Review of the HIVNET 012 Perinatal HIV Prevention Study (2005)

This chapter evaluates the HIVNET 012 trial efficacy endpoints (HIV-1 positivity and HIV-1-free survival¹ of infants) and safety endpoints—their design, their implementation, and the committee’s analysis.

The primary efficacy endpoints for HIVNET 012 were infant HIV-1 positivity and HIV-1-free survival at 6–8 weeks, 14–16 weeks, and 18 months of age. At the time the initial results were reported in The Lancet (Guay et al., 1999a), referred to in this report as Lancet I, there was insufficient duration of follow-up to enable reporting of 18-month efficacy; thus, the focus was on efficacy at the 6–8 week and 14–16 week evaluation times. A second paper published in The Lancet (Jackson et al., 2003), referred to in this report as Lancet II, reports results through month 18 of follow-up.

Samples for determination of infant HIV-1 infection status were obtained within 24 hours of birth, and at 6–8 weeks, 14–16 weeks, and 18

months of age. These timepoints for sampling were chosen to optimize the determination of antepartum, intrapartum, postpartum (breast feeding), and cumulative HIV-1 positivity. In addition, because infant mortality was of paramount interest in any potential intervention for mother-to-child transmission, a composite endpoint of HIV-1-free survival was included (Guay et al., 1999; Jackson et al., 2003).

HIVNET 012 used qualitative plasma HIV-1 RNA polymerase chain reaction (PCR) assay for determination of HIV-1 infection status of infants at 24 hours, 6 weeks, and 14 weeks of age. Positive test results were confirmed by quantitative plasma HIV-1 RNA PCR assay or HIV-1 culture. An HIV-1 EIA (enzyme immunoassay) test was performed at 18 months of age; a positive result was confirmed by HIV-1 Western blot (Guay et al., 1999; Jackson et al., 2003).

Qualitative plasma HIV-1 RNA PCR, quantitative plasma HIV-1 RNA PCR, and qualitative HIV-1 DNA PCR assays have been shown in multiple studies to be highly sensitive and specific for diagnosis of HIV-1 infection in young infants, especially after the first week of life. Simonds and colleagues (1998) reported a direct comparison of qualitative plasma HIV-1 RNA and DNA PCR assays performed on paired specimens from HIV-1-infected and uninfected infants less than 3 months of age. The sensitivity of the qualitative RNA assay was 38% at <7 days of age (95% confidence interval [CI], 22–56%), 97% at 7–41 days of age (95% CI, 88–100%), and 95% at 42–93 days of age (95% CI, 83–99%). Test specificity was 99% (95% CI, 97–100%). The authors concluded that the qualitative RNA assay was highly specific and more sensitive than qualitative HIV-1 DNA PCR for diagnosis of HIV-1 infection in young infants. These findings were confirmed in another comparative study reported by Cunningham et al. (1999).

HIVNET 012 included as one of three primary study endpoints the “safety/tolerance of oral nevirapine (NVP) and oral zidovudine (AZT, now ZDV) given to pregnant Ugandan women during labor and their neonates in the first week of life.” Both adverse events (AEs) and serious adverse events (SAEs) were recorded on case report forms (CRFs) and transmitted to the Statistical Center for HIV/AIDS Research and Prevention (SCHARP). SAEs also were reported directly to the Division of AIDS (DAIDS) through the adverse experience reporting mechanism in place at the time.

The HIVNET 012 study protocol specified that mothers be followed for AEs and SAEs for 6 weeks after delivery (HIVNET Group, 1997). Infants were followed for AEs until 6 weeks of age, and for SAEs until 18 months of age. For mothers, clinical and laboratory evaluations for safety were performed at delivery, and at 24–48 hours, 7 days, and 6–8 weeks postpartum. Infants were evaluated at the ages of 24 hours, 7 days, 6 weeks, 10 weeks, 14 weeks, 6 months, 9 months, 12 months, and 18 months of age. Laboratory testing, which included hematology and serum chemistries, was not required at every visit. The HIVNET 012, Study Specific Procedures (version 1.0, November 1997) state that toxicities were graded based on the DAIDS Toxicity Tables² for neonates, children, and adults, with a range from Grade 1 (mild) to Grade 4 (life-threatening) (SCHARP, 1997). A special grading system, outlined in the Study Specific Procedures manual was used for cutaneous/skin rash/dermatitis adverse experiences and maternal hemoglobin values (SCHARP, 1997). Dose modification for management of adverse events was not used in the study.

The HIVNET 012 protocol, version 1.0, defined an SAE as follows:

This definition was consistent with the definition of an SAE as it appeared at the time in the Code of Federal Regulations (21 CFR 312). After HIVNET 012 was initiated in its original form (November 1997), and about the time that enrollment was re-initiated after the placebo arms had been dropped (April 1998), the SAE definition in the Code of Federal Regulations (21 CFR 312) was revised. However, based on information from DAIDS and the U.S. Food and Drug Administration (FDA), HIVNET 012, which was already ongoing, was not required to adopt the revised regulations.

Infants were evaluated clinically during study visits, and were administered laboratory tests to determine HIV-1 infection status at established time points during their first 18 months of life. For infants who missed a scheduled study visit, survival status was confirmed by health visitors who conducted outreach and follow-up. For participants who were confirmed dead, this information was documented in the source documents and onto CRFs for transmittal to SCHARP. (As noted earlier, the HIVNET 012 study staff maintained records of adverse event experiences in all participants through 6 weeks, and of SAEs in infants through 18 months of age, later extended to 5 years of age as per protocol.) Adverse event information was generally recorded on source documents that were transcribed to CRFs and regularly transmitted to SCHARP for entry into the study database and analysis (Guay, 2004; SCHARP, 2004).

