Skip to content

Explain the assumptions of our ANC attendance model in more detail #1878

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
zmbc merged 4 commits into from
Dec 29, 2025
Merged

Explain the assumptions of our ANC attendance model in more detail #1878

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
948f01a
Explain the assumptions of our ANC attendance model in more detail
zmbc Dec 24, 2025
dc2de51
Be more specific about "a few" and "many"
zmbc Dec 29, 2025
5b30073
...
zmbc Dec 29, 2025
0b40779
Merge branch 'main' into anc_attendance
zmbc Dec 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
208 changes: 115 additions & 93 deletions ...ncept_models/vivarium_mncnh_portfolio/antenatal_care_module/module_document.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,7 +76,76 @@ and :ref:`our facility choice module documentation <2024_vivarium_mncnh_portfoli
- ANC propensity is correlated with LBWSG category propensity and IFD propensity as described in the the :ref:`correlated propensities <facility_choice_correlated_propensities_section>`
section of the facility choice model document. Currently we assume that there is no correlation of ANC with other factors.



2.2 Module Outputs
------------------

ANC attendance impacts hemoglobin exposure and facility choice in our model, and in order for the
outputs of this component to be compatible with the data needs of these two downstream components, we will need two different
outputs, one being polychotomous for the facility choice component and the other being dichotomous for the hemoglobin component.

.. list-table:: Module outputs
:header-rows: 1
:widths: 10 15 15

* - Output
- Value
- Dependencies
* - ANC attendance category
-
1. :code:`none`
2. :code:`later_pregnancy_only`
3. :code:`first_trimester_only`
4. :code:`first_trimester_and_later_pregnancy`

The categories of this polytomous variable are listed from highest risk (1) to lowest risk (4) in terms of ultrasound timing,
in accordance with the :ref:`special ordering of the categories section <facility_choice_special_ordering_of_categories_section>`
of the delivery facility choice model document: The categories need to be ordered 1 < 2 < 3 < 4 when sampling the ANC attendance
variable using the correlated ANC propensity in order to induce the correct correlations for the facility choice model.
- Used as an input for the :ref:`AI Ultrasound module <2024_vivarium_mncnh_portfolio_ai_ultrasound_module>`, for determining receipt and timing of ultrasound.
* - First trimester ANC attendance
-
- *True* when ANC attendance category is :code:`first_trimester_only` or :code:`first_trimester_and_later_pregnancy`
- *False* when ANC attendance category is :code:`none` or :code:`later_pregnancy_only`
- Used as an input for the :ref:`hemoglobin module <2024_vivarium_mncnh_portfolio_hemoglobin_module>`, for eligibility for receipt of oral iron in the first trimester.
This variable is dichotomous for each pregnancy.
* - Late pregnancy ANC attendance
-
- *True* when ANC attendance category is :code:`first_trimester_and_later_pregnancy` or :code:`later_pregnancy_only`
- *False* when ANC attendance category is :code:`none` or :code:`first_trimester_only`
- Used as an input for the :ref:`hemoglobin module <2024_vivarium_mncnh_portfolio_hemoglobin_module>`, for eligibility for anemia screening, as well as receipt of oral or IV iron in later pregnancy.
This variable is dichotomous for each pregnancy.

The categorical variable is simply a polytomous encoding of the 2x2 table implied by the dichotomous variables:

.. list-table:: ANC exposure options
:header-rows: 1

* -
- **Late pregnancy ANC attendance** = **True**
- **Late pregnancy ANC attendance** = **False**
* - **First trimester ANC attendance** = **True**
- :code:`first_trimester_and_later_pregnancy`
- :code:`first_trimester_only`
* - **First trimester ANC attendance** = **False**
- :code:`later_pregnancy_only`
- :code:`none`

