Current Epidemic Growth Status (Based on R_t) for States and Territories

Interpreting R_t

• R_t is a data-driven measure of disease transmission. R_t  is an estimate on date t of the average number of new infections caused by each infectious person. R_t accounts for current population susceptibility, public health interventions, and behavior.
• R_t > 1 indicates that infections are growing because, on average, each infected person is causing more than one new infection while R_t < 1 indicates that infections are declining.
• R_t can be a leading indicator of increases or decreases in cases, hospitalizations, or deaths, because transmission occurs before case confirmation, hospitalization, or death.
• The uncertainty range for each R_t estimate determines the probability that infections are growing. For example, if 75% of the uncertainty range falls above 1, then there is a 75% chance that the infections are growing in that location.
• When the data are sparse, the model used to generate R_t  estimates will tend to generate estimates nearer to 1 with wide credible intervals, which reflects uncertainty in the true epidemic trend during these time periods.

Caveats and limitations

• R_t estimates are sensitive to assumptions about the generation interval distribution.
• R_t estimates may be over or underestimated if the proportion of infections that result in hospitalizations changes abruptly. These estimates can be impacted by shifts in clinical severity, increased or decreased use of clinical testing, or changes in reporting.
• While these estimates are based on a single data source (hospitalizations), studies have indicated that any resulting biases are likely minor and that this is a robust approach to estimate R_t.

Methods

R_t is defined as the average number of new infections caused by each infected person at a particular time, t. When R_t > 1, infections are growing, and when R_t < 1, infections are declining. The color categories in the maps above were determined by estimating a distribution of possible R_t values based on the observed hospitalization data and model assumptions (formally, a “credible interval”). We then calculate the proportion of that credible interval where the R_t > 1. Credible intervals are determined using the EpiNow2 package, which uses a Bayesian model to estimate R_t, while adjusting for delays and reporting effects.

• If >90% of the credible interval distribution of R_t >1, infections are growing
• If 76%-90% of the credible interval distribution of R_t > 1, infections are likely growing
• If 26%-75% of the credible interval distribution of R_t > 1, infections have an uncertain trend or are stable (in this case, the credible interval spans across 1, and contains a mix of values above and below 1.)
• If 10%-25% of the credible interval distribution of R_t > 1, infections are likely declining; this is equivalent to 75%-90% of the credible interval of R_t ≤ 1
• If <10% of the credible interval distribution of R_t > 1, infections are declining; this is equivalent to >90% of the credible interval of R_t ≤ 1
• R_t was not estimated for states and territories in the following cases: 1. fewer than 10 laboratory-confirmed COVID-19 or influenza hospital admissions were reported in each of the prior 2 weeks, 2. there were detected anomalies in reported values, and 3. the model did not pass checks for reliability.

R_t  estimates are derived from daily counts of new COVID-19 or influenza hospitalizations. This blog post provides a more in-depth overview of the modeling approach used to estimate R_t, and the strategies CDC uses to validate the accuracy of estimates.

To estimate R_t, we fit Bayesian models to the data using the R packages EpiNow2, epinowcast, or using Stan models developed by the CDC Center for Forecasting and Outbreak Analytics. Following best practices, these models adjust for lags from infection to observation, incomplete observation of recent infection events, and  day-of-week reporting effects, in addition to uncertainty from all these adjustments.