Extraction of {epireview} parameters to provide consensus distribution

[adamkucharski]

During epireview x epiverse event on 11th July, we discussed having a function that could generate random draws from across the available distributions, either filtering only on parameters with known distributions, or defining an assumed distribution to allow samples to be generated from all distributions with enough data for extraction (e.g. by using extract_param and generate.

The below script gives a tally of how many parameters in {epireview} fall into different extraction categories. 95% CrI excluded as this refers to the parameter uncertainty rather than distributional uncertainty. 95% CI can be ambiguous, so need to include a check against sample size to decide which it refers to – perhaps using the relationship (sample sd) = sd/(sample size) and identify outliers? And use the 95% CI = 2*sd approximation similarly?

pathogen_select <- "ebola"
distn_select <- "Human delay - Symptom Onset/Fever to Death"

data_in <- load_epidata(pathogen = pathogen_select)
params_in <- data_in[["params"]] |> 
              dplyr::filter(article_qa_score>50) |>  # filter by score
              dplyr::filter(parameter_type==distn_select) # filter by type

tally_results <- params_in |> 
  dplyr::summarise(
    dist_known = sum(!is.na(distribution_type) & !is.na(distribution_par1_type) & !is.na(distribution_par2_type)),
    na_dist_par_sd = sum(is.na(distribution_type)  & parameter_uncertainty_singe_type == "Standard Deviation",na.rm=T),
    na_dist_par_range = sum(is.na(distribution_type) & is.na(parameter_uncertainty_singe_type) & parameter_uncertainty_type == "Range",na.rm=T),
    na_dist_par_95ci = sum(is.na(distribution_type)  & is.na(parameter_uncertainty_singe_type) & parameter_uncertainty_type == "95% CI",na.rm=T),
    na_dist_par_iqr = sum(is.na(distribution_type)   & is.na(parameter_uncertainty_singe_type) & parameter_uncertainty_type == "IQR",na.rm=T)
  )

print(t(tally_results))