Skip to contents

Calculate Net Initiation Probabilities (Synthetic Cohort)

Source: R/calculate_net_initiation.R
calculate_net_initiation.Rd

Calculates "Net Initiation" probabilities by simulating a synthetic cohort. This metric represents the net flow into the "Current Smoker" state (Initiation + Relapse - Quitting) relative to the non-smoking population at each age.

It solves the issue where high initiation rates at young ages are offset by high quit rates (experimentation vs. established smoking).

Usage

calculate_net_initiation(
  init_data,
  quit_data,
  relapse_data,
  pops,
  config,
  boot_mode = FALSE
)

Arguments

init_data

Data.table. Initiation probabilities.

quit_data

Data.table. Quit probabilities.

relapse_data

Data.table. Relapse probabilities (must contain 'time_since_quit').

pops

Data.table. Not used. The synthetic cohort is a fixed 1000 people per subgroup, so there is nothing to weight. Kept in the signature because process_country_wrapper() and run_bootstrap_pipeline() both pass it.

config

List. Must contain 'country' and 'path'. The uncertainty parameters this used to document ('kn', 'kn_samp', 'kR') are not used either: uncertainty is handled by run_bootstrap_pipeline() calling this once per bootstrap sample.

boot_mode

Logical. If TRUE, skips writing to disk and returns the data.table directly.

Details

Assumptions:

  • The cohort starts with 100 in the initiation data.

  • Nobody dies. Over ages 12 to 30 that is close enough to true, and it is what makes the denominator below equal to 1 - prevalence.

  • The cohort is synthetic within a single year: ages are iterated with the year held fixed, so the stocks at age a are those of a lifetime lived under that one year's rates. Under secular change this differs from a real cohort's stocks, which matters when comparing against a cohort-followed survey estimator - see the header of 22_validate_net_initiation.R for the direction and size.

  • Quit and relapse probabilities must cover every age present in the initiation data; the function stops if they do not. On the current pipeline they always do – the relapse table extends below 18 by carrying the age-18 values, the same convention as everywhere else, so this calculation assumes nothing about under-18 relapse that the main estimates do not.

p_start_net can be negative. Past the age where the cohort's smoking prevalence peaks, quitting runs ahead of initiation and relapse and the net flow turns negative. That is a real feature of the age profile and it is returned as it is. It used to be clamped at zero, which flattened the curve from about age 24 and hid the fact that the model has prevalence peaking there at all.

Time since quitting.

Quitters enter at time_since_quit 0, ongoing quitters move up one year at a time, and the top category absorbs. Every former smoker then has the relapse probability that actually applies to them.