Risk of leukaemia in children infected with enterovirus:
a nationwide, retrospective, population- based,
Taiwanese-registry, cohort study
Jiun-Nong Lin, Cheng-Li Lin, Ming-Chia Lin, Chung-Hsu Lai, Hsi-Hsun Lin, Chih-Hui Yang, Fung- Chang Sung, Chia-Hung Kao
Introduction
Leukaemia is the most common cancer in children, accounting for more than a third of childhood malignancies.1 Acute lymphoblastic leukaemia is the predominant leukaemia in children, accounting for 70–80% of cases, followed by acute myeloid leukaemia, which accounts for roughly 15–17%. Chronic myeloid leukaemia and other forms of myeloid leukaemia rarely exceed 4% of all types of leukaemia in children.1–3 However, the causes of leukaemia remain unclear. Several studies4,5 have proposed that genetic alterations in chromosomes and environmental factors are associated with leukaemia.
Molecular genetic alterations might drive mutation, leading to childhood acute lymphoblastic leukaemia with racial and ethnic disparities. Genetic abnormalities have been associated with chromosomal translocations.4 Mutations in the IKZF1 gene and other genes encoding kinase-activating proteins are important contributors to acute lymphoblastic leukaemia.4 Hispanic children are at a higher risk of developing acute lymphoblastic leukaemia but have poorer survival than children of non-Hispanic ethnic origin.5 The chromosomal translocations known to be associated with acute lymphoblastic leukaemia include t(12;21)(p13;q22) with ETV6-RUNX1 fusion, t(1;19)
(q23;p13) with TCF3-PBX1 fusion, t(9;22)(q34;q11) with BCR-ABL1 fusion, t(4;11)(q21;q23) with MLL-AF4 fusion, hyperdiploidy, or hypodiploidy.4,5
Viral infections have been regarded as a crucial environmental risk factor for leukaemia.6–11 Greaves12,13 proposed the delayed infection hypothesis in childhood leukaemia. According to this hypothesis, children who have a delayed exposure to common infections have a more vigorous immune response. Proliferative immunological
cells frequently undergo a second mutation leading to the development of leukaemia in children. Conversely,
exposure to common infections in early childhood is associated with a reduced risk of leukaemia.12,13 Kinlen14 proposed the population mixing hypothesis, suggesting that children might have an abnormal immune response to a common but unidentified infection; he suggested that when the infected population mixes with susceptible individuals, the risk of leukaemia increases.
Enteroviruses belong to the Picornaviridae family and include more than 90 distinct viral serotypes, such as polioviruses, coxsackieviruses, echoviruses, and
numerically named enteroviruses.15,16 Enterovirus infections are common in children and about 10–15 million children contract non-polio enterovirus infections in the USA every year.17 Enterovirus infections are prevalent in children in Taiwan, as they are worldwide. Multiple clinical
manifestations of enterovirus infections have been
recognised in human beings, including herpangina, handfoot- and-mouth disease, meningoencephalitis, acute
flaccid paralysis, haemorrhagic conjunctivitis, respiratory tract infection, myocarditis, and pericarditis.16
Although enterovirus infections are common in children, the association between infection and
leukaemia has not been assessed in a cohort study. We therefore aimed to establish whether the risk of
developing leukaemia was greater in children infected with enteroviruses by analysing data from the National Health Insurance Research Database of Taiwan.
Methods
Study population and design
The National Health Insurance (NHI) programme was implemented in 1995 and has information about up to 99% of the 23・74 million people living in Taiwan.18 We compiled data fi les for children (aged <18 years) from the NHI programme, which were established and
maintained by the National Health Research Institutes (NHRI). The dataset consisted of a randomly selected sample of half of all children in Taiwan who were insured from 1996 to 2008. Randomisation was done with use of computer-generated random numbers assigned to individual data fi les from the NHRI. To protect patient privacy, the NHRI had encrypted all personal
identifi cation numbers into unique numbers before releasing the data fi les to the public for research
purposes. The disease criteria were defi ned and classifi ed according to the diagnostic codes of the International Classifi cations of Diseases, Ninth Revision, Clinical Modifi cation (ICD-9-CM). To ensure the accuracy of disease diagnosis, the Bureau of NHI randomly reviewed the medical charts of one in 100 ambulatory and one in 20 inpatient claims.
