One year of modeling and forecasting COVID-19 transmission to support policymakers in Connecticut

Abstract

To support public health policymakers in Connecticut, we developed a county-structured compartmental SEIR-type model of SARS-CoV-2 transmission and COVID-19 disease progression. Our goals were to provide projections of infections, hospitalizations, and deaths, as well as estimates of important features of disease transmission, public behavior, healthcare response, and clinical progression of disease. In this paper, we describe a transmission model developed to meet the changing requirements of public health policymakers and officials in Connecticut from March 2020 to February 2021. We outline the model design, implementation and calibration, and describe how projections and estimates were used to support decision-making in Connecticut throughout the first year of the pandemic. We calibrated this model to data on deaths and hospitalizations, developed a novel measure of close interpersonal contact frequency to capture changes in transmission risk over time and used multiple local data sources to infer dynamics of time-varying model inputs. Estimated time-varying epidemiologic features of the COVID-19 epidemic in Connecticut include the effective reproduction number, cumulative incidence of infection, infection hospitalization and fatality ratios, and the case detection ratio. We describe methodology for producing projections of epidemic evolution under uncertain future scenarios, as well as analytical tools for estimating epidemic features that are difficult to measure directly, such as cumulative incidence and the effects of non-pharmaceutical interventions. The approach takes advantage of our unique access to Connecticut public health surveillance and hospital data and our direct connection to state officials and policymakers. We conclude with a discussion of the limitations inherent in predicting uncertain epidemic trajectories and lessons learned from one year of providing COVID-19 projections in Connecticut.

Keywords: SARS-CoV-2; SEIR epidemic model; case detection ratio; effective reproduction number; infection fatality ratio; infection hospitalization ratio; social distancing.

Figure 1:

Observed and estimated data used in model calibration and approximation of time-varying model parameters. Top row: A: daily new cases reported in Connecticut by the date of specimen collection among residents of non-congregate and congregate settings; B: COVID-19 hospitalization census; C: cumulative COVID-19 hospitalizations. In plots B and C, the total number (black line) represents observed data, while non-congregate (blue) and congregate (red) lines represent estimates. Middle row: D: cumulative COVID-19 deaths in Connecticut; E: hospital case fatality ratio (HFR) among hospitalized residents of non-congregate settings (estimated); F: average length of hospital stay among COVID-19 patients by month. Bottom row: G: normalized close interpersonal contact metric relative to the pre-epidemic period; H: proportion of cases 60+ years old among daily COVID-19 cases (only dataon the right of the dashed line is used in model parametrization); I: daily PCR testing volume.

Figure 2:

Schematic illustration of the model of SARS-CoV-2 transmission and COVID-19 disease progression; county map of Connecticut and county adjacency matrix. Individuals begin in the susceptible (S) compartment. Exposed individuals (E) may develop either asymptomatic (A), mild (I_M), or severe (I_S) infection. Asymptomatic and mild infections resolve without hospitalization and do not lead to death. Mild symptomatic cases self-isolate (R_M) shortly after development of symptoms, and transition to recovery (R) when infectiousness ceases. All severe cases require hospitalization (H) unless hospitalization capacity is exhausted, in which case they transition to $\overline{H}$ representing hospital overflow, then to recovery (R) or death (D). The model captures infection transmission in non-congregate settings, and excludes cases and deaths occurring in settings like nursing homes and prisons. It assumes a closed population without births and does not capture non-COVID-19 deaths. In the adjacency matrix, the dark gray cells correspond to counties that are adjacent.

Figure 3:

Model fit to observed data and estimates of epidemiologic features of SARS-CoV-2 transmission in Connecticut. Top row shows calibration results for: A: observed COVID-19 hospitalizations census, B: cumulative hospitalizations and C: cumulative deaths in Connecticut. Observed time series are shown as points and correspond to total hospitalizations and deaths among all Connecticut residents. The model is calibrated to estimated data series coming from non-congregate settings, and model projections are adjusted by the estimated difference to reflect the totals for congregate and non-congregate settings. Middle row: D: effective reproduction number, E: normalized measure of close contact relative to the pre-epidemic period along with the spline approximation (thick solid line), and F: contact function adjusted for estimated random effects that capture residual variation in transmission that is not explained by dynamics of close contact and other time-varying parameters. For comparison, black line in plot F shows smoothed normalized close contact metric unadjusted for random effects. Bottom row: G: cumulative incidence of SARS-CoV-2 infection, H: daily new infections, and I: estimated case detection ratio in non-congregate settings in Connecticut. Solid lines represent model-projected means and shaded regions represent 95% posterior predictive intervals. Some of the plots are overlaid with the key intervention dates (lockdown and phased reopening), as well as important event dates, including reopening of schools and universities and beginning of vaccination campaign.

Figure 4:

Posterior predictive performance of the transmission model calibrated using data up to the dashed line shown in each plot and projected forward for a period of two months. Solid lines represent model-projected means and shaded regions represent 90% posterior predictive intervals. Observed data are shown as points; lighter color points correspond to the data used in calibration.