Giters

mariabnd / measles-regional-immunity

Application of the Epidemic-Endemic framework to measles outbreaks.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

measles-regional-immunity

This repository contains the code and files to reproduce the analysis of our paper

Alexis Robert, Adam J Kucharski, and Sebastian Funk, The impact of local vaccine coverage and recent incidence on measles transmission in France between 2009 and 2018 (https://www.medrxiv.org/content/10.1101/2021.05.31.21257977v1)

This repository also contains scripts to generate a simulation study, aiming to assess the impact of aggregation on the Epidemic-Endemic framework.

Reproduce the analysis from the paper

The analysis presented in the paper uses case counts data reported to the European Center for Disease Prevention and Control, which is not publicly available. Therefore, in this repository, we first simulate the number of daily cases in each French department over a similar time span. Then, we fit the models developed using the Epidemic-Endemic framework with the same covariates as the models presented in the paper, and generate the figures. The data used to generate each covariate is located in the Data repository, and come from publicly available datasets. The details of each file and their sources is presented at the end of the README file.

Two scripts are needed to generate all the figures from the analysis (Main and Supplement):

  • R/generate_analysis_paper.R: Simulate all the data and fits needed to generate the figures in the main and the supplement:
    • Simulate the daily case counts.
    • Apply the report rate (i.e. keep 70% of the cases).
    • Fit both models (i.e. where all regions can infect each other, or only neighbours).
    • Simulate 1-year predictions using the model fits.
    • Generate the sensitivity analyses (impact of the inferred vaccine coverage + impact of the proportion of direct transmission in the model fits).
    • Compute the calibration scores 3,7,10, and 14 days ahead for both models, and the aggregated models.
    • Save all the elements listed above.
  • R/generate_all_figures_paper.R: Use the outputs from the script “generate_analysis_paper.R” to generate all the figures from the Main Paper, and from the Supplement.

Given the various analyses needed to generate all the figures, the script R/generate_analysis_paper.R can take several hours to run. Therefore, all the outputs are saved in the list Output/all_analysis.RDS, which is imported at the beginning of R/generate_all_figures_paper.R.

Two other scripts are needed to run the models and the analysis:

  • R/import_library.R: list and import all the libraries required in the analysis
  • R/import_covariates.R: Import all functions, generate the time series of each covariate (coverage, population, surface), compute the serial interval, compute the distance matrices. The covariates and the distance matrices were computed using publicly available datasets, imported from the folder “Data”.

The functions required to run each script are contained in the following files:

  • R/function_location_centroids.R: Defines the function function_centroids, which computes the weighted population centre of each French department using the R function zonal from the package raster, and the 1km2 European Grid dataset.
  • R/function_distance_population.R: Defines the function importation_distance and importation_pop_area. The function importation_distance generates the distance between population centroids using the files computed in function_centroids, or computes which regions are neighbours (if the input parameter neighbour is TRUE). The function importation_pop_area generates the number of inhabitant and the surface of each department using the Data file Data/age_structure.csv.
  • R/function_coverage.R: Contains the function importation_coverage, used to generate the three year average vaccine coverage in each department. The local vaccine uptake is imported from the Data file Data/Vacc_coverage_departement.csv. The missing values are then inferred using a Beta Mixed Model, and the function returns the time series of the local vaccine coverage.
  • R/function_serial_interval.R: Contains the function conv and serial_interval. conv is used to computed to convolution of a given function with itself. serial_interval is used to generate the serial interval of the disease, using mean and standard deviation as inputs, along with the proportion of transmission without missing generation. This function returns the distribution of the serial interval.
  • R/function_generate_outbreak.R: Contains the function generate_sim which generates one simulated outbreak using the Epidemic-Endemic framework, with the covariates used in the paper (coverage, population, recent incidence, surface, seasonality). Returns a list containing 8 elements: n_cases the number of daily cases per department, incid the level of recent incidence per department at each date, r0_ar, r0_ne, r0_en the value of the local predictor in each component, for each day, n_cases_ar, n_cases_ne, n_cases_en the average local number of cases stemming from each component at each date.
  • R/function_generate_all_outbreaks.R: Contains the function generate_all_sim, which generates a number of simulated outbreaks (defined in the input parameter n_sim). This function draws n_sim parameter sets, and uses generate_sim to simulate each outbreak. It returns a list containing 5 elements: sim a list containing the number of daily cases per department in each simulation, r_ar, r_ne, r_end a data frame describing the last value of each local predictor in each simulation, params_sim the parameter set used to generate each simulation, prop_comp: The proportion of cases stemming from each component for each simulation.
  • R/function_report_rate.R: Contains the function report_rate, used to remove a fraction of the generated cases, in order to mimic the partial detection of cases during an outbreak. Returns the number of daily cases reported per department.
  • R/function_interpret_control_daily.R: Contains a new version of the function surveillance:::interpretControl, re-defined to adapt the Epidemic-Endemic framework to the use of non-aggregated data.
  • R/function_exp_gravity.R: Contains the function W_exp_gravity_tot, used to implement an exponential gravity model to construct the weight matrix connecting the regions included in the Epidemic-Endemic model. Follows the structure of the function surveillance::W_powerlaw.
  • R/function_prepare_data_hhh4.R: Contains the function prep_data, used to prepare the covariate before running the Epidemic-Endemic models. This function generates the category of incidence at each date and the transmission potential using the simulated case counts data, and ensures the columns and rows order are the same in each covariate. It returns a list containing all the covariates used in the Epidemic-Endemic models
  • R/function_analysis_hhh4.R: Contains the functions function_hhh4_daily and function_hhh4_aggreg, used to fit a daily or aggregated Epidemic-Endemic model to a set of simulated data. The functions use prep_data to prepare the covariates. The models are then run. These functions return a list containing n_sim elements, with each element corresponding to one model fit.
  • R/function_predict_1y.R: Contains the functions term_predict, generate_1y_outbreak, simulation_main_figures, and coordinates, which are used to simulate future outbreaks after fitting the model, which will eventually be used to generate Figures 5 and 6. term_predict is used to integrate different conditions on the covariates (i.e. variations in vaccine coverage). generate_1y_outbreak is used to simulate one year of outbreak and returns the daily case counts per region. simulation_mains_figures uses generate_1y_outbreak to simulate the four simulation sets needed to generate Figures 5 and 6 (i.e. using the previous incidence, decreasing and increasing the vaccine coverage, and setting the level of previous incidence to the minimum). The simulations sets are run using the mean parameters and the covariance matrix from the model fits: the parameter n_param designates the number of parameter sets drawn, and n_sim_param describes the number of simulations per parameter set. Finally, coordinate is used to simulate the impact of group importations on transmission. It returns the coordinates of importation, using the regions of the imported cases as inputs. The coordinates can then be used as an input to figure_5_6 (parameter import_loc).
  • R/function_calibration.R: Contains the functions simulate_calib and calibration_model: In the daily model, the calibration study relies on simulating the number of overall cases over the calibration period at each date. simulate_calib is used to simulate the number of daily cases per department over the calibration period, and calibration_model calls this function for each date of calibration, and compute the scores describing the match between the simulations and the data. calibration_model is also used to generate the calibration scores for aggregated model, in this case it uses the function OneStepAhead from the surveillance package.
  • R/function_sensitivity.R: Contains the functions sens_SI, sens_vax, and sens_weekday: Generate the model fits obtained under different conditions. In sens_SI the proportion of direct transmission in the composite serial interval varies (parameter prop_gen1); in sens_vax various values of the missing coverage data are generated before fitting the model; in sens_weekday a covariate is added to each component, quantifying the impact of weekends on the number of daily cases.
  • R/function_generate_map.R: Contains the function generate_map, used to import the shapefile of the metropolitan French department.

The objects generated by all these functions are used to generate the figures from the Main Paper and the Supplement. The functions used to generate the figures are contained in the following scripts:

The analyses generated by the functions and scripts described above rely on various publicly available datasets, which are stored in the folder Data:

One file generated in the scripts is saved in the folder Output.

  • [Output /all_analysis.RDS](Output /all_analysis.RDS): List generated using the script R/generate_analysis_paper.R, contains all objects needed to generate the figures from the Main paper and the Supplement. Since running the entirety of this script can take several hours, this object can be used to generate the figures directly.

Analyse the impact of aggregation on the Epidemic-Endemic framework using a simulation study

The functions described in this folder were also used to generate the paper “Impact of aggregation on the Epidemic-Endemic framework: A simulation study”. The objective of this paper is to compare the fits obtained using a daily and an aggregated Epidemic-Endemic model, in order to assess the impact of aggregation on the parameter estimates, and the calibration. We generated 100 simulated case counts, fitted a daily and an aggregated model to each simulation, and compared the fits obtained with the input parameters. Most of the functions used in this project were described in the previous section. The scripts and functions developed specifically for this paper are:

  • R/generate_simulations.R: R script, can be run to generate and save a list containing 100 simulated case counts (saved as the object Output/simulation_set.RDS).
  • R/generate_analysis_simulations.R: R script used to fit aggregated and daily Epidemic-Endemic models to the 100 simulations. This can take several hours, the models are grouped in lists and saved in the Output folder.
  • R/generate_calibration_scores.R: R script used to compute the calibration scores of each model. Since we used a 10-day aggregation, we only compute the 10 days ahead calibration scores. The scores are computed over the last year of data (35 dates of calibration). This script can take several hours to run. The scores are therefore grouped in two lists, which are saved as RDS files in the Output folder.
  • R/function_figures.R: Contains the functions plot_simulations, transp distance_params, in_CI, pit_param, plot_analysis, plot_prop_comp, and plot_calib_scores used to generate the different figures from the paper. plot_simulations is used to generate figures 1 and 2; plot_analysis to generate figures 3, 4, 5, 6, and 8; plot_prop_comp to generate figure 7, and plot_calib_scores to generate figure 9.
  • R/generate_plots_simulations.R: R script used to generate all the figures included in the paper, using the simulations, fits, and calibrations computed during the previous scripts.

Given some scripts take several hours to run, some of the files generated are saved in the Output folder:

  • Output/simulation_set.RDS: List containing 6 elements describing the simulated data: sim the list containing all the daily case counts; r_ar, r_ne, and r_end the last local value of each predictor for each simulation, params_sim a data frame containing the parameter set used to generate each simulation, and prop_comp the proportion of cases stemming from each component in each simulation.
  • Output/models_aggregated.RDS: List containing the 100 aggregated models fitted to the simulations.
  • Output/models_daily1.RDS, models_daily2.RDS, models_daily3.RDS, models_daily4.RDS: Lists containing the 100 daily models fitted to the simulations. The 100 models were split in four lists, otherwise the object was too heavy for Github commits.
  • Output/scores_aggreg.RDS: List containing 6 data frames describing the calibration scores of the aggregated models: bias the value of bias for each calibration point, sharpness the value of sharpness for each calibration point, rps the ranked probability score for each calibration point, log the log score for each calibration point, px the probability that the number of cases in the simulation is equal to the data, pxm1 the probability that the number of cases in the simulation is equal to the data - 1.
  • Output/scores_day.RDS: List containing 7 data frames describing the calibration scores of the aggregated models: bias the value of bias for each calibration point, sharpness the value of sharpness for each calibration point, rps the ranked probability score for each calibration point, log the log score for each calibration point, px the probability that the number of cases in the simulation is equal to the data, pxm1 the probability that the number of cases in the simulation is equal to the data - 1, tot the median, 95% and 50% credible intervals for the number of cases generated in the calibration.

About

Application of the Epidemic-Endemic framework to measles outbreaks.

Languages

Language:R 100.0%