Background

Pre-treatment drug resistance (PDR) mutations have been demonstrated to be a major reason for virological failure (VF) after starting combination antiretroviral treatment (cART) in HIV-infected adults [1–6]. It is likely that they play a similar role in HIV-infected children. In vertically infected children, drug resistance mutations can be present prior to antiretroviral treatment due to the transmission of the resistant virus (transmitted drug resistance) from mothers [7] or the emergence of resistance mutations in wild-type virus transmitted from mothers as a consequence of pressure of maternal antiretroviral prophylaxis (drugs that have crossed the placenta and have a long half-life in infants) or neonatal prophylaxis [8–10].

In vertically infected children, exposure to single-dose nevirapine given for prevention of mother to child transmission (PMTCT) is associated with reduced efficacy of this drug when used for early treatment [11], and the response has been correlated with pre-existing resistance mutations [12–14]. Current treatment guidelines in the United States and Europe recommend genotypic resistance testing in all antiretroviral naive patients, including vertically infected children, to detect the presence of PDR mutations and to adapt their first-line treatment accordingly [15, 16]. Several studies have evaluated the prevalence of PDR mutations among HIV-infected children [6, 17–25]; however, few have determined their effect on virological response to first-line cART [6, 14, 23].

We assessed the prevalence of PDR mutations and their association with virological outcome in the first year of cART in children within a large collaboration of HIV observational cohort studies (EuroCoord-CHAIN-PENTA-EPPICC) in Europe and Thailand.

Methods

Study population

This study was conducted under the joint efforts of the Collaborative HIV and Anti-HIV Drug Resistance Network (CHAIN) and the EuroCoord network (CASCADE, COHERE, EuroSIDA and PENTA/EPPICC).

Cohorts participating through the EuroCoord network submitted defined dataset (patient demographics, use of cART, CD4 counts and HIV RNA measurements up to 16 months post-cART start, clinical (AIDS and death) events and genotypic resistance test) to their network-specific Coordination Centre, using the HIV Cohort Data Exchange Protocol [26].

Ethics approval was granted by the ethic committees of each of the participating cohorts according to local regulations. Written informed consent from participants’parents or legal guardians was obtained by the participating cohorts according to national legal and ethics requirements.

Maternal and infant prophylaxis data were not available for this analysis. Over the period of the study, no specific PMTCT protocol was recommended in Europe however some countries had their own PMTCT guidelines [27–31]. Practically, most cohorts used NVP or PI-based highly active antiretroviral therapy as the standard of care for mothers and 4–6 weeks zidovudine prophylaxis for infants. Zidovudine (ZDV) monotherapy with pre-labour caesarean section was an alternative for pregnant women who did not require treatment for their disease or in case of low viral load HIV. Threshold of CD4 varied over time and across countries (200 or 350 cells/mm3) as well as time to start antiretroviral prophylaxis. In Thailand national antiretroviral prophylaxis regimens have evolved from ZDV monotherapy for mothers and infants in 2000 to ZDV monotherapy plus single dose nevirapine (SD-NVP) in 2004 then to NVP-based cART for women with CD4 < 200 cells/mm3 or ZDV/SD-NVP for women with CD4 > 200 cells/mm3 with a 7-day post-partum tail regimen of ZDV/lamivudine to prevent NVP resistance in SD-NVP exposed women [32].

HIV-infected children aged <18 years were included in this study if they started cART between January 1,1998, and December 31, 2008 and if they had ≥1 sample for a genotypic test taken before the initiation of cART, whereby cART was defined as receiving at least three antiretroviral drugs. Genotypic resistance tests could have been performed retrospectively at each virology lab, i.e. test results may not necessarily have been used to guide first-line treatment. HIV genotyping assays were based on population sequencing techniques. We obtained nucleotide sequences for 93 % and amino-acid sequences for 7 % of the children. Alignment of nucleotide sequences was done centrally at the Copenhagen regional coordinating centre. If more than one genotypic test result was available mutations were cumulated.

Statistical analyses

Definition of PDR mutations and resistance

PDR mutations and resistance were defined in two steps. First, the World Health Organization 2009 list of NRTI, NNRTI and PI mutations for the surveillance of transmitted drug resistant HIV strains [33] was used to identify PDR mutations and distinguish children harbouring a virus with ≥1 PDR mutations and those harbouring a virus with no PDR mutation, referred to as ‘no PDR’.

