Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2406.12810 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911924522844160 |
|---|---|
| author | Safta, Cosmin Bridgman, Wyatt Ray, Jaideep |
| author_facet | Safta, Cosmin Bridgman, Wyatt Ray, Jaideep |
| contents | In this paper, we develop a method to estimate the infection-rate of a disease, over a region, as a field that varies in space and time. To do so, we use time-series of case-counts of symptomatic patients as observed in the areal units that comprise the region. We also extend an epidemiological model, initially developed to represent the temporal dynamics in a single areal unit, to encompass multiple areal units. This is done using a (parameterized) Gaussian random field, whose structure is modeled using the dynamics in the case-counts, and which serves as a spatial prior, in the estimation process. The estimation is performed using an adaptive Markov chain Monte Carlo method, using COVID-19 case-count data collected from three adjacent counties in New Mexico, USA. We find that we can estimate both the temporal and spatial variation of the infection with sufficient accuracy to be useful in forecasting. Further, the ability to "borrow" information from neighboring areal units allows us to regularize the estimation in areal units with high variance ("poor quality") data. The ability to forecast allows us to check whether the estimated infection-rate can be used to detect a change in the epidemiological dynamics e.g., the arrival of a new wave of infection, such as the fall wave of 2020 which arrived in New Mexico in mid-September 2020. We fashion a simple anomaly detector, conditioned on the estimated infection-rate and find that it performs better than a conventional surveillance algorithm that uses case-counts (and not the infection-rate) to detect the arrival of the same wave. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_12810 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Detecting Outbreaks Using a Latent Field: Part I -- Spatial Modeling Safta, Cosmin Bridgman, Wyatt Ray, Jaideep Applications In this paper, we develop a method to estimate the infection-rate of a disease, over a region, as a field that varies in space and time. To do so, we use time-series of case-counts of symptomatic patients as observed in the areal units that comprise the region. We also extend an epidemiological model, initially developed to represent the temporal dynamics in a single areal unit, to encompass multiple areal units. This is done using a (parameterized) Gaussian random field, whose structure is modeled using the dynamics in the case-counts, and which serves as a spatial prior, in the estimation process. The estimation is performed using an adaptive Markov chain Monte Carlo method, using COVID-19 case-count data collected from three adjacent counties in New Mexico, USA. We find that we can estimate both the temporal and spatial variation of the infection with sufficient accuracy to be useful in forecasting. Further, the ability to "borrow" information from neighboring areal units allows us to regularize the estimation in areal units with high variance ("poor quality") data. The ability to forecast allows us to check whether the estimated infection-rate can be used to detect a change in the epidemiological dynamics e.g., the arrival of a new wave of infection, such as the fall wave of 2020 which arrived in New Mexico in mid-September 2020. We fashion a simple anomaly detector, conditioned on the estimated infection-rate and find that it performs better than a conventional surveillance algorithm that uses case-counts (and not the infection-rate) to detect the arrival of the same wave. |
| title | Detecting Outbreaks Using a Latent Field: Part I -- Spatial Modeling |
| topic | Applications |
| url | https://arxiv.org/abs/2406.12810 |