Reduction in Mortality With Availability of Antiretroviral Therapy for Children With Perinatal HIV-1 Infection FREE

Clinical trials conducted among adults infected with human immunodeficiency virus type 1 (HIV-1) have shown that combination antiretroviral therapies are more efficacious than monotherapy in reducing the risk of acquired immunodeficiency syndrome (AIDS) and death.^1- 2 Since the adoption of combination therapy in clinical practice in the mid-1990s, data from observational population-based studies and surveillance systems have shown a reduction in the rate of progression to AIDS and HIV-related death among adults living in industrialized countries, thus confirming the effectiveness of these therapies at the population level.^3- 10

Clinical trials conducted among limited numbers of perinatally HIV-infected children have shown that combined antiretroviral therapy reduces viral load and increases CD4 lymphocyte counts^11- 14; however, to our knowledge, no studies have evaluated the effectiveness of combination therapy on survival among children in settings other than clinical trials.

The Italian Register for HIV Infection in Children,^15- 26 a large data set of children from the general population prospectively evaluated with homogeneous criteria since 1985, can provide important information on the course of pediatric HIV infection in an industrialized country. In the present study, we used the data from this register to estimate the effectiveness of antiretroviral therapy on the survival of perinatally infected children.

Data were provided by the Italian Register for HIV Infection in Children, which is involved with a nationwide multicenter study, instituted in 1985 by the Italian Association of Pediatrics, of children perinatally exposed to HIV.^15- 26 Data collection started on June 1, 1985; for children born before that date, data were collected retrospectively (approximately 13% of children for some data.) The register's data source is a network of 106 pediatric clinical centers located throughout Italy. These centers, which participate voluntarily, aim to enroll all children born to HIV-infected mothers and forward data to 2 coordinating centers, located at the Departments of Pediatrics of the University of Florence and the University of Turin. The study was approved by review boards and ethics committees for the participating institutions. The Italian Register for HIV Infection in Children and the present study involve epidemiological observational analyses that do not provide drug prescriptions, take blood samples, or perform any interventions with the children. The data are treated anonymously in a very rigorous manner.

Data regarding mother-child pairs are collected using specific individual forms for registration and follow-up.^15- 25 The registration form includes the following information: child's demographics, age at first observation at the pediatric clinical center, mother's risk factors for HIV infection, treatment with antiretroviral therapy during pregnancy, mother's clinical condition at the time of delivery, gestational age, birth weight, and type of feeding. Both the registration and follow-up forms are used to obtain information on infection status, HIV antibodies (investigated using enzyme-linked immunosorbent assay and Western blot technique), virus markers (proviral DNA, virus culture, and free and complexed p24 antigenemia), CD4 lymphocyte counts (measured using the standardized fluorescent-activated cell sorting technique),^16,25 disease stage according to the Centers for Disease Control and Prevention (CDC) pediatric classification,²⁷ fulfillment of criteria for AIDS diagnosis,²³ age at AIDS diagnosis, laboratory test results, HIV-related signs and age at the appearance of single signs, age at which the child entered the clinical and immunological categories of the CDC pediatric classification system, age at death, causes of death, dates at the beginning and end of treatment with antiretroviral therapy and specific antiretroviral drugs used, dates at the beginning and end of prophylaxis for Pneumocystis carinii pneumonia (PCP) and specific drugs used, and date of last follow-up visit. Any information concerning the mother's injecting drug use and treatment with antiretroviral therapy during pregnancy is confirmed by reviewing medical records from infectious disease centers, obstetrics wards, and services for drug addiction.²⁵ The register has only recently begun to collect information on viral load. Information on virus drug resistance in children, maternal viral load at the time of delivery, maternal concomitant infections prior to delivery, or therapies prior to pregnancy is not collected; maternal immunological condition at the time of delivery is known in a minority of cases.²⁵

Follow-up forms are filled out every 6 months by an appointed pediatrician at each center.¹⁷ Pediatricians representing participating centers meet at least annually to audit proceedings and standardize procedures.^15- 26 According to these procedures, infected children are examined at least every 2 months, and any clinical or immunological changes are reported on the follow-up form.^22- 23,25

