Incidence of cancer in children perinatally exposed to nucleoside reverse transcriptase inhibitors
AIDS:Volume 22(16)18 October 2008p 2165-2177
Benhammou, Valeriea; Warszawski, Josianea,b,c,d; Bellec, Stephaniec,e,f; Doz, Francoisg,h; Andre, Nicolasi,j; Lacour, Brigittek; Levine, Martinel; Bavoux, Francoisem; Tubiana, Rolandn; Mandelbrot, Laurento; Clavel, Jacquelinec,e,f; Blanche, Stephaneh,p; on behalf ANRS-Enquete Perinatale Francaise
aInserm, U822, Le Kremlin-Bicetre, France
bINED, Paris, France
cUniversite Paris-Sud, Faculte de Medecine Paris-Sud, Le Kremlin-Bicetre, France
dAssistance Publique - Hopitaux de Paris (AP-HP), Hopital Bicetre, Epidemiology and Public Health Service, France
eRegistre National des Hemopathies Malignes de l'Enfant (French National Registry of Childhood Hematopoietic Malignancies), INSERM U754, France
fINSERM U754, Villejuif, France
gInstitut Curie, Service de Cancerologie Pediatrique Paris, France
hFaculte de Medecine Rene Descartes, Equipe d'Accueil EA 3620, Universite Paris V, Paris, France
iAssistance Publique - Hopitaux de Marseille, Hopital de la Timone, Marseille, Service d'oncologie pediatrique, France
jINSERM UMR 911, Centre de Recherche en Oncologie Biologique et Oncopharmacologie, Universite Aix-Marseille, France
kRegistre National des Tumeurs Solides de l'Enfant (French Registry of Childhood Solid Tumors), Centre Hospitalier et Universitaire de Nancy, Nancy, France
lAP-HP, Hopital Robert Debre, Paris, Service d'Hematologie Pediatrique, France
mAP-HP Hopital Saint Vincent de Paul, Paris Centre de Pharmacovigilance Pediatrique, France
nAP-HP Hopital Pitie Salpetriere, Paris Service de Maladie Infectieuses, France
oAP-HP Hopital Louis Mourier, Colombes, Service de Gynecologie-Obstetrique, France
pAP-HP Hopital Necker-Enfants Malades, Paris, Service d'Immunologie Hematologie Pediatrique, Paris, France.
Context: Long-term studies of tolerance to perinatal exposure to antiretroviral nucleoside reverse transcriptase inhibitors are required, in view of the potential genotoxicity of some of these molecules.
Objective: To evaluate the incidence of cancers in uninfected children born to HIV-infected mothers.
Method: Cancers were detected in a nationwide prospective cohort of children born to HIV-infected mothers by standardized questionnaire during the prospective follow-up period of 2 years; thereafter, they were detected by spontaneous pharmacovigilance declaration and by crosschecking data with the national registries of childhood cancer. Standardized incidence ratio for incidence comparisons with general population.
Results: Ten cases of cancer were detected among the 9127 exposed HIV-uninfected children (median age: 5.4 years, 53 052 person-years of follow-up). The overall incidence did not differ significantly from that expected for the general population: 10 cases observed versus 8.9 and 9.6 expected depending on whether 1990-1999 or 2000-2004 national rates were used as reference [standardized incidence ratio of 1.1 (0.3-1.5) and 1.0 (0.5-1.9)]. Five cases of central nervous system cancer were observed (standardized incidence ratio of 3.1 [1.0-7.2] P = 0.05 and 2.4 [0.8-5.6], P = 0.12). The relative risk of cancer for children exposed to didanosine-lamivudine combination was higher than that for zidovudine monotherapy [hazard ratio: 13.6 (2.5-73.9)].
