Skip to contents

epi_growth_rate() estimates how fast an epidemic grows from counts of cases (or a line list of case dates) during its early, exponential phase. The daily (or weekly) incidence is modelled with a log-linear Poisson or negative binomial generalised linear model, $$\log E[C_t] = \alpha + r t,$$ so the growth rate \(r\) and its standard error come directly from the fitted model. From \(r\):

  • the doubling time is \(\log 2 / r\);

  • the reproduction number follows from the generation interval (Wallinga and Lipsitch 2007). For a gamma-distributed generation interval with mean \(\mu\) and standard deviation \(\sigma\), \(R = (1 + r \sigma^2/\mu)^{\mu^2/\sigma^2}\).

Confidence intervals are formed for \(r\) on the scale of the linear predictor (Wald) and carried through these monotone transformations. The fitted incidence and its confidence band are likewise computed on the link scale with the model's standard errors and back-transformed, so they are never negative. Choose window to cover the growth phase only: once susceptible hosts are depleted or control begins, growth slows and a single exponential no longer fits.

Usage

epi_growth_rate(
  data,
  date,
  cases = NULL,
  window = NULL,
  unit = c("day", "week"),
  family = c("poisson", "negbin"),
  generation_mean = NULL,
  generation_sd = NULL,
  level = 0.95
)

Arguments

data

A data frame: either a line list with one row per case, or counts per time with a cases column.

date

Column of dates (class Date) or numeric times.

cases

Optional column of case counts; if NULL, each row is one case.

window

Optional length-2 vector (same class as date) giving the first and last time of the growth phase to fit.

unit

Time unit of the counts and of \(r\): "day" or "week" (dates only).

family

"poisson" or "negbin" (negative binomial, for overdispersed counts; needs MASS).

generation_mean, generation_sd

Mean and standard deviation of the generation (or serial) interval, in units. If omitted, \(R\) is not computed.

level

Confidence level.

Value

A list of class tmr_epi_growth: estimates (a tibble with the growth rate, doubling time and \(R\), each with standard error or interval), fitted (incidence per time with the fitted value, its standard error on the link scale and confidence band), model (the fitted glm) and the settings. plot() draws the observed incidence with the fitted curve and band on a white background.

References

Wallinga, J. and Lipsitch, M. (2007). How generation intervals shape the relationship between growth rates and reproductive numbers. Proceedings of the Royal Society B, 274, 599-604.

WHO Ebola Response Team (2014). Ebola virus disease in West Africa: the first 9 months of the epidemic and forward projections. New England Journal of Medicine, 371, 1481-1495.

Examples

ebola <- outbreaks::ebola_sierraleone_2014
g <- epi_growth_rate(ebola,
  date = date_of_onset,
  window = as.Date(c("2014-06-01", "2014-09-30")), unit = "week",
  family = "negbin", generation_mean = 15.3 / 7, generation_sd = 9.3 / 7
)
g$estimates
#> # A tibble: 3 × 6
#>   quantity            estimate std_error  lower upper unit           
#>   <chr>                  <dbl>     <dbl>  <dbl> <dbl> <chr>          
#> 1 growth_rate            0.127    0.0193 0.0897 0.165 per week       
#> 2 doubling_time          5.44    NA      4.20   7.72  weeks          
#> 3 reproduction_number    1.30    NA      1.21   1.40  secondary cases
plot(g)