We identifi ed children aged younger than 18 years with newly diagnosed enterovirus infections (ICD-9-CM codes
008.67, 047, 048, 074, 079.1, and 079.2) as the cohort infected with enteroviruses. To avoid coding errors in the claims
data, we only included children who had at least three clinic visits with the diagnosis of enterovirus infection. The diagnosis date of the fi rst clinic visit for the enterovirus infection was defi ned as the index date for initiation of follow-up person-year measurement. For each child with enterovirus infection, one child without enterovirus
infection was randomly selected for the non-enterovirusinfected cohort with a frequency matching method to
ensure both cohorts had the same distributions for strata of sex, age (every 1 year span), urbanisation level, parental occupation, and index year of enterovirus infection.
Children with a history of cancer (ICD-9-CM codes
140–208) and with incomplete data for age or sex at baseline were excluded from both cohorts. This study was approved by the Institutional Review Board of China Medical
University Hospital (CMUH-104-REC2-115 and CRREC-103-048).
The sociodemographic variables in this study were age, sex, urbanisation level, and parental occupation (offi ce jobs, manual labour jobs, or other). The NHRI stratifi ed all city districts and townships in Taiwan (based on national administrative zones demarcation) into seven urbanisation levels based on population density (people per km2), proportion of residents with higher education, elderly and agricultural population, and the number of
physicians per 100 000 people in each area.19 Level 1 represented areas with the highest population density
and socioeconomic status and level 7 represented areas with the lowest. Because few people lived in rural areas classifi ed as levels 4–7, we grouped all these areas into the level 4 group. Offi ce workers were employees
characterised by indoor work, including public institutional workers, educators, and administrative personnel in business and industries. Manual labourers were characterised by increased hours of outdoor work, such as fi shermen, farmers, and industrial labourers.
Other occupations included mainly retired, unemployed, and low-income populations. Allergic diseases are
comorbidities of interest in this study. We defi ned allergic diseases as a group encompassing atopic dermatitis (ICD-9-CM code 691.8), allergic rhinitis (ICD-9-CM code 477), and bronchial asthma (ICD-9-CM code 493).
The study outcome was a diagnosis of leukaemia (ICD-9-CM codes 204–208) during the 9 year follow-up, according to the NHI catastrophic illness registry fi les.
Leukaemia is categorised as a catastrophic illness in the NHI programme and patients newly diagnosed with
leukaemia must apply for catastrophic illness certifi cation.
The certifi cation is issued by the government after a
stringent verifi cation process including reviews of medical records, images, and pathology reports by a panel of specialists and experts on the disease. All study patients were followed up until they developed leukaemia, they were lost to follow-up, they withdrew from the NHI programme, or until the end of the study (Dec 31, 2008) without leukaemia (censored).
Statistical analysis
We compared the distribution of sociodemographic
status and allergic diseases between the enterovirusinfected and non-enterovirus-infected cohorts using the
χ2 test. Continuous data were expressed as mean (SD) and compared between the two cohorts with Student’s t test. The incidence density rates (per 100 000 personyears) were calculated for each cohort according to the
leukaemia type. We used the Fine and Gray competing risks regression analysis20 to estimate the subhazard ratio (SHR) and 95% CI of leukaemia associated with
enterovirus infection. The SHR was calculated due to the competing risk in the analyses of survival and cumulative incidence and can analyse the time to the first event, irrespective of which one this was. We calculated the standardised incidence ratio (SIR) of leukaemia using the indirect method of Breslow and Day.21 The
multivariable models were simultaneously adjusted for age, sex, urbanisation-level, parental occupation, and allergic diseases. The age-specifi c, sex-specifi c,
urbanisation level-specifi c, parental occupation-specifi c, and allergy-specifi c incidence densities of leukaemia were estimated for both cohorts. We did Kaplan-Meier analysis for the visual inspection of the cumulative incidence of leukaemia in the two cohorts to compute the Aalen-Johansen estimator adjusting for competing
risks.22 A p value lower than 0・05 was deemed statistically signifi cant. We used SAS statistical package (version 9.3) to analyse the data. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or
writing of the report. C-LL had access to the raw data.
The corresponding author had full access to all the data in the study and had fi nal responsibility to submit for publication.
Results
Insurance claims data for 3 054 336 children younger than 18 years were randomly selected from the NHIRD (fi gure 1). Of these children, 285 597 were newly diagnosed with an enterovirus infection and included in the enterovirus-infected cohort between Jan 1, 2000, and Dec 31, 2007. We excluded 102 children with an
enterovirus infection and cancers before the index date and 3135 children with missing data for age or sex.
