search

ICRC-Models / HHCoM

Compartmental model of HIV and HPV heterosexual transmission, development of AIDS and cervical cancer, and interventions
3 stars 1 forks source link

Background mortality trends after 1985 #86

Closed carajbro closed 4 years ago

carajbro commented 4 years ago
carajbro commented 4 years ago

Darcy: Downloaded data from IHME using the GHDx results tool, which allowed me to output the number and rate of deaths by sex, age, group, and year 1990-2017 for all causes and HIV-specific causes. I then subtracted HIV deaths from all cause deaths to get the background mortality. I plotted the resulting data alongside the estimates currently used in the model for 1950 and 1985.

Two things stand out from this: 1) the mortality rate for the under 5 group has some strong discontinuities - the slopes from 1950-1985 and 1990-2017 are pretty disjointed. Here I joined them with a linear slope, but it is a bit strange. 2) There is a big dip in mortality rates for ages ~50+ from the late 1980s to the late 1990s. It’s strange that it comes back up again. This could be a reflection of more people reaching these ages and experiencing chronic diseases?

I think from this we have three options: 1) stick with what we’ve done previously, and assume that mortality rates remain at 1985 levels for the remainder of the model. But I don’t think that will get us the population distribution we want, since it looks like death rates are lower now. 2) Use these updated data and fit linear trends to them, perhaps with four joinpoints - 1950, 1985, and 1988, and 2017. So we would assume that mortality declines linearly from 1985 to 1998 and then decreases to 2017 levels. This would ignore the trough in the 1990s and smooth over it. It would also make the slope of the decline for under 5 a little less severe. 3) Use these updated data but try to fit the dip in the 1990s.

carajbro commented 4 years ago

Going with Option #2

Linear changes between observed data from 1950, 1985, 2000, and 2017. 1950 and 1985 being the same as what is currently in the model. Extrapolating trend between 2000 and 2017 to 2020.