Methods

Study design. The Simplification with Tenofovir-Emtricitabine or Abacavir-Lamivudine (STEAL) study was an academic, randomized, open-label, multicenter, non-inferiority, phase IV, 96-week trial of 350 eligible subjects randomly allocated 1:1 to switch existing NRTIs to either the TDF-FTC fixed-dose combination or the ABC-3TC fixed-dose combination, while continuing to receive their current nonnucleoside reverse-transcriptase inhibitor or protease inhibitor. Stratification was performed by baseline NRTI use (TDF without ABC, ABC without TDF, or other), current protease inhibitor use, and study site. A computerized random number generator with a blocking factor of 2 was used to produce a random number list. The randomization list was built into the electronic case record form to maintain blinding of physicians and participants until randomization occurred.

The study was approved by each site's Research Ethics Committee and was registered at Clinicaltrials.gov (NCT00192634). The study sponsor was the National Centre in HIV Epidemiology and Clinical Research (University of New South Wales; Sydney, Australia). Gilead Sciences, after follow-up was complete but before data analysis, committed to financially support a substudy of the trial evaluating serious non-AIDS event-related biomarkers.

Participants. Eligible participants were aged ⩾18 years, receiving stable antiretroviral therapy including ⩾2 NRTIs for ⩾12 weeks, had plasma HIV loads <50 copies/mL for ⩾12 weeks, had estimated glomerular filtration rates ⩾70 mL/min/1.73 m², and provided written, informed consent.

Exclusion criteria were HLA-B*5701 positivity (unless already receiving ABC), previous hypersensitivity or intolerance to any study drug, previous treatment failure with ⩾30 days of ABC or TDF, prior ABC-3TC or TDF-FTC fixed-dose-combination exposure (prior fixed-dose-combination components were permitted), prior nontraumatic fracture, receipt of current antiretroviral therapy without a protease inhibitor and/or nonnucleoside reverse-transcriptase inhibitor, current unboosted atazanavir, receipt of TDF for hepatitis B, receipt of current therapy contraindicated with either fixed-dose combination, cirrhosis, and creatinine clearance <50 mL/min.

Assessments and toxicity management. At screening or baseline, demographic characteristics and a complete medical history were recorded, physical examination was performed, and a blood sample was collected for HLA-B*5701 genotyping at a central laboratory and for locally determined HIV load, blood count, CD4 count, biochemistry, liver function tests, and metabolic measures.

Participants were seen at baseline, week 4, week 12, and then every 12 weeks until week 96. At each visit, we recorded safety, adverse events, concomitant medications, adherence, weight, biochemistry, and HIV load. Every 12 weeks, blood count, liver function tests, and CD4 count were assessed. Every 24 weeks, a symptom-directed, physical examination (including blood pressure measurement) was performed, and metabolic measures were collected after a ⩾10-h overnight fast. Body composition (bone and soft tissue) was measured using a common dual-energy X-ray absorptiometry (DEXA) protocol every 48 weeks [14]. All DEXA tests for each patient were performed at the same imaging facility. DEXA tests were not centrally analyzed. Serum and plasma samples were collected and stored at baseline and then every 12 weeks for future analysis.

Framingham scores estimating the 10-year risk of a fatal or nonfatal myocardial infarction were determined at baseline, at week 48, and at week 96 [15]. Baseline scores were only available for 281 participants because blood pressure was not collected at all sites when the study commenced. Framingham scores are usually not determined for patients with ischemic heart disease or diabetes mellitus (a cardiac “risk equivalent” in the general population). Therefore, we determined Framingham scores with the 2 following approaches: (1) excluding patients with prior ischemic heart disease or diabetes or (2) including all patients with risk data and assigning patients with diabetes mellitus or ischemic heart disease a high-risk value of 20%.

Study drug could be ceased at any time at the discretion of the investigator or participant. Participants who were withdrawn from treatment were followed up per protocol if possible. Dosage reduction of the fixed-dose combination was not permitted. A change in the dosing schedule of a background antiretroviral drug was permitted (eg, lopinavir twice daily to once daily). The protocol contained guidelines for the management of potential fixed-dose-combination-related adverse events (available on request).

End point definitions. The primary end point was the proportion of patients in each group who experienced virological failure (HIV load >400 copies/mL in plasma on 2 consecutive occasions ⩾1 week apart) over 96 weeks. Secondary clinical end points were death, AIDS, and serious non-AIDS events (cardiovascular events [ie, myocardial infarction, ischemic stroke, peripheral vascular disease, or therapeutic vascular procedure], non-AIDS-defining cancer, end-stage liver disease, end-stage renal disease, and long-bone and torso fractures), treatment-related serious or grade 3 or 4 adverse events, discontinuation of any antiretroviral component for >4 weeks, and ABC hypersensitivity. Serious non-AIDS event components were chosen before the study commenced. On the basis of the first presentation of results from the SMART trial [8], the composite serious non-AIDS event end point and composite cardiovascular end point were chosen as key secondary end points in February 2006, after the study commenced but before the database was unblinded and analyzed in August 2008. Supporting documentation for serious non-AIDS events was reviewed by a blinded reviewer after all follow-up was completed.

Lipid measurement end points were new fasting total cholesterol level >6.5 mmol/L, an increase in total cholesterol >2.0 mmol/L, new fasting high-density lipoprotein (HDL) cholesterol level <0.9 mmol/L, a reduction in HDL cholesterol >0.5 mmol/L (each measurement made on separate occasions ⩾4 weeks apart), and receipt of new lipid-lowering therapy. Glycemic end points were new diabetes mellitus and new therapy for diabetes mellitus. Hyperlactatemia was defined by a plasma lactate level >5.0 mmol/L or >2.0 mmol/L with relevant accompanying symptoms on 2 consecutive occasions ⩾1 week apart, with the participant being well hydrated and rested. Renal end points were an estimated glomerular filtration rate <60mL/min/1.73 m²and a serum phosphate level <0.65 mmol/L. Bone-related end points were new osteopenia (hip and/or spine t score <−1.0), osteoporosis (t score <−2.5), receipt of bone mineral density therapy, and fracture. Progression of lipoatrophy was defined by a ⩾10% decrease in limb-fat mass from baseline. Serious adverse events and grade 3 or 4 adverse events were summarized by treatment group and body system with use of MedDRA codes.

Sample size and data analysis. We estimated that the rate of virological failure would be 15% over 2 years. With a power of 90%, a 2-sided α risk of 5%, and a margin for non-inferiority of 15%, a sample size of 175 participants per group was necessary. Non-inferiority would be demonstrated if the lower bound of the 95% confidence interval (CI) for the difference between the primary end point was above the predefined margin. The proportions of patients with virological failure were compared by intent-to-treat, missing-equals-failure analysis after 96 weeks of follow-up, after permanent withdrawal, or after loss to follow-up. On-treatment analyses were also performed using only data for participants who were receiving randomized therapy and who had data available.

To assess differences in proportions, we used Fisher's exact test and calculated exact CIs. Hazard ratios (HRs) for comparison of event rates, adjusted for duration of randomized therapy as a time-dependent covariate, and predictors of serious non-AIDS events were assessed using Cox regression methods. For time-to-event analysis of composite end points, time to first event was determined and data were censored at the last time point with data available. For serious non-AIDS event components, HRs were only calculated if at least 1 treatment arm had more than 5 events. For continuous laboratory and body composition data, mean change from baseline to weeks 48 and 96 were calculated for participants with baseline and at least 1 follow-up visit. We conducted intention-to-treat (last observation carried forward) and on-treatment analyses and used T tests to compare groups, unless data were not normally distributed (ie, the data for change in triglyceride levels), in which case Wilcoxon's test was used. All analyses used a 2-sided α of 0.05. No adjustment was made for multiple comparisons. All analyses, unless otherwise specified, were determined a priori and were hypothesis driven. Statistical analyses were performed with STATA, version 10.1 (Statacorp).

An independent data and safety monitoring board reviewed virology and safety data by treatment allocation when all recruited subjects had completed 24 weeks of the study. The data and safety monitoring board made no recommendation to change the trial.

Stratified analysis for virological failure was not performed because of the small number of events. Prespecified subgroup safety analyses were conducted for the stratification variables of baseline NRTI type and use of protease inhibitor therapy at baseline. Differences in outcomes between assigned groups by these strata were assessed using linear regression tests for interaction.

Results

Participants. Of 441 patients screened from November 2005 through September 2006, 360 were randomly assigned, of which 357 commenced allocated therapy (Figure 1). Three ineligible patients were randomized in error (2 to the ABC-3TC arm); these participants were included in all analyses. Baseline char acteristics were well balanced between groups, except that smoking was more common in the ABC-3TC group, although Framingham risk scores were similar (Table 1).

