Infectious Causes of Cancer
Antibodies against malaria and Epstein-Barr virus in childhood Burkitt lymphoma: A case-control study in Uganda
Version of Record online: 13 NOV 2007
Copyright © 2007 Wiley-Liss, Inc.
International Journal of Cancer
Volume 122, Issue 6, pages 1319–1323, 15 March 2008
How to Cite
Carpenter, L. M., Newton, R., Casabonne, D., Ziegler, J., Mbulaiteye, S., Mbidde, E., Wabinga, H., Jaffe, H. and Beral, V. (2008), Antibodies against malaria and Epstein-Barr virus in childhood Burkitt lymphoma: A case-control study in Uganda. Int. J. Cancer, 122: 1319–1323. doi: 10.1002/ijc.23254
- Issue online: 21 JAN 2008
- Version of Record online: 13 NOV 2007
- Manuscript Accepted: 14 SEP 2007
- Manuscript Received: 30 MAY 2007
- Mulago Hospital and Makerere Medical School, Kampala, Uganda
- Imperial Cancer Research Fund (now Cancer Research UK)
- Centers for Disease Control and Prevention (CDC) Atlanta, United States and the International Agency for Research on Cancer, World Health Organization, Lyon, France
- Burkitt lymphoma;
- malaria antibodies;
- Epstein-Barr virus;
Burkitt lymphoma, a childhood tumor common in parts of sub-Saharan Africa, has been directly associated with Epstein-Barr virus (EBV) and indirectly with prevalence of malaria. We studied antibodies to both EBV and malaria in children diagnosed with this cancer in Uganda. We performed a case-control study of HIV-seronegative children (≤15 years) admitted to hospital. Cases were diagnosed with Burkitt lymphoma and controls with non-malignant conditions or non-lymphatic cancers. Interviews were conducted and serological samples collected and, when possible, tested for both EBV and malaria. Adjusted odds ratios (ORs) for Burkitt lymphoma were estimated using unconditional logistic regression adjusting for sex, age, residential district, household income and tribe. The mean age of cases was 7 years and 61% were male. Compared to controls, cases were more likely to be reported having received more frequent treatment for malaria in the past year (OR = 2.0; p = 0.001) and less likely to be living in a home where insecticides were used (OR = 0.2; p < 0.0001). Odds ratios for Burkitt lymphoma in children increased with increasing antibody levels against EBV (p < 0.0001) and malaria (p = 0.05). Findings were similar for children residing in districts close to the capital city and in remote areas. Cases were 5 times more likely than controls to have raised levels of both EBV and malaria antibodies (OR = 5.0; p = 0.003). Our findings suggest that EBV and malaria may act synergistically in the pathogenesis of childhood Burkitt lymphoma. Malaria prevention measures may also prevent this childhood cancer. © 2007 Wiley-Liss, Inc.
Burkitt lymphoma is a common tumor among children in parts of sub-Saharan Africa, including Uganda.1–7 Marked seasonal, temporal and geographical variations in incidence, with evidence of space-time clustering, have suggested an infectious aetiology.5, 8 Particular interest has focused on the role played by two infectious agents—the Epstein-Barr virus9, 10 and malaria.11–13 On the basis of the geographical distributions, holoendemic malaria has been suggested as a cofactor for Burkitt lymphoma14 but, unlike EBV,10 there has been no direct evidence that individual children diagnosed with this cancer are more likely to have had malaria. Associations with socio-economic and other environmental factors have also been suggested but comparatively little studied8 and the potential role played by maternal characteristics, such as a mother's exposure to infectious diseases, have yet to be addressed.
Here, we report on both childhood and maternal risk factors, including antibodies to EBV and malaria, from a case-control study of Burkitt lymphoma in Uganda, a country with endemic Plasmodium falciparum malaria.