2.3 Module Description
----------------------

Data strategy
~~~~~~~~~~~~~

Data to inform the exact breakdown of our categories are not available uniformly
across our modeled locations.
For example, the DHS survey only asks the date of the first visit and the number of total visits,
but doesn't ask the date of each visit, so we don't specifically know if a respondent went to ANC
in later pregnancy (they could have gone to multiple visits, all in the first trimester).

Rather than using DHS data directly, we use GBD covariates and other IHME modeling based on the DHS data.
The data we have are as follows:

.. list-table:: Additional input data
:header-rows: 1

Expand All @@ -94,141 +163,94 @@ and :ref:`our facility choice module documentation <2024_vivarium_mncnh_portfoli
- GBD covariate ID 8: :code:`get_covariate_estimates(location_id=location_id, release_id=16, year_id=2023, covariate_id=8)`
- Proportion of pregnant people receiving 4 or more antenatal care visits including 1 or more from a skilled provider

The use of ANC1 is straightforward: it determines how many people *never* attend ANC, vs attend ANC at all.
Similarly, ANCfirst determines how many people should attend in the first trimester.
Lastly, we use ANC4 as a proxy measure for attending ANC *throughout pregnancy*, though this could miss in both directions:
someone could attend only 2-3 visits and have these spread across time (more likely),
or they could attend 4+ visits all during one phase of pregnancy (less likely).
Due to the imperfectness of this proxy measure, we decided to treat it as an upper bound on attending throughout pregnancy,
which means it is only impactful when it is *lower* than ANCfirst (see next section for details).
In practice, in our modeled locations, this only occurs in Pakistan.

2.2 Module Description
----------------------
In the absence of better data, we assume that the DHS data used to produce the ANCfirst, ANC1, and ANC4 covariates applies to abortion/miscarriage/ectopic pregnancies
as well as pregnancies resulting in live birth or stillbirth.
We further assume that abortion/miscarriage/ectopic pregnancies end before they can attend later pregnancy ANC visits (though in fact abortions and miscarriages in particular could happen beyond this point).

ANC attendance will be modeled as a single variable with 4 possible exposure options:
.. note::

A. Attends ANC during first trimester AND later pregnancy
B. Attends ANC during first trimester but NOT later pregnancy
C. Attends ANC during later pregnancy but NOT first trimester
D. Does not attend ANC at all during pregnancy
As of `pull request #1690 <https://github.com/ihmeuw/vivarium_research/pull/1690>`_ we updated our strategy to
include the ANCfirst variable that the HS team processed and shared with us. Please see
`this JIRA ticket <https://jira.ihme.washington.edu/browse/SSCI-2474>`__
for more information on this strategy update and other options considered.

.. list-table:: ANC exposure options
:header-rows: 1
Vivarium modeling strategy
~~~~~~~~~~~~~~~~~~~~~~~~~~

* -
- Visit during late pregnancy
- No visit during late pregnancy
* - **Visit during first trimester**
- A
- B
* - **No visit during first trimester**
- C
- D
With the data and assumptions in the previous section, we can fully determine the proportions in the 4 categories of ANC attendance.
Here is the procedure for **live birth or stillbirth** pregnancies:

The below table describes what probability values to use for each exposure option outlined above, **for pregnancies resulting in live birth or stillbirth**.
1. First, assign probability of 1 - ANC1 to the :code:`none` category.
2. Then, assign probability to :code:`later_pregnancy_only` so that it and :code:`none`, taken together, equal 1 - ANCfirst.
(You'll never run out of probability on this step, since ANCfirst must be less than or equal to ANC1).
3. Then, assign probability to :code:`first_trimester_and_later_pregnancy` until
you have either assigned all remaining probability, or the probability in this category has become equal to ANC4 (which we treat as an upper bound).
4. Assign any remaining probability to :code:`first_trimester_only`. Note that in our actual data, this category only receives
non-zero probability when ANC4 > ANCfirst, which only occurs in Pakistan out of our modeled locations.

Mathematically, that means the probabilities are as follows, **for pregnancies resulting in live birth or stillbirth**.

