skip to main content

Title: STAN: spatio-temporal attention network for pandemic prediction using real-world evidence
Abstract Objective

We aim to develop a hybrid model for earlier and more accurate predictions for the number of infected cases in pandemics by (1) using patients’ claims data from different counties and states that capture local disease status and medical resource utilization; (2) utilizing demographic similarity and geographical proximity between locations; and (3) integrating pandemic transmission dynamics into a deep learning model.

Materials and Methods

We proposed a spatio-temporal attention network (STAN) for pandemic prediction. It uses a graph attention network to capture spatio-temporal trends of disease dynamics and to predict the number of cases for a fixed number of days into the future. We also designed a dynamics-based loss term for enhancing long-term predictions. STAN was tested using both real-world patient claims data and COVID-19 statistics over time across US counties.


STAN outperforms traditional epidemiological models such as susceptible-infectious-recovered (SIR), susceptible-exposed-infectious-recovered (SEIR), and deep learning models on both long-term and short-term predictions, achieving up to 87% reduction in mean squared error compared to the best baseline prediction model.


By combining information from real-world claims data and disease case counts data, STAN can better predict disease status and medical resource utilization.

 ;  ;  ;  ;  ;  ;  ;  
Publication Date:
Journal Name:
Journal of the American Medical Informatics Association
Oxford University Press
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract

    The strain on healthcare resources brought forth by the recent COVID-19 pandemic has highlighted the need for efficient resource planning and allocation through the prediction of future consumption. Machine learning can predict resource utilization such as the need for hospitalization based on past medical data stored in electronic medical records (EMR). We conducted this study on 3194 patients (46% male with mean age 56.7 (±16.8), 56% African American, 7% Hispanic) flagged as COVID-19 positive cases in 12 centers under Emory Healthcare network from February 2020 to September 2020, to assess whether a COVID-19 positive patient’s need for hospitalization canmore »be predicted at the time of RT-PCR test using the EMR data prior to the test. Five main modalities of EMR, i.e., demographics, medication, past medical procedures, comorbidities, and laboratory results, were used as features for predictive modeling, both individually and fused together using late, middle, and early fusion. Models were evaluated in terms of precision, recall, F1-score (within 95% confidence interval). The early fusion model is the most effective predictor with 84% overall F1-score [CI 82.1–86.1]. The predictive performance of the model drops by 6 % when using recent clinical data while omitting the long-term medical history. Feature importance analysis indicates that history of cardiovascular disease, emergency room visits in the past year prior to testing, and demographic factors are predictive of the disease trajectory. We conclude that fusion modeling using medical history and current treatment data can forecast the need for hospitalization for patients infected with COVID-19 at the time of the RT-PCR test.

    « less
  2. Abstract The early detection of the coronavirus disease 2019 (COVID-19) outbreak is important to save people’s lives and restart the economy quickly and safely. People’s social behavior, reflected in their mobility data, plays a major role in spreading the disease. Therefore, we used the daily mobility data aggregated at the county level beside COVID-19 statistics and demographic information for short-term forecasting of COVID-19 outbreaks in the United States. The daily data are fed to a deep learning model based on Long Short-Term Memory (LSTM) to predict the accumulated number of COVID-19 cases in the next two weeks. A significant averagemore »correlation was achieved ( r =0.83 ( p = 0.005 )) between the model predicted and actual accumulated cases in the interval from August 1, 2020 until January 22, 2021. The model predictions had r > 0.7 for 87% of the counties across the United States. A lower correlation was reported for the counties with total cases of <1000 during the test interval. The average mean absolute error (MAE) was 605.4 and decreased with a decrease in the total number of cases during the testing interval. The model was able to capture the effect of government responses on COVID-19 cases. Also, it was able to capture the effect of age demographics on the COVID-19 spread. It showed that the average daily cases decreased with a decrease in the retiree percentage and increased with an increase in the young percentage. Lessons learned from this study not only can help with managing the COVID-19 pandemic but also can help with early and effective management of possible future pandemics. The code used for this study was made publicly available on« less
  3. As the COVID-19 pandemic evolves, reliable prediction plays an important role in policymaking. The classical infectious disease model SEIR (susceptible-exposed-infectious-recovered) is a compact yet simplistic temporal model. The data-driven machine learning models such as RNN (recurrent neural networks) can suffer in case of limited time series data such as COVID-19. In this paper, we combine SEIR and RNN on a graph structure to develop a hybrid spatiotemporal model to achieve both accuracy and efficiency in training and forecasting. We introduce two features on the graph structure: node feature (local temporal infection trend) and edge feature (geographic neighbor effect). For nodemore »feature, we derive a discrete recursion (called I-equation) from SEIR so that gradient descend method applies readily to its optimization. For edge feature, we design an RNN model to capture the neighboring effect and regularize the landscape of loss function so that local minima are effective and robust for prediction. The resulting hybrid model (called IeRNN) improves the prediction accuracy on state-level COVID-19 new case data from the US, out-performing standard temporal models (RNN, SEIR, and ARIMA) in 1-day and 7-day ahead forecasting. Our model accommodates various degrees of reopening and provides potential outcomes for policymakers.« less
  4. Background Population mobility is closely associated with COVID-19 transmission, and it could be used as a proximal indicator to predict future outbreaks, which could inform proactive nonpharmaceutical interventions for disease control. South Carolina is one of the US states that reopened early, following which it experienced a sharp increase in COVID-19 cases. Objective The aims of this study are to examine the spatial-temporal relationship between population mobility and COVID-19 outbreaks and use population mobility data to predict daily new cases at both the state and county level in South Carolina. Methods This longitudinal study used disease surveillance data and Twitter-basedmore »population mobility data from March 6 to November 11, 2020, in South Carolina and its five counties with the largest number of cumulative confirmed COVID-19 cases. Population mobility was assessed based on the number of Twitter users with a travel distance greater than 0.5 miles. A Poisson count time series model was employed for COVID-19 forecasting. Results Population mobility was positively associated with state-level daily COVID-19 incidence as well as incidence in the top five counties (ie, Charleston, Greenville, Horry, Spartanburg, and Richland). At the state level, the final model with a time window within the last 7 days had the smallest prediction error, and the prediction accuracy was as high as 98.7%, 90.9%, and 81.6% for the next 3, 7, and 14 days, respectively. Among Charleston, Greenville, Horry, Spartanburg, and Richland counties, the best predictive models were established based on their observations in the last 9, 14, 28, 20, and 9 days, respectively. The 14-day prediction accuracy ranged from 60.3%-74.5%. Conclusions Using Twitter-based population mobility data could provide acceptable predictions of COVID-19 daily new cases at both the state and county level in South Carolina. Population mobility measured via social media data could inform proactive measures and resource relocations to curb disease outbreaks and their negative influences.« less
  5. Mathematical models are widely recognized as an important tool for analyzing and understanding the dynamics of infectious disease outbreaks, predict their future trends, and evaluate public health intervention measures for disease control and elimination. We propose a novel stochastic metapopulation state-space model for COVID-19 transmission, which is based on a discrete-time spatio-temporal susceptible, exposed, infected, recovered, and deceased (SEIRD) model. The proposed framework allows the hidden SEIRD states and unknown transmission parameters to be estimated from noisy, incomplete time series of reported epidemiological data, by application of unscented Kalman filtering (UKF), maximum-likelihood adaptive filtering, and metaheuristic optimization. Experiments using bothmore »synthetic data and real data from the Fall 2020 COVID-19 wave in the state of Texas demonstrate the effectiveness of the proposed model.

    « less