Conclusion: This study did not evidence an overall increase in cancer risk in nucleoside reverse transcriptase inhibitor exposed children until 5 years of age. Results suggesting associations with specific nucleoside reverse transcriptase inhibitor combinations need further investigations. A longer surveillance, including differential analysis of the different cancer sites and various nucleoside reverse transcriptase inhibitors administered is warranted.
The prevention of perinatal transmission of HIV by antiretroviral treatment during pregnancy is remarkably effective . This treatment was initially based on zidovudine monotherapy , but current recommendations propose the use of combinations of antiretroviral drugs generally consisting of at least two nucleoside reverse transcriptase inhibitors (NRTIs) [3,4]. The most widely used NRTI combination is zidovudine-lamivudine together with a protease inhibitor, but six NRTIs (zidovudine, lamivudine, didanosine, stavudine, abacavir and emtricitabine) and one nucleotide reverse transcriptase inhibitor (tenofovir) have now been licensed, leading to a large number of combinations. Among NRTIs, zidovudine is the antiretroviral molecule for which tolerance during pregnancy has been best studied [5,6]. Exposure to prophylactic zidovudine during pregnancy is associated with a series of biological perturbations, but only rare clinical events in children. The main feature is a usually mild and reversible anemia, but persistent perturbations of other hematological lineages have been described in several cohorts [7-9]. Being a nucleoside analogue, zidovudine has a significant affinity for DNA in mammals [10,11]. The potential genotoxicity of these drugs is supported by diverse experimental data: it differs between molecules and involves a complex network of events that may lead to genomic instability due to incorporation of the drug into the DNA, mutagenesis, clastogenesis and telomere attrition . Following in utero exposure to zidovudine in rodent models, a series of genotoxicity markers can be found in the offspring, including zidovudine integration into DNA, mutagenesis and shortened telomeres [12-16]. Dose-dependent tumor induction has been observed in CD-1 mice . This observation, which was not initially found with lower doses , has subsequently been confirmed in other studies in B6C3F1 mice and F344 rats . According to the International Agency for Research on Cancer, evidence for carcinogenicity from studies in laboratory animals is sufficient to consider zidovudine being classified as 'possibly carcinogenic to humans' (group 2B) . The other NRTIs have been much less studied. In addition to the individual genotoxicity profile intrinsic to each molecule, the NRTIs combinations could theoretically result in additive or synergistic toxicity . In chronically infected patients, there is no evidence of cancer induction by the use of NRTI sometimes for very long periods up to 20 years. On the contrary antiretroviral combinations induced a strong reduction of HIV-related cancers .
To date two studies with a limited number of children and follow-up found no cases of cancer in NRTI-exposed children [21-23]. Authors argued for larger and longer surveys based on long-term 'noninvasive' monitoring for these healthy children.
The French Perinatal Cohort [Enquete Perinatale Francaise (EPF)] has included more than 11 000 mother-child pairs since 1984, 80% of whom have been exposed to perinatal NRTIs. Our aim was to estimate the incidence of cancers in uninfected children in utero exposed to antiretroviral therapy, using both a system of active follow-up and crosschecking with registries.
The long-term tolerance of anti-HIV perinatal prophylaxis should be evaluated, but there are several obstacles to the implementation of an epidemiological surveillance system able to detect potential adverse events among uninfected children. After the first 2 years of life (standardized follow-up), the children diagnosed as HIV-uninfected are generally no longer followed by the medical team that managed the perinatal period. For understandable reasons, the parents do not necessarily reveal the history of this perinatal exposure, or do not see the utility of doing so. Crossmatching of children included in HIV perinatal cohorts with children notified to national specific registries, as proposed recently by UK team for cancer screening , may be an appropriate approach to detecting such events without an active follow-up.
With 10 cases of cancer observed to date in our cohort, the overall risk of cancer was not greater than that for the general population [25,28], as previously suggested on two smaller cohorts [21-23].
