Real-time pandemic surveillance using hospital admissions and mobility data

Spencer J. Fox, Michael Lachmann, Mauricio Tec, Remy Pasco, Spencer Woody, Zhanwei Du, Xutong Wang, Tanvi Ingle, Emily Javan, Maytal Dahan, Kelly Gaither, Mark E A Escott, Stephen I. Adler, S. Claiborne Johnston, James G. Scott, Lauren Ancel MeyersProceedings of the National Academy of Sciences 2022 · DOI

TL;DR

This paper introduces a robust, real-time forecasting model for COVID-19 transmission and healthcare demand, leveraging hospital admissions and mobility data. Developed by a multi-disciplinary task force in Austin, TX, the model has successfully guided local policy and public awareness, demonstrating high accuracy in predicting hospitalizations and ICU needs.

What Problem Does This Paper Address?

Forecasting the trajectory of the COVID-19 pandemic has been a significant challenge due to limitations in available data. Traditional metrics like case counts are often biased by testing availability and practices, while death counts lag considerably behind actual infections. Hospital census data, though more timely, can be influenced by changing admission criteria and patient demographics. This creates a critical gap in providing timely and reliable information for public health decision-making, resource allocation, and policy implementation. The authors aim to bridge this gap by developing a model that can accurately predict severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission rates and healthcare demand using more readily available and less biased data sources.

Methods

The core of the paper's methodology is a sophisticated epidemiological model, specifically an age- and risk-structured susceptible-exposed-infected-recovered (SEIR) model. This model incorporates asymptomatic and symptomatic transmission, hospitalization, and mortality. A key innovation is its dynamic coupling with real-time mobility data. The transmission rate ($\beta(t)$) is modeled as a function of mobility trends, captured by the first two principal components derived from eight mobility variables (e.g., time spent at home, visits to points of interest). Crucially, the relationship between mobility and transmission is allowed to change daily, reflecting evolving public behavior and policy impacts. This is achieved by estimating time-dependent coefficients that govern this relationship.

The model is fitted to daily COVID-19 hospital admissions, discharges, and in-hospital deaths using an iterated filtering algorithm and particle filtering to estimate parameters and smooth posterior distributions of model states. Projections are then generated by sampling from these posterior distributions, capturing uncertainty in both initial conditions and future behavioral dynamics. The authors also compare their model's performance against simpler statistical models like random walk and ARIMA, as well as a version of their own model that omits mobility data.

Datasets

The primary datasets used in this study are:

  • COVID-19 Hospital Admissions, Census, Discharge, and Mortality Data: Sourced from Austin Public Health for the Austin-Round Rock Metropolitan Statistical Area (MSA).
  • Cell Phone Mobility Data: Provided by SafeGraph, encompassing eight mobility variables (home dwell time, visits to universities, bars, grocery stores, museums/parks, medical facilities, schools, and restaurants) at the census block group level. This data was aggregated and baselined against prepandemic levels.
  • Seroprevalence Data: Texas-wide seroprevalence estimates from the CDC (adjusted for Austin's specific mortality burden) were used for validation.

The paper does not introduce new benchmark datasets but rather utilizes publicly available data to build and validate its model.

Key Insights and Ideas

  • Hospital Admissions as a Superior Indicator: The study strongly advocates for hospital admissions as a more reliable and timely indicator of SARS-CoV-2 transmission and healthcare demand compared to case counts or mortality data, due to their lower bias and shorter lag times.
  • Mobility Data's Dynamic Relationship with Transmission: The paper highlights that the link between mobility and transmission isn't static. As people adopt precautionary behaviors (like masking), the same level of mobility can lead to lower transmission. Conversely, relaxation of measures can strengthen this link. Their model captures this evolving relationship.
  • The Power of Multi-Disciplinary Collaboration: The success of the model is attributed to the collaboration between scientists, public health officials, policymakers, and hospital executives, ensuring the model's outputs are relevant, actionable, and integrated into real-world decision-making.
  • Forecasting for Policy Guidance: The model's forecasts have directly informed critical policy decisions in Austin, including the staged alert system, resource allocation for hospitals, and the establishment of alternative care sites, demonstrating its practical utility.