Ascertainment of Death, Definition of HIV-Related Death, and Representativeness of the Study Population. A large proportion of deaths occurred at the participating centers. When death occurred at nonparticipating hospitals or at home, ascertainment was carried out by the participating center in cooperation with the hospital or family pediatrician. We defined HIV-related death as any death attributable to HIV-related conditions included in the pediatric CDC classification system.²⁷ To determine the representativeness of the data set and the proportion of children lost to follow-up, data were cross-matched with 1998 data from the Italian National AIDS Registry, which is responsible for collecting mandatory reports of all AIDS cases diagnosed in health facilities nationwide.⁶ A code based on the child's name, date of birth, and sex was used. Of the 666 perinatally infected children with AIDS included in the Italian National AIDS Registry, 576 (86.5%) were also present in the Italian Register for HIV Infection in Children. Of these 576 children, 21 (3.6%) were still AIDS-free at the last reported visit to the pediatric clinical center participating in the Italian Register for HIV Infection in Children program (ie, children lost to follow-up). Furthermore, the Italian National AIDS Registry recorded 31 deaths (5.4% of the 576 matched children) not reported in the Italian Register for HIV Infection in Children.

Diagnostic Criteria and Treatment. The CDC pediatric classification system was used to define infection status and clinical and immunological conditions.²⁷ Infection was diagnosed via detection on at least 2 occasions of virus markers or the persistence of HIV antibodies after 18 months of life.^22- 25,27 All children were born to women with documented HIV infection at delivery.²⁰

Children followed up from birth were those born at 1 of the participating centers to mothers diagnosed as HIV infected before or at delivery, as previously defined.^19,24 Some of the children not prospectively followed up from birth were born in nonparticipating hospitals and transferred to a participating center a short time afterward to define infection status, usually before the onset of any HIV-related disease.²⁰ Definitions of preterm delivery (gestational age ≤36 weeks) and low birth weight (≤2500 g) used for this study have been reported elsewhere,^{18- 19,23- 25} as have the criteria for disease definition.^22- 25 The CDC classification for HIV infection in adults was used to define the mothers' clinical condition at the time of delivery.²⁸

The specific therapy offered is left to the discretion of the participating centers and is exclusively based on clinical and laboratory evaluation.^{17,22- 23,25} The criteria generally followed were initially those discussed during the periodical meetings of pediatricians from participating centers^15- 26 and subsequently those of the CDC²⁹ and the Italian²⁶ pediatric guidelines for antiretroviral therapy. For the purpose of the present study, a therapy was counted when it had been administered for at least 30 days.⁸ Therapy was categorized as 1 of the following: monotherapy (1 nucleoside reverse transcriptase inhibitor [NRTI]), double combination therapy (2 NRTIs), and triple combination therapy (2 NRTIs and 1 protease inhibitor [PI], or 3 NRTIs, or 2 NRTIs and 1 nonnucleoside reverse transcriptase inhibitor [NNRTI]).

We performed statistical analyses on the data collected through June 30, 1999, and relative to children with confirmed HIV infection born before January 1, 1998; children born after this date were excluded to reduce potential biases related to delayed enrollment in the study or still undetermined HIV infection (symptomatic children are more likely to have a confirmed diagnosis of infection earlier than asymptomatic children). The date of birth was assumed as the time of infection.³⁰ Descriptive analyses were performed to establish the type of combination therapy administered and the proportion of children treated over time.

The main outcome measure of the study was time to HIV-related death. Children whose death was not directly attributable to HIV infection were censored at the date of death. Those still alive were censored at the date of the last follow-up visit. We did not use the information on death from the Italian National AIDS Registry because this registry does not collect detailed information on therapy, and it is not possible to evaluate the treatment effect during the period between the last visit and the dates of events provided by the AIDS Registry.

