People infected with HIV have been found to be at increased risk of developing certain cancers 1 2 3 4 5 including Kaposi sarcoma, non-Hodgkin lymphoma and conjunctival carcinoma 6 . To date, however, the majority of evidence has been based on studies of adults; the impact of infection with HIV on the risk of cancer in children is less certain. Studies in developed countries, have suggested approximately 2.5% of children infected with HIV will develop cancer, lower than the proportion seen among infected adults 3 . Cancers described as being associated with HIV in children include Kaposi sarcoma, non-Hodgkin lymphoma and leiomyosarcoma 3 7 8 but to date, there are few data from sub-Saharan Africa, where the majority of HIV infected children live. Here, we examine the association between HIV infection and cancer among children in Blantyre, Malawi.

Out of a total of 586 suspected cases of childhood cancer, 541 were recruited into the study. All children whose diagnosis was unknown were excluded (n = 27, 4.6%). Of the children included in the study, 284 (52%) were diagnosed with Burkitt lymphoma, 53 (10%) with nephroblastoma (Wilms tumour), 34 (6%) with Kaposi sarcoma, 36 (7%) with non-Burkitt non-Hodgkin lymphomas, 28 (5%) with rhabdomyosarcoma and 22 (4%) with retinoblastoma (Table 1). Children diagnosed with other types of cancer and non-malignant conditions comprised 15% of the children included in the analysis (84 children). There were no children diagnosed with leiomyosarcoma.

There was a preponderance of male children in the sample (59% overall) and most children (69%) resided in rural regions of the country (Table 1). The majority of the children were diagnosed after their fifth birthday (65%), a pattern observed for most of the cancer groupings, except for those with nephroblastoma or retinoblastoma; 35 children had missing age data. Overall, 62% of the cases had laboratory confirmed diagnoses and 37% had the diagnosis made on clinical grounds alone. Ten percent of all children (n = 54) tested positive for HIV, although HIV prevalence varied by cancer type, ranging from 0% among children with retinoblastoma and rhabdomyosarcoma to 77% (n = 24) among children with Kaposi sarcoma. Of the 506 children whose age was known, 25 children were under 18 months, and all were HIV seronegative.

HIV was strongly and positively associated with Kaposi sarcoma (OR = 93.5, 95% CI 26.9 to 324.4, p < 0.001) and positively associated with non-Burkitt non-Hodgkin lymphoma (OR = 4.4, 95% CI 1.1 to 17.9, p = 0.04), but not with Burkitt lymphoma (OR = 2.2, 95% CI 0.8 to 6.4, p = 0.13; Table 2). No other cancer site or type was significantly associated with HIV infection; although to date, the numbers available for analysis remain small. None of the 8 leukaemia cases and only 1 of the 10 cases of Hodgkin lymphoma were HIV seropositive. Analyses were repeated restricting to those cases with laboratory confirmation of diagnosis: key findings remain unchanged although confidence limits were wider (see footnote to Table 2).

Relative to adults, there are few published analytical epidemiological studies of cancer in HIV infected children, particularly among those living in sub-Saharan Africa, and the spectrum of cancers affecting children may be different 4 . Unlike adults, the great majority of HIV-infected children acquire the virus in utero or in the first months of life when the immune system is at an early stage of development. The proportion of children with HIV infection who will go on to develop a malignancy is at present poorly defined 3 . This study, carried out in Malawi, has found clear evidence of positive associations between HIV infection and both Kaposi sarcoma and non-Burkitt non-Hodgkin lymphoma in children. The relative risk of Burkitt lymphoma was raised, although not significantly.

While Kaposi sarcoma in Malawian children was relatively rare prior to the HIV epidemic 9 , the number of cases has increased during the era of HIV 10 11 12 . In some sub-Saharan African countries, Kaposi sarcoma is now one of the most common cancers in children 13 14 15 16 17 18 19 20 . This increase is likely to reflect the prevalence of HIV, although it is also in an area with a high background prevalence of the underlying causal virus Kaposi's sarcoma-associated herpesvirus (KSHV) 21 .

The increased risk for non-Burkitt non-Hodgkin lymphoma in association with HIV in Malawian children is of similar magnitude to that reported for adults in South Africa (OR = 5.0, 95% CI 2.7 to 9.5, based on 128 cases), Rwanda (OR = 12.6, 95% CI 2.2 to 54.4, based on 26 cases), and Uganda (OR = 6.2, 95% CI 1.9 to 19.9, based on 21 cases) 22 23 24 25 . In these studies, the relative risk of non-Hodgkin lymphoma associated with HIV infection is an order of magnitude lower than that reported from developed countries 1 26 . The reasons for this difference are not clear. Possible explanations for the apparent lack of non-Hodgkin lymphoma among HIV infected people in Africa include under-ascertainment of the malignancy and competitive mortality from other HIV-associated illnesses. As with all African series, a proportion of diagnoses are made on clinical grounds alone, and so we cannot completely exclude the possibility that some cases were misdiagnosed (e.g. tuberculosis or other infection-associated lymphadenopathy).