In the EPF cohort, cancer cases were actively detected through the 2-year standardized follow-up, spontaneous declarations and crossmatching with national registries. Despite the large number of children included, the statistical power of the study was limited. With a power of 80%, and an _ risk at 5%, the size of the present study allowed detection of SIR higher than two for all cancers, three for leukemia and four for CNS tumors. The median age of exposed children was only 5.4 years old (IQR: 2.9-8.3); a longer follow-up is warranted.
Five CNS tumors were observed, whereas only 1.6-2.1 (according to the reference rate used) were expected, and the difference was close to statistical significance (P = 0.05 to 0.12). The analysis restricted to the period 2000-2004 for which the French register of childhood cancers was exhaustive gave similar results with two observed tumors versus 0.3-0.4 expected. We used the two different reference rates that were available in France. For CNS tumors, the national rates produced by the national registries in the recent period 2000-2004 are higher than those of the previous regional registries, which covered around 30% of the French territory in 1990-1999. This is probably due to the increase in quality of notification and registration that came with the development of computerized hospital files, but an increase in incidence cannot be ruled out. Both incidence rates are consistent with those published at the same period in other western countries. The annual age-standardized incidence rate estimated in the 1990s by the French regional registries was 29.1 cases per million  and 29.9 per million in Europe over the 1988-1997 period . The rates estimated by the French national registries for 2000-2004 (36.1 per million; unpublished data) were also close to the figures reported in the United States
(http://apps.nccd.cdc.gov/uscs/ChildhoodCancerData.aspx ) and Germany (http://info.imsd.uni-mainz.de/K_Krebsregister/english/ ) (32.9 and 31 per million, respectively) for the same period.
Description of the cohort
Overall, antiviral exposure (type, doses, timing) data were available for statistical analysis for 10 979 uninfected children live born at mainland French EPF sites between 24 September 1984 and 1 May 2007, that is 95.2% of children notified to the coordinating center. Of these children, 9127 (83%) were exposed to at least one nucleoside analogue at some time during pregnancy and/or the postnatal period (Figs 2 and 3). The uses of combinations progressively increased (e.g. 67% in 2000) and were given to nearly all mothers who delivered in 2006. The third molecule was a nonnucleoside reverse transcriptase inhibitor (NNRTI) in 1015 women and a protease inhibitor in 3792 women. Protease inhibitors were mainly nelfinavir-boosted (1828 women) and ritonavir-boosted lopinavir (1117 women).
The median duration of antiretroviral treatment during pregnancy was 11 [interquartile range (IQR): 7-18] weeks for mothers who received only zidovudine monotherapy (n = 2115) and 20 (IQR: 11-37) weeks for mothers who received combinations of drugs (n = 6644). Treatment began before or during the first trimester of pregnancy (before 14 weeks of amenorrhea) in 31.5% of mothers, during the second or third trimester (14 weeks of amenorrhea) and later) for 64.5% and was solely perpartum and postnatal for 3.0% of mother-child pairs (Fig. 2). The timing of treatment initiation during pregnancy was unknown (missing data) for 1.0% of the mother-child pairs. Among children exposed during pregnancy, 98% received postnatal prophylaxis consisting mostly (82%) of zidovudine monotherapy for 4 or 6 weeks, and of zidovudine-lamivudine for most other children.
Cancers detected and circumstances of their detection
By 1 May 2007, 10 cases of cancer had been identified in the cohort: three cases of acute lymphoblastic leukemia, two cases of retinoblastoma (one bilateral, one unilateral), two cases of pineoblastoma, three cases of glioma (one cerebellar pilocytic astrocytoma, one biopsied chiasmatic tumor and one intraventricular fibrillary astrocytoma) (Table 1). Diagnostic procedures and treatment were administered in specialized pediatric oncology centers for all children. The mean age at onset of cancer was 54 months (range: 10-132), with six boys and four girls affected. The child with the chiasma tumor suffered from familial neurofibromatosis type 1 (NF1 or von Recklinghausen's disease) which predisposes to this type of cancer. A constitutional mutation in the RB1 gene was identified in the child presenting with bilateral retinoblastoma. The two tumors in children under the age of 2 years were both declared as part of the active follow-up of the cohort. The spontaneous notification system notified the coordinating center about the other eight. No additional cases were identified by crosschecking with the national register after having checked the identifiers of 618 children with concordant date and place of birth and sex. Conversely, all cases of hematopoietic malignancy from 1990 and all solid tumors from 2000 notified directly to the EPF cohort were also notified by the national registries. No case was observed in the 1852 uninfected children included in the EPF cohort not exposed to perinatal antiretroviral therapy.
