Between 5 January 2004 and 28 October 2004, 600 individuals were randomized (). A year after enrolment, one participant (N) disclosed that they had previously taken 3 months of Triomune (co-formulated lamivudine/stavudine/nevirapine). As the number of other participants with concealed prior exposure is unknown, this participant was not excluded. Two further participants (2N) had tuberculosis at screening: one concealed a recent diagnosis from fear of not receiving ART, and one was diagnosed after randomization from a screening sputum sample; a further three participants (3A) had minor eligibility violations (two neutrophils <0.5 × 109/l, 1 ALT/AST >5 × ULN). Primary and secondary endpoint comparisons were identical if restricted to the strictly eligible population (297A, 297N). All but one participant started allocated trial drugs within 3 days of randomization; this last participant started on day 8.
Baseline characteristics were broadly similar between the two groups (). Fourteen of the 15 women who reported previous ARVs to prevent mother-to-child HIV transmission had taken single-dose nevirapine: the remaining woman did not know which drug(s) she had received. One hundred and twenty-five (42%) abacavir and 120 (40%) nevirapine participants were taking cotrimoxazole prophylaxis or started it at randomization (median CD4 104 cells/mm3): only 11 (2%) and one (0.2%) participants were taking fluconazole or isoniazid prophylaxis, respectively.
Five hundred and seventy (95%) participants (289A, 281N) completed 24 weeks: 24 participants died before 24 weeks (9A, 15N), five were lost to follow-up (1A, 4N) and one (A) formally withdrew consent (). Participants reaching week 24 were transferred to open-label nevirapine or abacavir and continue to be followed-up in DART.
There was no difference between randomized groups in the proportion with >95% adherence according to 4-weekly pill counts to combivir (global P = 0.84), blinded abacavir/abacavir placebo (P = 0.60) or blinded nevirapine/nevirapine placebo (P = 0.96). In both groups and for all drugs, the proportion reporting >95% adherence increased over time, from 85% of participants at week 4 to 92% at week 12 and 94% at week 24. However, the percentage of participants reporting that they had never missed a dose decreased from 73% at 4 weeks to 61% and 49% at 12 and 24 weeks with no difference between arms (global P = 0.58), although the percentage reporting that they had missed a dose in the last week remained stable at 4–6%, again with no difference between arms (global P = 0.55).
Serious adverse reactions (SARs, primary endpoint)
A total of 37 SAEs occurred in 36 participants (14A, 22N; a second SAE occurred in one participant on open-label nevirapine following discontinuation of blinded abacavir for HSR), all on blinded drug or within 30 days of stopping blinded trial drugs. Twenty (6A, 14N) were considered to be SARs at independent review (definitely/probably related or uncertain whether related to blinded trial drugs, ) occurring in six (2.0%) and 14 (4.7%) participants receiving abacavir and nevirapine respectively [HR = 0.42 (95% CI 0.16–1.09), logrank P = 0.06, ]. The 20 SARs were one life-threatening event (N), two hospitalizations (2N), and 17 other important medical conditions (6A, 11N); only 10 (2A, 8N) were grade 3/4. Three (3N) and one (N) SAR were also considered uncertain whether related to cotrimoxazole and isoniazid respectively. The six abacavir SARs were in three men and three women, whereas the 14 nevirapine SARs were in three men and 11 women (P = 0.19 for heterogeneity with sex). Seven additional SAEs (3A, 4N) were originally reported as uncertain whether related to blinded trial drugs by the clinical investigator, but were judged unlikely to be related at independent review ().
Grade 4 and serious AEs and ARs
Time to first serious adverse reaction (serious adverse event definitely/probably related or uncertain whether related to blinded trial drugs).
In 19 of 20 participants, the SAR was consistent with a suspected HSR: six abacavir (2.0%, 95% CI 0.7–4.3%) vs. 13 nevirapine (4.3%, 95% CI 2.3–7.3%). Broadly similar proportions of those with suspected HSR on abacavir and nevirapine had fever (4/6A, 9/13N); respiratory (4A, 7N), constitutional (3A, 7N), or gastrointestinal (2A, 5N) symptoms; rash (6A, 10N) and oral/mucosal involvement (3A, 6N): but hepatic involvement was only seen in the nevirapine group (4N) (). All participants with respiratory symptoms in the nevirapine group had either rash (n = 4) or hepatic involvement plus constitutional symptoms and fever (n = 3). Fifteen of the 16 rashes were disseminated; three were grade 4 (3N), one grade 3 (A) and 11 (5A, 6N) grade 1/2. No participant had Stevens–Johnson syndrome. The remaining SAR was asymptomatic grade 4 hepatitis (N). In total, five of the 14 nevirapine participants with SARs had grade 4 LFT elevations (including the one with asymptomatic hepatitis alone) and two had grade 3 LFTs, compared with none of the six abacavir participants.
