|
|
|
|
|
|
| Re: Maternal Mortality [message #19765 is a reply to message #19764] |
Wed, 12 August 2020 11:03   |
Trevor-DHS
Messages: 803
Registered: January 2013
|
Senior Member |
|
|
Hi Placid
First I should note that DHS no longer calls this the maternal mortality ratio, but rather the pregnancy-related mortality ratio. In newer surveys we calculate both the pregnancy-related mortality ratio, and an updated calculation of the maternal mortality ratio. Please see the Guide to DHS Statistics chapter 16 for more information about the calculation of both the pregnancy-related mortality ratio and the maternal mortality ratio. In particular, the calculation of the numerator differs between the two:
Numerators:
1) Pregnancy-related mortality: Number of female siblings of respondents who died during pregnancy, delivery or within two months of delivery (mm1 = 2 & mm2 = 0 & mm9 in 2:6) in the period 0-6 years prior to the interview by five-year age group at time of death (this is what you have currently)
2) Maternal mortality: Number of female siblings of respondents who died during pregnancy, delivery or within 42 days of delivery (mm1 = 2 & mm2 = 0 & mm9 in 2:6 & mm12 in 100:141,198,199), excluding due to accidents or violence (mm16 ≠ 1 & mm16 ≠ 2), in the period 0-6 years prior to the interview by five-year age group at time of death.
Older surveys do not include variables mm12 and mm16 and it is only possible to calculate the pregnancy-related mortality ratio.
For more information on the difference between pregnancy-related mortality and maternal mortality, see the Youtube video
To answer your questions, you get the unweighted numbers by using the same commands, but without the sample weight (wt), as follows:
* Deaths from table 15.4 - restrict to pregnancy-related deaths
tab agegrp [iw=deaths] if mm1 == 2 & mm9 >= 2 & mm9 <= 6
* Exposure from table 15.4 (expo is in months, division by 12 to give years) - restrict to women only
tab agegrp [iw=expo/12] if mm1 == 2
For the pregnancy-related (or the maternal) mortality ratio, this is presented per 100,000 live births, so in the example here, the 357 means 357 women per 100,000 live births. So actually a relatively small (but important) number.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Re: MMR & PRMRatio (pregnancy related mortality ratio) [message #20458 is a reply to message #20455] |
Thu, 05 November 2020 10:48 |
Trevor-DHS
Messages: 803
Registered: January 2013
|
Senior Member |
|
|
Hi Placid
AM_gfr.do appears to run ok because all it is doing is defining up the 'program's that are being called by AM_rates.do. When AM_rates.do is running it calls the programs that are defined in AM_gfr.do, and it is n the code defined in AM_gfr.do that the error appears. The message about using the old merge syntax is not important - it is using older syntax, but that part still works properly, as you can see by the tab _merge output that follows. We need to find the actual command that it is failing on, which is buried in the code.
Line 1346 is main_adult_mm_mortality, which calls several other programs, and it is in these programs, many of which are in AM_gfr.do that the problem occurs. For example, you can see in your output that temp1.dta is saved. This happens in line 330 of AM_gfr.do. After this the programs setup_GFR, make_exposure_and_births, and calc_rates are run. Setup_gfr is fairly simple and seems to have run. I think the issue is in the make_exposure_and_births program. Can you try commenting out the recast command on line 202? I think it is possible that the recast command does not exist in your version of Stata, and it also looks like it is not needed. Please send me your output after trying this. Good luck.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Re: MMR & PRMRatio (pregnancy related mortality ratio) [message #20478 is a reply to message #20471] |
Mon, 09 November 2020 07:41 |
Bridgette-DHS
Messages: 3199
Registered: February 2013
|
Senior Member |
|
|
Following is a response from DHS Research & Data Analysis Director, Tom Pullum:
The "micro.dta" file has one record for each sibling in the sibling histories. It's analogous to the BR file, which has one record for each child in the birth histories. It's built into the program mainly for data quality checking and simple tabulations. If you want to use it for data analysis, you need to take into account that there tends to be some rounding (and probably displacement) in ages of siblings, years since death, and age at death (for those who died). There may be some omission, especially for siblings who were much older than the respondent and died when young. And characteristics of the respondent, such as level of education and use of health services, cannot safely be attributed to the sibling. These are the reasons why DHS doesn't recommend estimating adult and maternal mortality for earlier time periods and subpopulations. Even so, you may find the file useful. You can add the respondent's IR variables onto micro.dta using v001 v002 v003 and a normal merge command.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Re: Maternal Mortality [message #25164 is a reply to message #25163] |
Thu, 08 September 2022 11:21 |
Bridgette-DHS
Messages: 3199
Registered: February 2013
|
Senior Member |
|
|
Following is a response from DHS staff member, Tom Pullum:
All of the demographic rates in DHS reports and on STATcompiler--fertility, under-5 mortality, adult and maternal mortality--are calculated for aggregates of women, or children, or siblings, and for exposure + events that occur in an interval of time, such as the last 3 years or the last 5 years or the last 7 years. There are some differences but basically they are based on a pooling of exposure or risk, and events, for an aggregate. The calculations can be repeated within aggregates or subpopulations such as the wealth quintiles, but cannot control for other potential covariates unless you go down to smaller and smaller subpopulations.
The GitHub programs to calculate the rates do include logit or poisson regressions with no covariates but basically as a shortcut to calculate proportions or means and to get standard errors.
If you want to do multivariate (multivariable?) analysis with controls, interactions, etc., you need to shift to a statistical framework that gives results in terms of coefficients rather than rates. You can do logit regressions that adjust for exposure, e.g. with hazard models, or poisson regressions, etc., depending on the outcome. Bridging demographic and statistical approaches is a methodological challenge.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
