To explore the time trend and geographical distribution of childhood leukaemia incidence over the territory of the Italian region of Sardinia.

All hospitals departments, diagnostic centres and social security agencies in Sardinia were regularly screened in 1974–2003 to identify, register and review the diagnoses of incident cases of haematological malignancies (HM).

The whole child population aged 0–14 resident in Sardinia.

Incidence and time trend of childhood HM and childhood acute lymphoblastic leukaemia (ALL) over the study period, and use of Bayesian methods to plot the probability of areas with excess incidence on the regional map.

Overall, 675 HM cases, including 378 ALL cases, occurred among children aged 0–14 years resident in Sardinia in 1974–2003, with an incidence rate of 6.97×10^{-5} (95% CI 6.47 to 7.51) and 3.85×10^{-5} (95% CI 3.48 to 4.26), respectively. Incidence of HM and ALL showed an upward trend along the study period especially among females. Three communes out of the 356 existing in 1974, namely Ittiri, Villa San Pietro and Carbonia, stand out as areas with excess incidence of HM and ALL in particular and another, Carloforte, for ALL only.

Our results might serve as convincing arguments for extending the coverage of routine cancer registration over the whole Sardinian population, while prompting further research on the genetic and environmental determinants in the areas at risk.

We explored time trends of childhood haemolymphatic malignancies in Sardinia, Italy over 30 years using a regional registry.

Bayesian statistics allowed us to describe the geographical pattern of childhood haemolymphatic malignancies in Sardinia.

The causes of the observed increasing incidence and clustering of childhood acute lymphoblastic leukaemia are still unknown.

Childhood leukaemia is known to manifest itself in time/space clusters, which has led to speculations whether chance or specific aetiological agents might be responsible.

The region of Sardinia is the second largest Mediterranean island, and it is well known for the genetic peculiarities of its population, resulting from millennia of isolation and pressure from malaria.

A cancer registry was active from the 1990s only in northern Sardinia, covering approximately one-third of the regional population. Therefore, from the very beginning of the ‘A. Businco’ Oncology Hospital in 1974 in Cagliari, the capital and major urban centre in Sardinia, the chief oncohaematologist registered and updated all cases of oncohaematology malignancies incident in the regional territory up to 2003, with the collaboration of all the clinical, surgical and pathology departments, social security agencies and health authorities, as described elsewhere.

A detailed description of the database of haemolymphatic malignancies we used in this study can be found elsewhere.

The resident population of each commune by gender and age groups was available from the 1971, 1981, 1991 and 2001 population censuses. To estimate the resident population in the intercensal years, we extended each census data 4 years onwards and 5 years backwards. The standardised annual incidence rate of all HM, and ALL among the resident child population was calculated along the study period using the person-years of the total regional population for each gender, and age group (0–4, 5–14 years) as the standard. Annual incidence rates were plotted by year, and the Pearson’s correlation coefficient was calculated based on the equation describing the linear regression.

To explore the geographical distributions of HM and ALL, the most common cancer among children under the age of 14, we have used a Bayesian approach to characterise the incidence in each commune, the term used in some European countries to indicate the smallest local administrative unit, that is, town, city or village, which has its own council and governs a territory. The map of the territorial borders of the 356 individual communes existing in 1974 over the region of Sardinia is available online (https:\\umap.geonue.com/en/map/confini-e-dati-statistici-dei-comuni-ditalia_297#8/40.102/8.973). Another 21 were created along the 30 years of follow-up by separation from the original administrative unit, which we kept considering as one for consistency with the initial map. The objective is to identify those communes for which there is a high probability that the incidence rate is higher than average. We interpret this to be the communes for which the incidence probability exceeds some critical value determined from data for the region.

The Bayesian approach

where

The prior summarises what is known about the standardised incidence rate, η, considering solely the regional data. In the definition, the following information is used:

The incidence probability is a number in the interval between 0 and 1.

The condition is rare, implying a low value for the incidence probability.

It is acceptable to use data from the whole island when analysing data at the individual commune level.

A few communes might have incidence probabilities higher than the rest of the island for genetic or geographical factors.

This information is summarised with a

Plot of the prior probability distribution of the haematological malignancies in Sardinia, Italy, the likelihood function, and the posterior probability of a commune taken as an example. There is not a vertical scale to stress the relevance of the shape of the curve. The likelihood function has been rescaled. The thin vertical line indicates the critical value, and it shows that a substantial portion of the posterior probability curve lies beyond it.

Whether an individual child succumbs or not to the disease in a given year is a binomial process. Therefore, the number of cases observed k, in a population of size n, measured in child-years, is best described by a binomial distribution, which, since

where

The posterior probability combines the prior knowledge with the information that can be obtained from the data via the likelihood function, according to Bayes rule, as shown in

If we carry out the same type of analysis for the whole region, the posteriors are almost indistinguishable for any reasonable choice of prior. So, we used a uniform prior to generate the posterior function. Different priors at commune and regional levels are used because different information is available at the two levels. At the commune level, we can use information that was derived at regional level. We then defined the critical value as the value such that the probability that the incidence rate for the region is less than the critical value is 0.999, that is,

where D is the data for the whole region. The data sets d referred to earlier are for the communes and are not identical to the equivalent data D for the region. The choice of 0.999 is arbitrary, but the subsequent analysis is not sensitive to the value chosen. In

Critical values

Group | Critical value of η ( | Critical incidence rate |

0–4 female | 0.0000850 | 8.50 per 100 000 |

5–14 female | 0.0000735 | 7.35 per 100 000 |

0–4 male | 0.0001110 | 11.10 per 100 000 |

5–14 male | 0.0000930 | 9.30 per 100 000 |

We can now calculate the probability for the hypothesis

For example in

The calculation of the probability of exceeding the relevant critical incidence rate for all childhood haemolymphatic malignancies and childhood acute leukaemia for each commune was carried out using bespoke python code. To allow simpler coding and calculations the critical rates for each group based on the regional data are calculated first, followed by the processing of the commune data. The results are plotted in the regional map (

Map of 1974–2003 childhood haemolymphatic cancer incidence in Sardinia, Italy. P values of exceeding the threshold incidence rate are presented by commune with the following colour scales: white <0.17, pale grey 0.17–0.5, medium grey 0.5–0.75, dark grey 0.75–0.95, black >0.95. (A) All haemolymphatic cancer; (B) acute lymphoblastic leukaemia.

No patient or public was involved.

Incident cases of haemolymphatic cancer among children of age 0–14 in Sardinia, Italy in 1974–2003, by histology and gender

Histology | Males N (%) | Females N (%) | Total N (%) |

Acute lymphatic leukaemia | 209 (53.0) | 169 (60.2) | 378 (56.0) |

Acute myeloid leukaemia | 42 (10.7) | 29 (10.3) | 71 (10.5) |

Hodgkin lymphoma | 49 (12.4) | 27 (9.6) | 76 (11.3) |

Non-Hodgkin lymphoma | 81 (20.6) | 37 (13.2) | 118 (17.5) |

Other lymphoproliferative diseases | 6 (1.5) | 11 (3.9) | 17 (2.5) |

Other myeloproliferative diseases | 7 (1.8) | 8 (2.8) | 15 (2.2) |

Total | 394 (100.0) | 281 (100.0) | 675 (100.0) |

Over the 30 years of observation, the annual age-standardised incidence rate of childhood haemolymphatic cancer was 6.97×10^{-5} (95% CI 6.47 to 7.51), 7.93 (95% CI 7.19 to 8.75) among males and 5.96 (95% CI 5.30 to 6.69) among females, with a male/ female ratio of 1.41. The 378 childhood ALL cases corresponded to an overall incidence rate of 3.85×10^{-5} (95% CI 3.48 to 4.26), 4.14 (95% CI 3.62 to 4.75) among males and 3.55 (95% CI 3.05 to 4.12) among females, with a male to female ratio of 1:24.

Graphs in ^{-5} per year on average along the study period (p=0.004). The increasing trend was observed particularly among females (0.18×10^{-5} per year, p=0.015). ALL incidence did not increase in the overall population and among males (p=0.868, and p=0.171, respectively), but it did show an increase among females aged 5–14 (0.09×10^{-5} per year, p=0.039).

