Table 1.
Epidemiological/compartmental US-CDC COVID-19 forecasting models—their description, features employed, methodology and assumptions they make regarding public health interventions.
| Model | Features used for forecasting | Author | Method | Assumptions |
|---|---|---|---|---|
| TTU-Squider (8) | Takes into account power-law incident rate, separate compartments for silent spreaders, quarantine/hospital isolation of infected individuals, social contact restrictions, possible loss of immunity for recovered individuals. | Hussain Lab, Texas Tech University | SIR | Effects of interventions are reflected in observed data and will continue going forward. |
| JHU-IDD (9) | Accounts for uncertainty in epidemiological parameters including R0, spread of more transmissible variants, infectious period, time delays to health outcomes and effectiveness of state-wide intervention policies. | JHU IDD Working Group | Meta population SEIR | Current interventions will not change during the period forecasted. |
| IowaStateLW-STEM (10) | A non-parametric space-time disease transmission model for epidemic data to study the spatial-temporal pattern of COVID-19. | Iowa State- Lily Wang's Group | Non-parametric spatiotemporal model | − |
| BPagano-RtDriven (11) | The effective transmission ratio, Rt, drives the model's projections. To forecast how Rt will change with time, the model analyzes Rt change data through the pandemic and applies a model of that characteristic behavior to forecast infections. | BPagano | SIR | Effects of interventions are reflected in observed data and will continue going forward. |
| UCLA-SuEIR (12) | An SEIR Model variant that takes into consideration the effects of re-openings. Assumes a transition from a virtual “Quarantined" group to the “Susceptible" group at a specific rate for the states that have reopened/ partially reopened. Most notable feature is that it can infer untested cases as well as unreported cases. | UCLA Statistical Machine Learning Lab | Modified SEIR | Contact rates will increase as states reopen and calculate the increase in contact rates for each state. |
| COVID19Sim-Simulator (13) | Uses a validated compartment model defined using SEIR with continuous-time progression to simulate the trajectory of COVID-19 at the state level. | COVID-19 Simulator | SEIR | Based on assumptions about how in the future, the levels of social distancing may evolve. |
| USACE-ERDC_SEIR (14) | Bayesian Inference calculates model parameters from observations of total number of cases. A prior probability distribution over the model parameters. The accumulated observations & subject matter knowledge are then coupled with a statistical model of model-data mismatch to generate a posterior probability distribution across model parameters. To make forecasts, parameters maximizing posterior probability density are used. | US Army Engineer Research & Development Center | Process-based classic SEIR model with compartments for unreported infections/ isolated individuals. | (i) Current interventions don't change during forecast period. (ii) Modeled populations are large enough that disease states fluctuations grow slower than average. (iii) Recovered individuals are not infectious/ susceptible to infections. |
| Microsoft-DeepSTIA (15) | Deep Spatio-temporal network with intervention under the assumption of Spatio-temporal process in the pandemic of different regions. | Microsoft | SEIR model on spatiotemporal network | Current interventions will not change during the period forecasted. |
| CovidAnalytics-DELPHI (16) | Introduces new states to accommodate for unnoticed cases, as well as an explicit death state. A non-linear curve reflecting government reaction is used to adjust the infection rate. Also, a meta-analysis of 150 factors is used to determine key illness parameters, while epidemiological parameters are fitted to historical death counts & identified cases. | MIT Covid-Analytics | Augmentation of SEIR model | |
| Columbia_UNC-SurvCon (17) | Considers transmission throughout pre-symptomatic incubation phase, employing a time-varying effective R0 to capture the temporal trend of transmission & change in response to a public health intervention. Uses permutation to quantify uncertainty. | Columbia_UNC | − | − |
| CU-select, CU-nochange, CU-scenario_low, CU-scenario_mid, CU-scenario_high (18) | Produces different intervention scenarios, each assuming various interventions & rates of compliance are implemented in the future. (i) Presents the weekly scenario believed to be most plausible given current observations & planned intervention policies. (ii) Current contact rates will remain unchanged in the future. Assumes relatively (iii) low transmission, (iv) moderate transmission, & (v) high transmission | Columbia University | Metapopulation county-level SEIR |