The audit and remonitoring of HIVNET 012 conducted by DAIDS in 2002 reported no major deficiencies related to the laboratory, from the collection and storage of samples, to the laboratory source documentation. Westat’s limited laboratory audit reported that although the laboratory was not yet certified, it had participated in several proficiency programs, there was documentation of ongoing quality assurance programs, and laboratory standard operating procedures were documented and implemented (Chamberlin et al., 2002). The DAIDS remonitoring reported that the laboratory was operating according to Good Laboratory Procedures and that laboratory values in the clinic source file matched the actual laboratory source documents (DAIDS, NIAID, 2003a).

Laboratory data completeness and accuracy was monitored by SCHARP, which would e-mail the site for resolution of any inconsistencies. Inconsistencies that would generate such an e-mail included, for example, a positive HIV-1 PCR test followed by a negative test, missing laboratory results from a scheduled visit, a positive HIV-1 test result without confirmation and inconsistent results between an HIV-1 antibody test and PCR (SCHARP, 2004).

In the implementation of HIVNET 012, the investigators had to contend with the practical implications of identifying and reporting to the

study database a great many adverse events (AEs and SAEs), the majority of which represented co-morbid conditions common among HIV-1-infected and HIV-1-uninfected women and infants in Uganda, for example, malaria, anemia, or tuberculosis. With a principal objective of capturing all clinically relevant events while at the same time reducing the background noise introduced by the high prevalence of co-morbid conditions, the investigators consistently interpreted the SAE definition contained in the 1996 and 1997 Code of Federal Regulations and the protocol so as to use hospitalization as the primary determinant of seriousness for capture of SAEs. If a study participant experienced a condition that required hospitalization, that condition was considered an SAE. To the extent possible, investigators attempted to capture unifying conditions or diseases by name (e.g., gastroenteritis), rather than as a collection of individual symptoms or signs (e.g., vomiting and diarrhea). The investigators interpreted the last phrase of the SAE definition (“… or is otherwise judged to be serious by the onsite clinician”) in a specific manner; rather than using this criterion to capture a wide range of conditions that were neither life-threatening nor triggering hospitalization, the investigators used it narrowly, to capture infrequent occurrences that had not resulted in hospitalization. (See Tables 4-1 and 4-2 for a breakdown of reported SAEs by mother and infant.) As a result of this interpretation, in practice some conditions that may have been considered serious in a different clinical context but that did not result in hospitalization in the local clinical setting were not identified as SAEs.³

The investigators’ interpretation of the definition of SAEs in the Code of Federal Regulations and protocol led to concerns on the part of some auditors and monitors that the use of hospitalization as the principal determinant of seriousness of clinical events may have resulted in undercounting of SAEs. In their 2002 site visit, Westat auditors reported that HIVNET 012 study staff informed them about the SAE definition being used in the study. Westat site visit staff expressed their concern that the protocol definition seemed in practice to make hospitalization the primary threshold for SAEs, which if true, would mean that other serious conditions that were not considered “hospitalizable” might potentially be missing. (As noted earlier in this chapter, the committee finds that the interpretation of SAE used by the HIVNET 012 investigators was reasonable under the circumstances.)

The remonitoring effort overseen by DAIDS in 2002 also expressed

concern about an apparent difference between the protocol (both the original versions and the long-term follow-up amendment) and the “algorithm used by investigators” (DAIDS, NIAID, 2003a). (See committee’s conclusion below, that the definition of SAE had not been altered but that the interpretation of its criteria functioned to make hospitalization the primary threshold for reporting an SAE.) The remonitoring team reported that investigators defined SAEs as “those clinical events leading to hospitalization or death,” noted that this interpretation of the definition was not formally approved by DAIDS or submitted to the Investigatinal New Drug (IND) file, and identified several events unrelated to the study drug that they believed should have been considered SAEs (LaMontagne, 2004; Jackson et al., 2003). While the protocol definition of SAEs used by HIVNET 012 was properly reviewed and approved by the relevant IRBs, protocols do not usually spell out the planned interpretation of each definition.

HIVNET 012 study staff documented clinical and laboratory findings in the source documents and transcribed these to CRFs for each participant. Westat and remonitoring teams reported that some AEs and SAEs were not consistently noted on the proper forms, including CRFs. Also, they reported the absence of documentation about the resolution of some SAEs (Chamberlin et al., 2002; DAIDS, NIAID, 2003a).

According to the 2002 Description of Study Procedures (Guay et al., 2002), data on study participants were collected on a variety of source documents appropriate to the nature of the visit, and key data were transcribed on CRFs that were later transmitted to SCHARP via the DataFax system for input into the study database. A source file (binder) was created for each mother/infant pair of HIVNET 012 study participants. The source file contained primary documentation of clinic visits, including clinical and laboratory information, and scheduled and unscheduled visit forms, secondary documentation such as a “hospital admissions form” tracking a

study participant’s hospitalization at Mulago Hospital (later, actual copies of original hospital admission records), and notes based on patient’s statements about other clinical encounters outside of the study clinic. CRFs were filed in a parallel set of binders for each mother/infant pair. There was no requirement to report unscheduled visits in the CRFs, unless the infant had an AE or SAE or a laboratory test was performed.

The HIVNET 012 investigators created a series of CRFs to capture adverse events. Questions about deaths were asked on the following four CRFs: the Delivery Form (DF-1 for stillbirth), the Illness/AE Form for mothers and infants (AE-N), the Missed Visit Form for mother and infant (MV-01), and Status Change Notice for mothers and infants (SCN-01). Specific questions about adverse events were asked on the following six forms: Delivery Form (DF-2), Follow-up Form (MFU-1), the Concomitant Medications Log for the mother and infant (CM-1), the Birth Form (IB-1), the Infant’s Follow-up Form (IF-1) and the Illness/AE Form (AE-N). Physical exam findings were asked about on the Mother’s Enrollment Form (ME-2), the Delivery Form (DF-1 for meconium in amniotic fluid, chorio-amnionitis, and blood loss and DF-2 for physical exam findings), Follow-up Form (MFU-1), Birth Form (IB-1), and Infant’s Follow-up Form (IF-1). Abnormal laboratory tests and severity grading were noted on the Laboratory Results Form for both mothers and infants (LR-1). The AE-N form also assigned grades, relatedness to study drug, and outcome. Thus, deaths, adverse events, abnormal physical exam findings, and abnormal laboratory data could be captured on several different forms. These were used by SCHARP to assess internal consistency (SCHARP, 2004).

