Risk Factors for Pediatric Human Immunodeficiency Virus–Related Malignancy FREE

Malignancy risk is increased in individuals with human immunodeficiency virus (HIV) infection.¹ The occurrence of Kaposi sarcoma in young male homosexuals was a sentinel indicator of the AIDS epidemic. Pediatric and adolescent AIDS cases with malignancy were first reported in hemophiliacs who had received contaminated blood products.^2- 4 For many of these children, malignancy was the first indication of HIV infection.

Although the prevalence of HIV infection is lower in children than in adults, cancer incidence rates for HIV-infected children appear disproportionately lower. Prior to the era of highly active antiretroviral therapy, estimates of the incidence of malignancy in HIV-infected children ranged from 0.66 cases per 1000 children per year in the United States¹ to 4.18 per 1000 children per year from an Italian registry⁵ while the incidence of malignancy in adults appeared to be much higher.⁶ The discrepancy in the risk of developing malignancy between adults and children may be related to differences in the mode of HIV transmission and other age-related factors including concomitant viral infections, such as Epstein-Barr virus [EBV]. Epstein-Barr virus infection and seropositivity are nearly ubiquitous among adults and are present in 78% of HIV-infected children by 3 years of age.⁷ At present, the specific risk factors for cancer in children infected with HIV have not been identified.

We investigated risk factors for malignancy in HIV-infected children in a multicenter case-control study. We hypothesized that HIV-induced immunodeficiency combined with EBV-induced chronic B lymphocyte stimulation increases the risk of cancer while coinfection with cytomegalovirus (CMV) and human herpesvirus 6 (HHV-6) and prior treatment with antiretroviral agents do not increase the risk.

Cases were HIV-infected pediatric patients with a new malignancy. Before anticancer therapy was initiated, biological specimens were evaluated at central laboratories to confirm the malignancy diagnosis, HIV-infection status, and virologic and immunologic characteristics. Human immunodeficiency virus–infected control patients who were free of malignancy were identified by institutions that had enrolled 1 or more cases; they were matched based on the duration of infection (±3 months for patients <2 years, ±6 months for those aged 2-4 years, ±12 months for those aged >4 years). Blood specimens from controls were evaluated at central laboratories for virologic and immunologic characteristics. At the time of enrollment into the protocol, demographic information was obtained by interview with a parent or guardian. Other information was obtained by direct clinical examination of the patient and from the medical record.

The Pediatric Oncology Group was a consortium sponsored by the National Cancer Institute and comprised approximately half of all centers in North America that treat childhood cancer. The Pediatric Oncology Group established the Pediatric AIDS Lymphoma Network, which included a cancer registry and a case-control protocol designed to characterize the clinical, immunologic, and viral characteristics associated with increased risk of malignancy in HIV-infected children. The protocol was approved by the institutional review board at each participating center and written informed consent was obtained from the parents or guardians of all study patients, with additional written assent from older children. All patients were centrally registered at the Pediatric Oncology Group Statistical Office in Gainesville, Fla. Case and control patients were enrolled between January 1992 and July 1998. The accrual pattern was similar for case and control patients; the average registration occurred 3.48 years from the time the study opened for cases and 3.26 years for controls (P = .54, 2 sample t test).

Human immunodeficiency virus infection was defined by a positive HIV enzyme-linked immunosorbent assay result with confirmation by immunoblot for children aged at least 15 months. Patients had to be younger than 21 years at the time of HIV infection. Malignancy diagnoses for all cases were confirmed by centralized histopathology review by one of the authors (V.V.J. or R.E.H.). Non-Hodgkin lymphomas (NHLs) were classified using the working formulation⁸ into the categories of small noncleaved cell, Burkitt type, and diffuse large cell. Other lymphomas not classifiable by the working formulation included mucosa-associated lymphoid tissue lymphoma; primary central nervous system lymphoma, not further classified; other NHL, not further classified; and Hodgkin disease (HD). Non-Hodgkin lymphoma cases were B-cell phenotype. Eligibility was restricted to patients who had not received prior antineoplastic therapy.