All the children with tumors were exposed to nucleoside analogues during pregnancy: four were exposed to zidovudine monotherapy, three to a combination of zidovudine and lamivudine, two to a combination of zidovudine, didanosine and lamivudine and one to a combination of didanosine and lamivudine. Concerning the other classes of antiretroviral treatments, two mothers received nelfinavir and one ritonavir. Two children received two doses each of nevirapine.
Comparison of cancer incidence among exposed children included in Enquete Perinatale Francaise cohort with that in the general population
The present analysis was based on all 9127 children exposed at any time (prepartum, intrapartum, postpartum) to at least one NRTI between September 1984 and May 2007 (Acknowledgements); this represents 53 052 exposed person-years for ages 0-15 years old, and 10 746 for the period 2000-2004 when the national registries RNHE and RNTSE were both exhaustive.
The number of cancer cases observed in the cohort (10 cases) did not differ significantly from the expected numbers based on regional (8.9 cases; SIR = 1.1 [0.5-2.1]) or national (9.6 cases; SIR = 1.0 [0.5-1.9]) reference rates (Table 2). The result was similar when the analysis was restricted to children born and followed up during the period 2000-2004 with three cases of cancer observed for 2.1 or 2.3 expected. However, five of the tumors involved the central nervous system (CNS), whereas only 1.6 or 2.1 would be expected using the 1990-1999 regional and 2000-2004 national references [SIR = 3.1 (1.0-7.2), P = 0.05 and 2.4 (0.8-5.6), P = 0.12, respectively] (Table 2). Similar estimates were obtained for the period 2000-2004 when crosschecking with national registries became possible. Two of the five tumors of the CNS were pinealoblastomas. A third tumor was unusual: an intraventricular fibrillary astrocytoma. Conversely, 4.1 cases of other cancers (sympathetic nervous system tumors, renal tumors or other tumors) were expected over the study period and none was observed.
Factors associated with risk of cancer among children exposed to prenatal antiretroviral treatment
Among the 8853 children exposed to NRTIs during the prepartum phase, cancer was not associated with maternal geographical origin, drug addiction, last viral load before delivery or sex of the child. Prematurity was associated with cancer (P-logrank = 0.0018), with a hazard ratio (HR) of 10.0 (95% CI 2.1-48.5) for severe prematurity (<33 weeks) compared with full-term delivery. Combinations including both zidovudine and lamivudine, with or without other molecules except didanosine, were not associated with a higher risk of cancer than zidovudine monotherapy. The risk of cancer was significantly higher [HR: 12.5 (2.4-66.1)] in the 365 children exposed simultaneously to NRTI combinations including at least both lamivudine and didanosine, than the 2147 children exposed only to zidovudine monotherapy (Table 3 and Fig. 4). In utero exposure to the protease inhibitors nelfinavir or lopinavir was not significantly associated with a higher risk of cancer. Prematurity and lamivudine-didanosine combinations remained independently associated with cancer in a multivariate Cox model (Table 3). Adjustments for maternal CD4 cell count as a continuous variable did not change the results.
