R package for the Evaluation Platform in COPD (EPIC), an agent-based whole-disease model for projection of health and economic outcomes and COPD interventions.
noticed that 73% of my cohort is female. Based on clinical trials for triple therapy I was expecting the opposite, about 70% male. I haven't added anything to EPIC for this analysis that stratifies by sex so I don't think it's anything I've done.
I then ran a general simulation (1 million agents over 10 years) from the master branch, and the population is about 75% female there as well (based on n_alive_by_ctime_sex). Agent sex is set in the create_agent module by this line:
(*ag).sex=rand_unif()<input.agent.p_female;
I checked the input file and p_female is 0.5.
I then thought maybe something is wrong with mortality rates, maybe a disproportionate number of males are dying, but about 70-80% of the deaths are from females.
I've looked into this further. From the export_figures() output, sex-specific incidence, prevalence, and mean FEV1 all look OK. However, females have a higher rate of total exacerbations, and consequently a higher rate of severe/very severe exacerbations and higher COPD-related mortality. This explains why I'm getting 75% females in the triple therapy cohort. I've attached the export figures file. To clarify, I've ran this on the main epicR package, not my triple therapy branch.
Looking at the R input file, the sex regression parameter for the exacerbation rate equation is zero. But, the fev1 parameter = -1.5 i.e. a higher fev1 gives a lower rate. But females have a lower mean fev1 compared to males. Is this possibly causing the issue? Looking at the original MDM paper, fev1 wasn't in the exacerbation rate equation; I know it's been updated since then but not sure at what point it got added in.
the exacerbation rate and exacerbation severity model parameters are the same as reported in the MDM paper up to v0.30.1 (2 versions ago). Then the parameters change for both equations in v0.31.0. Then the parameters change for just the rate equation in v0.32.0 (the current version). I have uploaded a document that compares the values. I've also attached the export_figures() output for v0.30.1 and v0.31.0.
The rate of all exacerbations was higher for males up until the most recent version, which is when the fev1 parameter got changed from zero. Which supports my theory that this is essentially creating a higher intercept for females. Note also that in v0.31.0, a gold1 parameter got added which isn't, statistically speaking, correct. Gold grade is a factor variable with 4 levels so gold 1 should be accounted for in the intercept.
The intercept1, intercept2, and fev1 parameters in the severity equation changed quite drastically in v0.31.0. However, the rate of severe / very severe exacerbations is higher for females across all versions, even using the values in the MDM paper. So perhaps that is just how it's supposed to be? If you think there's an error in versions further back that may have caused this,
From Rachael: