Designing a Study of Correlates of Risk for Ebola Vaccination

M. Elizabeth Halloran, Ira M. Longini, Peter B. GilbertbioRxiv 2018 · DOI

TL;DR

This paper proposes a statistical framework and power calculations for identifying "correlates of risk" (CoRs) for Ebola vaccine protection during outbreak settings. The goal is to establish measurable immune responses that predict vaccine effectiveness, crucial for licensing new vaccines and guiding future development, especially when traditional placebo-controlled trials are infeasible.

What Problem Does This Paper Address?

Developing and licensing new vaccines, particularly for rapidly evolving threats like Ebola, relies on demonstrating their effectiveness. While the rVSV Ebola vaccine has shown high efficacy, there's a critical gap: we don't know why it works for some individuals better than others. Identifying "correlates of risk" (CoRs) – specific immune responses that predict protection – is essential. This is challenging because:

  1. Outbreak Dynamics: Conducting large, randomized placebo-controlled trials during an active Ebola outbreak is ethically difficult and logistically complex.
  2. Efficacy vs. Correlates: High overall vaccine efficacy (VE) doesn't tell us about individual-level protection. We need to understand what biological markers predict a stronger or weaker response to the vaccine.
  3. Licensure: Regulatory bodies often require evidence of correlates of protection for vaccine licensure, especially for vaccines intended for widespread use.

This paper tackles the problem of how to design studies to find these CoRs in real-world outbreak scenarios, even when traditional trial designs are not possible.

Methods

The authors adapt existing methods for assessing correlates of risk (CoRs) in clinical efficacy trials to an outbreak setting. The core approach is a nested case-control sampling plan within the vaccinated population.

  • Blood Sampling: Blood samples are collected from vaccine recipients at specific time points (e.g., baseline and a post-vaccination visit, "Visit 1"). These samples are stored.
  • Case Identification: During the follow-up period, individuals who develop Ebola (cases) are identified.
  • Control Selection: A subset of individuals who remain healthy (controls) are randomly selected from those who completed follow-up. A 5:1 control-to-case ratio is suggested for efficiency.
  • Immune Assays: Immune assays are performed on the stored blood samples from both cases and controls to measure potential immune response markers.
  • Statistical Analysis: The analysis focuses on estimating the relative risk (RR) of Ebola endpoint occurrence after Visit 1, comparing vaccine recipients with different levels of immune response markers. This can be done for continuous or trichotomous (low, medium, high) markers.
  • Correlates of VE: The paper also discusses how a CoR in vaccinees can be interpreted as a correlate of vaccine efficacy (VE) by making assumptions about the risk in a hypothetical placebo group.

A key aspect is the development of power calculations to determine the sample size and number of endpoints needed to detect a significant CoR under various assumptions about VE, marker variability, and assay precision.

Datasets

Not discussed in the paper. The paper focuses on the design and power calculations for a hypothetical study, referencing past Ebola outbreaks (e.g., DRC May-July 2018) as contexts for potential data collection.

Key Insights and Ideas

  • Feasibility in Outbreaks: It is possible to study correlates of risk even without a placebo group, by focusing on vaccinated individuals and making assumptions about hypothetical placebo risk.
  • Importance of Endpoints: The power to detect a CoR is highly dependent on the number of Ebola endpoints (cases) occurring in the vaccinated population. Even with high VE, a sufficient number of cases are needed.
  • Marker Quality Matters: The precision of the assay used to measure immune markers (represented by parameter $\rho$) significantly impacts the ability to detect a CoR. Markers with high signal-to-noise ratios ($\rho \approx 0.9$) are crucial.
  • Distribution of Responses: The distribution of immune responses within the vaccinated population is critical. Having a substantial proportion of individuals at the extreme ends of the immune response spectrum (very high or very low) increases the power to detect a CoR.
  • Sample Size is Key: Storing blood from a large number of vaccine recipients (e.g., at least 2500) is recommended to achieve adequate power for plausible effect sizes, especially if multiple markers are to be analyzed.