At the time of study enrollment, anticoagulated blood (in EDTA or acid-citrate-dextrose) was collected and shipped at room temperature overnight for processing at a central laboratory in San Antonio, Tex. Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll-Paque (Sigma, St Louis, Mo) and stored at −70°C in aliquots of 1 × 10⁶ cells. Cells were resuspended in 50 µL of polymerase chain reaction (PCR) lysis buffer (0.45% NP-40, 0.45% Tween 20, 10 mmol tromethamine hydrochloric acid [Tris-HCl] {pH 8.3}, 50 mmol potassium chloride [KCl], 0.01% gelatin, 200 µg/mL proteinase K) and lysed at 55°C for 1 hour. DNA from tumor samples was extracted using standard methods.

Semiquantitative PCR was used to determine HIV, EBV, CMV, and HHV-6 DNA genome copies per 10⁵ PBMCs. Briefly, PCR was performed in a total volume of 20 µL using 5 µL of sample containing 10⁵ cells. All samples were amplified in duplicate. Published primers and methods were used for amplification of HIV,⁹ EBV,⁹ CMV,¹⁰ and HHV-6.¹¹ The signal was detected using a PhosphorImager (Molecular Dynamics, Sunnyvale, Calif) and quantitated using a polynomial standard curve generated for each PCR run. As an internal control, primers PCO4 and GH20 amplifying a conserved β-globin sequence were included in each reaction.¹² The β-globin signal was used to adjust the PCR signal of tumor to a quantitative value per 10⁵ PBMCs.

Serologic testing was performed. Immunofluorescence assays were used to determine antibodies to EBV,¹³ CMV (Gull Laboratories, Salt Lake City, Utah), and HHV-6.¹⁴

Cases and controls were compared by demographic, clinical, HIV-infection, and coviral infection characteristics using contingency table analysis. Simple 2 × 2 analyses were performed using exact methods for categorical variables (StatXact 4 for Windows, CYTEL Software Corp, Cambridge, Mass). Quantitative covariates were compared using independent sample t tests or the Wilcoxon rank sum test for nonnormally distributed variables. Two-sided significance tests are reported. The magnitude of association between putative risk factors and case-control status was summarized as the odds ratio (OR) with 95% confidence intervals (CIs). Even though controls were matched to cases by HIV infection duration, multivariate results are presented for unconditional logistic regression analysis. These results were similar to those obtained by conditional logistic regression analysis (data not presented). Preselected first-order interaction terms were included in the regression models. The Statistical Analysis System version 8 (SAS Institute Inc, Cary, NC) was used to estimate adjusted ORs with 95% CIs.

A total of 43 cases and 74 controls were accrued from 26 institutions. Table 1 shows the distribution of malignancy diagnoses for cases by route of HIV infection. Of the case patients, 4 had B-cell acute lymphoblastic leukemia (ALL); 28, non-Hodgkin lymphoma (NHL); 1, Hodgkin disease (HD); 8, leiomyosarcoma; 1, hepatoblastoma; and 1, schwannoma. Lymphoid malignancies (B-cell ALL, NHL, and Hodgkin lymphoma) comprised 76.9% of cancers in vertically acquired cases and 76.5% in postnatally acquired cases. Within the lymphoid malignancy diagnoses, there was 1 case of B-cell ALL, 19 cases of NHL, and no cases of HD in the vertically acquired case group, but 3 cases of B-cell ALL, 9 cases of NHL, and 1 case of HD in the postnatally acquired case group (P = .10, exact χ² test).

Cases were slightly younger than controls at registration (Table 2) although the percentage of cases and controls younger than 4 years was similar (23.3% vs 23.0%, respectively). The percentage of boys in the case and control group was similar, 55.8% vs 60.8%, respectively. There were slightly more white Hispanics among cases (27.9%) vs controls (18.9%). Vertical HIV transmission was the predominant source of infection for cases, 74.4%, and for controls, 60.8% (P = .13, exact χ² test). No patients reported acquisition of HIV infection through sexual transmission or intravenous drug use. All patients whose transmission was nonvertical had acquired their infection through contaminated blood or blood products. The mean (SD) duration of HIV infection at the time of registration was 4.8 (3.1) for cases and 5.7 (3.6) years for controls (P = .32). Cases had significantly lower CD4 cell counts than controls (median CD4 cell counts 124 µL vs 337 µL, respectively; P = .006). Prior zidovudine use (P = .31), mother's HIV infection status (P = .15), and father's HIV infection status (P = .37) were not significantly associated with cancer risk.