Key Results

  • Accurate Forecasting: The model achieved high retrospective accuracy, with its 95% credible intervals containing 93.6%, 89.9%, and 87.7% of reported data for 1, 2, and 3 weeks ahead forecasts, respectively.
  • Transmission Dynamics: The estimated local reproduction number ($R_t$) peaked at 5.8 early in the pandemic and dropped to a low of 0.65 after the summer 2020 surge.
  • Case Detection Rates: Estimated case detection rates varied significantly, from 17.2% early on to a high of 70% in January 2021.
  • Weakened Mobility-Transmission Link: By February 2021, mobility-associated transmission was estimated to be 62% lower compared to March 2020, indicating increased public safety measures.
  • Model Performance: The authors' mobility-driven model outperformed simpler statistical models and a version of their own model without mobility data, particularly during pandemic surges, offering better accuracy and precision.

Limitations and Future Work

  • Superspreading Events: The model does not explicitly incorporate superspreading events, which could lead to underestimation of future risks, especially in settings like long-term care facilities.
  • Population Mixing Assumption: The model assumes Austin is a well-mixed population, neglecting important heterogeneities like specific facility types (e.g., long-term care) and significant geographic segregation that could influence transmission dynamics.
  • Infection Hospitalization Rate Uncertainty: Estimates are sensitive to assumed infection hospitalization rates, which vary by age and health status and remain uncertain.
  • Future Work: The authors suggest incorporating mechanistic behavioral dynamics, organizational decision-making, and feedback loops between sociological and epidemiological dynamics into future models. They also emphasize the need to integrate data on vaccine-acquired immunity and to adapt models for other cities, considering their unique demographic and epidemiological characteristics. The ongoing need for immediate, reliable, and comprehensive access to SARS-CoV-2 surveillance data is also highlighted.

Summary updated

Abstract

Forecasting the burden of COVID-19 has been impeded by limitations in data, with case reporting biased by testing practices, death counts lagging far behind infections, and hospital census reflecting time-varying patient access, admission criteria, and demographics. Here, we show that hospital admissions coupled with mobility data can reliably predict severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission rates and healthcare demand. Using a forecasting model that has guided mitigation policies in Austin, TX, we estimate that the local reproduction number had an initial 7-d average of 5.8 (95% credible interval [CrI]: 3.6 to 7.9) and reached a low of 0.65 (95% CrI: 0.52 to 0.77) after the summer 2020 surge. Estimated case detection rates ranged from 17.2% (95% CrI: 11.8 to 22.1%) at the outset to a high of 70% (95% CrI: 64 to 80%) in January 2021, and infection prevalence remained above 0.1% between April 2020 and March 1, 2021, peaking at 0.8% (0.7-0.9%) in early January 2021. As precautionary behaviors increased safety in public spaces, the relationship between mobility and transmission weakened. We estimate that mobility-associated transmission was 62% (95% CrI: 52 to 68%) lower in February 2021 compared to March 2020. In a retrospective comparison, the 95% CrIs of our 1, 2, and 3 wk ahead forecasts contained 93.6%, 89.9%, and 87.7% of reported data, respectively. Developed by a task force including scientists, public health officials, policy makers, and hospital executives, this model can reliably project COVID-19 healthcare needs in US cities.
BibTeX
@inproceedings{fox2022real,
  author = {Spencer J. Fox and Michael Lachmann and Mauricio Tec and Remy Pasco and Spencer Woody and Zhanwei Du and Xutong Wang and Tanvi Ingle and Emily Javan and Maytal Dahan and Kelly Gaither and Mark E A Escott and Stephen I. Adler and S. Claiborne Johnston and James G. Scott and Lauren Ancel Meyers},
  title = {{Real-time pandemic surveillance using hospital admissions and mobility data}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2022},
  doi = {10.1073/pnas.2111870119},
}
Figure