Health visitors’ (outreach workers) logs and notes, developed after visits to participants who missed scheduled visits, or required follow-up, were an additional source of information, especially in identifying participant deaths (DAIDS, NIAID, 2003a).

The Westat site visit team conducted a limited assessment of the quality of record keeping and the conditions of record storage (Chamberlin et al., 2002). The Westat team reported that source primary records and CRFs were generally well organized, and with a few exceptions, were stored in adequately secure locations, but they expressed concern about “hospital admission forms,” which were re-creations by study staff of hospital records. In addition, each time a patient was admitted to Mulago Hospital, the patient was assigned a new unique hospital chart number rather than a consistent unique identifier, thus making it difficult to readily identify all of a given patient’s hospital admissions without a manual search of hospital files. These issues later were clarified as the study staff indicated a goal of

maintaining a comprehensive profile on every study participant, and the investigators received permission from the hospital to gather all records of participants and organize them alphabetically, thus giving HIVNET 012 staff access to the primary hospital files.

Other concerns raised by the Westat audit team included the completeness and accuracy of data recording, and quality assurance mechanisms, such as regular investigator review of CRFs (Chamberlin et al., 2002). In a subsequent visit to the site, the DAIDS remonitoring team reported finding that source files were not consistently dated and signed, and also that changes, inconsistencies, and explanations for protocol violations were not recorded. The Westat team also concluded that corrections on CRFs or other documents were sometimes not properly made (i.e., a single line drawn through) and were not consistently initialed and dated (DAIDS, NIAID, 2003a).

In response to some of the auditors’ concerns, HIVNET 012 investigators took additional steps to ensure adherence to good data-collection and record-keeping practices, including hiring additional staff to conduct quality control and assurance activities, developing standard operating procedures (e.g., for the submission of regulatory documents), and ensuring regular implementation of proper dating and initialing procedures on study forms (Guay and Jackson, 2003).

The Westat site visit team expressed concern about the fact that HIV-positive children who had completed the HIVNET 012 study regimen were enrolled beginning at ages 6 to 7 months in a vitamin A study seeking to determine the supplement’s effects on HIV-related deaths and illnesses. DAIDS agreed that such co-enrollment could “potentially complicate the analysis of long-term safety data,” but found that the co-enrollment was permitted by the protocol which allowed any opportunity for treatment to HIV-infected children participating in HIVNET 012 (DAIDS, NIAID, 2003a).

A small number of infants from both the active arms and the early placebo arms of HIVNET 012 were enrolled in the vitamin A study, including 33 HIV-positive infants from the ZDV arm and 23 infants from the NVP arm. Of the 56 children from the active arms of HIVNET 012 (infants who received placebo were not included in most analyses), only 24 (16 ZDV and 8 NVP) received vitamin A and the remainder received placebo (SCHARP, 2005).

The committee notes that participation in the vitamin A (or any other) trial following infant HIV infection does not have any impact on the

HIVNET 012 primary efficacy endpoints of HIV-1 positivity and HIV-1-free survival, as these endpoints would have occurred prior to enrollment into the vitamin A trial. Additionally, participation in the vitamin A study would have no impact on (nonserious) AEs captured in HIVNET 012 because those were only collected until 6 weeks of age, whereas participation in the vitamin A study did not begin until at least 6 months of age. However, it is possible, in theory, that participation in the vitamin A study could have affected the incidence of SAEs captured in HIVNET 012, since SAEs were collected until 18 months of age. A total of 33 ZDV and 23 NVP infants from HIVNET 012 participated in the vitamin A study. Of these, 16 (ZDV) and 14 (NVP) had at least one SAE after enrolling in the vitamin A study. Therefore, the SAEs occurring after 6 months in these 30 infants, representing 5% of the HIVNET 012 infants, could have been related to vitamin A (Mwatha, 2005a; Mwatha, 2005b).

The incidence of SAEs in the vitamin A study was somewhat lower in infants receiving vitamin A than in those receiving placebo, suggesting either no effect or a possible protective effect of vitamin A on SAE risk. Since more ZDV infants than NVP infants participated in the vitamin A study, a protective vitamin A effect would have the consequence of biasing the HIVNET 012 SAE data in favor of ZDV, since more ZDV infants participated than did NVP infants. If vitamin A had no effect on SAE rates, then participation would neither affect nor bias the HIVNET 012 SAE results.

Thus, while the possibility exists that participation in the vitamin A study may have affected the number of SAEs that were captured in HIVNET 012, the opportunity for this is limited because of the small number of infants that participated, and we see no evidence or reason that such participation could have led to an understated relative safety of NVP in the HIVNET 012 study.

In its report, the Westat site visit team stated that it found that one of the health visitors’ log books, containing notes about follow-up visits to participants’ homes, appeared to be a recent transcription. Upon asking study staff about that, Westat team members were informed that the origi-

nal had been damaged by flooding caused by a plumbing problem. Those log books were secondary materials not used as source documents for the study, and it appears that much of the information contained in the damaged notebook(s) was legible and copied into a new notebook or notebooks (Guay, 2004).

A second concern about the state of study documentation arose when study staff obtained hospitalization records from Mulago Hospital on study participants and began to reorganize those records. Study staff found that a small number of the records had been slightly damaged by rodents or insects, but not to an extent that rendered them unusable (Guay, 2004). None of the participants’ source documents was affected by this event. Based on its review of secondary sources (Westat and DAIDS remonitoring reports) and information heard during the investigators’ presentation before the committee, this committee has concluded that the extent and significance of missing documents was quite limited and has no bearing on the integrity of the study.