282 360 children were included in the enterovirusinfected cohort and 282 355 were included in the nonenterovirus- infected group (control group). Compared
with the non-enterovirus-infected cohort, we noted that the enterovirus-infected cohort had a lower number of deaths (89 [0・03%] vs 218 [0・08%]) and loss to follow-up (1419 [0・50%] vs 4759 [1・69%]) than the control group.
Baseline data for both groups were similar, with the exception of allergic diseases, which were more prevalent in the enterovirus-infected cohort than the control group (table 1). The median follow-up was 5・66 years
(IQR 3・60–7・55) in the enterovirus-infected group and 5・63 years (3・71–7・55) in the control group.
We compared the month of occurrence (fi rst
diagnosis) of acute lymphoblastic leukaemia in tertiles between the two cohorts. Acute lymphoblastic leukaemia was reported much earlier in the enterovirus-infected group (median 25・5 months, IQR 6・31–42・7) than in the non-enterovirus-infected children (median
37・9 months, IQR 28・8–60・8). The analysis of seasonality between enterovirus infection and acute lymphoblastic leukaemia showed a correlation
coeffi cient of 0・45 (95% CI –0・17 to 0・81; p=0・15;
appendix).
Kaplan-Meier analysis showed that the cumulative
incidence of leukaemia was signifi cantly lower in the enterovirus-infected cohort than in the non-enterovirusinfected
cohort after accounting for deaths and loss to
follow-up as the competing risks (fi gure 2). The SIR for leukaemia calculated with the indirect method for enterovirus-infected children compared with the general population was 0・57 (95% CI 0・43–0・75; p<0・0001). The incidence density rates of leukaemia were 3・26 per
100 000 person-years for the enterovirus-infected group and 5・84 per 100 000 person-years for the nonenterovirus- infected cohorts (table 2). The risk of
leukaemia was signifi cantly lower in the enterovirusinfected cohort than in the non-enterovirus-infected
cohort (adjusted SHR 0・44, 0・31–0・60; p<0・0001).
Analyses of leukaemia by type showed that children with enterovirus infection had a signifi cantly lower risk of both lymphocytic leukaemia and acute myeloid leukaemia than that of children without enterovirus infection.
We also estimated the risk of leukaemia by subtypes of enterovirus infection (table 3). Lower risk of leukaemia was noted in children with herpangina and hand-footand- mouth disease; there were no other signifi cant
associations. The incidence rates of leukaemia for both cohorts were stratifi ed by sex, the age at the diagnosis of enterovirus infection, urbanisation level, parental
occupation, and allergic status (table 4). The gap in leukaemia incidence between the enterovirus-infected cohort and the non-enterovirus-infected cohort was greater for children aged 2 years or younger than for children of older age. The incidence diff erences of leukaemia between the two cohorts were greater in boys than in girls, in children living in less urbanised areas (level 3), and in children with parents with offi ce jobs (table 4). For children without allergic diseases, the incidence of leukaemia was signifi cantly lower in the
enterovirus-infected cohort than in the non-enterovirusinfected cohort. The analysis of interaction between the
enterovirus exposure and each of the factors showed that age (p=0・27), sex (p=0・37), urbanisation level (p=0・80), and parental occupation (p=0・18) did not have signifi cant interactions with the enterovirus infection. A statistically signifi cant interaction only existed between enterovirus infection and allergic diseases (p=0・04; table 4).
Discussion
In this study, we identifi ed that the risk of leukaemia was signifi cantly lower in the enterovirus-infected cohort than in the non-enterovirus-infected cohort. Herpangina and hand-foot-and-mouth disease were the main diseases associated with the reduced risk of leukaemia. The risk of leukaemia for children without allergic disease was signifi cantly lower in the enterovirus-infected cohort than in the non-enterovirus-infected cohort.
The incidence of leukaemia might diff er between people from groups and countries of diff erent ethnic origin.23–25 The incidence of childhood leukaemia ranges from 1・64 per 100 000 person-years in low-income
countries to 4・09 per 100 000 person-years in highincome coutries.26 The Taiwan Cancer Registry, a
government-supported and population-based cancer registry system, has reported that the average incidence of leukaemia was 4・3 per 100 000 person-years
(95% CI 4・14–4・46) in children aged 0–14 years from 1996 to 2010.27 In our study, children aged 0–18 years with an enterovirus infection had lower incident leukaemia
than the Taiwan Cancer Registry measured value. Moreover, the SIR of leukaemia for enterovirus-infected
children compared with the general population aged 0–18 years also showed a reduced risk of leukaemia for children with enterovirus infection.
