is each sibling's death pregnancy-related, by DHS coding? — is_preg_related

The DHS records this in one coded item, sib.died.pregnant (mm9):

Usage

is_preg_related_dhs(sib_df, na.action = NULL, prmr.accident.recode = FALSE)

Arguments

sib_df: the prepped sibling dataset
na.action: retained for symmetry with is_maternal_dhs and ignored here. It used to decide how a missing mm12 was treated; mm12 is no longer consulted, so it has no effect on this column
prmr.accident.recode: apply the documented-but-unexecuted 2016 PRMR rule, which drops a death during pregnancy (mm9 = 2) that is reported as violence or an accident. Default FALSE, matching the reference implementation. See Details

Value

a logical vector, one entry per row of sib_df

Details

`mm9`	meaning	pregnancy-related	maternal
2	died while pregnant	yes	yes
3	died during delivery	yes	yes
4	since delivery – never assigned in practice	yes	yes
5	within six weeks of a delivery	yes	yes
6	between six weeks and two months of a delivery	yes	no

Code 6 is what makes this the two-month quantity rather than a 42-day one, and it is exactly the line The DHS Program draws between this and is_maternal_dhs, which stops at code 5.

No other condition is applied. In particular the time-since-delivery band mm12 is not used: the reference implementation (DHS-Indicators-Stata, Chap16_AM/AM_rates.do:725) counts mm9 >= 2 & mm9 <= 6 and nothing else, and its header states plainly that "mm12 is not needed" – it is dropped before the roster is reshaped.

Changed in this version

This function previously required mm9 to be 2, 3, 4 or 5 and mm12 to fall in the band 100–141 (0–41 days) or be 997/998. Both conditions were wrong:

excluding code 6 dropped postpartum deaths that the function's own documentation described as in scope, and
the mm12 band imposed a 42-day cut on a quantity defined over two months, and applied a postpartum timing test even to deaths that occurred during pregnancy or delivery.

This changes DHS results, in some surveys substantially. Which code a survey uses for postpartum deaths is a property of its questionnaire, so the old behaviour lost anywhere from none to about 48% of pregnancy-related deaths depending on the survey. On Rwanda 2010 it gave 51.2 deaths against a published 91; it now gives 90.7. See the DHS validation plan in the package repository.

The 2016 PRMR redefinition

AM_rates.do:307-320 carries, inside a comment block and therefore never executed, a rule introduced as "Important for redefinition of Pregnancy Related Mortality Ratio (PRMR) in surveys from 2016 onwards":

If mm9=2, and mm16=1 or 2, recode mm9 to 1

That is: a death during pregnancy that is reported as due to violence or an accident stops counting as pregnancy-related. The DHS Program documents the rule (see blog.dhsprogram.com/mmr-prmr/) but the code it ships does not apply it, so published figures produced with that code do not reflect it either.

prmr.accident.recode = TRUE applies it. The default is FALSE, which is what the shipped reference code does and therefore what reproduces published tables. It can only bite on the five surveys that carry mm16 at all.