A case-control study from Uganda was the first to report an association between Burkitt lymphoma and infection with HIV among children living in an area where the tumour is relatively frequent 23 (note: a small subset of preliminary data from this study previously reported no association 27 ). In 1994, no cases of Burkitt lymphoma were found in 78 HIV positive children in Côte d'Ivoire 28 , and no increase in the incidence of childhood Burkitt lymphoma after the onset of the HIV epidemic was noted in Zambia 16 . A preliminary analysis of data from Malawi, published in early 2008, identified an excess risk of Burkitt lymphoma - based on 11 HIV infected cases and 228 without HIV (OR = 12.4, 95% CI 1.3 to 116.2) 29 . In the updated analyses reported here, the odds ratio is lower and no longer statistically significant, although the adjusted prevalence of HIV is similar to that of non-Burkitt non-Hodgkin lymphoma. However, the number of HIV positive children with Burkitt lymphoma reported in the literature to date is small and there remains substantial uncertainty about the role of HIV (if any) in the aetiology of this common malignancy among children in parts of sub-Saharan Africa. Our results and those of other African studies contrast with those from Western populations, where adults and children with AIDS are at least 1,000 times more likely to develop Burkitt-type lymphoma than the general population 1 23 30 31 , probably reflecting different pathogenetic mechanisms.

The overall HIV seroprevalence of 10% found among children included in this analysis was as expected in a population of children admitted to hospital in Malawi 32 , and is similar to that seen in children with cancer in Uganda 23 . For some cancers, the number of cases available at the time of writing was too small to draw reliable conclusions. Our results are broadly in line with those reported elsewhere in Africa 9 13 23 33 34 35 , although data on cancer in HIV-infected children remains scant. In developed countries, both primary brain lymphomas and leiomyosarcomas tend to occur in children with HIV infection 3 36 37 38 39 40 . Brain tumours are difficult to diagnose in developing countries, although 4 cases with cranial tumours were reported here (all HIV seronegative). No cases of leiomyosarcoma were found.

Leiomyosarcoma is a very rare tumour, occurring in less than 2 cases per 10 million HIV negative children per year 41 , although it occurs at a much higher than expected frequency in HIV infected children in developed countries 3 7 8 38 . There is paucity of data from African studies regarding the occurrence of leiomyosarcoma in HIV infected children. It is possible that leiomyosarcoma is under-diagnosed in this setting. Soft tissue tumours are diagnostically challenging, with histopathological criteria that are continuously changing. It is therefore possible that leiomyosarcomas may be diagnosed as rhabdomyosarcoma or other soft tissue tumours. Even in settings where more advanced tumour morphology, immunophenotyping and other ancillary investigations are done, discrepancies in diagnoses occur 42 .

The results reported here are subject to the potential problems of incomplete diagnostic verification. Laboratory verification of cancer diagnosis by histology and cytology was available for 64% of all cancers. This proportion varied by cancer site; being higher if the tumour was easily accessible for biopsy. These results are typical of studies in developing countries, where laboratory services are limited, and compare favourably with other cancer series reported from Africa 13 14 43 . On the other hand, our study had the advantage of a high ascertainment of HIV status among all children with cancer (97%), which was consistently high across all cancer types, in contrast to a previous review of childhood cancer in Malawi 13 . It is possible, however, that children with both cancer and HIV were likely to die before a diagnosis could be made, potentially biasing our findings.

The impact of HIV on the risk of cancers other than Kaposi sarcoma and non-Burkitt non-Hodgkin lymphoma remains uncertain and is the subject of further research. As data accrue from Malawi and elsewhere, the impact of HIV on the risk of other cancer types among children should be clarified.

HIV: (Human Immunodeficiency Virus); AIDS: (Acquired Immunodeficiency Syndrome); OR: (Odds ratio); CI: (Confidence Interval); KSHV: (Kaposi's sarcoma-associated herpesvirus).

The authors declare that they have no competing interests.

NM, RN, EMM and LMC conceived of the study, and participated in its design and coordination and helped to draft the manuscript. WTJ performed the statistical analysis. SK, EB, NM, GNL, MB and HWJ participated in drafting the manuscript. All authors read and approved the final manuscript.