Material and methods
Cases considered for inclusion were all children aged 15 years or less diagnosed with Burkitt lymphoma identified from a study of childhood cancer conducted at the Uganda Cancer Institute. Cases were admitted to one of the 3 main hospitals in Kampala (Mulago, Mengo or Nsambya) between December 1994 and October 1999. Of the 399 children diagnosed during the study period, human immunodeficiency virus type 1 (HIV) sero-status was determined for 349. As HIV might be associated with antibody titers for other infections, we restricted analyses to 325 children known to be HIV sero-negative. Controls were the 579 HIV-negative children aged 15 years or less concurrently being treated as hospital in-patients, or in follow-up clinics for benign, non-infectious, surgical or orthopaedic conditions, at Mulago, Mengo and Nsambya hospitals and comprised 447 children with non-malignant surgical conditions and 132 with cancers other than lymphoma and leukaemia (49 with nephroblastoma, 30 with Kaposi sarcoma, 10 with rhabdomyosarcoma, 9 with osteosarcoma, 4 with neuroblastoma, 3 with retinoblastoma and 27 with other or uncertain malignancy). No restriction was placed on recruitment of children into the study of childhood cancer. Analyses presented in a previous report were not restricted to HIV-negatives but were limited to 128 children with cancer and 190 children with nonmalignant conditions who lived in Kampala prior to becoming ill.15
Mothers of cases and controls, when available, were interviewed by 1 of 5 trained staff to obtain information about demographic and socio-economic factors, their experience at (or around) the child's birth, the child's siblings, household characteristics and their sexual and reproductive histories. Additional questions regarding child's tribe, use of insecticides in the home, and history of childhood illnesses and symptoms were added after the study had begun, in August 1996. Responses were recorded by interviewers using standardized questionnaire. Results are presented for 92 HIV sero-negative mothers of cases and 260 HIV sero-negative mothers of controls. When the mother was not present for interview (9 cases and 9 controls) information on socio-demographic and other factors was obtained from the child's father or guardian. Ethical approval for this study was obtained from the Committee on Human Research (VA Medical Centre and University of California San Francisco) and by the Uganda National Council for Science and Technology.
Preliminary clinical diagnoses of cancer were made by one of the investigators (EM) and confirmed by histology, cytology or other laboratory investigations where possible. All 399 cases of Burkitt lymphoma were subsequently histologically verified. Laboratory tests for EBV and malaria were only possible for children and mothers whose serological samples were transported successfully to the Centers for Disease Control and Prevention (CDC), Atlanta, GA, USA. As a result of losses in transit, EBV tests for current study members could only be performed for 173 of cases and 102 of controls. Subsequent serological testing was only possible for malaria for 126 cases and 70 controls. Corresponding test results in mothers were obtained for EBV for 39 cases and 30 controls, and for malaria in 46 cases and 76 controls.
Serological testing for EBV was performed at the CDC and, for malaria, in a US commercial laboratory. Serologic testing for EBV was done using a commercial kit, VCA IgG by DiaSorin (Stillwater, MN), and the manufacturer's instructions were followed for assay interpretation. Optical density readings (OD) were used as a surrogate measure of EBV antibody titer. For the purposes of analysis, children were categorized into one of three groups: low (OD < 2), medium (OD ≥ 2) or high antibodies (OD ≥ 3.5) while mothers were categorized as either: low (OD < 2) or medium/high (OD ≥ 2). Antibodies to Plasmodium falciparum were detected by an indirect immunofluorescence assay.16 Unless stated otherwise, children and mothers with positive malaria test results were categorized as either low (antibody levels of 1:64 or 1:256) or high (antibody levels of 1:1,024 or higher).
Data were entered using EPI-INFO software17 by trained computer clerks in Uganda. These were subsequently independently checked against the original questionnaires in Oxford where further data cleaning procedures were carried out. All data analyses were performed using STATA computer software (version 8.0).18 Odds ratios (ORs) were estimated by maximum likelihood using unconditional logistic regression. All ORs relating to data on children were adjusted for child's sex and age (under 5 years, 5–7 years, and 8 years and over) while those relating to data on mothers were adjusted for mother's age (under 30 years, 30 years and over). To minimize potential bias from differential patterns of ascertainment of cases and controls noted previously,15 all ORs were additionally adjusted for usual residential district (largely urban districts of Kampala and the two adjacent (Mukono and Mpigi), and largely rural districts elsewhere). Adjustment was also made for any additional socio-demographic variables for which cases and controls were found to differ significantly at the 1% level.