Treatment outcomes. Eighteen (10%) patients in the ABC-3TC group and 16 (9%) patients in the TDF-3TC group discontinued randomized therapy. Four ABC-3TC recipients were diagnosed with ABC hypersensitivity, 2 of whom were patch-tested with ABC after week 96 with no skin reaction. Of 41 participants who initiated lipid-lowering therapy after baseline, 36 (88%) did so before the first report of an association between ABC and myocardial infarction [7]. One participant discontinued ABC at week 96 soon after the initial presentation (in February 2008) of data reporting an association between ABC and myocardial infarction.

Virological efficacy. At week 96, by intent-to-treat, missing-equals-failure analysis, virological failure occurred in 10 participants (5.6%) in the ABC-3TC group and 7 participants (3.9%) in the TDF-FTC group (difference, 1.7%; 95% CI, −2.8 to 6.1; P=.62) (Figure 2). In the intent-to-treat non-completer-equals-failure analysis, virological failure occurred in 23 participants (12.8%) in the ABC-3TC group and 17 participants (9.6%) in the TDF-FTC group (difference, 3.3%; 95% CI, −3.3 to 9.9; P=.40). In the on-treatment analysis, virological failure occurred in 2 participants (1.1%) in each group (difference, 0%; 95% CI, −2.2 to 2.2; P>.99). Times to virological failure by intent-to-treat, missing-equals-failure analysis were similar (3.08 and 2.15 events per 100 patient-years in the ABC-3TC and TDF-FTC groups, respectively; HR, 0.70; 95% CI, 0.27–1.83; P=.47). The proportions of patients with viral loads >50 copies/mL, >400 copies/mL, and >1000 copies/mL in plasma specimens were also not significantly different (difference, 2.2% [95% CI, −3.7 to 8.1%]; 4.5% [95% CI, −0.1 to 9.0]; 3.9% [95% CI, −0.5 to 8.3], respectively). Genotypes of virological failure were not analyzed because too few participants developed virological failure with a viral load >1000 copies/mL to allow for meaningful analysis.

AIDS and serious non-AIDS events. No participant developed AIDS. Eighteen (5.0%) participants developed a serious non-AIDS event (rate, 2.79 events per 100 person-years; 95% CI, 1.76–4.43]) (Table 2). TDF-FTC was associated with significantly fewer serious non-AIDS events than ABC-3TC (1.2 vs 4.4 events per 100-patient years; HR, 0.26; 95% CI, 0.08–0.79; P=.018). The difference in the serious non-AIDS event rate was not altered by adjustment for smoking at baseline or for time receiving assigned therapy.

No baseline parameter affected the significant association between randomization to ABC-3TC and the serious non-AIDS event rate. There was no evidence that the relationship between serious non-AIDS events and randomized therapy differed in participants receiving ABC at screening, compared with those who commenced ABC after randomization (P=.994, for interaction). Compared with patients with a baseline Framingham score of <10%, those with higher scores had a non-statistically significantly greater risk of a serious non-AIDS event (HR for baseline score 10%-19%, 1.60; 95% CI, 0.47–5.48; P=.45; HR for baseline score >20%, 3.32; 95% CI, 0.69–15.97; P=.14).

Four serious non-AIDS events were fatal (all cancers). No type of cancer occurred in >1 participant (1 carcinoma each of lung, kidney, prostate, and unknown primary location; 1 metastatic melanoma; 1 testicular cancer, and 1 Hodgkin lymphoma).

The most common serious non-AIDS event was cardiovascular disease (9 events, including 4 myocardial infarctions, 2 peripheral arterial disease [1 with angioplasty], 1 coronary artery bypass surgery, 1 ischemic stroke, and 1 deep venous thrombosis), which did not cluster at any particular time point during follow-up. Serious non-AIDS cardiovascular events occurred significantly less frequently in those receiving TDF-FTC than in those receiving ABC-3TC (1 vs 8 events; 0.3 and 2.2 events per 100 patient-years, respectively; HR, 0.12; 95% CI, 0.02–0.98; P=.048).

Metabolic outcomes. ABC-3TC was associated with greater increases in total, low-density lipoprotein, and HDL cholesterol levels than TDF-FTC by intent-to-treat analysis (Table 3); the ratio of total to HDL cholesterol was significantly lower in the TDF-FTC group at week 48 but not at week 96. Lipid events were significantly more common in the ABC-3TC group (Table 4). There was no significant between-group difference for any glycemic or renal parameter or for lactate levels. Per-protocol analyses yielded similar results (data not shown).

