Gilbert Peter B
Vaccine and Infectious Disease and Public Health Sciences Divisions, Fred Hutchinson Cancer Research Center, Seattle, Washington, USA.
Department of Biostatistics, University of Washington, Seattle, Washington, USA.
Stat Commun Infect Dis. 2019;11(1). doi: 10.1515/scid-2019-0003. Epub 2019 Jul 27.
Four randomized placebo-controlled efficacy trials of a candidate vaccine or passively infused monoclonal antibody for prevention of HIV-1 infection are underway (HVTN 702 in South African men and women; HVTN 705 in sub-Saharan African women; HVTN 703/HPTN 081 in sub-Saharan African women; HVTN 704/HPTN 085 in U.S., Peruvian, Brazilian, and Swiss men or transgender persons who have sex with men). Several challenges are posed to the optimal design of the sequel efficacy trials, including: (1) how to account for the evolving mosaic of effective prevention interventions that may be part of the trial design or standard of prevention; (2) how to define viable and optimal sequel trial designs depending on the primary efficacy results and secondary "correlates of protection" results of each of the ongoing trials; and (3) how to define the primary objective of sequel efficacy trials if HIV-1 incidence is expected to be very low in all study arms such that a standard trial design has a steep opportunity cost. After summarizing the ongoing trials, I discuss statistical science considerations for sequel efficacy trial designs, both generally and specifically to each trial listed above. One conclusion is that the results of "correlates of protection" analyses, which ascertain how different host immunological markers and HIV-1 viral features impact HIV-1 risk and prevention efficacy, have an important influence on sequel trial design. This influence is especially relevant for the monoclonal antibody trials because of the focused pre-trial hypothesis that potency and coverage of serum neutralization constitutes a surrogate endpoint for HIV-1 infection. Another conclusion is that while assessing prevention efficacy against a counterfactual placebo group is fraught with risks for bias, such analysis is nonetheless important and study designs coupled with analysis methods should be developed to optimize such inferences. I draw a parallel with non-inferiority designs, which are fraught with risks given the necessity of making unverifiable assumptions for interpreting results, but nevertheless have been accepted when a superiority design is not possible and a rigorous/conservative non-inferiority margin is used. In a similar way, counterfactual placebo group efficacy analysis should use rigorous/conservative inference techniques that formally build in a rigorous/conservative margin to potential biases that could occur due to departures from unverifiable assumptions. Because reliability of this approach would require new techniques for verifying that the study cohort experienced substantial exposure to HIV-1, currently it may be appropriate as a secondary objective but not as a primary objective.
四项关于候选疫苗或被动注入单克隆抗体预防HIV-1感染的随机安慰剂对照疗效试验正在进行中（在南非男性和女性中开展的HVTN 702试验；在撒哈拉以南非洲女性中开展的HVTN 705试验；在撒哈拉以南非洲女性中开展的HVTN 703/HPTN 081试验；在美国、秘鲁、巴西和瑞士的男性或与男性发生性行为的跨性别者中开展的HVTN 704/HPTN 085试验）。后续疗效试验的优化设计面临若干挑战，包括：（1）如何考虑可能作为试验设计一部分或预防标准的不断演变的有效预防干预措施组合；（2）如何根据每项正在进行的试验的主要疗效结果和次要“保护相关因素”结果来定义可行且最优的后续试验设计；（3）如果预计所有研究组中的HIV-1发病率都非常低，以至于标准试验设计的机会成本过高，如何定义后续疗效试验的主要目标。在总结正在进行的试验后，我将讨论后续疗效试验设计的统计学考量，包括一般情况以及针对上述每项试验的具体情况。一个结论是，“保护相关因素”分析的结果对后续试验设计有重要影响，该分析确定了不同的宿主免疫标志物和HIV-1病毒特征如何影响HIV-1风险和预防效果。由于试验前有针对性的假设，即血清中和的效力和覆盖范围构成HIV-1感染的替代终点，这种影响在单克隆抗体试验中尤为相关。另一个结论是，虽然评估针对虚拟安慰剂组的预防效果存在偏差风险，但这种分析仍然很重要，应该开发结合研究设计和分析方法来优化此类推断。我将其与非劣效性设计进行类比，非劣效性设计在解释结果时需要做出无法验证的假设，因此存在风险，但当无法采用优效性设计且使用严格/保守的非劣效性界值时，它仍被接受。类似地，虚拟安慰剂组疗效分析应使用严格/保守的推断技术，正式纳入因偏离无法验证的假设而可能出现的潜在偏差的严格/保守界值。由于这种方法的可靠性需要新技术来验证研究队列是否大量接触了HIV-1，目前将其作为次要目标可能合适，但不作为主要目标。
