INSERM U897 & CIC-EC 7, University Victor Segalen Bordeaux 2, School of Public Health (ISPED), Bordeaux, France.
Correspondence to Laura Richert, INSERM U897, University Victor Segalen Bordeaux 2, 146, rue Léo Saignat, 33076 Bordeaux cedex, France. Tel: +33 557 57 48 12; fax: +33 557 57 11 72; e-mail: firstname.lastname@example.org
We read with great interest the publication by Pulido et al.  who reported the results of a randomized trial evaluating a lopinavir–ritonavir monotherapy for maintenance in HIV-infected patients. On the basis of the primary endpoint, the authors concluded that maintenance with lopinavir–ritonavir monotherapy is non-inferior to a triple therapy in the studied population. The authors also acknowledged the limitations of their results, in particular the fact that non-inferiority was not demonstrated for all secondary endpoints.
The present report illustrates some of the methodological difficulties in the design and analysis of non-inferiority trials for HIV treatment strategies in general.
First, we are concerned by the apparent absence of a consensus regarding the choice of the primary endpoint in trials comparing different strategies of antiretroviral treatment. Pulido et al.  chose a composite endpoint to define therapeutic failure as follows: confirmed HIV RNA higher than 500 copies/ml, or loss to follow-up, or treatment discontinuation, or change of randomized therapy other than reinduction. According to the provided definition, cases in the monotherapy group with confirmed virological failure (two measurements of HIV RNA > 500 copies/ml separated by at least 2 weeks) are not considered failures, if HIV RNA is resuppressed successfully after reindroduction of nucleosides. To our knowledge, this is an uncommon choice as compared with the endpoints of other randomized trials evaluating simplification regimens in HIV-infected patients [2–4]. Yet, if reinduction in the monotherapy group is not considered as therapeutic failure, non-inferiority of the two treatment strategies is more likely to be demonstrated. For example, one could assume that due to early virological failure, a number of the patients would receive reinduction treatment shortly after the switch to monotherapy. Consequently, these patients would receive the same treatment as the comparator group for almost the entire length of the trial, which in turn would downsize the difference between the two groups over the time of the trial. Thus, we believe that the secondary analyses reported by the authors, in which treatment modification was considered as failure, constitute a more cautious choice. In that case, the authors could not conclude consistently that the simplification strategy was non-inferior to a triple therapy in the studied population.
Second, we suggest that some aspects concerning the treatment of missing data and the statistical approach in non-inferiority trials should be further clarified. In some of the analyses reported by Pulido et al. , the authors considered missing data to be failures. It is noteworthy that this approach tends to equalize outcomes in the compared groups. This effect is deliberate in superiority trials, but it may be inappropriate in non-inferiority analyses as it minimizes the difference between groups . Pulido and colleagues thus tested the robustness of their results by performing an as-treated analysis. The results of additional sensitivity analyses would be more convincing by using the worst-case methodology to quantify the potential for bias due to missing data, that is considering missing data to be failures in the intervention group, but successes in the comparator group and vice versa. Indeed, a per-protocol (or as-treated) analysis in non-inferiority and equivalence designs might also bias the results towards a smaller difference between groups [5–7]. The worst-case method, by contrast, may provide a truly conservative assessment of the robustness of a binary endpoint in a non-inferiority trial and its broader application should be discussed for future trials.
Third, there is a need for a large consensus regarding the non-inferiority margin in trials evaluating maintenance strategies in treatment-experienced patients with suppressed HIV replication. For an assumed failure rate of 10%, Pulido et al.  defined a non-inferiority margin of 12%, without commenting on the latter choice. The lack of rationale for the non-inferiority margin seems indeed common in HIV trials , and its relevance remains to be properly assessed. According to the authors' premise, a failure rate of up to 22% is accepted in pretreated patients in whom viral replication is controlled prior to randomization. We postulate that the acceptability of this assumption should be scrutinized . A consensual, clinically relevant non-inferiority margin should be defined for a given response rate and be applied to all non-inferiority trials in this population, as has been proposed in other research areas .
In summary, some key aspects of non-inferiority trials in HIV-infected patients warrant thorough methodological deliberation. We need an international consensus to help design future non-inferiority trials in HIV patients, as these trials are more and more common, given the potency of current antiretroviral drugs.
1. Pulido F, Arribas JR, Delgado R, Cabrero E, González-García J, Pérez-Elias MJ, et al
. Lopinavir-ritonavir monotherapy versus lopinavir-ritonavir and two nucleosides for maintenance therapy of HIV. AIDS 2008; 22:F1–F9.
2. Cameron W, da Silva B, Arribas J, Myers R, Bellos NC, Gilmore N, et al.
A two-year randomized controlled clinical trial in antiretroviral-naive subjects using lopinavir/ritonavir (LPV/r) monotherapy after initial induction treatment compared to an efavirenz (EFV) 3-drug regimen (Study M03-613). In: Program and abstracts of the XVI International AIDS Conference; 13–18 August 2006; Toronto. Canada. Abstract THLB0201.
3. Delfraissy JF, Flandre P, Delaugerre C, Ghosn J, Horban A, Girard PM, et al
. Lopinavir/ritonavir monotherapy or plus zidovudine and lamivudine in antiretroviral-naive HIV-infected patients. AIDS 2008; 22:385–393.
4. Girard PM, Cabié A, Michelet C, Verdon R, Katlama C, Mercié P, et al.
TenofovirDF + efavirenz (TDF+EFV) vs tenofovirDF+ efavirenz + lamivudine (TDF+EFV+3TC) maintenance regimen in virologically controlled patients (pts): COOL Trial. In: Program and abstracts of the 46th Interscience Conference on Antimicrobial Agents and Chemotherapy; 27–30 September 2006; San Francisco, California. Abstract H-1383.
5. Piaggio G, Elbourne DR, Altman DG, Pocock SJ, Evans SJ. CONSORT Group. Reporting of noninferiority and equivalence randomized trials: an extension of the CONSORT statement. JAMA 2006; 295:1152–1160.
6. Jones B, Jarvis P, Lewis JA, Ebbutt AF. Trials to assess equivalence: the importance of rigorous methods. BMJ 1996; 313:36–39.
7. Brittain E, Lin D. A comparison of intent-to-treat and per-protocol results in antibiotic noninferiority trials. Stat Med 2005; 24:1–10.
8. Parienti JJ, Verdon R, Massari V. Methodological standards in noninferiority AIDS trials: moving from adherence to compliance. BMC Med Res Methodol 2006; 6:46.
9. Garattini S, Bertele' V. Noninferiority trials are unethical because they disregard patients' interests. Lancet 2007; 370:1875–1877.
10. Committee for Proprietary Medicinal Products. Note for guidance on the evaluation of medicinal products indicated for treatment of bacterial infections. London: EMEA; 2004. CPMP/EWP/558/95 rev 1.