Because of the various and somewhat inconsistent reports about the quality and completeness of the HIVNET 012 study data, the committee undertook its own evaluation of HIVNET 012 for the purpose of assessing the quality and completeness of source documents, the consistency between information in source documents and the CRFs, the information captured in the SCHARP data sets, and the timeliness and accuracy with which information was transferred from the source documents/CRFs to the SCHARP database. We focused on infant survival status, HIV-1 PCR/EIA results, adverse events, SAEs, and hospitalizations. Because of the importance of survival status information, the committee examined a subsample of the random sample of 80 mother/infant pairs identified by the EMMES Corporation for the DAIDS remonitoring effort, as this was intentionally selected to oversample infants that died.

Based on its detailed examination of study data, the committee found no evidence of misrepresentation or inappropriate manipulation of the reporting of the original study results.

The committee asked the HIVNET 012 study investigators to provide copies of all source documents on file at the study site in Uganda for 80 mother/infant pairs that were previously identified through a weighted random sample by the EMMES Corporation (EMMES Corporation, 2002). In its review, the committee used a sequential sampling procedure to evaluate a subset of these 80 mother/infant pairs. Only infant records were included in the committee’s review. Because of the greater frequency of adverse events and scheduled PCR/EIA evaluations than SAEs, deaths, and hospitalizations, it was decided that the committee would conclude its review of source documents when information was collected for at least 100 adverse events occurring within the first 6 weeks of life in no fewer than 20 infants. The rationale for evaluating at least 100 adverse events in source documents was that this would allow an estimation of the probability that these would be captured in the analysis database with adequate precision (estimated standard error of 0.03).

Based on this strategy, source documents representing 47 mothers and 49 infants (two sets of twins), were reviewed. Twenty-three deaths and 26 hospitalizations were recorded among the infants included in this cohort. Twenty-seven infants received ZDV, 17 received NVP, and 5 received no study drug. The imbalance in the number of ZDV and NVP infants is a result of the sampling design, which oversampled infant deaths, and the fact that there were more infant deaths in the ZDV arm of HIVNET 012 than in the NVP arm.

Copies of the source documents were transferred from the study site in Uganda to Baylor College of Medicine in Houston, Texas, for processing and review. Approval for copying and review of study documents was obtained from Human Subjects committees at Mulago Hospital/Makerere University and Johns Hopkins University.

Copies of CRFs and data files corresponding to the requested source documents were obtained from SCHARP. All study patient identification numbers were redacted from copies of source documents, CRFs, and data files before review by any committee member or consultant assisting the committee with the review. New, unique committee review identification numbers were assigned to each record to allow linkage between source and SCHARP records to maintain confidentiality. Three sources of information were reviewed for each record:

1. Source documents for each mother/infant pair

2. CRFs for each mother/infant pair

3. SCHARP data file for each mother/infant pair

Two consultants to the committee performed primary review and abstraction of information from these documents. Nancy R. Calles, B.S.N., R.N., A.C.R.N., is a pediatric HIV/AIDS nurse-specialist with more than 14 years of experience as a study coordinator for a wide variety of pediatric and perinatal HIV/AIDS clinical trials. Meg Ferris, M.P.H., has worked for 12 years in pediatric and perinatal HIV/AIDS clinical trials and health-professional education and training. She served for 5 years as a clinical trials specialist for the National Institute of Allergy and Infectious Diseases (NIAID)-sponsored Pediatric AIDS Clinical Trials Group. Ms. Calles conducted primary reviews of CRFs. Ms. Ferris was primary reviewer for the source documents and SCHARP data files.

For each set of records reviewed, the following information was abstracted from source documents and CRFs, and compared to information contained in the corresponding SCHARP data files:

• Date of birth of the infant

• Dates of all scheduled or unscheduled study visits by the infant

• Infant adverse events and dates of occurrence

• Infant clinical serious adverse events and dates of occurrence

• Study drug assignment (ZDV or NVP)

• Date and result of each PCR or EIA assay

• Date of death

• Date study site became aware of death

• Source of verification of death

For the purposes of this review, study definitions of clinical adverse events and serious adverse events were employed. Infant adverse events were recorded through 6 weeks of life; serious adverse events were recorded through 18 months of life.

The committee evaluated the records of the sample of 49 infants to assess the completeness of ascertainment and verification of survival status in the source documents, the timeliness by the site in ascertaining infant deaths, the degree to which survival status information in source documents was accurately transferred to the SCHARP database, and the timeliness in reporting survival status to the SCHARP database. Because of the dynamic nature of information gathering during the conduct of a trial, our comparisons of the source and analysis databases were based on the status of both in June 1999, when the study database was “frozen,” or locked, in preparation of the first publication (Guay et al., 1999a) of the study results,

and in April 2001, when the analysis database was frozen for the second major publication (Jackson et al., 2003).

To assess the completeness and accuracy with which infant deaths (in the sample of 49 infants) were captured by the HIVNET 012 staff, the committee examined the sources of information that led to the ascertainment/verification of infant deaths as well as the completeness of scheduled visits.

Of the 23 deaths noted in the source documents, 16 (70%) occurred in a clinic or hospital, and thus the death and date of death were directly observed and verified. The remaining 7 infants died at home, with the site learning about the deaths a median of 34 days later. In 4 of the 7 cases, the death was verified by the infant’s mother (3 infants) or other relative (1 infant). In 1 case, the source of verification was from neighbors, resulting from a home visit. In 2 cases, the source documents did not indicate the basis for verifying the infant death, but indicated that the site learned of the deaths 34 and 48 days later.