Symptoms occurring in participants with suspected HSRs
The 17 (9A, 8N) SAEs considered unlikely to be related to blinded drug were anaemia (n = 7, one fatal), pancytopenia/haematemesis (one fatal), sepsis/haematemesis (one fatal), death from sudden/unexpected cause/death at home [two at weeks 4 (N) and 9 (A), neither with any recorded interruption in blinded drugs], indeterminate cerebral disease (fatal), head injury, DVT, duodenal ulcer/haematemesis, fever (uncertain whether related to rabies vaccination), and rash (related to open label nevirapine following discontinuation of blinded abacavir for a previous SAR) ().
Grade 4 adverse events
There were significantly more grade 4 AE in the nevirapine than in the abacavir arm [98 (33%) vs. 65 (22%) participants, rates 93 and 61 per 100 person years, P = 0.003]. The majority were neutropenia (47A, 73N) or anaemia (19A, 15N). Only nine participants (9N) ever had grade 4 (>10 × upper limit of normal) elevations in LFTs (including one with asymptomatic hepatitis and four with suspected HSR reported as SARs). Four resolved after substitution of tenofovir for nevirapine (all four reported as SARs), and five resolved without substituting for nevirapine (4 had one grade 4 measurement only, the remaining one (reported as a SAR) was maintained on open-label nevirapine). Of the 196 grade 4 AE, only 31 (13A, 18N) were considered by the clinical investigator to be definitely/probably or uncertain whether related to blinded trial drugs [in 13 (4%) vs. 18 (6%) participants, respectively], with no evidence of a difference between randomized groups (rates 9.9 and 14.5 per 100 person-years, P = 0.30).
Discontinuation of blinded trial drug
Blinded trial drug was prescribed for 96.8% and 92.3% of person-time to 24 weeks in the abacavir and nevirapine arms, respectively. In total 14 (4.7%) and 30 (10.0%) participants discontinued blinded trial drug, respectively (exact P = 0.02, P = 0.04 in strictly eligible population). Twenty-one discontinuations were for toxicity [6A (2.0%) vs. 15N (5.0%), exact P = 0.07] and included all 20 participants with SARs: the toxicity was rash/suspected HSR (6A, 13N) and/or hepatotoxicity (4N) in all cases. The remaining 19 patients were to start tuberculosis therapy [6A (2.0%) vs. 13N (4.3%), respectively], or for other reasons [pregnancy (1N), personal (2A, 1N)]. However, two of the 13 discontinuing blinded trial drug to start tuberculosis therapy in the nevirapine arm had tuberculosis which could have been identified at screening. Time to AE causing discontinuation did not differ significantly between the two groups [median 32 days (IQR 25–59, range 8–85) in the nevirapine vs. 44 days (12–66, 4–112) in the abacavir group; ranksum P = 0.70].
Twenty-two participants (6A, 16N) were unblinded before 24 weeks – all but one (N, pregnancy) because of toxicity leading to discontinuation of blinded trial drugs described above. Sixteen (4A, 12N) of the 22 substituted tenofovir according to protocol and six (2A, 4N) substituted open-label nevirapine. Of the four participants who therefore remained on active nevirapine, one unblinded for pregnancy, one for fever/constitutional symptoms subsequently judged not related to blinded trial drugs and uncertain whether related to rabies vaccine (), and two for suspected HSRs (grade 1 and 2 reactions, respectively).
Other substitutions and discontinuations whilst retaining blinding
A further 20 (7A, 13N) participants substituted open-label tenofovir for blinded trial drugs without unblinding (most following tuberculosis diagnosis). Only two participants (1A, 1N) stopped all antiretrovirals for >31 days (both for personal reasons). Forty-five participants (22A, 23N) substituted stavudine for zidovudine, because of anaemia (16A, 17N) and/or neutropenia (5A, 8N), or myopathy (1A). No other substitutions occurred during the 24-week study, and no participant switched to second-line therapy.
Changes in laboratory parameters of toxicity
The nevirapine arm had significantly greater increases from baseline in absolute levels of liver function tests (ALT global P = 0.003, AST P = 0.07) compared with the abacavir arm, and significantly higher grades of liver function test toxicity (ALT P = 0.006, AST P = 0.05). In contrast, the nevirapine arm had small but significantly greater increases from baseline in glomerular filtration rate calculated by Cockcroft–Gault (GFR) compared with the abacavir arm (global P = 0.01, e.g. mean +1 vs. +9 in the nevirapine arm at week 12, t-test P = 0.007), and lower grades of GFR toxicity (P = 0.03). Early haemoglobin decreases were smaller in the abacavir arm (mean −0.3 g/dl vs.−0.6 g/dl in the nevirapine arm at week 4, t-test P = 0.02) although comparable by week 24 (global P = 0.09) and there was no overall difference in toxicity (P = 0.90). Similarly, there was a non-significant trend towards smaller decreases in neutrophils in the abacavir arm (P = 0.43), and significantly lower grades of toxicity (P = 0.03). There was no difference between nevirapine and abacavir arms in other laboratory parameters of toxicity.