Second, for children harbouring a virus with ≥1 PDR mutations, the Stanford algorithm version 6.0.5 [34] was used to classify children into 2 groups: children receiving fully active cART (Stanford levels 1 and 2, corresponding to susceptible and potential low-level resistance, for all prescribed drugs) and those harbouring resistant HIV strain (Stanford levels 3 4, 5 corresponding to low-level, intermediate and high-level resistance, respectively) affecting ≥1 of the prescribed drugs. Among children with PDR mutations, those receiving fully active cART were referred to as ‘PDR and fully-active cART’ and children with a resistant strain were referred to ‘PDR and resistant’. Children with PDR receiving a fully active cART were regrouped with children harbouring a virus without PDR for the analysis of virological failure because there was no event among those receiving a fully active cART. Thus, two groups were considered: one group including all children with no PDR or ‘PDR and fully-active cART’ and one group including all children ‘PDR and resistant’.

Virological response

Virological failure was defined as the first of two consecutive viral loads >500 copies/mL after 6 months of cART (window 6–16 months), considering the date of first viral load >500 copies/mL as failure date. Children were censored if they died, stopped cART or were lost-to-follow-up. The time to virological failure was described by Kaplan-Meier curves and analysed by Cox proportional Hazards models. Baseline is defined as date of cART initiation.

All multivariable models were adjusted for the following potential confounders chosen a priori: sex, age, pre-treatment viral load (log10 transformed) and CD4 count, subtype (B, non B, unknown), region of origin (African, European, Asian, other/unknown), year of treatment start (1998–1999, 2000–2002, 2003–2004, 2005–2006, 2007–2008), previous AIDS diagnosis (yes, no, unknown) and HIV transmission risk group (vertical, heterosexual, injection drug use, other/unknown) and initial cART regimen (NNRTI plus ≥ 2NRTIs, boosted PI plus ≥ 2NRTIs, unboosted PI plus ≥ 2NRTIs, other). We conducted a stratified analysis according to initial cART regimen (NNRTI plus ≥ 2NRTIs, unboosted PI plus ≥ 2NRTIs). Proportionality assumptions were checked graphically by depicting the (log-log (Survival Probability) according to log(survival time) and by testing an interaction term between the covariables and the survival time. Analyses were performed using SAS 9.2 and 9.3 (SAS Institute, Inc., Cary, NC). P < 0.05 was considered significant.

Posteriori power calculation

Based on the number of virological failures observed among the 476 HIV-infected children, we could achieve 90 % power in a two-sided test with type I error of 5 % to detect a hazard ratio (HR) of 1.95 if risk groups were well balanced (i.e. 50 % of patients with high risk and 50 % with low risk). The HR limit would be 2.17, 3.05 and 4.64 if risk groups were not balanced whatever the direction (25–75, 10–90 and 5–95 % respectively) [35].

Results

Study population and baseline characteristics

A total of 476 children had sufficient follow-up and resistance data to be included in the analysis. They were enrolled in 18 cohort studies in 11 countries. At baseline, 246 children (51.7 %) were female and a large majority was infected through vertical transmission of HIV (Table 1). The median age at cART initiation was 6.6 years (interquartile range (IQR), 2.1–10.1), median baseline CD4 cell count 297 cells/mm3 (IQR, 98–639), and median HIV RNA load 5.2 log10 copies/mL (IQR, 4.7–5.7) (Table 1).

Table 1 Characteristics at cART initiation
Full size table

Two hundred thirty two (48.7 %) children initiated cART with 2 nucleoside reverse transcriptase inhibitors (NRTIs) and 1 nonnucleoside reverse transcriptase inhibitor (NNRTI) and 41 (8.6 %) with 3 NRTIs and 1 NNRTI, 139 (29.2 %) with 2 or more NRTIs and 1 unboosted protease inhibitor (PI) (mostly nelfinavir), 43 (9 %) with 2 or more NRTIs and 1 boosted PI, and 21 (4.8 %) with other combinations of three or more drugs (Table 1).

Pre-treatment resistance mutations

At least one PDR mutation was identified in 37 children (7.8 %; 95%CI, 5.5–10.6) of whom 76 % harbored a virus with NRTI, 46 % with NNRTI and 8 % with PI resistance mutations and 30 % had PDR mutations to 2 ARV classes (Table 2). The proportion of children with PDR mutations was higher among children below 2 years of age than among those aged 2 years or more; 12.2 % (14 of 155) versus vs 6.4 % (23 of 361) (P = 0.043).

