1,2Department of Computer Applications, SRM Institute of Science and Technology, Kattankulathur, Chennai, Tamil Nadu, India.
1Email: wl6649@srmist.edu.in
2Email: alberts@srmist.edu.in
Clinically significant sequelae of COVID-19 such as respiratory infections and enduring pulmonary vulnerability have become clinically significant challenges to surveillance, triage, and resource planning on the healthcare system level. This paper suggests a machine learning model based on stacking to evaluate personalized risks of developing post-COVID-19 respiratory infections and other unfavorable events through combinations of regularly measured clinical, laboratory, radiological and longitudinal follow-up variables. The methodology is based on combining heterogeneous base learners who are randomly chosen to represent complementary linear, non-linear and interaction effects into a meta-learner which optimally aggregates out-of-fold predictions to enhance generalization and minimize model variance. An overview of a reproducible pipeline is presented, that is, cohort definition during the post-acute period, feature harmonization across care settings, missingness treatment, temporal leaks management, and calibration to attain clinically interpretable risk probabilities. Discrimination, calibration, and decision-analytic measures are used to conduct model performance, and subgroup analysis is used to test performance across the age, sex, vaccination status, comorbidity burden, and severity of acute disease domain. To facilitate clinical adoption, explainability is also added with global and patient level attribution, which makes it possible to identify risk drivers to act upon, including inflammatory signatures, pulmonary impairment markers, oxygenation indices, and previous secondary infection hints. The presented stacking method is expected to provide better predictive results compared to single models without sacrificing transparency and operational viability to be implemented in post-COVID clinics and hospital follow-up routes. The article adds a systematic roadmap of the creation, testing, and reporting of stacking-based risk assessment frameworks that should be used in early intervention, targeted surveillance, and informed decision making in high-risk individuals on the risk of post-COVID respiratory illnesses. External validation among institutions is pointed out in order to make transportability and to reduce site specific bias.
Keywords: NA
How to cite this article: Mary LW, Raj SAA. Risk Assessment of Post-COVID-19 Respiratory Infections Using Stacking-Based Machine Learning Techniques. Int J Drug Deliv Technol. 2026;16(16s): 143-154. DOI: 10.25258/ijddt.16.16s.15
Source of support: Nil.
Conflict of interest: None