Real-time pandemic surveillance using hospital admissions and mobility data

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.

Keywords: COVID-19; epidemiological data; forecasting; healthcare usage.

Fig. 1.

Fidelity and timeliness of COVID-19 data sources in the Austin, TX, MSA. (A) Scaled 7-d rolling averages of confirmed COVID-19 cases (8), COVID-19 hospital admissions, COVID-19 hospital census, COVID-19 ICU census, and COVID-19 mortality. Time series are scaled from zero to one. (B) Time-lagged correlations between all candidate predictors and ICU census at five lag intervals (–14, –7, 0, 7, and 14 d). Error bars indicate the 95% CI of the correlation coefficient for the specified lag. Negative x-axis lag values mean that the predictor leads the target (desirable); positive values mean the predictor lags the target.

Fig. 2.

Estimated COVID-19 pandemic healthcare, mobility, and epidemiological trends in the Austin–Round Rock, TX, MSA from February 18, 2020 to February 28, 2021. (A) Median fitted (line) and observed (points) daily COVID-19 hospital admissions, with gray ribbon indicating the 95% prediction interval. (B) First two principal components derived from eight cell phone mobility variables provided by SafeGraph (19). Yellow, orange, and red shading in the top graphs indicate the timing of COVID-19 alert stages 3, 4, and 5, respectively, in the Austin MSA (46). (C) Estimated 7-d average reproduction number (R_t) with gray 95% credible band. (SI Appendix, Fig. S4 shows the full range of values early in the pandemic.) Text boxes mark key policy changes and epidemiological events (details in SI Appendix, Table S7), with the following abbreviations: SH-WS indicates the March 24, 2020 Stay Home–Work Safe order; UT indicates the University of Texas at Austin; AISD indicates Austin’s largest public school system, Austin Independent School District; ACS indicates the Alternative Care Site established in a convention center to expand healthcare capacity; and GA-32 indicates a Texas order restricting elective surgeries, bars, and restaurants according to COVID-19 healthcare usage. (D) Transmission rates relative to baseline behavior from February 19, 2020 to March 1, 2021. Our model continually estimates this relationship between mobility and transmission, since increases and decreases in precautionary behavior can change this relationship. The graph compares the estimated transmission rate at each point in time to a hypothetical transmission rate that assumes no behavioral changes (i.e., the relationship between mobility and transmission remains fixed at a value estimated prior to wide adoption of COVID-19 safety measures). A positive (negative) value indicates that the observed transmission rate was higher (lower) than would be expected if precautionary behavior had remained constant. Shading indicates 95% credible bands. (E) Comparison between our projections for SARS-CoV-2 seroprevalence (black line with gray 95% credible bands) and estimates from a Texas-wide seroprevalence survey scaled to Austin (red points with 95% CIs) (48). (F) Estimated weekly case reporting rate, with gray 95% credible band. Values correspond to the proportion of cases infected on the given day (x axis) that are eventually reported. We indicate a data anomaly, in which thousands of backlogged cases were reported on a single day, and the impact of a catastrophic winter freeze that disrupted citywide testing and reporting operations (–51).

Fig. 3.

Retrospective validation of Austin area COVID-19 hospitalization and ICU projections from March 12, 2020 through February 1, 2021. (A and B) A comparison of predicted and observed (A) COVID-19 hospital census and (B) COVID-19 ICU census. Blue lines and points provide 2-wk-ahead projections with 95% prediction intervals at 14 time points throughout the pandemic; black points are reported data. The black tick marks along the bottom indicate our 30 worst forecasts, that is, dates with large differences between the observed value and our 2-wk-prior prediction of the value. (C and D) Predicted (median) versus observed (C) hospital or (D) ICU COVID-19 census. Colors indicate the time horizon of each prediction; the diagonal line indicates that the predicted value equals the observed value.

Fig. 4.

Comparison of 1-wk-ahead COVID-19 ICU projections for four models, from April 1, 2020 through February 1, 2021. Observed data (black points) are superimposed on forecasts using (A) a random walk model, (B) an autogenerated ARIMA model, (C) a simplified version of our model omitting the mobility covariate, and (D) the full version of our model. Blue lines and shading represent medians and 95% prediction intervals, respectively, across 1,000 stochastic projections. The tick marks on the x axis indicate days on which the observed ICU usage fell outside of the 1-wk-ahead 95% prediction interval. The horizontal dashed line indicates the estimated ICU capacity of 200 beds for the Austin metropolitan area.