Standard time-to-event techniques were used. The Kaplan-Meier product limit method incorporating late entries^31- 32 was used to estimate the cumulative probability of survival at different ages. The estimates were calculated for birth cohort and calendar period and grouped according to the distribution of predominant type of antiretroviral therapy administered over time (in part reflecting availability): 1980-1989 (reference group), 1990-1995, and 1996-1998 (1996-1997 only for birth cohort because those born later were excluded from this analysis). For calendar period, the time from birth to HIV-related death (or to the date of the last follow-up visit if still alive) of each child was divided into several periods, thus allowing children to enter the risk set at the date of birth, date of entry in the Italian Register for HIV Infection in Children, or date of the beginning of each calendar period, whichever was the latest. Analogously, we allowed children to exit the risk set at the date of death (or date of last visit if still alive) or at the date of the end of the calendar period, whichever was earliest. Comparisons among cumulative probability curves were tested for statistical significance using the log-rank test.

Multivariate Cox proportional hazards models were then applied. In a first analysis, we evaluated temporal changes in the risk of death adjusted for the effect of birth cohort and of calendar period, in addition to other potential determinants, such as mother's clinical condition at delivery,³³ treatment with antiretroviral therapy during pregnancy,²⁵ gestational age,³⁰ and birth weight.²³

All these determinants were included as categorical variables; if there was missing information, we added "information not known" as a specific category for that determinant. The calendar period was introduced in the model as a time-dependent variable. In a second analysis, we assessed the effectiveness of antiretroviral therapy (ie, monotherapy or double or triple combination therapy) by modeling each treatment regimen as a time-dependent covariate. Each covariate was initially set to 0 and became 1 when the specific therapy was started. If a child interrupted that specific therapy, the covariate remained set at 1. Adjustment was made for the same variables as in the first analysis, as well as for immunological and clinical category, and PCP prophylaxis (all as time-dependent variables). Finally, in a third analysis, the 2 previous models were combined to evaluate whether temporal changes in the risk of death were associated with the effectiveness of the therapy regimens used. The proportionality assumption of the Cox model was tested through the inclusion of time-dependent variables for the other determinants previously described. Finally, to assess potential biases, we repeated all the analyses excluding children born before 1986 or those not followed up from birth. All of these models were stratified by clinical center.

Overall, 4564 children born to HIV-infected mothers before January 1, 1998, were enrolled; of these children, 3121 (68.3%) were prospectively followed up from birth. During a median follow-up time of 3.1 years (range, 0.1-17.4 years), 3072 children (67.3%) seroreverted, 350 (7.7%) still had an indeterminate infection status at the last available visit, and 1142 (25.0%) were infected.

The distribution of infected children according to year of birth is shown in Table 1. Of these, 599 (52.5%) were girls, 651 (57.0%) were born to a mother who was an injecting drug user or a sexual partner of an injecting drug user, and 450 (39.4%) were prospectively followed up from birth. Overall, 560 children (49.0%) were enrolled within 3 months of birth; 664 (58.1%), within 6 months of birth; and 765 (67.0%), within 1 year of birth.

Regarding the mother's clinical category at the time of delivery, 731 children (64.0%) were born to a mother belonging to the CDC adult classification²⁸ category A; 119 (10.4%), to a mother in category B; and 29 (2.5%), to a mother in category C. This information was missing for 263 children (23.0%). Forty-five (3.9%) children were born to a mother who had received antiretroviral therapy during pregnancy: 42 of these women received zidovudine monotherapy; 2 women, zidovudine and lamivudine; and 1 woman, a triple combination therapy that included zidovudine.

During a median follow-up time of 5.9 years (range, 0.1-17.4 years), 644 children (56.4%) entered the CDC pediatric classification clinical category C at a median age of 1.6 years (range, 0.2-15.6 years), 554 (48.5%) entered immunological category 3 at a median age of 2.7 years (range, 0.2-13.4 years), and 421 (36.9%) died of HIV-related causes at a median age of 3.3 years (range, 0.1-14.4 years). Six additional children died of causes not related to HIV.