Ninety-five percent confidence intervals for adjusted ORs were derived from the variance-covariance matrix of the logistic regression estimators as were the tests of statistical significance for ORs examining presence or absence of risk factors.19 The latter resulted in one-degree of freedom χ2-tests derived from Wald statistics while those examining factors with three or more ordered categories resulted in one-degree of freedom χ2 statistics for linear trend. All p-values reported are obtained using two-sided tests of statistical significance. When more than two groups are compared, variances are estimated by treating the ORs as floating absolute risks.20 Use of floating methods does not alter the odds ratio estimates, but yields floated standard errors and confidence intervals (FCI) that enable valid comparisons to be made between any 2 exposure groups, even if neither is the baseline group.
About two-thirds of the 325 HIV-negative children with Burkitt lymphoma (cases), and a similar proportion of the 579 HIV-negative controls, were boys. The ages of cases ranged from 18 months to 15 years [mean age = 7.0 (standard deviation (SD) 3.0 years] and 23% were aged under 5 years. Cases were more likely than controls to come from poorer households and those outside the more urban areas of Kampala and neighboring districts, Mukono and Mpigi (Table I). The geographical distribution of 247 cases with known residential district (76%) is shown in Figure 1. Outside the more urban areas of Kampala, Mukono and Mpigi, the majority of cases came from Iganga district (21 cases) or Luwero (16 cases). While data on the child's tribe were limited, cases were less likely than controls to originate from the Baganda (29% vs. 57%) who predominantly reside in Kampala, Mukono and Mpigi and neighboring districts.
|Cases (n = 325)||Controls (n = 579)|
|Female||127 (39)||195 (34)|
|Male||198 (61)||384 (66)|
|Age group (years)|
|8 and over||120 (37)||208 (36)|
|5–7||130 (40)||188 (32)|
|Under 5||75 (23)||183 (32)|
|Usual residential district1|
|Kampala||19 (8)||222 (40)|
|Mukono/Mpigi||71 (29)||172 (31)|
|Other||158 (64)||165 (30)|
|Usual monthly household income (UgSh)1|
|45,000 or more||89 (40)||369 (68)|
|Less than 45,000||136 (60)||174 (32)|
|Baganda||53 (29)||302 (57)|
|Other||129 (71)||227 (43)|
After adjusting for child's sex, age, usual residential district, monthly income and tribe, households in which cases lived were less likely to report using insecticides in the home (adjusted OR = 0.2, p < 0.0001) (Table II). In addition, compared to controls cases were more likely to report having been treated three or more times for malaria in the past year (adjusted OR = 2.0, 95% CI 1.3–3.1; p = 0.001). For both reported treatment for malaria and insecticide use, associations with Burkitt lymphoma were found to persist when additional adjustment was made for the other factor. For reported treatment for malaria, the odds ratio estimate resulting from this additional adjustment was 1.8 (95%CI 1.2–2.8; p = 0.005) and for reported insecticide it was 0.2 (95% CI 0.1–0.4; p < 0.001). Cases were also more likely to have 3 or more siblings (adjusted OR = 1.5; p = 0.02) and sleep in a shared bed (adjusted OR = 1.6; p = 0.04) (Table II). No association was found for reported symptoms such as cough, fever or diarrhea (results not shown).