Serum alanine aminotransferase and (unfractionated) alkaline phosphatase levels increased modestly but significantly 4 weeks after initiating TDF-FTC and thereafter remained stable. There were two grade 3 or 4 increases in serum transaminase levels but no grade 3 or 4 clinical or hepatic event in patients receiving TDF-FTC.

Body composition. TDF-FTC significantly lowered hip and spine t scores at both weeks 48 and 96, whereas both t scores increased in the ABC-3TC group (Table 3). Bone events (incident osteopenia or osteoporosis or receipt of antiresorptive therapy) were significantly more common in the TDF-FTC group (Table 4). TDF-FTC was also associated with modest decreases in weight and lean mass. There was no between-group difference for limb-fat mass, which increased in both groups.

Subanalyses. Most safety subanalyses found no difference in outcome by treatment strata. The decrease in triglyceride levels was greater in those who switched to TDF-FTC from ABC (P=.031, for interaction). Decreases in low-density lipoprotein and total cholesterol levels were greater in those assigned to TDF-FTC who were receiving a protease inhibitor at baseline (P⩽.02, for interaction).

Discussion

In this population, TDF-FTC and ABC-3TC had similar high virological efficacy, but TDF-FTC was associated with fewer adverse effects, fewer serious non-AIDS events (particularly cardiovascular disease), and less receipt of lipid-lowering therapy than ABC-3TC, but caused more bone mineral density loss. Fatal and non-fatal serious non-AIDS events were more prevalent than AIDS.

Our study is the first randomized study to support cohort data reporting an increased risk of myocardial infarction with receipt of ABC [7, 8]. Our data suggest that this risk is not HLA-B*5701-mediated, as only 1 HLA-B*5701-positive patient receiving ABC was randomized (the patient did not experience a cardiovascular event). This risk also appears unrelated to the changes in clinical characteristics or metabolic parameters, because changes in Framingham risk score and in the ratio of total to HDL cholesterol levels were similar between groups at week 96, and blood pressure and glycemic measures were similar at all time points. It is notable that the similar changes in the ratio of total to HDL cholesterol levels between groups at week 96 occurred despite significantly more receipt of new lipid-lowering therapy in the ABC-3TC group.

We did not find an increased risk of myocardial infarction with ABC, but rather we found an increased risk of overall car diovascular disease, which was a composite of myocardial infarction, coronary artery surgery, peripheral vascular disease, ischemic stroke, and deep venous thrombosis. Two large, cohort studies have found an association between recent ABC use and myocardial infarction, whereas 1 smaller cohort study (involving younger patients) found no such association [6–8]. Also, the D:A:D study, the largest such study, found no association of ABC with stroke, although this study made no distinction between ischemic and hemorrhagic stroke. Our finding of more cardiovascular disease among those who received ABC had a relatively wide CI; a larger study with more cardiovascular events would have yielded a more definitive result.

TDF-FTC resulted in significant reductions in hip and spine bone mineral density, compared with ABC-3TC, which was actually associated with an increase in bone mineral density. The clinical significance of these differences are uncertain, because patients in the TDF-FTC group did not experience more fractures, and the decrease in bone mineral density was only observed in the first year of TDF therapy. TDF-FTC resulted in small decreases in lean mass and weight; whether these changes are related to bone mineral density loss remains to be determined. Both fixed-dose combinations were associated with similar improvements in limb fat.

We found no evidence that tenofovir resulted in declining renal function. However, 30% of participants were receiving tenofovir at baseline, and patients with prior intolerance to tenofovir (or any study drug) were excluded from the study. We did not find, however, a decrease in any renal parameter in the subset of patients who were not receiving tenofovir at baseline.

No participant developed AIDS. This suggests that the modest difference in CD4 count in favor of ABC was not clinically important in this population.

Our study has limitations. Approximately 50% of participants were not naive to ABC and TDF, and more participants were receiving TDF at baseline than were receiving ABC. Both of these factors may have influenced treatment outcomes, including the composite serious non-AIDS event end point, although there was a very limited effect of NRTI stratum on treatment outcomes. ABC better penetrates the central nervous system, compared with tenofovir, which has been associated with better neurocognitive function, a parameter that we did not analyze [16]. Almost all participants were white men, and only 25% were receiving a protease inhibitor; thus, the generalizability of our study may be limited. Too few cardiovascular events occurred to permit a risk factor analysis, although other studies suggest that the most risk will occur in those with greater underlying cardiovascular risk [7, 8]. The results cannot explain the pathogenesis of the increased cardiovascular event rate observed with ABC-3TC; analysis of cardiovascular biomarkers in stored serum samples from the STEAL study is planned. Lastly, a non-naive study population is probably not the best for testing of virological non-inferiority.

