In silico clinical studies for optimal COVID-19 vaccination schedules in patients with cancer

Abstract

This study introduces a tailored COVID-19 model for patients with cancer, incorporating viral variants and immune-response dynamics. The model aims to optimize vaccination strategies, contributing to personalized healthcare for vulnerable groups.

This study introduces a tailored COVID-19 model for patients with cancer, incorporating viral variants and immune-response dynamics. The model aims to optimize vaccination strategies, contributing to personalized healthcare for vulnerable groups.

Main text

Despite the declining number of deaths and hospitalizations, the combination of waning immunity and multiple circulating viral variants means that there is an ongoing need for a rational approach to the design of routine COVID-19 vaccination schedules.¹ Determining the optimal timing of additional vaccine doses in individuals with varied characteristics and comorbidities as well as the need for vaccine updates to account for evolving viral variants is of crucial importance. Only recently, clinical studies provided the first sets of human data showing the ability of a booster dose of BNT162b2 vaccine to induce spike-binding antibodies to the Omicron subvariants BA.2, BA.3, BA.4, BA.5, and XBB.1.^2,3,4,5,6 However, by the time these studies were published, vaccination protocols had largely been changed to incorporate the new bivalent vaccines designed to be more specific for the BA.4/5 subvariants. Thus, both viral and vaccine characteristics continue to evolve, and this remains an ongoing challenge in the design of vaccination schedules. This is particularly evident for patients with significant comorbidities, such as cancer, who may be underrepresented in general vaccine cohorts. Unfortunately, clinical data on the long-term effectiveness of booster doses for patients with cancer are limited and rarely available coincident with identification of a novel viral variant. More rigorous and scientifically founded guidelines may also help reduce the now prevalent “vaccine fatigue.”⁷

Given these challenges, there is an urgent need for the development of rapidly adaptable tools to predict the effectiveness of COVID-19 vaccines over the long term, accounting for individual characteristics and comorbidities as well as the emergence of new viral variants. We propose that in silico clinical studies, i.e., the use of computer simulations for the evaluation of a medicinal product or intervention, are a feasible solution. We developed a mechanistic model of COVID-19 infection that consists of a series of linear and non-linear differential equations that describe the dynamics of infection of epithelial cells in the lung by SARS-CoV-2, the innate immune response to infection, including the production of pro- and anti-inflammatory cytokines, and the activation of the coagulation cascade. The model has been extensively described previously.^8,9,10 The model further accounts for interactions between the virus and immune cells including neutrophils, B cells, and T cells. We explicitly model the separate mechanisms of mRNA vaccines. The vaccines, as nanoparticles in the case of mRNA vaccines, enter host cells, and then translation produces viral peptide antigens. Subsequently, vaccine-induced peptides exit the cells and interact with dendritic cells to produce antigen-presenting cells. These subsequently activate T cells and B cells to create CD4⁺ and CD8⁺ effector and memory T cells as well as short-lived and long-lived plasma (antibody-secreting) B cells. Details of the full equations governing these processes in the model have been previously provided.^8,9,10 Furthermore, the specific characteristics of novel variants, including immunogenicity and replicative potential, can also be incorporated. We have performed in silico studies using our modeling framework to predict the long-term effectiveness of COVID-19 vaccines in healthy individuals and those who have cancer or suppressed immune responses.⁹ The challenge with such mechanistic predictive models is that they depend upon a large number of parameters whose values are not explicitly known for the SARS-CoV-2 virus. Therefore, robust validation of model predictions against newer clinical data, whenever they become available, is critical.

In Figures S1A–S1F, we present validation data in which we compare the predictions previously made by our model with newly published data.^2,3,4,5,6 For this validation study, model parameter values were fixed at those values determined based on previous datasets available at the time of model development with no additional fitting performed for the newly available validation data.^9,10 To quantify the quality of the fit, we provide the χ² values calculated for the model predictions vs. the clinical data; these values are χ² < 0.0081 (caption of Figures S1A–S1F). The good agreement of our in silico model predictions with these new data highlights the usefulness of mathematical tools for optimizing vaccine design and schedule. To estimate the optimal schedules of vaccinations for cancer patients, we conducted simulations for the risk of severe disease as a function of time since the last injection for both healthy individuals and patients with cancer with monovalent and bivalent booster doses (as illustrated in Figures S1F and S1G). As expected, we find that variants that result in lower antibody-virus affinity than XBB.1 can cause severe infections, with high viral loads following a monovalent booster dose. Additionally, a bivalent booster dose resulted in lower viral loads. The bivalent booster provides greater protection against the Omicron variant and its related variants (BA.1 to BA.5) than the monovalent booster and supports the prioritization of bivalent booster vaccinations for patients with cancer or reduced immunity. These findings are in accord with the April 18, 2023 approval by the U.S. Food and Drug Administration (FDA) of an additional booster 6 months after the last booster for people above 65 and/or who are immunocompromised.

Preliminary data on newer variants such as EG.5 and BA.2.86 suggest these variants might possess distinct transmission dynamics and/or potential for immune evasion. While clinical data on these variants are still sparse, our adaptive modeling approach presumes that, based on the current trajectory of the virus, such variants might demonstrate transmission dynamics and potential immune escape surpassing that of the previous variants BA.5 and XBB.1. Hence, they could lead to more severe infections accompanied by elevated viral loads. Such assumptions underscore the pressing need for updated COVID-19 vaccines that align closely with currently circulating variants, ensuring enhanced protection against grave COVID-19 outcomes like hospitalizations and fatalities. This viewpoint is further corroborated by the recent September 11, 2023 approval from the U.S. FDA for an additional vaccine dose.

An advantage of our in silico approach is the ability to rapidly assess vaccine response in unique populations, such as those with hematological malignancies. The immune landscape of these patients is intricate, and prior reports suggest that patients with hematologic malignancies have lower serologic response rates to COVID vaccination, as underscored by recent findings.¹¹ Adapting to this critical insight, we evolved our modeling framework, ensuring a holistic representation that now includes the nuanced characteristics of hematological malignancies, especially those with depressed humoral immunity,¹² by setting to zero immune cell perturbations to production of naive CD8⁺, CD4⁺ T cells, and B cells. Other components and mechanisms of immunity were not modified. A patient population can have unique, heterogeneous immune landscapes, and our aim is to capture the diverse spectrum of immune capabilities, acknowledging that individual responses can vary widely depending on the type of hematological malignancy and treatment received. We observed that not only do variants with diminished antibody-virus affinity like XBB.1 pose a risk, but emerging variants such as EG.5 and BA.2.86 might also exacerbate the severity of infections, leading to heightened viral loads in these uniquely vulnerable patients (Figure S1H).

Importantly, our in silico analysis agrees well with the conclusions of the recent study by Khoury et al.¹³ based on aggregated clinical data (Figure S1A). This study predicted that boosting with ancestral vaccines can significantly enhance protection against both symptomatic and severe disease caused by SARS-CoV-2 variants, but variant-modified vaccines offer additional protection even if they do not perfectly match with the circulating variants. Therefore, choosing ancestral-based versus variant-modified vaccines for booster shots is crucial, as these options offer different degrees of protection against emerging variants of the virus.

In conclusion, despite the large number of parameters needed to construct a truly mechanistic model of SARS-CoV-2 infection and vaccination, the close correspondence between our model predictions and subsequently available clinical data supports the utility of such models and suggests that they might serve as a beneficial tool for predicting the effectiveness of booster doses for different vaccine and viral variants. We hope that our study will contribute to the development of effective vaccination strategies for vulnerable populations, including patients with cancer.