Abstract
The SARS-CoV-2 driven disease, COVID-19, is presently a pandemic with increasing human and monetary costs. COVID-19 has put an unexpected and inordinate degree of pressure on healthcare systems of strong and fragile countries alike. In order to launch both containment and mitigation measures, each country requires accurate estimates of COVID-19 incidence as such preparedness allows agencies to plan efficient resource allocation and design control strategies. Here, we have developed a new adaptive, interacting, and cluster-based mathematical model to predict the granular trajectory COVID-19. We have analyzed incidence data from three currently afflicted countries of Italy, the United States of America, and India, and show that our approach predicts state-wise COVID-19 spread for each country with high accuracy. We show that $R_0$ as the basic reproduction number exhibits significant spatial and temporal variation in these countries. However, by including a new function for temporal variation of $R_0$ in an adaptive fashion, the predictive model provides highly reliable estimates of asymptomatic and undetected COVID-19 patients, both of which are key players in COVID-19 transmission. Our dynamic modeling approach can be applied widely and will provide a new fillip to infectious disease management strategies worldwide.
Introduction
Standard epidemiological models, such as the classic Susceptible-Exposed-Infected-Removed (SEIR) framework, often struggle to capture the reality of a pandemic in large, diverse nations. These models typically assume population homogeneity—the idea that every individual in a country interacts identically—and treat the basic reproduction number ($R_0$) as a fixed biological constant. In practice, COVID-19 transmission is a patchwork of local outbreaks influenced by population density, local lockdowns, and inter-state migration. Traditional models frequently overestimate peak infections because they fail to account for these granular shifts and the "closed population" assumption.
The AICSEIR Framework
To bridge the gap between theoretical overestimates and the likely underestimates of clinical testing, researchers developed the Adaptive Interacting Cluster-based SEIR (AICSEIR) model. This framework introduces three critical innovations:
- Clustering and Interaction: Populations are divided into sub-populations (states or regions). Interaction between these clusters is modeled using a stochastic migration parameter and an inter-cluster restriction parameter ($C$), where $C=0$ represents a total travel ban and $C=1$ represents free movement.
- Adaptive $R_0$: The model recognizes that policy changes, such as lockdowns, alter transmission rates. It re-estimates $R_0$ every 14 days by fitting the model to the most recent observed data.
- Weighted Prediction Correction ($L$): Inspired by Kalman filters, a weighting factor ($L$) is used to reconcile model predictions with clinical observations. This allows for a more realistic estimate of the true infection burden, including undetected asymptomatic cases.
Results and Discussion
The AICSEIR model demonstrated superior accuracy across datasets from Italy, the United States, and India. While traditional SEIR models overestimated peak infections by as much as 5x for Italy, the cluster-based approach provided significantly closer estimations.
Dynamic $R_0$ and Spatial Spread
The study confirmed that $R_0$ is a dynamic reflection of human behavior rather than a fixed constant. In Italy and the USA, $R_0$ distributions were broad (ranging from 2–14 and 4–12, respectively), whereas India showed a narrower range (2–6) during early stages, likely due to early interventions. Furthermore, modeling migration as a function of population and distance allowed the framework to successfully predict outbreaks in "virgin territories"—states with no initial cases.
Quantifying Intervention Impact
The model quantitatively captured the impact of public health measures. In Italy, mobility restrictions caused $R_0$ to drop toward unity, effectively "flattening the curve." By providing granular, state-level predictions (e.g., for Veneto or Maharashtra), the AICSEIR framework offers a practical tool for local resource allocation and targeted policy design.
Conclusion
The AICSEIR model moves beyond the "one-size-fits-all" approach of traditional epidemiology. By accounting for spatial heterogeneity, adaptive behavioral changes, and the gap between clinical data and true infection rates, it provides a robust framework for managing the current pandemic and future infectious disease outbreaks.
Summary updated