Clinical Model-Based Meta-Analysis applied to Vaccine Development for SARS-CoV-2 Variants
Anna Largajolli (1), Akshita Chawla (2), Bhargava Kandala (2), Soumya Perinparajah (1), Nele Plock (1), Kenny Watson (1), Niharika Gandhapuneti (1), Raj Thatavarti (1), S. Y. Amy Cheung (1), Rik de Greef (1), Jeffrey R. Sachs (2)
(1) Certara, Princeton, NJ, USA; (2) Quantitative Pharmacology and Pharmacometrics, Merck & Co., Inc., Rahway, NJ, USA
Objectives: Kandala et al [1] showed how the model based meta-analysis (MBMA) approach can be successfully applied to enable decision making on COVID-19 vaccine candidates. In this work, an MBMA model was built based on aggregate level clinical data of 6 vaccine candidates (BNT162b2, mRNA-1273, Ad26.COV2.S, 2019nCoV, Gam-COVID-Vac, ChAdOx1 nCoV-19) against the wild type strain and validated on only 2 vaccine candidates (CoronaVac and BBV152). This model quantified the relationship between clinical immunogenicity, represented by serum neutralizing titers (SN), and protection data and supported previous findings that showed how SN are a correlate of protection in humans [2]. Here, we have extended this work, by applying the identified MBMA model on an updated dataset that includes new clinical studies containing information around different SARS-CoV-2 strains (Alpha (B.1.17), Beta (B.1.351) and Delta (B.1.617.2)) as well as studies on other vaccines that were published after the MBMA model was first developed.
Methods: A literature review based on pre-specified inclusion/exclusion criteria was conducted using PubMed-LitCovid, BioRxiv and MedRxiv databases. Summary level data of new clinical studies published up to November 2022 were curated and added to the database in addition to the eight studies already available. From the identified 27 clinical studies, only 19 were retained in the actual analysis dataset as they presented both immunogenicity and efficacy information. Moreover, data from Cromer et al [3] were curated and included, as this reference contained additional variant-specific information. In total data of 15 vaccines was retained in the final analysis dataset. MBMA was performed to assess the relationship between SN titers after vaccination and incidence rate of symptomatic disease (IR) as an efficacy endpoint to indicate severity of infection. The different SARS-CoV-2 virus strains were evaluated as potential covariates during model development. To standardize assay differences across clinical studies, SN titers were normalized to the reported geometric mean convalescent titers (HCS) from their respective assay [2, 4]. Non-linear mixed-effects modeling, as implemented in the nlme R package, was used to quantify the relationship in humans between SN titers after vaccination and IR.
Results: The new data was first overlayed on top of the previously identified SN and IR relationship, a sigmoidal function that decreases with increase of the titers. New data from previously included vaccines aligned with the identified relationship between SN and IR previously predicted by the model. Data from all the seven newly added vaccines (e.g., Ad5-nCoV or ZyCoV-D) were not well predicted. The MBMA model was then re-fitted to the new data, and it confirmed the exploratory findings: when considering newly added vaccine data, the model predictions changed substantially; when considering the same subset of eight vaccines used in the initial analysis, the model estimates were comparable. When focusing on this subset of data, the virus strain was found to be a significant covariate on EC50. Specifically, EC50 hints that it is harder to protect against non-wild type virus strains.
Conclusions: In this work, we have applied the previously identified MBMA clinical model to an updated analysis dataset containing publications up to late 2022. The identified model relationship between immunogenicity and efficacy was confirmed on new data pertaining to the initial eight vaccines included in Kandala et al [1] and the virus strain was identified as a significant covariate. The newly introduced vaccines did not follow the previously identified SN-IR relationship. Some model limitations that could cause this are the fact that the HCS was collected earlier in time with respect to the time of the studies and that the population under analysis may not be naïve to covid anymore.
References:
[1] Kandala, Bhargava, et al. "Accelerating model-informed decisions for COVID-19 vaccine candidates using a model-based meta-analysis approach." EBioMedicine 84 (2022).
[2] Earle, Kristen A., et al. "Evidence for antibody as a protective correlate for COVID-19 vaccines." Vaccine 39.32 (2021): 4423-4428.
[3] Cromer D, Steain M, Reynaldi A, et al. “Neutralising antibody titres as predictors of protection against SARS-CoV-2 variants and the impact of boosting: a meta-analysis." Lancet Microbe. 2022;3(1):e52-e61.
[4] Khoury DS, Cromer D, Reynaldi A, et al. “Neutralizing antibody levels are highly predictive of immune protection from symptomatic SARS-CoV-2 infection.” Nat Med. 2021;27(7):1205-1211.