estimate_dist: Estimate a delay distribution using primarycensored

In EpiNow2: Estimate and Forecast Real-Time Infection Dynamics

Estimate a delay distribution using primarycensored

Description

Fit a delay distribution accounting for primary and secondary event censoring (double interval censoring) and right truncation. Estimation is done via MCMC using a Stan model that vendors likelihood functions from the primarycensored package. For more flexible delay distribution modelling (e.g. time-varying delays, partial pooling, or regression on covariates), see the epidist package. If you use this function, please cite primarycensored in addition to EpiNow2.

Usage

estimate_dist(
  data,
  dist = "lognormal",
  priors = switch(dist, lognormal = list(meanlog = Normal(1, 1), sdlog = Normal(0.5,
    0.5)), gamma = list(shape = Normal(2, 2), rate = Normal(0.5, 0.5)), normal =
    list(mean = Normal(5, 5), sd = Normal(1, 1)), exp = list(rate = Normal(0.5, 0.5)),
    weibull = list(shape = Normal(2, 2), scale = Normal(5, 5))),
  primary = "uniform",
  primary_params = numeric(0),
  stan = stan_opts(),
  max_value = NULL,
  obs_time_threshold = 2,
  verbose = FALSE
)

Arguments

data	A data.frame with date columns: pdate_lwr (required): lower bound of primary event date pdate_upr (optional): upper bound of primary event date (default: pdate_lwr + 1) sdate_lwr (required): lower bound of secondary event date sdate_upr (optional): upper bound of secondary event date (default: sdate_lwr + 1) obs_date (optional): observation/censoring date (default: max(sdate_upr)) n (optional): observation count/weight (default: 1)
dist	Character string, which distribution to fit. One of "lognormal" (default), "gamma", "normal", "exp", or "weibull".
priors	A list of ⁠<dist_spec>⁠ objects specifying priors for the distribution parameters. Names must match the parameters of the chosen distribution. Defaults depend on dist: lognormal: list(meanlog = Normal(1, 1), sdlog = Normal(0.5, 0.5)) gamma: list(shape = Normal(2, 2), rate = Normal(0.5, 0.5)) normal: list(mean = Normal(5, 5), sd = Normal(1, 1)) exp: list(rate = Normal(0.5, 0.5)) weibull: list(shape = Normal(2, 2), scale = Normal(5, 5))
primary	Character string specifying the primary event distribution. One of: "uniform" (default): uniform distribution over the primary window "expgrowth": exponential growth distribution. Requires primary_params to supply a fixed growth rate.
primary_params	Numeric vector of parameters for the primary distribution. Only used when primary = "expgrowth", in which case it should be a single numeric value for the growth rate. The growth rate is passed as fixed data to Stan, not estimated.
stan	A list of stan options as generated by stan_opts(). Defaults to stan_opts(). Can be used to override data, init, and verbose settings if desired.
max_value	Numeric, maximum delay value for PMF. If not provided, inferred from data.
obs_time_threshold	Numeric, multiplier for the obs-time-to-Inf heuristic. Observations where relative_obs_time > max(delay_upr) * obs_time_threshold are treated as untruncated. Default 2, following epidist. Set to Inf to disable.
verbose	Logical, print progress messages? Defaults to FALSE.

Details

The model fits an interval-censored delay distribution while accounting for:

• Primary event censoring (e.g., daily reporting of exposure)

• Secondary event censoring (e.g., daily reporting of symptom onset)

• Right truncation (observation window effects)

• Per-observation truncation times (via obs_date)

When a data frame with date columns is provided, observations are aggregated by unique combinations of ⁠(delay_lwr, delay_upr, pwindow, relative_obs_time)⁠ to reduce the number of likelihood evaluations. Observations where the relative observation time is much larger than the maximum observed delay are treated as untruncated (observation time set to infinity).

The primarycensored Stan functions are vendored (included in the package), so the model is pre-compiled and runs without needing primarycensored at runtime.

Limitations

• Delay distributions are limited to lognormal, gamma, normal, exponential, and weibull.

• The primary event distribution is limited to uniform or exponential growth with a fixed rate. Primary event parameters are not estimated.

• Left truncation is not yet exposed (internally set to -Inf).

Value

An ⁠<estimate_dist>⁠ object (inheriting from ⁠<epinowfit>⁠) with components:

fit

The Stan fit object.

args

The Stan data list used for fitting.

data

The input data.

Use get_parameters() to extract the fitted ⁠<dist_spec>⁠.

References

Park SW, et al. (2024) "Estimating epidemiological delay distributions for infectious diseases." doi:10.1101/2024.01.12.24301247

Charniga K, Park SW, et al. (2024) "Best practices for estimating and reporting epidemiological delay distributions of infectious diseases." PLoS Comput Biol 20(10): e1012520. doi:10.1371/journal.pcbi.1012520

Please cite primarycensored if you use this function; see citation("primarycensored").

See Also

vignette("estimate_dist_workflow", package = "EpiNow2") for a worked example, and primarycensored::primarycensored-package for the underlying censoring methodology.

Examples

# Fit lognormal distribution from date-based linelist
if (requireNamespace("primarycensored", quietly = TRUE)) {
  set.seed(1)
  n <- 100
  D <- 30
  pdate_lwr <- as.Date("2023-01-01") + rpois(n, 5)
  delays_sim <- primarycensored::rprimarycensored(
    n = n, rdist = rlnorm,
    meanlog = log(5), sdlog = 0.5,
    pwindow = 1, D = D
  )
  linelist <- data.frame(
    pdate_lwr = pdate_lwr,
    sdate_lwr = pdate_lwr + delays_sim,
    obs_date = pdate_lwr + D
  )
  result <- estimate_dist(linelist, dist = "lognormal")
}

EpiNow2 documentation built on June 17, 2026, 1:07 a.m.