Figure 1 shows the distribution of the type of antiretroviral therapy administered by calendar year. The first antiretroviral therapy (zidovudine monotherapy) was administered in 1987. The use of monotherapy steadily increased until 1993. The most frequently administered drugs were zidovudine (696 children [73.6%]) and didanosine (154 children [16.4%]). Double combination therapy was first used in 1993; the most frequently administered combinations were zidovudine and lamivudine (265 children [35.3%]) and zidovudine and didanosine (193 children [25.7%]). Double combination therapy became the most common antiretroviral therapy used after 1995. In 1998, triple combination therapy became the most common therapy; the most frequently used combinations were stavudine, lamivudine, and ritonavir or nelfinavir or indinavir (191 children [35.3%]) and zidovudine, lamivudine and ritonavir or nelfinavir or indinavir (117 children [21.6%]). In a minority of children, zalcitabine (n=107) among NRTIs (for double combination therapy), nevirapine (n=18) (for triple combination therapy) (no other NNRTIs were used), and saquinavir (n=77) among PIs were also used. Overall, for 719 children (62.9%), monotherapy was initiated (106 in CDC pediatric classification clinical category N, 116 in CDC category A, 293 in category B, and 204 in category C; 376 in immunological category 1, 184 in category 2, and 159 in category 3); for 551 children (48.2%), double combination therapy was initiated (69 in clinical category N, 86 in category A, 204 in category B; and 192 in category C; 149 in immunological category 1, 158 in category 2, and 244 in category 3); and for 371 children (32.4%), triple combination therapy was initiated (36 in clinical category N, 46 in category A, 128 in category B, and 161 in category C; 83 in immunological category 1, 83 in category 2, and 205 in category 3). It is of note that triple combination therapy was started as first antiretroviral therapy in 47 children (12.7%). The median ages at initiation of monotherapy and double and triple combination therapy were 2.1 years (range, 0.1-12.4 years), 6.0 years (range, 0.1-15.6 years), and 7.7 years (range, 0.2-16.8 years), respectively. The clinical and immunological staging of the children not receiving antiretroviral therapy (not administered because of being asymptomatic or only mildly symptomatic²⁶; none had an advanced immunological status) in 1998 and during the first 6 months of 1999 was as follows: N1 or A1 (69.7%), B1 (28.5%), and C1 (1.8%). With regard to PCP prophylaxis, 864 oral trimethoprim-sulfamethoxazole or aerosolized pentamidine isethionate treatments were prescribed to 657 children (57.5%).

Figure 2A shows the Kaplan-Meier curves of the cumulative probability of survival by birth cohort. The probability of survival did not significantly differ between the 1980-1989 and 1990-1995 birth cohorts (P=.15), whereas it was significantly higher for the 1996-1997 birth cohort compared with the 1980-1989 (P=.01) and 1990-1995 (P=.01) birth cohorts.

In evaluation of the calendar period effect (Figure 2B), the cumulative probability of survival did not significantly differ between children at risk in 1980-1989 and 1990-1995 (P=.75), whereas probability of survival was significantly higher for 1996-1998 vs 1980-1989 (P=.003) and 1990-1995 (P=.001). Because of the varying mix of entries in each period (see "Methods" section), survival curves are not concordant with those for the birth cohort.

The crude and adjusted relative hazards (RHs) of death by birth cohort and calendar period are shown in Table 2. The crude RH of death vs the 1980-1989 birth cohort for the 1990-1995 birth cohort was not significantly different from 1.0, whereas it was significantly lower than 1.0 for the 1996-1997 birth cohort. Similar results were seen for calendar periods 1990-1995 and 1996-1998 vs 1980-1989 (Table 2). The adjusted RH of death by birth cohort compared with the 1980-1989 cohort was close to 1.0 for the 1990-1995 cohort and less than 1 for the 1996-1997 cohort, although the difference was not significant. The same level of statistical association was found for calendar periods 1990-1995 and 1996-1998 (Table 2). However, in a multivariate model using as comparison the period 1980-1995 (ie, combining the 2 earliest periods), the 1996-1997 birth cohort had an RH of 0.57 (95% confidence interval [CI], 0.22-1.47), not statistically different from 1.0 (P=.27), but the RH for the calendar period 1996-1998 was 0.63 (95% CI, 0.47-0.85), significantly different from 1.0 (P=.002).