The possibility of unreported infant deaths (false negatives) was assessed by determining each infant’s last study visit prior to the June 1999 and April 2001 data freeze dates for the main study publications. Missed visits immediately prior to the freeze dates could, in theory, reflect unrecognized infant deaths. There were three infants that became lost to follow-up prior to their 18-month visit, and missed at least one study visit prior to either the Lancet I or Lancet II freeze dates. One infant’s last visit was the 12-month visit, which coincided with the last scheduled visit before the June 1999 (in preparation for Lancet I) freeze date. Her/his (missed) 18-month visit would have occurred after the Lancet I but before the Lancet II freeze dates. Another infant’s last visit was the 6-week visit. Her/his missed 9-month (18-month) visit would have occurred prior to the Lancet I (II) freeze date. The third infant was lost to follow-up immediately after birth. Her/his missed 9-month (18-month) visit occurred prior to the Lancet I (II) freeze date. Thus, the opportunity for an unrecognized death at the time of the Lancet I freeze date is limited to two infants, and the opportunity for an unrecognized death at the time of the Lancet II freeze date is limited to three infants.

All deaths, death dates, and last dates known alive for infants were accurately reported to the SCHARP database. The median time between the site awareness of the 23 infant deaths and the recording of the death in the (SCHARP) database was 19 days. For each of the 26 infants that did not die, the last date known to be alive by the site was transferred to SCHARP prior to the earlier Lancet data freeze following the study visit.

For the assessment of infant HIV-1 infection in the 49 sampled infants, we examined the degree to which scheduled PCR/EIA assessments were completed by the site, and whether the information about PCR/EIA positivity/negativity in the source documents/site CRFs was verified by inclusion of laboratory slips. We also compared this information to that in the SCHARP database to determine whether infant HIV-1 positivity was correctly captured in the analysis dataset. The primary analyses of HIV-1 positivity in HIVNET 012 were at 6–8 weeks, 14–16 weeks, and 18 months of age. The initial study publication (Guay et al., 1999a) considered only the first two of these time points, as well as the Day 1 sample, because not enough time had elapsed from the initiation of the trial to allow a thorough assessment at 18 months of age. The second study publication (Jackson et al., 2003) examined all four time points. Thus, we focused on the completeness of the HIV-1 PCR/EIA data at these four time points, and the fidelity and timeliness with which the data were transferred to SCHARP relative to the Lancet I and Lancet II data freeze dates.

It is important to note here that the rates of retention and follow-up of study participants were high, so there were few missing blood specimens. As noted in the HIVNET 012 remonitoring report, “these health visitors knew each patient individually and used culturally sensitive methods of making the contact. As a result of their efforts, maternal and infant follow-up overall for the first six weeks of the study was 97.4% for those who received AZT [ZDV] and 98% for those in the NVP group. The 18 month follow-up completion rates of the study were also high, with 93.8% for the AZT [ZDV] group and 96.1% for the NVP group” (DAIDS, NIAID, 2003a).

Of the 49 infants whose charts were reviewed, 43 (88%) had all of their scheduled HIV-1 tests (PCR or EIA). Of the 131 possible PCR/EIA tests that could have been done (excluding visit dates occurring after an infant died or became HIV-1 positive), a total of 11 (8%) were not completed.

In 47 (96%) of the 49 infants, all available PCR/EIA information was transferred to the SCHARP database, and done so before the following (Lancet I or Lancet II) data freeze. In two infants, a week 14–16 PCR result was found in the source records that was not transferred to SCHARP. However, both had previously been found to be HIV-1 infected by PCR and this information had been transferred to SCHARP. Thus, in terms of information used in the analysis of HIV-1 positivity, all available relevant information for the 49 infants was transferred to SCHARP.

The committee’s review of AEs, SAEs, and hospitalizations for the subset of 49 infants began with a review of source documents, followed by review of corresponding CRFs and database information. Information was collected from the source documents for 106 individual AEs. More than one AE often occurred concomitantly. Eleven of the 106 AEs noted in the source documents were not found in the corresponding CRFs. These AEs are shown in Table 4.3 as they appear in the source documents.

One other AE (infant number 22, “septic cord”) was reported in the CRFs but was not found in the source documents. All of the AEs that were found in the CRFs were found in the SCHARP data files. Two infants (1 NVP, 1 ZDV) were reported to not have had any AEs when in fact the source documents report an AE. The proportions of infants with an unreported AE, among those with at least one AE, was not significantly different (p=0.23) between the NVP (5/13) and ZDV (4/25) arms.

Excluding deaths, 87 individual clinical SAEs were found in the source documents of the 49 infants that were included in this review. Seventeen of those 87 SAEs were not entered into the corresponding CRFs. On every occasion on which one or more SAEs occurred, at least one of those concurrent SAEs was reported in the case report form. For subjects who had a

single SAE occurring in isolation, that event was always recorded in the case report form. In no case did the review reveal a failure to record in the CRF an SAE that occurred in isolation or all SAEs occurring concomitantly. All hospitalizations found in the review of the source documents were recorded in the CRFs. The 17 SAEs that were found in the source documents but not in the corresponding CRFs are shown in Table 4.4.

Only 2 of the 23 infant deaths identified in the source documents were recorded in the CRFs as serious adverse events. However, all deaths were recorded in the corresponding CRFs as a change of status. All of the SAEs found in the CRFs (including all deaths) were found in the SCHARP data files. All SAEs that were reported on the CRFs were also reflected in the source documents.

In assessing the implications of unreported AEs and SAEs in HIVNET 012, several points should be noted and considered. First, unreported events

had no implications for the care of the infants that participated in the study, but rather, could possibly affect the published study results. Second, because the sample of charts reviewed by the committee was weighted to over-represent infants that died, the numbers and rates of AEs and SAEs identified in the sample do not reflect those in the entire study.

It is also important to note that assessments of whether an AE or SAE may be due to a study drug are often difficult. Subjects have morbidities due to underlying disease or general environmental factors. As a result, it is common (and appropriate) to report all AEs and SAEs that occur in a trial, regardless of whether or not they are believed to be related to study drug. In a randomized comparative study, such as HIVNET 012, this provides a valid assessment of the relative safety of the treatments being compared. Nondifferential missed AE/SAE rates between the ZDV and NVP arms will not cause biased safety comparisons. However, because the overall rates of AEs and SAEs reflect both side effects of treatment and background comorbidities, one cannot in general extrapolate these overall rates to settings with different rates of background AEs and SAEs.

In assessing the effects of underreporting of some AEs on safety comparisons between the treatment arms, an important consideration is whether the rate of underreporting differs by treatment arms. If not, then underreporting will not affect the Type I error⁴ when comparing AE rates between treatment arms; that is, it will not increase the probability of declaring a treatment difference in safety rates when one does not exist, and thus such comparisons remain unbiased. Underreporting can, however, decrease the power of a study to detect real differences in AE rates between the treatment groups. In our review of the reporting practices in HIVNET 012, we saw no reason that could cause differential underreporting of AEs and SAEs, and in the sample we reviewed, there were no significant differences between the non-reporting AE rates of the NVP and ZDV groups. On this basis, the committee concludes that any underreporting of non-serious AEs and concomitant SAEs did not affect validity (that is, Type 1 error) of the comparisons of AEs between the NVP and ZDV arms in HIVNET 012, though it could have decreased the power to detect a real difference.

Concerns were raised in a DAIDS IND safety report issued on April 8, 2003, to FDA about a possible high frequency of neonatal hyperbilirubinemia⁵ (jaundice) in HIVNET 012 that was “probably related to the study drugs” (DAIDS, NIAID, 2003c). The IOM committee determined that it was important to evaluate the appropriateness of toxicity values used in the IND safety report to assess hyperbilirubinemia and to determine whether there was an increased incidence of hyperbilirubinemia among all infants enrolled in HIVNET 012. Consultants to the committee, Thomas Newman, M.D., M.P.H., a pediatrician and professor of epidemiology and biostatistics, with expertise in hyperbilirubinemia and other pediatric conditions, and Valerie Flaherman, M.D., M.P.H., a pediatrician and epidemiologist, conducted this assessment.

The HIVNET 012 investigators used the Harriet Lane Handbook (Barone, 1996) as a source for the reference value for an upper limit of normal (ULN) bilirubin in U.S. infants, specifically 7 mg/dL (Guay [on behalf of HIVNET 012 protocol team], 2003). Study infant bilirubin levels were then assigned a grade of severity based on DAIDS tables grading serious adverse events as multiples of the ULN. Using these criteria, the authors reported one SAE of hyperbilirubinemia among all participants (Jackson et al., 2003). When the data hyperbilirubinemia data were reevaluated in the April 8, 2003, IND safety report, DAIDS initially applied an incorrect upper limit of normal of 1.2 mg/dL for all infants ≥ 7 days of age as the criterion for hyperbilirubinemia. Based on that incorrect crite-

rion, the IND safety report stated that there were 63 Grade 4 bilirubin abnormalities in the ZDV group and 24 Grade 4 abnormalities in the NVP group. This report was subsequently retracted by DAIDS (DAIDS, NIAID, 2003b) when the error in defining the criterion for hyperbilirubinemia was recognized. When the bilirubin levels were evaluated relative to age-appropriate bilirubin normal values, DAIDS found no increased incidence of Grades 3 and 4 hyperbilirubinemia.

In the absence of data on bilirubin levels that would be of concern among Ugandan newborns, clinically significant hyperbilirubinemia was defined as a total serum bilirubin (TSB) level at which the American Academy of Pediatrics recommends phototherapy for infants in the U.S. However, the committee modified these thresholds to reflect a possible higher risk of bilirubin toxicity in the Ugandan infants. Thus, bilirubin levels of term, normal birth weight study infants were analyzed as if they were U.S. infants with one risk factor for bilirubin toxicity, such as prematurity or hemolysis. In addition, preterm or low birthweight study infants were analyzed as if they were U.S. infants with two or more risk factors.

The committee obtained all bilirubin levels for study infants in the first 2 weeks of life. Using the definition of significant hyperbilirubinemia described above, there was no significant difference in the incidence of significant hyperbilirubinemia between study infants who received NVP (4/319) and study infants who received ZDV (8/310) (P=0.26 by Fisher’s exact test, risk difference –0.013, 95% CI for risk difference (–0.035, 0.008)). Furthermore, mean maximum bilirubin levels measured at days 1 and 7 were significantly lower in the NVP arm (5.32 ± 3.13 mg/dL) when compared to both the placebo arms (6.87 ± 4.14 mg/dL; P<0.05) and the ZDV arm (6.58 ± 3.52 mg/dL; P<0.001).⁶

The committee and consultants compared the efficacy and safety results of HIVNET 012, which was the first study to look at this regimen of nevirapine, with findings from other trials with similar NVP and ZDV arms. The results from this review are described below (see Appendix B for a discussion of methods).

The review identified five randomized controlled trials that included single-dose NVP-only arms for prevention of mother-to-child transmission of HIV (Kiarie et al., 2003; McIntyre et al., 2004; Moodley et al., 2003; Taha et al., 2003; Taha et al., 2004). The proportions of infants infected in the antepartum period were similar between HIVNET 012 and these five studies. The review of the five randomized studies also showed that rates of Grades 3 and 4 adverse events in HIVNET 012 infants were similar to those of other studies.

Rates of transmission in the intrapartum and immediate postpartum periods were lower in HIVNET 012 (3.9%) compared to the other five studies that reported this variable (8.0%). However, after excluding the NVAZ study (Taha et al., 2003), in which mothers were not treated with NVP, this proportion was lower (6.0%) and thus more similar to the HIVNET 012 results. The NVAZ study was a Malawi study that randomized infants of women who presented in late stages of labor with unknown HIV status and did not receive NVP. The newborns were randomized to receive one dose of NVP or one dose of NVP plus a week of ZDV. The investigators subsequently analyzed the subset of infants born to mothers who were found to be infected with HIV. Since the mothers did not receive NVP, the infants were likely less protected against intrapartum transmission than those whose mothers did receive NVP. Hence, the intrapartum plus early postpartum transmission rates from this study are not directly comparable to those of HIVNET 012 or the other four studies.

The ZDV-only arm of the HIVNET 012 trial was less directly comparable to other randomized controlled trials for prevention of MTCT that included ZDV. Of the five trials reviewed, all included antepartum treatment of the pregnant women prior to labor for varying periods during pregnancy with ZDV, while HIVNET 012 ZDV-only arm included treatment of women during labor. Previous studies have suggested that the duration of antenatal treatment with ZDV is a strong predictor of efficacy of prevention of antepartum transmission (Shaffer et al., 1999). However,

when excluding transmission during the antepartum period (as measured by infant infection at 1 to 3 days of age), the rate of intrapartum and early postpartum transmission found in the two studies conducted in breastfeeding populations (10.5%) was similar to that found in HIVNET 012 (10.3%). These findings suggest that the findings in both arms of the HIVNET 012 trial are consistent with other studies that have used similar interventions, although over much shorter time periods.

Observational studies have also suggested similar rates of transmission and adverse events when the HIVNET 012 NVP regimen has been employed. For example, Stringer and colleagues (2003) have followed two observational cohorts of HIV-exposed infants who received single-dose nevirapine and whose mothers received intrapartum NVP in Zambia. They found transmission rates of 11.7% at 4 to 6 weeks (Stringer et al., 2004) and 11.2% at 6 to 8 weeks of age (Stringer et al., 2003). Other published observational studies which examined the effectiveness of NVP in practice in Kenya (Quaghebeur, 2004) and South Africa (Sherman et al., 2004) found higher rates of transmission (13% at 6 weeks, 18.1% at 14 weeks, respectively), but a study in Cameroon (Ayouba et al., 2003) found a lower rate of transmission (10.6% at 6 to 8 weeks). No maternal or infant complications or adverse effects were reported from Cameroon (Ayouba et al., 2003).

In conclusion, the findings from both the NVP and ZDV arms from HIVNET 012 appear to be consistent with findings on HIV transmission from other randomized controlled trials that tested similar treatment regimens. Additionally, the observed rates of serious adverse events were similar to those observed in randomized controlled trials that tested similar NVP regimens, randomized controlled trials that used NVP plus other antiretrovirals, a randomized postnatal prophylaxis trial, HIVNET-023 (Shetty et al., 2003), and observational studies.

Ayouba A, Tene G, Cunin P, Foupouapouognigni Y, Menu E, Kfutwah A, Thonnon J, Scarlatti G, Monny-Lobe M, Eteki N, Kouanfack C, Tardy M, Leke R, Nkam M, Nlend AE, Barre-Sinoussi F, Martin PM, Nerrienet E. 2003. Low rate of mother-to-child transmission of HIV-1 after nevirapine intervention in a pilot public health program in Yaounde, Cameroon. Journal of Acquired Immune Deficiency Syndromes 34(3): 274-280.

Barone, M. 1996. The Harriet Lane Handbook, 14th ed. Table 6.1. St. Louis, MO: Mosby-Year Book, Inc.

Chamberlin J, Gustavson SA, Hensley M, Lander S. 2002. Site Visit Report, Kampala Uganda, February 18-28, 2002. Westat Corporation, MD.

Cunningham C, Charbonneau T, Song K, Patterson D, Sullivan T, Cummins T, Poiesz B. 1999. Comparison of human immunodeficiency virus 1 DNA polymerase chain reaction and qualitative and quantitative RNA polymerase chain reaction in human immunodeficiency virus 1-exposed infants. Pediatric Infectious Diseases Journal 18(1):30-35.

DAIDS, NIAID (Division of AIDS, National Institute of Allergy and Infectious Diseases). 2003a. HIVNET 012 Monitoring Report.

DAIDS, NIAID. May 19, 2003b. Letter to All Relevant Parties.

DAIDS, NIAID. 2003c. Investigational New Drug (IND) Safety Report Issued on April 8, 2003 Re: Hyperbilirubinemia.

EMMES Corporation. 2002. Clinical Site Data Audit Report Stage I, NIH/NIAID/DAIDS/HIVNET 012 Re-Monitoring Study. The EMMES Corporation.

Guay L. 2004. Presentation to IOM’s Committee on Reviewing the HIVNET 012 Perinatal HIV Prevention Study. Transcript of September 30, 2004, Data-Gathering Meeting.

Guay L (on behalf of the HIVNET 012 protocol team). May 12, 2003. Letter to Fanning M.

Guay L, Mmiro F, Musoke P, Fanning M. 2002. HIVNET 012: Phase IIB Trial to Evaluate the Efficacy of Oral Nevirapine and the Efficacy of Oral AZT in Infants Born to HIV-Infected Mothers in Uganda for Prevention of Vertical HIV Transmission. Description of Study Procedures.

Guay LA, Musoke P, Fleming T, Bagenda D, Allen M, Nakabiito C, Sherman J, Bakaki P, Ducar C, Deseyve M, Emel L, Mirochnick M, Fowler MG, Mofenson L, Miotti P, Dransfield K, Bray D, Mmiro F, Jackson JB. 1999a. Intrapartum and neonatal single-dose nevirapine compared with zidovudine for prevention of mother-to-child transmission of HIV-1 in Kampala, Uganda: HIVNET 012 randomised trial. Lancet 354(9181): 795-802.

HIVNET 012 Investigators. 2003. Makerere University–Johns Hopkins University Quality Management Plan.

HIVNET Group. Protocol Version 1.0, 2 June 1997. A Phase III Placebo-Controlled Trial to Determine the Efficacy of Oral AZT and the Efficacy of Oral Nevirapine for the Prevention of Vertical Transmission of HIV-1 Infection in Pregnant Ugandan Women and Their Neonates.

Jackson JB, Musoke P, Fleming T, Guay LA, Bagenda D, Allen M, Nakabiito C, Sherman J, Bakaki P, Owor M, Ducar C, Deseyve M, Mwatha A, Emel L, Duefield C, Mirochnick M, Fowler MG, Mofenson L, Miotti P, Gigliotti M, Bray D, Mmiro F. 2003. Intrapartum and neonatal single-dose nevirapine compared with zidovudine for prevention of mother-to-child transmission of HIV-1 in Kampala, Uganda: 18-month follow-up of the HIVNET 012 randomised trial. Lancet 362(9387):859-868.

Kiarie JN, Kreiss JK, Richardson BA, John-Stewart GC. 2003. Compliance with antiretroviral regimens to prevent perinatal HIV-1 transmission in Kenya. AIDS 17(1):65-71.

LaMontagne J. September 7, 2004. Re: FW: HIVNET 012 follow-up. E-mail to Kington R.

Last JM. 1995. A Dictionary of Epidemiology, 3rd edition. New York: Oxford University Press.

McIntyre J, Martinson N, Investigators for the Trial 1413, Boltz V, Palmer S, Coffin J, Mellors J, Hopley M, Kimura T, Robinson P, Mayers D. 2004. Addition of short course combivir (CBV) to single dose viramune (sdNVP) for prevention of mother to child transmission (MTCT) of HIV-1 can significantly decrease the subsequent development of maternal NNRTI-resistance virus. eJournal of the International AIDS Society.

Moodley D, Moodley J, Coovadia H, Gray G, McIntyre J, Hofmyer J, Nikodem C, Hall D, Gigliotti M, Robinson P, Boshoff L, Sullivan JL. March 1, 2003. A multicenter randomized controlled trial of nevirapine versus a combination of zidovudine and lamivudine to reduce intrapartum and early postpartum mother-to-child transmission of human immunodeficiency virus type 1. Journal of Infectious Diseases 187(5):725-735.

Mwatha A. March 27, 2005a. Re: Please disregard previous messages [regarding vitamin A study coenrollment]. E-mail to Baciu A on behalf of the IOM’s Committee on Reviewing the HIVNET 012 Perinatal HIV Prevention Study.

Mwatha A. March 28, 2005b. Re: Co-enrollment in vit A study. E-mail to Baciu A on behalf of the IOM’s Committee on Reviewing the HIVNET 012 Perinatal HIV Prevention Study.

Mwatha A. March 9, 2005c. Re: FW: An inquiry to SCHARP [regarding breakdown of serious adverse events for infants and mothers in HIVNET 012]. E-mail to Gable A on behalf of IOM’s Committee on Reviewing the HIVNET 012 Perinatal HIV Prevention Study.

Quaghebeur A, Mutunga L, Mwanyumba F, Mandaliya K, Verhofstede C, Temmerman M. 2004. Low efficacy of nevirapine (HIVNET 012) in preventing perinatal HIV-1 transmission in a real-life situation. AIDS 18:1854-1856.

SCHARP. 2005. Number of Reported SAEs by Vitamin A Study Arm and HIVNET 012 Study Arm (Table).

SCHARP. 2004. Data Collection, Management, and Quality Control of HIVNET 012.

SCHARP. 1997. HIVNET 012 Study-Specific Procedures: Phase III Efficacy Trial of Oral AZT vs Oral Nevirapine in Pregnant Ugandan Women and Their Neonates for the Prevention of Vertical Transmission of HIV-1 Infection.

Shaffer N, Chuachoowong R, Mock P, Bhadrakom C, Siriwasin W, Young N, Chotpitayasunondh T, Chearskul S, Roongpisuthipong A, Chinayon P, Karon J, Mastro T, Simonds R. 1999. Short-course zidovudine for perinatal HIV-1 transmission in Bangkok, Thailand: A randomised controlled trial. Bangkok Collaborative Perinatal HIV Transmission Study Group. Lancet 353(9155):773-780.

Sherman GG, Jones SA, Coovadia AH, Urban MF, Bolton KD. 2004. PMTCT from research to reality—results from a routine service. South African Medical Journal 94(4):289-292.

Shetty A, Coovadia H, Mirochnick M, Maldonado Y, Mofenson L, Eshleman S, Fleming T, Emel L, George K, Katzenstein D, Wells J, Maponga C, Mwatha A, Jones S, Abdool KS, Bassett M, HIVNET 023 Study Team. 2003. Safety and trough concentrations of NVP prophylaxis given daily, twice weekly, or weekly in breast-feeding infants from birth to 6 months. Journal of Acquired Immune Deficiency Syndromes 34(5):482-490.

Simonds RJ, Brown TM, Thea DM, Orloff SL, Steketee RW, Lee FK, Palumbo PE, Kalish ML. 1998. Sensitivity and specificity of a qualitative RNA detection assay to diagnose HIV infection in young infants. Perinatal AIDS Collaborative Transmission Study. AIDS 12(12):1545-1549.

Stringer JS, Sinkala M, Goldenberg RL, Kumwenda R, Acosta EP, Aldrovandi GM, Stout JP, Vermund SH. 2004. Universal nevirapine upon presentation in labor to prevent mother-to-child HIV transmission in high prevalence settings. AIDS 18(6):939-943.

Stringer JS, Sinkala M, Chapman V, Acosta EP, Aldrovandi GM, Mudenda V, Stout JP, Goldenberg RL, Kumwenda R, Vermund SH. 2003. Timing of the maternal drug dose and risk of perinatal HIV transmission in the setting of intrapartum and neonatal single-dose nevirapine. AIDS 17(11):1659-1665.

Taha TE, Kumwenda NI, Hoover DR, Fiscus SA, Kafulafula G, Nkhoma C, Nour S, Chen S, Liomba G, Miotti PG, Broadhead RL. 2004. Nevirapine and zidovudine at birth to reduce perinatal transmission of HIV in an African setting: A randomized controlled trial. Journal of the American Medical Association 292(2):202-209.

Taha TE, Kumwenda NI, Gibbons A, Broadhead RL, Fiscus S, Lema V, Liomba G, Nkhoma C, Miotti PG, Hoover DR. 2003. Short postexposure prophylaxis in newborn babies to reduce mother-to-child transmission of HIV-1: NVAZ randomised clinical trial. Lancet 362(9391):1171-1177.