|Cases/Controls (325/579)||Adjusted1 OR (95% CI or FCI) χ2 for heterogeneity2|
|Number of times treated for malaria in past year3|
|Three times or more||87/201||2.0 (1.3–3.1)|
|Less than three times||66/281||1.0|
|χ = 11.4; p = 0.001|
|Use insecticides in home3|
|χ = 35.6, p < 0.0001|
|Number of siblings|
|Three or more||205/291||1.5 (1.1–2.2)|
|Less than three||99/272||1.0|
|χ= 5.3, p = 0.02|
|Child sleeps in shared bed3|
|χ = 4.3, p = 0.04|
|Epstein-Barr virus antibodies4|
|χ = 15.7, p < 0.0001|
|χ= 3.9, p = 0.05|
Cases with serological results for EBV (n = 173) and malaria (n = 126) were of the same age as all cases (n = 325) with each group having a mean age of 7.0 years (SD = 3.1 and 3.2 respectively). Cases and controls were also very similar with respect to the socio-demographic characteristics listed in Table I (results not shown). Despite the limited availability of serological test results, ORs for Burkitt lymphoma were found to be strongly associated with increasing titre of antibodies against EBV viral capsid antigens (Table II). In addition, the ORs increased with increasing antibodies against malaria (p = 0.05).
Because cases and controls differed in their usual residential district, associations observed in Table 2 were examined separately for the 2 main groups of residential districts: the more urban areas of Kampala, Mukono and Mpigi and other more rural areas (“Elsewhere”). While the number of cases in each stratum was relatively small, the ORs for Burkitt lymphoma for risk factors measured at interview (reported treatment for malaria in the past year and lack of use of insecticides in the home) were raised in both the more urban areas and elsewhere (Table III). Associations with EBV and malaria serology were also evident both in children coming from the more urban and more rural areas (Table III).
|Risk factor||Residential district|
|Cases/Controls||Adjusted1 OR (95% CI or FCI)||Cases/Controls||Adjusted1 OR (95% CI or FCI)2|
|Number of times treated for malaria in past year:|
|Three times or more||30/159||1.3 (0.7–2.4)||57/42||3.0 (1.7–5.4)|
|Less than three times||28/197||1.0||38/84||1.0|
|Use insecticides in home:|
|Yes||5/203||0.1 (0.03–0.2)||16/48||0.3 (0.2–0.7)|
|High||10/4||4.4 (1.1–17.0)||27/4||7.0 (2.3–21.3)|
|Medium||26/28||2.4 (1.3–4.4)||44/11||4.3 (2.1–8.5)|
|Low||9/25||1.0 (0.4–2.4)||17/19||1.0 (0.5–2.1)|
|High||8/8||2.2 (0.6–7.7)||27/5||9.1 (3.1–26.8)|
|Low||19/19||1.7 (0.9–3.1)||36/16||3.5 (1.9–6.3)|
|Negative||6/10||1.0 (0.2–4.2)||4/4||1.0 (0.2–5.1)|
The joint effects of EBV and malaria antibodies were examined by estimating adjusted ORs for children with raised levels of EBV antibody only, malaria antibody only, and both EBV and malaria antibodies, compared with children with low levels of both. Cases were five times more likely than controls to have raised levels of both EBV and malaria antibodies (OR = 5.0, 95% CI 2.8–8.9; p = 0.003) but not more likely to have increased levels of either antibodies to EBV alone or malaria alone but the confidence intervals associated with these latter OR estimates were wide (Table IV).
|EBV antibodies1||Malaria antibodies||Cases/Controls||Adjusted2 OR (95% FCI)|
|Low||Negative or 1:64||12/17||1.0 (0.4–2.4)|
|1:256 or more||15/16||1.1 (0.5–2.4)|
|Medium/High||Negative or 1:64||22/19||1.0 (0.5–2.2)|
|1:256 or more||77/18||5.0 (2.8–8.9)|
Cases whose mothers were available for interview were slightly younger than other cases (mean age 6.7 vs. 7.1 years), more likely to be male (64 vs. 58%), to live in Kampala and neighboring districts (40 vs. 32%) and be from the Baganda tribe (37 vs. 23%). Similar differences were also observed for controls (results not shown). Mothers of cases available for interview were similar to those of controls with respect to age [mean age = 30.6 (SD 7.1) and 30.7 (SD 7.5) years, respectively]. After adjustment for mother's age, usual residential district and household income, the ORs for Burkitt lymphoma was lower for children whose mothers reported being 18 years or more when first married (adjusted OR = 0.5, p = 0.008) (Table V). Apart from this finding, there was no significant association with any indicators of reproductive and sexual history examined (age at first sex, number of children, children having different fathers or number of lifetime sexual partners). There was also no significant association with the mother's antibody level for EBV (p = 0.4). If anything, mothers of cases seemed to have lower malaria antibodies than those of controls but this apparent negative association was not statistically significant (p = 0.1).