In summary, our data suggest that serious non-AIDS events should be routinely assessed and reported in all antiretroviral switch trials. ABC-3TC is an effective treatment for HIV infection, but its use should be considered carefully in those with high underlying cardiovascular risk.

STEAL Study Group

Protocol steering committee and writing committee. Allison Martin, Mark Bloch, Janaki Amin, David Baker, David Cooper, Sean Emery, Andrew Carr (Principal investigator).

Site nurses and investigators. Holdsworth House Medical Practice, Sydney: Mark Bloch, Shikha Agrawal, Kate Beileiter (n=81); St Vincent's Hospital, Sydney: Andrew Carr, David Cooper, Karen Macrae, Richard Norris, Robert Fielden (n=69); East Sydney Doctors, Sydney: David Baker, Robyn Vale (n=31); Taylor Square Private Clinic, Sydney: Robert Finlayson, Robyn Richardson, Sophie Dinning, Isabel Prone (n=23); The Alfred Hospital, Melbourne: Jennifer Hoy, Christine Alveras, Rachel Liddle (n=19); Melbourne Sexual Health, Melbourne: Tim Read, Julie Silvers, Helen Kent (n=14); Burwood Medical Centre, Sydney: Nicholas Doong, Jeff Hudson (n=11); Prahran Market Clinic, Melbourne: Norman Roth, Helen Lau (n=10); The Carlton Clinic, Melbourne: Jonathan Anderson, Richard Moore, Kaye Lowe, Paul Cortissos, Sian Edwards (n=10); Gold Coast Sexual Health, Queensland: John Chuah, Denise Lester, Tammy Schmidt, Fiona Clark (n=9); Royal Melbourne Hospital, Melbourne: Alan Street, Janine Roney (n=8); Royal Adelaide Hospital, Adelaide: David Shaw, Lyndal Daly (n=7); Gladstone Road Medical Centre, Queensland: David Orth, David Youds (n=7); AIDS Medical Unit, Brisbane: Mark Kelly, Paul Negus, Peita-Lee Ambrose (n=7); Lismore Hospital, NSW: David Smith, Denni Pearson, Cherie Mincham (n=6); Royal Perth Hospital, Perth: David Nolan, Claire Forsdyke (n=5); Flinders Medical Centre, Adelaide: Mark Boyd, David Gordon, Robyn Gilligan, Michelle Wall, Rachel Wundke (n=5); The Centre Clinic, Melbourne: Nicholas Medland, Ban Kiem Tee, Helen Lau (n=5); Westmead Hospital, Sydney: Dominic Dwyer, Maggie Piper (n=5); Fremantle Hospital, Western Australia: John Dyer, Jacqueline Kerth, Samantha Libertino (n=4); Monash Medical Centre, Melbourne: Ian Woolley, Michelle Giles, Pauline Galt (n=4); Royal North Shore Hospital, Sydney: Stephen Davies, Linda Dayan, James Baber, Victoria Hounsfield (n=4); O'Brien Street Practice & The Care & Prevention Program, Adelaide: William Donohue, Michael Curry, Joy Oddy (n=3); Doll's House Sexual Health Clinic, Queensland: Darren Russell, Christine Remington, Laura Foy (n=3); Holdsworth House Medical Practice, Byron Bay: Mark Bloch, Debra Hayhoe, Bernie Monaghan, Nicky Cunningham (n=3); Prince of Wales Hospital, Sydney: Jeffrey Post, Suzanne Ryan (n=2); Liverpool Hospital, Sydney: John Quinn, Helen Best, Catherine Magill (n=2); Albion Street Centre, Sydney: Don Smith, Jason Gao, Jega Sarangapany (n=2); Royal Brisbane Hospital, Brisbane: Anthony Allworth, Janelle Zillman, Anne Sleat, Holly Asher (n=1).

Study coordination. Sean Emery, Allison Martin, Janaki Amin, Wilma Goodyear, Kymme Courtney-Vega, Simone Jacoby, Hila Haskelberg, Cate Carey, Ally MacDonald, Lina Safro, David Courtney-Rodgers, Maja Berilazic, Aurelio Vulcao, Maria Ariaga, Tian Erho, Kat Marks, Julie Yeung, Kate Merlin (National Centre in HIV Epidemiology and Clinical Research, University of New South Wales, Sydney).

End point review. Gail Matthews.

Data safety monitoring board. Alan Winston, Steve Wesse-lingh, Deborah Black.