There were no statistically significant quantitative differences in HIV DNA viral burden in peripheral blood between cases and controls (Table 3). The median number of HIV DNA (provirus) copies per 10⁵ PBMCs was 21.7 for cases and 37.0 for controls (Wilcoxon signed rank test, P = .23). Although newer HIV RNA assays have greater sensitivity, 4 cases did not have detectable HIV DNA in PBMCs. In contrast, cases presented with much higher EBV DNA viral loads than controls; the median number of EBV DNA copies per 10⁵ PBMCs was 9.9 for cases vs 3.8 for controls (Wilcoxon signed rank test, P = .03). The percentage of cases with 50 EBV DNA copies per 10⁵ PBMCs or more was 31.6% compared with 11.3% for controls (exact χ² test, P = .18). There were no statistically significant differences between cases and controls with respect to detectable viral loads for CMV DNA (12.5% cases vs 3.4% controls; exact χ² test, P = .18) or for HHV-6 DNA (29.0% cases vs 38.9% controls; exact χ² test, P = .48).

Most cases (95.2%) and controls (87.7%) were EBV viral capsid antigen (VCA)–IgG seropositive at the time of study entry (Table 3). Detectable early antigen (EA) antibodies were found in 76.2% of case and 72.3% of control patients. Early antigen–diffuse pattern antibodies were present in 59.4% of case and 53.2% of control patients. A nonsignificantly higher proportion of case (40.6%) vs control patients (29.8%) had detectable anti-EA-restricted pattern antibodies (P = .12). No case patients had anti-EA-diffuse and restricted antibodies but 8 (17.0%) of 47 controls did (P = .01). There were no statistically significant case-control differences in CMV (P = .11) or HHV-6 (P = .61) seroprevalence or antibody titers. Among a subset of the current study population, we have previously reported that CMV and HHV-6 infections are common among children with HIV infection and cancer and that geometric mean titers of serum anti-CMV antibodies, but not anti-HHV-6 antibodies, were higher in case vs control patients.¹⁵ However, in this larger analysis, case-control differences in anti-CMV antibodies were diminished and not statistically significant (P = .08).

We assessed ever-reported HIV-related illnesses as well as illnesses that had occurred in the 12-month period prior to enrollment into this study. Results using either of these 2 referent times were nearly identical. Cases were more likely than controls to have splenomegaly, 40.5% vs 11.1% (OR, 5.5; 95% CI, 1.8-17.8) as well as hepatomegaly, 52.8% vs 20.3% (OR, 4.4; 95% CI, 1.6-11.8), respectively. There were no significant differences in the rates of previous lymphadenopathy or parotid gland enlargement. A higher proportion of cases (17.7%) compared with controls (3.2%) were likely to have had lymphoid interstitial pneumonitis and oral candidiasis (OR, 6.5; 95% CI, 1.1-68.7). There were no statistically significant case-control differences in history of disseminated tuberculosis, other mycobacterial infection, Pneumocystis jiroveci pneumonia, toxoplasmosis, cryptosporidiosis, isosporiasis, cryptococcosis, coccidioidomycosis, nocardiosis, or viral hepatitis.

Immunization rates for case and control patients ranged from 96% to 100% for diphtheria, tetanus, pertussis, inactivated poliovirus vaccine (cases), oral poliovirus vaccine (controls), measles, mumps, rubella, and Haemophilus influenzae B. Routine childhood immunizations did not appear to be associated with cancer risk. However, cases were less likely to have received pneumococcal polysaccharide vaccine than controls, 54.6% vs 83.3%, respectively (OR, 0.24; 95% CI, 0.06-0.94). Likewise, fewer cases received influenza vaccination in the previous 12 months than did controls, 77.8% vs 97.4%, respectively (OR, 0.09; 95% CI, 0.002-0.87).