1974–2003 annual incidence rate of total childhood haematological malignancies (cHM) and acute lymphoblastic leukaemia (cALL) in Sardinia, Italy, by gender (A, B) and overall (C1, C2).

The same three communes standing out for HM, also showed a probability above 95% that the incidence of ALL was above the critical rate. Such probability was 0.976 for Carbonia (based on 18 cases), 0.983 for Ittiri (based on 9 cases) and 0.973 for Villa San Pietro (based on 5 cases,). For ALL, a probability above 95% was also observed for Carloforte (0.952, based on six cases). Another six showed a probability between 75% and 94%. Again, these were scattered all over the regional territory, and did not show a tendency to cluster in specific areas.

Geographical distribution of the 1974–2003 probability of childhood acute lymphoblastic leukaemia incidence above the threshold by commune in Sardinia, Italy. (A) 0–4 years (M+F); (B) 5–14 years (M+F); (C) females (age 0–14); (D) males (age 0–14).

Our results show that the incidence of childhood cancer of the haemolymphopoietic system, and particularly ALL, increased in the Italian region of Sardinia among females; among male children we observed and upward time trend of all haemolymphatic malignancies, while that of ALL was weaker and not significant. The slope of the regression line was apparently steeper for cases occurring between age 5 and 14, which would point out to a postnatal exposure, or to internal migration effects. Apart from chance, we do not have a clear explanation for the gender-specific excesses we observed in several communes.

The excess ALL incidence in Carbonia, Ittiri and Villa San Pietro had been the object of specific cluster analyses,

Other possible explanations include polymorphisms of genes implicated in the pathways leading to childhood leukaemia. Most genetic susceptibility studies have focused on immune function, response to infection, one-carbon metabolism, membrane transport, xenobiotic phase I and phase II metabolism of environmental carcinogens, reactive oxygen species deactivation and DNA repair enzymes.

However, although Sardinia is well known for the genetic peculiarities of its population,

One possible alternative explanation of the clustering of ALL cases in specific areas and within specific time frames might be the special ability of the local family physician/s in detecting the disease, and in referring the sick children to the haematology departments. Such ability might have increased along the study period, generating a spurious increasing trend,

Concern in interpreting our results might be raised because of the wide range in population size between communes, from less than 100 to more than 150 000, which would be reflected by the childhood population at risk. To minimise the number of false positive findings related to random variations in the observation of a small number of events, we first set at p<0.001 the critical value in the prior probability distribution of the standardised incidence rate with reference to the prior expectation. Then, in each commune, we considered the posterior probability of exceeding that critical value.

An advantage of our study is that the diagnoses were all reviewed by the same expert haematologist (GB), thus preventing bias due to the varying diagnostic ability by time and geographical area, and minimising and spreading equally the probability of misdiagnosis over the whole region and along the study period.

Along the period covered by our follow-up study, a cancer registry was active only in northern Sardinia, covering two local health units and approximately 30% of the regional population. A second cancer registry started operating in 2006, 3 years after the end of recruiting incident childhood HM cases for the data base we herein analysed, covering two local health units in the central-eastern area of the region and extending the coverage to 43% of the regional population.

This is the first report describing time trends over 30 years and the geographic pattern of haemolymphatic malignancies in Sardinia, a special region for studying the interaction between gene polymorphisms and environmental factors. Our results will hopefully prompt further research, and might serve as convincing arguments for extending the coverage of cancer registration over the whole Sardinian population, and for distributing the necessary paediatric haematology resources to better match the local needs.

The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

The depiction of boundaries on this map does not imply the expression of any opinion whatsoever on the part of BMJ (or any member of its group) concerning the legal status of any country, territory, jurisdiction or area or of its authorities. This map is provided without any warranty of any kind, either express or implied.

None declared.

Obtained.

The Ethics Committee of the Cagliari University Hospital approved the use of these data for the purposes of scientific publication (Protocol No. PG 2019/18070, 18 December 2019).

Not commissioned; externally peer reviewed.

Data are available on reasonable request. Data are preserved in the archives of the department of Medical Sciences and Public Health of the Cagliari University in aggregated form, and they are publicly available as such. Please contact PC (pcocco@unica.it) for any request.