Patients and methods
The prospective Enquete Perinatale Francaise cohort
The French perinatal survey (EPF) is a prospective multicenter study following HIV-infected pregnant women and their children . Uninfected children, who now account for almost 99% of neonates born to HIV-infected mothers, are followed from birth until the age of 2 years through standardized questionnaires completed by physicians according to national recommendations for the management of children born to HIV-infected mothers . An open system of pharmacovigilance has been established for after the end of the active follow-up period and investigators are asked to notify the coordinating center (together with the health authorities) of all severe events that they learn of in children from the cohort. In parallel, the investigators are asked to make parents aware of the importance of notifying any new medical team treating their child of that child's perinatal history. Among the 12 074 live born infants included in the metropolitan EPF sites cohort between 24 September 1984 and 1 May 2007, 11 553 were considered as HIV-1 uninfected. Among them, 9127 children were exposed to at least one NRTI at during one or more of the prepartum, perpartum or postpartum phases since 1990 (274 during the per and postpartum phases exclusively) (Fig. 1).
Participation in the study is voluntary and requires parental consent. The study protocol has been approved by the ethics and regulatory committees, as laid down in French law covering such research.
Participating physicians are asked to follow national guidelines for the treatment of pregnant women with antiretroviral drugs; they are, however, free to make their own decisions concerning the treatment of pregnant women in the best interests of the woman and child's health. Successive French recommendations have been similar to those in North America and in other European countries [3-4].
Prophylaxis was initially based on zidovudine monotherapy. Treatment gradually became more complex, due to the use of combinations of nucleoside analogues (mostly zidovudine-lamivudine), and then with the development of multitherapies (combinations of two nucleoside analogues and a protease inhibitor or a nonnucleoside inhibitor of reverse transcriptase). Regardless of the treatment received by the mother during pregnancy, zidovudine perfusion is recommended during delivery and the child is treated with zidovudine for 4-6 weeks postpartum. The treatment of the child is sometimes strengthened by the addition of one or several other antiviral molecules in cases of a high residual risk of transmission.
We looked for cases of cancer: standardized follow-up questionnaire during the 24 first months; spontaneous notification by the clinicians participating in the EPF cohort or by the network of pediatric hemato-oncologists who were targeted by an information campaign; crosschecking with the French national childhood cancer registries: the National Registry of Childhood Hematopoietic Malignancies (RNHE), which started cancer registration in 1990, and the National Registry of Childhood Solid Tumors (RNTSE), which started in 2000  to identify whether cancers notified in registries occurred in eligible children born in EPF sites. The registries include nominative information, but the coordinating EPF center does not. Concordance of date of birth, sex and place of birth was used to crosscheck between children notified in registries and those in the EPF database. The 618 potential concordant entries were then individually checked by contacting each maternity or pediatric EPF site to identify whether or not the two sets of data corresponded to the same child. At the time of this analysis, the national registries were considered exhaustive for all pediatric (0-15 years) cases of cancer diagnosed in France up to 31 December 2004.
A group of pediatric cancer experts (including the coauthors F.D., N.A., J.C. and B.L.) was set up to collect the information required for the validation of each case. Classifications of detected cancer are presented according to the International Classification of Childhood Cancer . The last evaluation of cancer detection was made on 1 May 2007
The total follow-up time for uninfected children, expressed in person-years, was calculated for the 9127 uninfected live-born children exposed to nucleoside analogues, alone or in combination. We compared the incidence of cancers in the EPF cohort with that in the general population by indirect standardization. The standardized incidence ratios (SIRs) were obtained by dividing the number of observed cases in the cohort by the number of expected cases calculated by applying the reference rate for each year of age to the cohort person-years of this age. We took as reference rates the two different sets of incidence rates that were available in France, that is those estimated by the French regional pediatric cancer registries (one-third of the French territory over the decade 1990-1999 ), and the first national estimates provided for the 2000-2004 period by the French national childhood cancer registries RNHE and RNTSE (unpublished data). Confidence intervals (95% CI) for SIR were estimated using Byar's method .