The crude association between EBV load (≥50 EBV DNA copies per 10⁵ PBMCs) and cancer risk was significant (OR, 3.63; 95% CI, 1.14-12.08; P = .02), as was the crude association between CD4 cell count (<200/µL) and cancer risk (OR, 2.75; 95% CI, 1.10-6.88; P = .02). We further examined the combined effects of EBV viral burden and CD4 cell count. Figure 1 shows the distribution of EBV DNA copies per 10⁵ PBMCs for cases and controls stratified by low (<200/µL) and high (≥200/µL) CD4 cell counts. Although EBV viral load was similar for cases and controls with low CD4 cell counts, median (interquartile range [IQR]), 4 (1-36) vs 7 (undetectable-29) EBV DNA copies per 10⁵ PBMCs, EBV viral load was higher for cases vs controls for children with high CD4 cell counts, 135 (7-337) vs 5 (undetectable-20) EBV DNA copies per 10⁵ PBMCs. In a multivariate analysis that included the covariates EBV viral loads and CD4 cell counts, there was a statistically significant interaction (P = .03) between these variables. We then stratified the regression analysis by CD4 cell count (<200/µL vs ≥200/µL). High EBV viral loads were strongly associated with cancer risk but only for children with CD4 cell counts of at least 200/µL; there was more than an 11-fold increased risk of cancer in children with high EBV viral load and with high CD4 cell counts (OR, 11.33; 95% CI, 2.09-65.66; P = .001). In contrast, there was no association between EBV and cancer risk for children with CD4 cell counts lower than 200/µL (OR, 1.12; 95% CI, 0.13-9.62; P = .99). To determine whether these results were confounded by malignancy diagnosis (lymphoid malignancy vs soft tissue sarcoma), we excluded leiomyosarcoma, which has been shown to be associated with EBV infection of smooth muscle cells,^9,16 and EBV still was associated with cancer risk for children with CD4 cell counts of at least 200/µL (OR, 11.96; 95% CI, 2.64-53.47).

We examined other putative risk factors and found no significant associations with cancer risk. Case-control comparison of quantitative viral load for CMV DNA (P = .23, Wilcoxon rank sum test) and HHV-6 DNA (P = .20, Wilcoxon rank sum test) failed to show a significant difference in cancer risk. Prior antiretroviral therapy with zidovudine was not significantly associated with cancer risk in either the high CD4 cell-count group (OR, 0.81; 95% CI, 0.22-3.09; P = .77) or the low CD4 cell-count group (OR, 0.27; 95% CI, 0.04-1.46; P = .16). Route of HIV infection (vertically vs postnatally acquired) was not associated with cancer risk in either the high CD4 cell-count group (OR, 0.32; 95% CI, 0.08-1.27; P = .10) or low CD4 cell-count group (OR, 1.31; 95% CI, 0.37-4.64; P = .68).

As in adults with AIDS, the risk of malignancies is increased in HIV-infected children, but the precise incidence of cancer has been difficult to define. In a survey of Children's Cancer Group institutions using only diagnoses of NHL or Kaposi sarcoma as AIDS-defining malignancies, Mueller et al¹⁷ estimated that there was at least a 100-fold increased risk of malignancy in HIV-infected children. From a US surveillance cohort of 4954 children with AIDS, 124 children (2.5%) were identified as having a malignancy with an estimated relative risk of 651 for children who are 2 or more years beyond their AIDS diagnosis.¹ These studies demonstrate that children infected with HIV are at significantly increased risk for developing cancer.

To our knowledge, this is the first reported case-control study designed to identify clinical and molecular risk factors for pediatric HIV-related malignancies. The distribution of malignancy diagnoses in our case group was similar to other reported pediatric series.^1,17- 18 Lymphoid malignancies occurred most commonly followed by leiomyosarcomas and other solid tumors. The route of HIV infection was not associated with the type of cancer that developed.

Univariate analysis showed that both high EBV viral load and immunosuppression, as reflected by CD4 cell count, were associated with increased cancer risk. An independent role for EBV in HIV-infected children is suggested by the fact that there is no evidence that EBV infection decreases CD4 cell count in HIV-infected children,⁷ and by our regression results for which CD4 cell count was not an independent risk factor after adjusting for EBV viral load. Our stratified analysis revealed that EBV viral load was higher only among children with cancer and higher CD4 cell counts and that the EBV viral load among children with lower CD4 cell counts, regardless of cancer status, was comparable with that of children with higher CD4 cell counts without cancer (Figure 1). These findings suggest that there is a critical level of lymphopenia below which EBV replication, and hence the EBV viral load, decreases. This effect modification might be the result of decreased target cells (ie, B-cell lymphocytes) or other processes that culminate in a less favorable cellular milieu for EBV replication. This also suggests multiple paths for HIV-associated malignancy, including one characterized by high EBV viral load with moderate immunocompromise. However, the wide CIs and the fact that this was a post hoc observation suggest further study is required.