Table 2 Prevalence of resistance mutations using the WHO 2009 list for surveillance of transmitted drug resistance mutations
Full size table

Of these 37 children, seven were in the ‘PDR and fully-active cART’ group and 30 children in the ‘PDR and resistant’ group. In this latter group, 11 children had PDR mutations to 2 antiretroviral drug classes (Table 2). Prevalence of PDR mutations was 7.1, 7.7, 7.8 and 9.8 % in children from African, European, Asian or unknown origin. There was no statistical difference of PDR according the geographic origin (P = 0.93). Furthermore, baseline characteristics were not different between children with and without PDR mutations (Additional file 1: Table S1).

Analysis of virological failure and risk factors

Overall, median follow-up was 12 months (10; 14) corresponding to 449 person years (PY) of follow-up; children with fully active treatment were followed-up over a median of 12 (10; 14) months (424 PY of follow-up), and children with PDR and receiving a treatment to which the virus was resistant were followed over a median of 11 (8; 13) months (26 PY of follow-up). Median number of viral load measurements after baseline was 4 (IQR: 2; 8) and there was no difference in median viral load measurements between the group no PDR/PDR susceptible and the group PDR and resistant with 4 (2;8) and 4 (2;7) measurements, respectively. Virological failure occurred in 83 of 446 children who started with fully active cART (no PDR or having “PDR and fully-active cART”) and in 9 of 30 children of the “PDR and resistant” group. The overall estimated cumulative Kaplan-Meier risk of virological failure at 12 months of cART was 19.8 % (95%CI, 16.4–23.9). Children in the ‘PDR and resistant’ group tended to have higher cumulative risk of virological failure than those with no PDR or having ‘PDR and fully-active cART’, at 32.1 % (17.2–54.8) versus 19.4 % (15.9–23.6) (p = 0.095) (Fig. 1). Crude incidence rates of virological failure were 35 per 100 PY of follow-up in the ‘PDR and resistant’ group compared to 19 per 100 PY of follow-up in the no PDR/‘PDR and fully active cART’ group.

Fig. 1
figure 1

Kaplan-Meier estimate for proportion of children with virological failure. Risk of failure in children harboring virus fully “susceptible” to cART (black continuous line) versus risk of failure in children with virus “resistant” to at least one drug (red continuous line) (Log-rank test: P = 0.0954). The dotted lines correspond to 95 % confidence intervals. In total, 21 patients stopped cART between cART initiation and 6 months of follow-up and were censored at cART stop. This explains why at 6 months there are 427 participants at risk in the no PDR/PDR and susceptible group and 28 participants at risk in the PDR and resistant group

Full size image

In univariable analysis, younger age, year at cART initiation (1998–1999 vs 2007–2008), initial cART regimen (Unboosted PI plus ≥ 2NRTIs vs NNRTI plus ≥ 2NRTIs) and region of origin (African vs Europe) were significantly associated with VF (Table 3). PDR and resistance to ≥ 1 prescribed antiretroviral drugs was not significantly associated with a higher risk of virological failure in univariable analysis (HR, 1.78; 95 % CI, 0.89–3.54; P = 0.101) and also after adjustment for covariables (aHR, 1.42; 95 % CI, 0.62–3.24; P = 0.400).

Table 3 Univariable and multivariable analysis of risk factors for virological failure
Full size table

Upon multivariable analysis, age remained associated with VF and the risk of VF decreased by 12 % per year older at cART initiation (HR 0.88; 95 %CI, 0.82–0.95; P < 0.001). The association between initial cART regimen and virological failure persisted in the multivariable model. In stratified analysis according to the initial treatment regimen, the association between PDR and resistance to ≥ 1 prescribed antiretroviral drugs was not significantly associated with a higher risk of virological failure neither in the > =2NRTI + 1NNRTI stratum nor in > =2NRTIs + an unboosted PIs stratum (Additional file 2: Figure S1).

Discussion

In this large international pediatric multicohort analysis, the prevalence of PDR mutations in antiretroviral naïve children was 7.8 % (95 %CI, 5.5–10.6) and was similar to that reported in a large European multicohort study, 9.5 % (95 % CI, 8.9–10.1) of 10,056 patients, mostly adults [6]. Most of mutations were associated with resistance to NRTIs and NNRTIs reflecting the drug classes used over that period. Other studies conducted in different settings with smaller population size reported a frequency of drug resistance mutations ranging between 5.7 and 100 % depending on the age of children at time of testing, the birth period and the genotyping technique sensitivity [7, 9, 10, 12, 17–25].

