JMIR Publications

ABSTRACT

Background:

Despite immunization efforts, severe COVID-19 is a worldwide health problem since dexamethasone and remdesivir fail many hospitalized patients. These drugs target inflammation and viral replication, but they seldom improve immune function, especially in high-risk patients. New therapy involves adopting SARS-CoV-2-specific T cells (CoV-2-STs) produced ex vivo from convalescent or vaccinated donors. Cellular treatments may immediately ameliorate viral-specific adaptive immunity, humoral responses, and severe lymphopenia. In randomized clinical studies, CoV-2-STs improved clinical recovery, survival, and in vivo T cell growth without safety issues. COVID-19 progression heterogeneity owing to age, comorbidities, immunological status, and baseline biomarker profiles makes it challenging to determine which individuals may benefit most from targeted immunotherapies. Using multimodal clinical and immunological data, machine learning (ML) may forecast illness severity, prioritize patients, and optimize COVID-19 treatment options. Machine learning is used for imaging-based diagnosis, mortality prediction, and hospital admission categorization. However, machine learning employing early cellular immunology data to recommend treatment choices for novel medicines like CoV-2-STs is restricted. This must be fixed to allow precision medicine, provide patients the most modern treatments, and minimize unneeded procedures.

Objective:

The purpose of this research was to stratify high-risk individuals using early immunological and clinical indicators and create a prediction tool that would allow for individualized treatment choices.

Methods:

We performed a post hoc analysis of data from a randomized phase 1-2 trial comparing CoV-2-STs plus standard-of-care versus standard-of-care alone in severe COVID-19 patients. Longitudinal clinical and biomarker data from days 0 and 5 post-treatment were analyzed. Linear Discriminant Analysis was used to identify key biomarkers and construct predictive models. Cross-validation and Monte Carlo simulations assessed model robustness and hypothetical treatment scenarios. Model performance was evaluated using accuracy, precision, and sensitivity metrics.

Results:

Key immune cell populations and inflammatory markers, such as CD4+, CD8+, CD56+ cells, IL-6, LDH, and CRP, demonstrated significant differences between treatment arms by day 5. The Linear Discriminant Analysis model predicted patient outcomes with over 87% accuracy. Precision per class ranged from 85% to 100% depending on subgroup. Simulations indicated that up to 30% of patients in the standard-of-care group might have improved survival with CoV-2-STs treatment. Conversely, mortality could increase by 22% if CoV-2-STs-treated patients received standard-of-care alone. A therapeutic decision-support tool was developed to guide treatment based on patient-specific early biomarker profiles.

Conclusions:

We present a robust computational tool for individualized risk stratification and treatment selection in severe COVID-19. By integrating clinical and immunological data, it enables early prediction of treatment outcomes and supports precision medicine strategies. Future validation in prospective studies is warranted to confirm its clinical utility.