TL;DR
This paper uses Bayesian inference to estimate the basic reproduction number ($R_0$) and herd immunity threshold (HIT) for COVID-19 at the state level in the US. They found significant variation in $R_0$ across states, with New Jersey having the highest and Wyoming the lowest, indicating that reaching herd immunity will be a more challenging task in some states than others. Crucially, their analysis suggests that as of September 2021, no US state had achieved herd immunity, even when accounting for vaccination and infection data, and considering the increased transmissibility of the Delta variant.
What Problem Does This Paper Address?
Understanding the dynamics of infectious disease spread, particularly for a novel pathogen like SARS-CoV-2, is critical for effective public health interventions. A key metric for this is the basic reproduction number ($R_0$), which represents the average number of secondary infections caused by a single infected individual in a completely susceptible population. $R_0$ is directly related to the herd immunity threshold (HIT), the proportion of a population that needs to be immune to prevent sustained transmission. However, estimating these parameters accurately, especially at a granular level like individual states within a large and diverse country like the US, is challenging. Previous studies often provided national estimates, which can mask significant regional variations, or used simpler models that might not fully capture the complexities of disease transmission, including the impact of interventions. This paper aims to fill this gap by providing state-level $R_0$ and HIT estimates using a robust Bayesian modeling approach, allowing for a more nuanced understanding of the pandemic's progression and the challenges of achieving herd immunity in different parts of the US.
Methods
The researchers employed a Bayesian inference framework built upon a compartmental SEIR (Susceptible-Exposed-Infectious-Recovered) model, specifically adapted for COVID-19 dynamics. This model accounts for non-pharmaceutical interventions like social distancing. They used a next-generation matrix approach to derive the basic reproduction number ($R_0$) from the model's parameters. Bayesian inference was used to estimate region-specific model parameters, including the transmission rate ($\beta$), by fitting the model to daily confirmed case data. This approach allows for the quantification of uncertainty in the estimates through posterior distributions. They then calculated the herd immunity threshold (HIT) using the formula HIT = 1 - 1/$R_0$. To account for the evolving pandemic, they also adjusted $R_0$ and HIT estimates for the increased transmissibility of the SARS-CoV-2 Delta variant.
Datasets
The primary dataset used for training and inference was daily confirmed COVID-19 case reports for all 50 US states, obtained from The New York Times' COVID-19 data repository. This data spanned from January 21, 2020, to June 21, 2020. For assessing herd immunity progress, they also incorporated vaccination data from the CDC and seroprevalence survey data from various sources.
Key Insights and Ideas
- Significant State-Level Heterogeneity: The most striking insight is the substantial variation in $R_0$ across US states. This highlights that a "one-size-fits-all" approach to pandemic control is insufficient and that tailored strategies are needed for different regions.
- No State Achieved Herd Immunity (as of Sept 2021): Even with the Delta variant's increased transmissibility factored in, the analysis strongly suggests that no state had reached herd immunity by September 2021. This underscores the ongoing threat posed by the virus and the limitations of current vaccination and infection-induced immunity levels.
- Importance of Bayesian Inference: The use of Bayesian inference provides not just point estimates but also credible intervals, offering a crucial measure of uncertainty. This is vital for understanding the reliability of the $R_0$ and HIT estimates and for making informed public health decisions.
- Delta Variant's Impact: The study explicitly quantifies how the increased transmissibility of the Delta variant significantly raises the HIT, making herd immunity even more challenging to achieve.
Key Results
- $R_0$ Estimates: Maximum a posteriori (MAP) estimates for the basic reproduction number ($R_0$) for ancestral strains ranged from a low of 2.3 in Wyoming to a high of 7.1 in New Jersey. The population-weighted mean $R_0$ for the US was estimated at 4.7.
- HIT Estimates: Correspondingly, Delta-adjusted HIT estimates ranged from 82% to 94%.
- Herd Immunity Status: Based on their analysis of vaccination data, seroprevalence surveys, and Delta-adjusted HITs, the authors concluded that no US state had achieved herd immunity as of September 20, 2021.
- Model Validation: The model's predictive inferences for daily confirmed case counts showed good agreement with observed data for the states analyzed (New Jersey, Wyoming, Florida, and Alaska).
Limitations and Future Work
The authors acknowledge several limitations:
- Potential Downward Bias in HIT: The awareness of COVID-19's impact in other countries might have led some US populations to change their behavior before the study's observation period, potentially underestimating the true disease burden and thus the HIT.
- Dependence on Seroprevalence Data: The estimation of "percent progress toward herd immunity" is heavily reliant on seroprevalence data, which itself carries uncertainty, particularly regarding the detection of all infections.
- Model Simplifications: The compartmental model assumes well-mixed populations and lacks age structure, which is a simplification of real-world dynamics.
- Emergence of New Variants: The study was conducted before the emergence of the Omicron variant, which is known to be significantly more transmissible than Delta, further complicating the path to herd immunity.
Future work could involve incorporating more detailed demographic data (like age structure), refining models to better capture population heterogeneity, and continuously updating estimates as new data on variants and vaccine effectiveness become available. The authors also suggest that policies encouraging infection in children and vaccinated individuals might be considered, though they acknowledge the significant drawbacks of such an approach.
Summary updated