Cumulative incidence of virological failure tended to be higher among children starting cART with PDR and resistance ≥1 drug prescribed than among those starting with a fully active regimen (Log-rank test: P = 0.095) but no direct association between PDR mutations and VF was found, in contrast to what was reported in the large European multicohort study [6]. This may be a consequence of the smaller sample size of our population and the relatively low prevalence of PDR mutations. Based on our population, we could achieve 75 % power to detect a HR of 3.13 [35]. Another possible reason may have been the long time interval (6 years or more in over half the population) between perinatal infection and the onset of treatment. In this study, higher frequency of PDR mutations was indeed observed in children less than 2 years. During this interval, minor populations of drug-resistant virus, in particular populations that might emerge following maternal or neonatal prophylaxis with either zidovudine as monotherapy or single-dose nevirapine, might have diminished or disappeared altogether, so that individuals with minor resistant variants might have been categorized as entirely without mutations [12, 36, 37]. Indeed high-sensitivity methods for measurement of drug resistance were not used in this study. However the usefulness of using high-sensitivity methods over consensus sequencing to predict virological failure in children is unclear [12, 13].

High virological failure rates on first-line cART have been reported in children in different settings [17, 38–42], including within clinical trials [11, 43, 44]. Among independent predictors of virological failure identified were use of nevirapine vs efavirenz or ritonavir [41, 43–46], poor adherence to ART [47], prior exposure to single dose of nevirapine presence of baseline resistance [12, 14], and younger age [39]. In our study VF was more likely to occur when cART was started at younger ages (HR 0.88). This association of outcome with age is probably due to several factors. Treatment of younger children during the decade that ARVs were started depended heavily on liquid forms of ARVs that, particularly for the more potent protease inhibitors, are often unpalatable. Also during the time of this study information about the pharmacokinetics of several important drugs, particularly nevirapine and nelfinavir, was limited in children aged less than 2 years and required dose adjustment [48, 49]. Fixed drug combinations suitable for children were not available during the period of this study, so large pill burdens or unpalatable liquid formulations [50] were a common problem in pediatrics [51–54]. For those reasons, adherence and therefore efficacy of cART were often low. The risk of virological failure is correlated with the proportion of missed doses, but impact of nonadherence on viral resistance depends on pharmacology (regimen potency, pharmacokinetics, drug interactions) and viral (viral fitness, and resistance barrier to ARV) factors. Suboptimal or poor adherence (missed/late doses) to drugs can result in sub-therapeutic plasma concentrations of drugs and subsequent development of drug resistance to one or more drugs in a given regimen, and possibly cross-resistance to other drugs of the same class. In the absence of drug pressure as in treatment interruptions or discontinuations, resistance mutations are unlikely to be selected.

Initial cART regimen was significantly associated with virological failure, however these results should be interpreted with caution as children were not randomized to initial treatment and thus results may be subject of indication bias, e.g. children with more advanced disease status (higher proportion of children with a previous AIDS diagnosis) were more likely to receive boosted PI regimen.

Our findings of higher risk of VF in younger children support the 2013 WHO recommendation to treat all children <3 years of age with boosted-PI [27]. In addition, due to the higher barrier to drug resistance of boosted PIs the effect of PDR might be of limited importance today for clinical practice. Nevertheless pre-existing NNRTI resistance may still compromise response to PI-based first line cART. Indeed PI resistance can emerge in children with PDR as evidenced by the presence of multidrug resistance mutations in single HIV genome in children failing PI-based first line [55].

Our study has some limitations. It was an observational study and some resistance results were obtained retrospectively. However, all children starting cART with a sample for HIV genotypic resistance testing were included in the analysis and testing was performed with no knowledge of the outcome. Association of outcome with age may be different at the era of WHO PMTCT option B+. Indeed, in that context more pregnant women receive NNRTI-based cART and children who may become infected with resistant virus selected during treatment failure may have NNRTI resistance mutations that can persist longer than those selected by single dose-nevirapine. Another limitation is that we were not able to assess the effect of the pretreatment CD4 percentage since this was measured in only 49 % of children and to address adherence and specific HIV subtypes effect. Lastly, despite the large study population it was not possible to specifically analyze more substrata as the power was already limited for the overall question of this analysis.

Conclusions

Our study shows that the risk of VF decreased with age at treatment initiation. We hypothesize that this finding might be due to fading of PDR mutations over time although the direct association between those resistance mutations and VF did not reach statistical significance potentially due to lack of power. Furthermore, misclassification due to the relatively low sensitivity of population sequencing in detecting mutations in viral subpopulations long after initial infection or exposure to ARVs for the prevention of mother to child transmission shortly prior to infection could have occurred diluting the effect. Also, lack of appropriate formulations, in particular for the younger age group, which may possibly have resulted in poorer adherence, might be an important determinant of virological failure.