|Cases/Controls (92/260)||Adjusted1 OR (95% CI or FCI) χ2 for heterogeneity2|
|Age at first sex|
|16 years or more||47/170||0.7 (0.4–1.1)|
|Less than 16 years||42/86||1.0|
|χ = 2.1; p = 0.1|
|Age at marriage|
|18 years or more||35/149||0.5 (0.3–0.8)|
|Less than 18 years||52/102||1.0|
|χ = 7.0; p = 0.008|
|Number of children|
|Five or more||57/131||1.5 (0.8–2.9)|
|Less than five||32/125||1.0|
|χ= 1.6; p = 0.2|
|Children have different fathers|
|χ = 0.8; p = 0.4|
|No of lifetime sexual partners|
|Three or more||53/140||1.4 (0.8–2.3)|
|Less than three||36/117||1.0|
|χ = 1.4; p = 0.2|
|χ = 0.8, p = 0.4|
|χ = 2.8, p = 0.1|
This case-control study reports on both childhood and maternal risk factors for Burkitt lymphoma among HIV sero-negative children in Uganda. Specific attention has focused on the potential role played by infectious diseases in children and their mothers. Previous evidence for an association between Burkitt lymphoma and malaria has rested on ecological data with no direct measurement of antibody levels in infected children. To our knowledge this is the first study to show a relationship between high titers of malaria antibodies in children and increased risk of Burkitt lymphoma. An additional finding was the strong inverse relationship with reported use of household insecticides suggesting that the use of these insecticides may prevent not only malaria but also Burkitt lymphoma for which it is also a risk factor.
That the majority of cases originated from outside the capital city and neighboring districts appears consistent with the lower incidence of Burkitt lymphoma associated with residence in urban areas noted previously.14 Controls were mainly children with nonmalignant surgical conditions recruited from the largest of the hospitals contributing cases and much more likely than cases to come from the urban districts located in and around the capital city. It is possible that this reflects referral bias with children with Burkitt lymphoma more likely to be referred from rural districts to major hospitals in the capital whereas control children are likely to receive care more locally. Controls were also more likely to be Baganda, a tribe mainly residing in Kampala and adjacent areas whose reduced risk of Burkitt lymphoma has been noted previously.13
In accord with results from previous epidemiological studies from sub-Saharan Africa,21–23 we found clear evidence that the odds ratio for Burkitt lymphoma increased with increasing titer of antibodies against EBV viral capsid antigens in the child. No association was found with anti-EBV antibody titer in the mother. In addition, we were able to demonstrate for the first time clear evidence of an association with past exposure to malaria as measured from antibodies in serological samples. Moreover, not only were the relative odds of the tumor found to increase with increasing anti-malaria antibody titers in the child but they were also found to increase with increasing reported number of recent treatment episodes for malaria.