The crude RHs of death, stratified by clinical center, for children treated with monotherapy or double or triple combination therapy compared with no therapy, were 3.29 (95% CI, 2.53-4.28), 0.81 (95% CI, 0.57-1.14), and 0.42 (95% CI, 0.20-0.89), respectively. Table 3 shows the adjusted RH of death for different types of therapy. Compared with nontreated children, the adjusted RH for those who had received monotherapy and double combination therapy was lower than 1.0, but the difference was not statistically significant (P=.13 and P=.07, respectively), whereas the RH for children receiving triple combination therapy was significantly lower than 1.0 (P=.003). Triple therapy had a significantly greater effect than double therapy (P=.005), whereas there was no statistically significant difference between double therapy and monotherapy (P=.49). Table 3 also shows that when birth cohort and calendar period were combined with type of therapy in the same model, the RH for the 1996-1997 birth-cohort and 1996-1998 calendar period did not significantly differ from 1.0 (P=.19 and P=.83, respectively). These analyses were repeated including only children followed up from birth or only those born after 1985, and similar results were obtained (unpublished data from the study).

With regard to the effect of prior zidovudine monotherapy in children subsequently given double or triple combination therapy, we have not observed any deaths among 47 children who started highly active antiretroviral therapy as naive; 4 deaths occurred among the 263 who previously received monotherapy.

Although clinical trials provide the most accurate estimate of the biologic efficacy of therapy,⁴ they are carried out under very stringent conditions, including selection for compliant patients and/or families and monitoring for compliance.⁴ Furthermore, the surrogate end points used (ie, CD4 lymphocyte counts and viral load depletion) can result in misleading conclusions,³ whereas clinical end points such as disease progression and survival require long observation periods and may provide obsolete information, given that new therapies continue to be developed.³ Observational studies provide information on effectiveness at the population level, thus allowing temporal changes in relevant end points to be evaluated.^3- 4

Several observational studies have documented the effectiveness of combination therapy among HIV-infected adults^3- 10; however, for perinatally HIV-infected children, we are unaware of studies on combination therapy that take place outside the setting of controlled trials.^11- 14 Our study shows that although the survival of HIV-infected children in Italy remained unchanged up to 1995, it has significantly improved since 1996, with a more than 30% reduction in the adjusted risk of death for children at risk in the period 1996-1998 vs those at risk in the period 1980-1989. This can be explained by the fact that in 1996 the use of double combination therapy became widespread and triple combination therapy was introduced in the care of HIV-infected children. The incremental protective effect of therapy was also evident, in that the risk of death decreased by 30% for double combination therapy (although statistically nonsignificant) and by approximately 70% for triple combination therapy. Finally, the estimated decrease in the risk of death since 1996 disappeared after adjusting for type of therapy, suggesting a causal relationship between decreased risk of death and use of combination therapy. Thus, overall, our results indicate that antiretroviral combination therapy reduced, at the population level, the risk of death in perinatally infected children.

All therapies were associated with a reduced RH of death, and a gradient effect was evident: triple combination therapy was the most effective, followed by double combination therapy and monotherapy. Differences were statistically significant between triple and double combination therapies but not between double combination therapy and monotherapy. This latter finding is in contrast with the results of a large clinical trial in adults,¹ although it is possible that the insufficient period of observation (the introduction of the double combination therapy was rapidly followed by that of triple combination therapy) led to our study having limited statistical power. The small number of antiretroviral-naive children treated with triple combination therapy did not allow us to evaluate the hypothesis that previous treatment can reduce the efficacy of triple combination therapy.

Population studies can be biased in various ways: reporting delay and differences in the timing of diagnosis may change the size of the population at risk,³⁴ only basic data can be collected,^9,34 and the study group may be too small or insufficiently representative of typical patients and/or of typical physicians caring for these patients.^5,34 Inequalities in access to care and drugs may exist,^4,34 and homogeneous determinants of outcome are not discriminant inclusion criteria.⁷ Furthermore, patients with more severe disease are more likely to be lost to follow-up because of a switch to terminal care, which leads to informative censoring,^7,34 and the fact that patients died before the introduction of antiretroviral combination therapy may have led to a survivor treatment selection bias.^6- 7