Greaves12,13 proposed the delayed infection hypothesis for childhood leukaemia, and many epidemiological studies6–11,28–30 have supported this hypothesis. Common hygiene-related infections, such as infections of the respiratory system and gastrointestinal tract, occur frequently in the day-care centre;28 therefore, the
attendance of young children in these settings is often regarded as an indicator of early exposure to infections.
The UK Children’s Cancer Study, a large population
based case-control study, used day-care and social activity in the children’s fi rst year of life as proxies for potential exposure to infection to test the Greaves hypothesis. This UK study showed a dose–response association between increasing levels of social activity and a reduced risk of acute lymphoblastic leukaemia. The Northern California Childhood Leukemia Study,29 another large populationbased case-control study, also identifi ed that extensive
contact with other children in daycare settings resulted in a signifi cantly reduced risk of acute lymphoid lymphoma. A meta-analysis30 shows that day-care attendance, particularly in children aged 2 years or younger, is associated with a reduced risk of acute lymphoblastic leukaemia. The ESTELLE study9 done in France reported that early common infections through active day-care activities before the age of 1 year, breastfeeding, and regular contact with pets were also associated with a reduced risk of acute lymphoblastic leukaemia. Our study showed an inverse association between enterovirus infection and childhood leukaemia risk. The present study results have provided an adequate dataset with valid evidence to support the delayed
infection hypothesis as a potential explanation for childhood leukaemia.13
In the population mixing hypothesis, Kinlen14
emphasises the association of leukaemia with exposure to previously unencountered infections. However, little is known about leukaemogenic infections. Marcotte and colleagues10 analysed the association of childhood acute lymphoblastic leukaemia from the California Cancer Registry with the respiratory syncytial virus season surveillance reports from the US Centers of Disease
Control and Prevention and California Department of Public Health Influenza Surveillance Program. They
identifi ed that the risk of acute lymphoblastic leukaemia
was higher in children with the infection of influenza or respiratory syncytial virus at 9–12 months of age than in children with the infections during the fi rst 3 months of life.10 However, no direct records of infl uenza or respiratory syncytial virus infection were obtained from the database in their study. A study from New Zealand31 assessed the association between several infections and childhood acute lymphoblastic leukaemia using interviews or questionnaires. Their study identifi ed that children infected with influenza during the first year of life were at an increased risk of childhood acute lymphoblastic
leukaemia (odds ratio 6・8, 95% CI 1・8–25・7). However, no association was reported between childhood acute lymphoblastic leukaemia and measles, whooping cough, rubella, chickenpox, oral infection, eye infection, ear infection, persistent cough, and diarrhoea and vomiting.
The information about serological tests were available in the study for measles IgG, Epstein-Barr virus capsid antigen IgG, cytomegalovirus IgG, and poliovirus type 1–3 antibody tests.31 An increased risk of childhood acute lymphoblastic leukaemia was associated with only poliovirus type 1 serology, after controlling for age, sex, and other confounding variables. But few cases were available for the analyses. Our study showed a reduced risk of leukaemia in children with enterovirus infection.
However, we did not examine the association between other types of viral infections and leukaemia. To our knowledge, no study strongly supports the asso ciation between leukaemia risk and a unique microorganism so far.
The incidence of childhood leukaemia is associated with socioeconomic status.32 Findings of the association were heterogeneous.33,34 An inverse association between childhood leukaemia and family income and parental education has been reported in studies in the USA that used data obtained from interviews or self-administered questionnaires for analysis. Conversely, a positive association between childhood leukaemia and a higher
socioeconomic status was identifi ed in record-based case-control studies from Europe.33,34 In our study, we examined the leukaemia risk in association with children’s parental occupation and urbanisation level where they lived. However, these statuses did not play an important part in the association between enterovirus infection and leukaemia in this study.