Univariate and multivariate survival analyses were performed for the subpopulation of children born to mothers who received antiretroviral therapy during pregnancy (N = 8853), to identify perinatal factors associated with cancer occurrence. Survival curves were compared by using logrank tests. The cumulative incidence of cancer was estimated and plotted by the Kaplan-Meier method. Noncolinear variables found to be associated with cancer with a logrank P-value less than 0.10 were included in a multivariate Cox proportional hazards model. Statistical analyses were performed using the Stata software (Stata Corp., College Station, Texas, USA).
The observation of two cases of pineoblastoma among the five CNS tumors, together with two cases of retinoblastoma, is worrying. Pineoblastoma is a rare tumor in children, with a mean of only three cases per year reported in the French national registry of solid tumors. It is also one of the secondary tumors that may affect children with hereditary retinoblastoma . Another tumor observed in this cohort, an intraventricular fibrillar astrocytoma or 'holoventricular low-grade glioma', is extremely rare .
Unexpectedly higher risk of cancer was observed for children who were exposed to combinations including didanosine-lamivudine than for those exposed to zidovudine monotherapy (adjusted HR = 13.6 [2.5-73.9]). Didanosine-lamivudine was given to less than 4% of treated women of the cohort and is associated with one-third of tumors. Given the small number of cancer events seen, interpretation of this difference must be cautious. We used a global logrank test with a single P-value to study the association between cancer incidence and the type of NRTI combinations ever administered during pregnancy, and not multiple tests to compare each of the five combination to the reference category (exclusive monotherapy). This P-value was very low (<0.0001) and remained lower than the Bonferroni corrected P-value threshold (0.05/8 = 0.006) taking into account eight statistical tests as we also tested separately the association between cancer incidence and seven molecules. Moreover, the hazard ratio associated with 'didanosine and lamivudine' was very high (12.5) and the lower bound of 95% CI strongly higher than the unity: 95% CI 2.4-66.1. Interestingly, a rare neuroectormic tumor (melanotic progonoma tumor) was recently diagnosed in a child exposed to a didanosine-lamivudine combination but not included in the EPF cohort (D. Plantaz, MD, Grenoble, France, personal communication).
A scale of genotoxicity for several NRTI combinations has been recently suggested on the basis of data from in-vitro mutagenesis assays for hypoxanthine-guanine phosphoribosyl transferase (HPRT) and thymidine kinase. The combinations were ranked as follows in decreasing order: zidovudine-didanosine > didanosine-lamivudine > zidovudine-lamivudine , but no comparative assay of these combinations in animal models of in utero exposure for tumor induction has been published. The combination of didanosine and lamivudine has been rarely prescribed for pregnant women. However, didanosine is now sometimes used as secondary or subsequent treatment and an increasing number of exposed children might be expected. In chronically HIV-1 infected adults and children, there is no evidence that zidovudine or other NRTIs alone or in combination contribute to the development of cancers despite 20 years of use. However, fetal and neonatal exposure may be a particular situation. Fetal and neonatal life is characterized by intense DNA replication activity and the activation of many oncogenes and tumor suppressor genes. Exposure to a genotoxic agent during this period may, therefore, have consequences different from those observed in adults . In newborn and infants, exposure to zidovudine in utero leads to biological abnormalities, reflecting this genotoxicity. Thus, the integration of zidovudine into the DNA of neonates exposed to the drug in utero  may lead to mutagenesis  or other abnormalities, such as an increase in the frequency of micronuclei in erythrocytes  or an increase in heterochromatin dispersion . The effects of other NRTIs alone or in combination have been less studied experimentally; nevertheless, some of them show effects in vitro similar to those of zidovudine and/or induce an additive or synergistic effect in combination. Our findings certainly highlight the need for further data concerning the genotoxicity of the various NRTI combinations.
Antiretroviral combinations are extremely powerful prophylaxis for HIV mother-to-child transmission but an effort to select genotoxicity-free combinations is justified.