Nonetheless, we believe these potential biases did not greatly affect our results. The Italian Register for HIV Infection in Children is one of the world's largest cohorts of HIV perinatally infected children; the children are prospectively evaluated with homogeneous criteria.^15- 26 Participating centers are located throughout the country, with their geographic distribution reflecting the distribution of perinatal HIV infection in Italian children.^19- 21 Furthermore, the large proportion of data overlapping with the Italian National AIDS Registry data indicates that this data set is very likely to be representative of HIV-infected children in Italy. Since participating pediatricians attempt to enroll all exposed children,^19- 21,25 changes in disease outcome are not biased by changes in perinatal HIV incidence.²⁵ Data are available on an individual patient basis,^15- 25 allowing the link between changes in disease outcome and therapy to be investigated. The practice of filling out follow-up forms every 6 months¹⁷ reduces the time lag between the disease outcome and the report reaching the coordinating centers; participating pediatricians directly provide care for children; and access to care and drugs is uniform and universally covered under public programs. Clinical and laboratory diagnostic criteria,^17,19,23,25 laboratory methods,^16,22,25 and the timing of follow-up^22- 23,25 are standardized during periodical meetings of participating pediatricians.^15- 24 The network of centers throughout the country minimizes the proportion of children who would be lost to follow-up because of transfer to another care facility,²¹ and related potential informative censoring bias³⁴ is further reduced by standing access to and cross-matching with surveillance data.⁶ In our study, children born after 1997 were excluded to eliminate potential biases related to still undetermined HIV infection status or delayed enrollment; all models were stratified by clinical center and adjusted for known potential determinants of outcome^23,25,33; and the use of time-dependent covariates addressed the survivor treatment selection bias.⁷

One goal of this study is to monitor the trends over time of characteristics of interest in children born to HIV-infected mothers, and thus all children observed at any time after birth are reported and prospectively followed up to evaluate the course of HIV infection. This inclusion of late entries explains why only 39.4% of the HIV-infected children were prospectively followed up from birth. Although this percentage is low and could thus affect the representativeness of the results, this is unlikely because of the type of statistical analysis performed (ie, survival analysis with late entry) and the high degree of overlap of our data set with that of the Italian National AIDS Registry; furthermore, similar results were obtained when we restricted the analysis to children prospectively followed up from birth or born after 1985.

We considered maternal antiretroviral treatment during pregnancy as a covariate because, in a previous study, we observed a more rapid disease progression in a selected group of children (born 1992-1997 and followed up from birth) whose mothers received zidovudine prophylaxis during pregnancy compared with age-matched children of untreated mothers.²⁵ However, the influence of this determinant on the comparison of the 3 calendar periods has probably been small given that only 3.9% of children were born to a mother who received antiretroviral treatment during pregnancy.

Theoretically, several factors could negatively influence the effectiveness of antiretroviral therapy in children. The natural history of perinatal infection is characterized by early onset of symptoms, rapid disease progression, rapid CD4 lymphocyte loss, and severe clinical course.^17,22- 23 The HIV load is usually higher in children than in adults,³⁵ probably because of the consequences of viral replication within the expanding lymphoid mass in infancy.³⁶ It is difficult to achieve sustained suppression of viral replication in children even with triple combination therapy,^26,37 and the regeneration of CD4 lymphocytes (which is probably a thymus-dependent process in children as well as in adults)^38- 39 could be impaired by the detrimental effects on the thymus of HIV infection acquired very early in life.³⁹ Finally, adherence and compliance to therapy may be reduced in children, particularly in the case of combined therapy.²⁶ The lack of pediatric formulations and poor palatability of antiretroviral drugs cause difficulties in treatment of infants and young children.²⁶ Parents of older children may be reluctant to accept therapy, because drugs taken several times a day generate problems in terms of discretion, and therapy may disclose HIV infection status both to the child and others.²⁶

Notwithstanding the biologic characteristics of perinatal HIV infection and difficulties in treating infants and children, our study shows the effectiveness at a population level of triple combination therapies in the pediatric age group. The effectiveness in infants and children is at least similar,^3,5- 6,9 or even greater,^4,8 than that observed in adults.