Although EBV seroprevalence is nearly ubiquitous for older adults, the seroprevalence is slightly lower for young children. In this study, most cases and controls were EBV seropositive (95.2% and 87.7%, respectively) and these rates were similar to the EBV seroprevalence among children born to HIV-infected mothers reported previously.⁷ However, lower seroprevalence of EBV alone is unlikely to completely explain the discrepancy between pediatric and adult cancer occurrence in HIV-infected populations.

We have previously reported that EBV infection of smooth muscle cells is present in leiomyosarcomas of children and young persons with AIDS,^9,16 and we have documented EBV positivity in 9 (36%) of 25 lymphoid tumors in children with AIDS.¹⁹ Since EBV is not found to be associated with all cases of AIDS malignancy, either in adults or children, EBV cannot provide the sole explanation for the increased risk of cancer. Indeed, in our HIV-infected study population, increased EBV burden alone does not appear to be a sufficient cause for malignancy given that risk was not independently increased in children with low CD4 cell counts. In fact, Van Baarle et al²⁰ recently reported that, among a cohort of homosexual men studied longitudinally, EBV viral load in PBMCs displayed considerable fluctuation over time and was not predictive of the development of AIDS-associated NHL.

For our case-control study, we hypothesized that the route of HIV infection, either vertical transmission or postnatally acquired, may be associated with the spectrum of malignancy diagnoses that occur, yet we found no evidence to support this. Age at study entry (the time of malignancy diagnosis for cases), sex, race, and mother's or father's HIV infection status were also not associated with cancer risk. Most patients received the recommended childhood immunizations. With nearly ubiquitous vaccination rates, we did not observe an association between routine childhood immunizations and cancer risk. Case patients were less likely than control patients to have received pneumococcal polysaccharide and influenza vaccinations within the previous 12 months. The prevalence of other AIDS-defining conditions was extremely low in our study population. Only patients with a history of oral candidiasis and lymphoid interstitial pneumonitis appeared to be at greater risk of developing cancer. These are likely proxies for more advanced stages of AIDS rather than independent risk factors.

Exposure to antiretroviral drugs may also modify cancer risk; zidovudine has demonstrated carcinogenic potential in mice and monkeys.²¹ However, in our study, prior zidovudine use was not associated with cancer risk. Treatment of pregnant women with zidovudine was shown to be highly effective at blocking maternal transmission of HIV, demonstrated in the AIDS Clinical Trials Group 076, which led to its widespread use.²² However, none of the mothers enrolled on our study were participants in that protocol. We did not collect other maternal antiretroviral use information during pregnancy, but widespread use of zidovudine during pregnancy to prevent transmission and the more recent advent of highly active antiretroviral therapy were not common until long after our study patients were born, thus minimizing this as a potential confounder.

This study had a number of limitations. First, it was not possible to ascertain the proportion of all-incident cancer cases in the population that was represented in our study or the proportion of potentially eligible matched controls. Hence, selection bias may have influenced our results. However, this is unlikely to explain entirely our results given the magnitude of the association between EBV and cancer risk and the fact that demographic and other clinical factors were similarly distributed for case and control patients. Second, given the rarity of AIDS malignancies in children and the relatively small sample size of our case series, we had limited power to look at individual cancers or to test the association of less common putative risk factors. The risk of certain malignancies may be greater than others in HIV-infected children, and risk factors may differ. Third, case patients were evaluated after the diagnosis of malignancy and the effect of the malignancy on immunosuppression in our results is unknown.

Our study had a number of strengths. All viral assays and lymphocyte assays were performed in central laboratories, and all malignancies underwent centralized pathology review. All assays were performed prior to the start of cancer treatment. Finally, this epidemiologic study represents the largest group of children with HIV-related malignancies that have been uniformly characterized with respect to their malignancy diagnosis, immunologic status, and viral burden of HIV and other coinfecting viruses.

A prospective cohort study may further delineate the risk factors and pathogenesis of these malignancies, since such a design would allow for assessment of the temporal interrelationships among EBV replication, viral load, and immune function. Given the low incidence of these pediatric malignancies in developed countries and the difficulties in performing molecular assays on fresh biological specimens in high seroprevalence areas in developing countries, such a prospective cohort study would be daunting to mount and most likely require an international effort.