Skip to contents

Stability analysis of QUEST thresholds via iterative refitting

Source: R/run_loop_thresholds.R
run_loop_thresholds.Rd

Evaluates the robustness of the HMM states and the QUEST thresholds by repeatedly refitting the model over an expanding window of the time series.

Usage

run_loop_thresholds(
  obs_data,
  n_states = 2,
  type = "rate",
  seasonal = FALSE,
  quantiles = c(0.05, 0.7, 0.9, 0.99),
  epidemic_state_indices = NULL,
  step = 7,
  window_summary = 12
)

Arguments

obs_data

A data.frame formatted for run_hmm().

n_states

An integer. Number of hidden states (2 to 4).

type

A character string. Either "rate" (Gaussian) or "perc" (binomial).

seasonal

A logical. If TRUE, treats seasons as independent.

quantiles

A numeric vector. Quantile levels for threshold calculation.

epidemic_state_indices

An integer vector. HMM states considered as "epidemic" (see the section 'Defining the "epidemic" state(s)' in run_threshold_computation).

step

An integer. The number of units to advance the cutoff date in each iteration. Defaults to 7 (one week).

window_summary

An integer. The number of most recent iterations to include when calculating the final summary statistics.

Value

An object of class epiquest_loop. This is a list containing:

  • thresholds: A data frame of all thresholds calculated at each cutoff date.

  • states: A data frame showing how state means and standard deviations evolved.

  • summary: A summary table (median, mean, min, max) of thresholds within the final window_summary iterations.

  • data: The original input data.

Details

Stability analysis: To ensure the HMM has enough data to converge to a stable estimate, the function starts with a minimum training set of 50 step units. It then adds data in increments of step (default 1 week of 7 days), refitting the model and recalculating thresholds at every stage.

This process mimics how the model would perform in a real-time surveillance setting as new data arrives each week. If the thresholds or state means show high volatility, it may indicate that the model is overfit or that the surveillance signal is too noisy for the chosen number of states.

See also

create_loop_plots, run_threshold_computation

Examples

if (FALSE) { # \dontrun{
# Stability analysis takes time as it refits the model repeatedly
fit_loop <- loop_thresholds(df_sari_be, n_states = 2)

# Look at results
summary(fit_loop)

# Visualize results
create_loop_plots(fit_loop)
} # }