Of particular interest is the evidence suggesting that EBV and malaria antibodies act synergistically in the development of Burkitt lymphoma. Compared with those who had low levels of both EBV and malaria antibodies, cases were five times more likely than controls to have raised levels of both EBV and malaria antibodies but not more likely to have increased levels of antibodies against either EBV alone or malaria alone. This finding, albeit based on small numbers, suggests that EBV and malaria may act together to increase the risk of Burkitt lymphoma in children. If real, the exact nature of the interaction between EBV and malaria is unclear. In a study of children living in two regions of Kenya, EBV viral load in blood was highest in the children living in an area of holoendemic malaria compared with those living in an area of sporadic malaria transmission.24 A Ugandan study showed that EBV load in plasma, but not in saliva, fell following treatment of children with malaria.25 Two main hypotheses have been advanced to explain the interaction between the infections: malaria leading to reactivation of memory B cells which are latently infected with EBV and/or suppression of EBV-specific T-cell immunity.26
While it has been suggested that those of low socio-economic status might be at increased risk of Burkitt lymphoma, the epidemiological evidence for this has been fairly limited.8 We also found that children living in households reporting lower monthly income were at greater risk of Burkitt lymphoma. It is possible that the association with low socio-economic status is a surrogate for the higher intensity of malaria in poor rural areas and perhaps lack of money to purchase household insecticides or treatment for malaria. Unfortunately, questions on use of insecticides, treatment for malaria and some household characteristics were not introduced until after the start of the study.
With regards to maternal risk factors, we found that the odds ratio for Burkitt lymphoma in the child declined with increasing maternal malaria antibodies although this relationship was not statistically significant. A study from Tanzania suggested that passively transferred maternal antibodies to malaria tend to protect young infants from clinical malaria by increasing the rate at which blood stage infections are cleared.27 While our findings appear consistent with the view that low maternal antibodies to malaria may increase the risk of Burkitt lymphoma the data were sparse and did not measure maternal malaria antibody levels either during gestation or shortly after birth.
In summary, our case-control study of Ugandan children strengthens the evidence associating Burkitt lymphoma with both EBV and malaria. Associations found with reported recent treatment for malaria, and use of insecticides in the home, suggest that malaria prevention also has the potential to decrease the risk of this important paediatric malignancy in Africa.
Serological tests for EBV were performed at CDC by Dr. Scott Schmid. The study was conducted by the Uganda Kaposi's Sarcoma Study Group which includes named authors and V Sembajwe, M Kalinaki, R Byansi, C Rwatooro, S Nambooze, B Tushimiere, N Byabazaire (deceased), E Bitamazire, E Katabira, J Mugerwa (deceased), D Tindyebura, J Whitworth, B Richardson, G Reeves, R Weiss and K de Cock. JZ was on secondment from the Department of Veterans Affairs and the University of California, San Francisco, to the International Agency for Research on Cancer, France. LC became involved in this research while employed by the UK Medical Research Council Programme on AIDS in Entebbe, Uganda. Data analyses were carried out by DC.
- 6Distribution of tumours in Uganda. In: TempletonAC, ed. Tumours of a Tropical Country. Recent Results in Cancer Research, No. 41. Berlin: Springer Verlag, 1973. 1–22., .
- 8Burkitt's lymphoma. In: SchottenfeldD andFraumeniJF, eds. Cancer epidemiology and prevention. Philadelphia: WB Saunders. 1982. pp 779–94..
- 9Virus particles in cultured lymphoblasts from Burkitt's lymphoma. Lancet 1964; i: 702–3., , .
- 10Is Burkitt's lymphoma related to perinatal infection by Epstein-Barr virus? Lancet 1977; i: 335–38..
- 11An epidemiologic approach to the lymphomas of African children and Burkitt's sarcoma of the jaws. Perspect Biol Med 1964; 36: 435–49., , , .
- 14Epidemiological evidence for the role of falciparum malaria in the pathogenesis of Burkitt's lymphoma. In: LenoirGM,O'ConorGT,OlwenyCLM, eds.Burkitt's lymphoma: a human cancer model. Lyon: IARC Scientific Publication, no. 60, 1985..
- 17Epi Info, Version 5: a word processing, database and statistics program for epidemiology on micro-computers. Atlanta, Georgia, USA: Centers for Disease Control, 1990., , , .
- 18StatCorp. 2003. Stata Statistical Software: Release 8.0, 2003. College Station, TX: Stata Corporation.
- 19Applied logistic regression. New York: Wiley, 2001., .