperiods, 1990 -1996 and 1997 -2003, favoured a positive association between CL incidence and population movements. Using linear interpolation between the two censuses to estimate the populations on the commune scale between 1990 and 1999 and over 1999 did not change the results. Similarly, the results were unchanged when merged or split (about 200) communes were excluded.

DISCUSSION
Nationally, CL incidence was associated with the level of migrants in 'isolated' communes, more markedly above a given threshold of population density. Although of smaller magnitude, similar associations were seen in low-density 'non-isolated' communes. The associations were also more pronounced at ages 0 -4 years, but were not specific to common B-cell ALL. Findings in the two subperiods were mostly similar, more marked in 1997 -2003. Sensitivity analyses showed stability of the results with respect to the population estimates on the commune scale and their changes over time.
The present study examined the data for 14 years of national CL registration, judged as 99.2% complete (Clavel et al, 2004), so missing data were too small (0.2%) to have affected the results. It also benefited from INSEE's provision of details by urban status, and between-census population movements. The 10-year interval between censuses may have allowed any rapid and temporary changes in communes (e.g., wartime evacuation) to go undetected and introduce misclassifications, but there is no obvious reason E ¼ expected number of cases; SIRR ¼ SIR (standardised incidence ratio) ratio. SIRR estimated by Poisson regression. a X100 km for all communes; X60 km for 'isolated' communes and X120 km for 'non-isolated' communes. b 0.07 for all communes; 0.08 for 'isolated' communes; 0.06 for 'non-isolated' communes. c 0.12 for all communes; 0.14 for 'isolated' communes; 0.11 for 'non-isolated' communes. d adjusted for the proportion of migrants. e 0.42 for all communes; 0.33 for 'isolated' communes; 0,45 for 'non-isolated' communes. f 0.59 for all communes; 0.44 for 'isolated' communes; 0,60 for 'non-isolated' communes.
Childhood leukaemia in France S Bellec et al why these would be differential and related to migration distance. An increase in the total population may be more important under the population mixing hypothesis than a sustained high proportion of migrants. The measures of population movements considered in the present study, mostly based on changes in address between 1990 and 1999, provided no indication of whether a high level of migrants was new. Any of these areas with a substantial proportion of migrants at some time before 1990 would have been considered with areas where the arrival of migrants was new, although they may not be strongly expected to show an excess following the arrival of other migrants between 1990 and 1999. With regard to the infectious hypothesis, population movements may have a greater impact when they occur around the time of birth. Using place of residence at diagnosis may therefore be less accurate and might partly explain the weaker associations than those of Rudant et al (2006). Many investigations have been conducted of CL incidence in rural isolated areas that had been subject to extreme and unusual population influxes (Kinlen et al, 1990(Kinlen et al, , 1993(Kinlen et al, , 2002Kinlen and John, 1994;Kinlen and Petridou, 1995;Kinlen, 1988Kinlen, , 2006Dickinson and Parker, 1999;Koushik et al, 2001;Boutou et al, 2002;Wartenberg et al, 2004). Those studies, based on specific rural population increases, were mainly consistent in finding an increased risk of CL with an increasing proportion of incomers.
Unlike those studies that focused on new population increases, the present study considered the migrants who moved from a commune to another between the last two censuses with no account of whether the high levels of migration were new. Besides, the population variations between the census years were so skewed on the commune scale that this measure could not be considered in the present study. Several other studies have shown that areas with high proportions of migrants or marked changes in population were at a significantly higher risk of CL than other areas (Langford, 1991;Stiller and Boyle, 1996;Alexander et al, 1997;Dickinson et al, 2002;Labar et al, 2004;Nyari et al, 2006;Rudant et al, 2006). In the UK, Dickinson et al found an increased risk of disease only in urban census wards. Two large studies did not demonstrate any significant association with the overall proportion of people who changed address in the year before a census (not specifically in rural areas), but found instead a negative association with an index of mixing diversity (Parslow et al, 2002;Law et al, 2003).
In the present study, a positive nonsignificant association was found only for the subgroup of isolated communes with a low population density. The highly skewed distribution of the proportions of migrants, with an inter-quartile distance of about 10%, could have hampered detection of a potential association. Moreover, considering the overall proportion of individuals who moved from one commune to another, regardless of distance, may be too imprecise a measure.
Only three studies have investigated migration distance in relation to CL, either directly (Stiller and Boyle, 1996) or by considering the proportions of migrants from distant areas (other district, other region, outside the country) (Dickinson et al, 2002;Rudant et al, 2006); all found a positive association with migration distance.
We considered migration distance using three measures that were significantly correlated. The further the migrants came, the higher was CL incidence, particularly at ages 0-4 years in 'isolated' areas. These results were in line with the literature and provide some support for the infectious hypothesis of Kinlen. This involvement of migration distance may reflect the fact that the further the migrants had travelled, it was less likely that there was much previous contact with the local population and therefore of having a similar immune status with respect to the hypothetical viral agent.
The associations were found in both isolation status groups but only in communes with a population density below 394 inh km À2 for the 'non-isolated' commune group. This may stem from possible inaccuracy in the definition of the isolation status. Communes that were located in the vicinity of an urban area, and thus dependent on it for employment, were considered 'nonisolated', regardless of their population size or population density. It was considered, perhaps erroneously, that the people who lived in those communes were likely to share a common immunity with the people who lived in the urban pole. If the association between CL incidence and migration distance was mainly in rural or isolated locations, as implied by Kinlen, the results found in the 'non-isolated' communes could reflect misclassification with respect to the real isolation status of the communes.
The involvement of population density may also indicate that the evaluation of the isolation status of French communes, which was intended to identify communities with low background immunisation, may not have been sufficiently pertinent to the infectious hypothesis.
However, a striking increase in CL incidence was found in relation to the weighted average migration distance in the 'isolated' group, but only in communes with a density greater than 45 inh km À2 . This result, in line with the results of a French birth cohort study (Rudant et al, 2006), seems to favour a role for population density. There may be a threshold below which the suspected infection cannot spread because of the limited number of person-to-person contacts.
As in a previous study (Stiller and Boyle, 1996), no association with the proportion of commuters in the communes was found, although the definitions of commuting were slightly different. An investigation of increases of commuting in relation to CL found an 80% excess in locations with the greatest increases . With respect to the infectious hypothesis, an increase in commuting patterns over time may be more important than a sustained high level.
In contrast to Kinlen's work, based on specific rural influxes that increased total population and therefore its density, our study was based on migration measures (irrespective of population increases) over a 10-year period on a national scale, and took account of both isolation status and population density of the communes. The association of population movements with CL incidence in young children in 'isolated' areas at the time of diagnosis is consistent with the possible involvement of viral agents in this disease.