Strachan35 proposed the hygiene hypothesis, and
identifi ed an inverse association between hay fever and eczema risk and the number of older children in the household. The hygiene hypothesis is also implicated with childhood leukaemia. The hygiene hypothesis suggests that a small family size and improved sanitation lower the tendency for cross infection in early childhood and thus results in a higher prevalence of allergies.35 This hypothesis has been supported by several epidemiological studies36–39 that reported a reduced risk of allergy in
children with a higher birth order or early day-care attendance. Moreover, childhood acute lymphoblastic leukaemia is inversely associated with a higher birth order and early day-care attendance, as in the case of the aforementioned delayed infection hypothesis of
leukaemia.6–11 Therefore, a positive correlation between allergy and leukaemia can be expected. However, studies inspecting the association between allergy and leukaemia showed confl icting results. Most studies reported an inverse association between allergic disorders and the risk of childhood leukaemia.40–43 But Chang and
colleagues analysed the data from the NHIRD of Taiwan and reported an increased risk of acute lymphoblastic leukaemia in children with allergic disorders occurring before 1 year of age, less than 1 year before acute lymphoblastic leukaemia diagnosis, and more than 1 year before acute lymphoblastic leukaemia diagnosis.44 Acute lymphoblastic leukaemia and allergic disorders possibly shared a mutual biological cause factor, the overactive and dysregulated immune reaction in response to infection.44 Tests for interaction between age, sex,
urbanisation level, parental occupation, and allergic diseases on one hand, and enterovirus infection on the other hand, showed that the interactions were not statistically signifi cant apart from the interaction of allergic diseases with the enterovirus infection (p=0・04).
Our study showed that the risk of leukaemia for children without allergic disease was signifi cantly lower in the
enterovirus-infected cohort than in the non-enterovirusinfected cohort. This fi nding was not noted for those
with allergic diseases, indicating that allergic diseases had no role in the causes of childhood leukaemia. The results in our study were not consistent with the fi ndings from the study by Chang and colleagues.44 These
diff erences might be explained by the dissimilar study designs. In our study, we focused on the relation of enterovirus infection to leukaemia and we did not examine the temporal sequence of the diagnoses of allergy and leukaemia. However, the exact reasons for the equivocal results between allergy and childhood
leukaemia are unclear.
The main strengths of this study include the large nationwide sample and comprehensive demographic characteristics. However, several limitations of this study exist. First, several risk factors for childhood leukaemia, such as air pollution, parental occupational chemical exposures, tobacco smoking, and pesticides in the household, have been reported.3 However, the NHIRD does not have this information. Second, because of the absence of information about the T-cell or B-cell subtypes of lymphocytic leukaemia, the association of enterovirus infections with these subtypes could not be analysed.
Third, laboratory confi rmation of enterovirus infection, such as serology or viral cultures, was unavailable in the NHIRD. However, the most common forms of
enterovirus infections—namely, herpangina and handfoot-
and-mouth disease—accounted for 277 859 (98%) of all 282 360 enterovirus infections in our study. Both
these enterovirus infections are easily and reliably
diagnosed solely on the basis of clinical manifestations.
Fourth, children with mild symptoms of enterovirus infection might not seek health care, or might only have sought health care once or twice, and so were not included in the enterovirus cohort. Thus only those with more severe clinical manifestations were included.
Whether children with benign or subclinical infections are associated with the risk of leukaemia development is unclear. Fifth, the precision of diagnoses based on the ICD-9-CM codes in the database might aff ect the study fi ndings. However, the Bureau of NHI regularly reviews the charts and assesses the accuracy of claims files.
Incorrect coding of diseases will result in no reimbursement, and the institutions are fined
accordingly, and the high accuracy of data in the NHIRD has been proven by several studies.45,46 Finally, although the results of our study are consistent with the reports on the possible role of childhood infection in the reduction of leukaemia risk in children, biological plausibility and experimental explanation still does not exist. Further studies to understand the mechanism and pathogenesis between enterovirus infection and
leukaemia are warranted.
Overall, enterovirus infection is a crucial issue in children worldwide. Some highly virulent serotypes of enterovirus, such as EV71, have posed a serious challenge to public health because of their severe clinical
manifestations and complications.15,16 To our knowledge, no similar investigation has examined the association of enterovirus infection with leukaemia in the scientifi c literature. This large nationwide retrospective cohort study showed a reduced risk of leukaemia in children infected with enteroviruses, particularly in those who had no allergic diseases. Additional investigations are needed to unravel the pathogenesis of enterovirus infections and childhood leukaemia.
