corinne-riddell
commented
8 years ago @sbh4th In support of reproducibility we'll need to report exactly how we arrived on our cut of data, and specifically what causes of death are included in each category. Can you describe here how the cause of death categories were created (and in particular, what is captured by the non-communicable category)?
From above, we didn't really close this discussion, but from my preliminary look at the data it appears that the non-communicable category is an important one. For example you can see that in 1969 in Alabama the % of death due to non-communicable disease is low (first figure) but in 2013, there is a much higher % of non-communicable disease, especially in the older age brackets (second figure):
This issue covers our email discussion (July 22, 2016) on choosing how to aggregate causes of death into categories. In future these discussion should all happen in real-time on GitHub, so that we have a log of the discussion that we can easily search and view when writing up the manuscript. I'm just going to copy and paste the discussion from email below as a record:
SH: Another quick clarification on causes of death. After defining: 1) CVD, 2) Cancers, 3) Infections, 4) Injuries, what SEER*Stat has left are the following groups (plus a residual):
The question is: should we collapse all of these residual causes into a single “all other causes” group, or is there some rationale for, say, a separate category of “chronic” diseases that would include Diabetes, COPD, Alzheimer’s, ulcers, and liver/kidney disease and the remainder as all other causes.
We are already including hopelessly different diseases in the other broad categories, so another category of “other non-communicable diseases” is just as arbitrary, but another category of disease will also mean many more cells/missing.
Not sure what you guys think.
JK: You can die of a stomach ulcer? OK, I don’t think that any of these warrant an additional category. I’d just dump them all into a residual (which for adults will be mostly COPD, Alzheimer’s and Kidney disease)
JK: Actually, a lot of the liver and kidney disease will be infectious (e.g. hepatitis) or injury (drug overdose that doesn’t kill you will knock out your liver, etc). Not to mention that at least half of the “Alzheimer’s” deaths are actually vascular in origin. I would just leave it in a big residual.
SH: Here’s the breakdown by cause, excluding those values that are suppressed:
So that residual (non-communicable) category that includes liver/kidney dx, diabetes, etc. is actually bigger than the communicable or injury category, and would push up the “residual” to >20% of our deaths, which is a little unsatisfying but I get that it is an odd grouping.
JK: But having the residual category be the 3rd largest cause of death seems wrong. Can we combine diabetes with CVD to make a “CVD and metabolic disease” category? What else is big in that category? Since you have categories for communicable and non-communicable, I don’t understand what is left!
SH: Those are just category names, but we could probably be more specific.
Here are all the deaths from 1999-2014 by ICD chapter (from CDC Wonder):
You can see there are a considerable number that fall under “Mental/behavioral” and “Diseases of the nervous system” that I don’t think the SEER coding scheme captures, plus the R codes for “not elsewhere classified” is also not small. Basically, I think these are ill-defined mental and nervous system disorder, plus all of the other weird things that are, apparently, hard to classify.
I don’t know where this leaves us, but likely no matter how we do this there is going to be a healthy “residual/all other causes” component, I’m afraid.