In n8thangreen/ltbiScreenLite: ltbiScreenLite

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Introduction

This document explains how to run the cost-effectiveness model in the LTBIscreeningproject R package. There are a large number of files in this package but many of them are not essential to running the model. They are for model checking or plotting of outputs. Scripts are in ./scripts/ and functions are in ./R/.

Top-level

The model running consists of scripts (which can be sourced) and functions (which are available when the package is loaded). The highest level script is 000-programme-level-params-scenario-runner.R This simply loads required packages, input values and then runs the model using run_model().

Before this can be done though, we need to prepare the input values using create_input_workspace.R.

Create input workspace

There are 2 terms used to define a model run. They are ordered in terms of if and how they are modified between model runs.

\color{red}intervention\color{black} These are the higher-level simulation settings. Some of these are alway kept fixed and others can be \color{blue}varied\color{black}. Any given set of intervention values we say define a policy. These are discrete options.
N.mc: Number of Monte Carlo samples; Default 1.
no_students: TRUE/FALSE; default: FALSE.
force_everyone_stays: TRUE/FALSE; default: FALSE.
screen_with_delay: Rather than screen everyone on entry screen at random 0-5 years from entry.
MAX_SCREEN_DELAY: 5 years after entry to country.
FUP_MAX_YEAR: Time horizon for active TB progression; default 100 years.
screen_age_range: e.g. 18-35 years old.
year_cohort: 2012 is most recent complete year; largest cohort, corresponds with Pareek () LTBI risk.
min_screen_length_of_stay: Default 0 i.e. everyone included.
discount_rate: 3.5%
\color{blue}incidence_list\color{black}: which incidence by country of origin groups to target for LTBI screening. WHO categories per 100,000 are $[0,50),[50,150),[150,250),[250,350),>350$
\color{blue}endpoint\color{black}: calculate QALYs and costs including those that exit EWNI or not i.e. time horizon
\color{blue}LTBI_test\color{black}: type of test QFT, QFT-plus, TSPOT.TB
\color{blue}treatment\color{black}: 6 months or 3 month LTBI treatment
\color{red}scenario\color{black} These are varied within policies. These can be discrete, deterministic values or defined distibutionally. We don't vary any health state utilities.
p: screening pathway branch probabilities
cost: screening pathway per individual costs

Therefore, the script create_input_workspace.R

Load raw cohort data 051206 - IMPUTED_sample.RData. This is from Aldridge() in Lancet.
Create and save policy data using data-prep_policies.R. This is a list of different model run inputs created using create_and_save_policies().
Create list of intervention parameter values using interv_constructor().
Create cost and QALY input lists unit_costs.RData, cost_effectiveness_params.RData, synthetic_cohort_params.RData using 01b-data-prep_cost-effectiveness.R.
Create list of dataframes each representing a scenario using create_and_save_scenarios() in 01-data-prep_scenario.R. This reads from an Excel workbook.
Clean the raw individual level data using 01c-data-prep_modelling.R and save as sample_cleaned.RData. This basically remove individuals with inconsistent event times or missing data.
This is the main working script in the data prep stage. A TB progression curve is estimated in active-TB-extrapolation.R. This is used in 04a_3-include-new-tb-events.R to extrapolate the times to TB progression for the total sample, under other event contraints. We can also now estimate the QALYs gained for disease-free, case fatality and cured for each patient.

`run_model()` and `run_policy()`

run_model() is a wrapper for run_policy(), iterating over all policies. run_policy() has these main steps:

1. Set-up

Define output folders, with setup_folders()
The specific policy and intervention values are loaded in to the workspace and the cohort modified accordingly in data-prep_constants-policy.R, using policy_interv() and policy_cohort()
The particular intervention and policy probability, cost and health state value are substitutes in to the screening decision tree using prep-decisiontree.R.

2. Modelling

parallel_decision_tree() is the parallelised wrapper around the cost-effectiveness calculations for a decision tree decision_tree_cluster().
The output of decision_tree_cluster() includes the probability of an LTBI individual being successfully treated to cured. This is used in activetb_qaly_cost() which calculates the population QALYs and costs due to active TB.
The costs and QALYs from the decision tree model and the population model are combined to give a total cost-effectiveness using combine_popmod_dectree_res().