Key Results

The paper presents power calculations under various scenarios:

  • Overall VE = 75%:
    • With 50 Ebola endpoints in vaccinees, good power can be achieved.
  • With 2500 vaccinees and blood storage, power is highly dependent on the distribution of VE across immune response groups. For example, with VE high = 96%, VE medium = 75%, and VE low = 0%, 80% power can be achieved.
  • Overall VE = 90%:
    • With 20 Ebola endpoints in vaccinees, good power can be achieved under certain assumptions.
  • With 2500 vaccinees and blood storage, higher VE values in the high response group (e.g., 0.985) and lower VE in the low response group (e.g., 0.3) are needed to achieve high power, with stricter constraints on the proportion of individuals in the low VE group.
  • Noise Level ($\rho$): Power drops off rapidly with decreasing $\rho$, highlighting the importance of reliable assays.

Limitations and Future Work

Limitations:

  • Assumed Attack Rate: Power calculations are sensitive to the assumed attack rate in a hypothetical placebo group. Changes in this assumption linearly affect the required number of stored blood samples.
  • Indirect Effects of Ring Vaccination: The attack rate in vaccinated individuals might be lower than assumed due to the indirect effects of ring vaccination, potentially requiring more stored blood to identify enough cases.
  • Constraints on Proportions: With high overall VE (75% or 90%), there are narrow constraints on the proportions of individuals that can be in the low, middle, and high immune response groups for trichotomous biomarkers.
  • Control Definition: Controls are defined as individuals who complete follow-up without Ebola. The follow-up period can vary between outbreaks, and controls must ideally come from the same outbreak as the case, requiring careful consideration.
  • Early Samples: The study design might miss crucial information if early immune responses (e.g., within days of vaccination) are not captured, as some cases can occur very soon after vaccination.

Future Work:

  • Standardization: Standardizing the time for blood draws across outbreaks is recommended.
  • Early Samples: Collecting early samples from individuals, especially those who become cases, could provide valuable information about correlates.
  • Multiple Outbreaks: The study could be conducted over multiple outbreaks to accrue sufficient endpoints if one outbreak is insufficient.
  • Sampling Design: If individuals with predicted low or high immune responses can be identified, over-sampling them could increase power.
  • External Data: Utilizing data from animal challenge studies or previous immunogenicity trials to pre-select promising immune response markers for further investigation.
  • Further Power Calculations: The paper mentions supplementary materials with additional power calculations exploring variations in $\rho$, sample size, and CoR relative risk.

Summary updated

Abstract

Abstract The rVSV Ebola vaccine was shown to be very efficacious in a novel ring vaccination trial in Guinea. However, no correlates of vaccine protection have been established for Ebola vaccines. Several Ebola vaccine candidates are available, but conducting randomized trials of additional candidates in outbreak situations has become difficult. Establishing correlates of vaccine protection would be useful in helping vaccine candidates become licensed. In this note, we explore power and sample calculations to study potential correlates of risk (protection) during an Ebola vaccination campaign in an outbreak situation under a number of assumptions. At an overall vaccine efficacy of 75%, 50 Ebola endpoints in the vaccinees provided good power. At an overall vaccine efficacy of 90%, 20 Ebola endpoints gave good power under certain assumptions. In the May – July 2018 Ebola outbreak in DRC, over 3000 individuals were vaccinated, with no reported cases in vaccinated individuals. To be feasible, this type of study need Ebola endpoints in vaccinated individuals.
BibTeX
@misc{halloran2018designing,
  author = {M. Elizabeth Halloran and Ira M. Longini and Peter B. Gilbert},
  title = {{Designing a Study of Correlates of Risk for Ebola Vaccination}},
  year = {2018},
  doi = {10.1101/397547},
  howpublished = {bioRxiv},
}
Figure