.. list-table:: ANC exposure probability values for pregnancies resulting in live birth or stillbirth
:header-rows: 1

* - ANC exposure option
- Description
- Probability value
- Notes
* - A
- Attends ANC during first trimester AND later pregnancy
* - :code:`first_trimester_and_later_pregnancy`
- ``min(ANCfirst, ANC4)``
- Assume that attending ANC in first trimester reflects "active care seeking behavior" and that it is unlikely
for someone who attends first trimester ANC to attend no subsequent visits.
* - B
- Attends ANC during first trimester but NOT later pregnancy
* - :code:`first_trimester_only`
- ``ANCfirst - min(ANCfirst, ANC4)``
- Prevalence of first trimester visit ONLY (and no late pregnancy visit) only occurs if ANCfirst > ANC4
* - C
- Attends ANC during later pregnancy but NOT first trimester
* - :code:`later_pregnancy_only`
- ``ANC1 - ANCfirst``
-
* - D
- Does not attend ANC at all during pregnancy
* - :code:`none`
- ``1 - ANC1``
-

.. note::

As of `pull request #1690 <https://github.com/ihmeuw/vivarium_research/pull/1690>`_ we updated how we assign our ANC exposures to
include the ANCfirst variable that the HS team processed and shared with us. Please see `these slides <https://uwnetid.sharepoint.com/:p:/r/sites/ihme_simulation_science_team/_layouts/15/Doc.aspx?sourcedoc=%7BADD6223E-9FCA-40BB-BB7F-FE44F377CCDB%7D&file=ANC%20visit%20timing%20data%20strategy%20options.pptx&action=edit&mobileredirect=true>`_
for more information on this strategy update.

The above probabilities are to be implemented pregnancies resulting in live birth or stillbirth only.
Abortion/miscarriage/ectopic pregnancies are assigned
probabilities differently because we assume their pregnancies end before they can attend later pregnancy ANC visits.
The below table describes what probabilities to use for each exposure option **for abortion/miscarriage/ectopic pregnancies**:
Abortion/miscarriage/ectopic pregnancies are similar, except that we make it impossible for them to attend later
pregnancy ANC visits, resulting in the following (simpler) probabilities:

.. list-table:: ANC exposure probabilities for abortion/miscarriage/ectopic pregnancies
:header-rows: 1

* - ANC exposure option
- Description
- Probability value
- Notes
* - A
- Attends ANC during first trimester AND later pregnancy
* - :code:`first_trimester_and_later_pregnancy`
- 0
- Assumption
* - B
- Attends ANC during first trimester but NOT later pregnancy
* - :code:`first_trimester_only`
- ``ANCfirst``
-
* - C
- Attends ANC during later pregnancy but NOT first trimester
* - :code:`later_pregnancy_only`
- 0
- Assumption
* - D
- Does not attend ANC at all during pregnancy
- Probability equal to ``1 – ANCfirst``
-


2.3 Module Outputs
------------------

As mentioned earlier, ANC attendance impacts hemoglobin exposure and facility choice in our model, and in order for the
outputs of this component to be compatible with the data needs of these two downstream components, we will need two different
outputs, one being dichotomous for the hemoglobin component and the other being polychotomous for the facility choice component.

.. list-table:: Module outputs
:header-rows: 1
:widths: 10 15 15

* - Output
- Value
- Dependencies
* - First trimester ANC attendance
* - :code:`none`
- ``1 – ANCfirst``
-
- *True* for groups A and B
- *False* for groups C and D
- Used as an input for the :ref:`hemoglobin module <2024_vivarium_mncnh_portfolio_hemoglobin_module>`.
This variable is dichotomous for each pregnancy.
* - Late pregnancy ANC attendance
-
- *True* for groups A and C
- *False* for groups B and D
- Used as an input for the :ref:`hemoglobin module <2024_vivarium_mncnh_portfolio_hemoglobin_module>`.
This variable is dichotomous for each pregnancy.
* - ANC attendance category
-
1. :code:`none` for group D
2. :code:`later_pregnancy_only` for group C
3. :code:`first_trimester_only` for group B
4. :code:`first_trimester_and_later_pregnancy` for group A

The categories of this polytomous variable are listed from highest risk (1) to lowest risk (4) in terms of ultrasound timing,
in accordance with the :ref:`special ordering of the categories section <facility_choice_special_ordering_of_categories_section>`
of the delivery facility choice model document: The categories need to be ordered D < C < B < A when sampling the ANC attendance
variable using the correlated ANC propensity in order to induce the correct correlations for the facility choice model.
- Used as an input for the :ref:`AI Ultrasound module <2024_vivarium_mncnh_portfolio_ai_ultrasound_module>`.


3.0 Assumptions and limitations
++++++++++++++++++++++++++++++++

* We assume that the prevalence of attending both first trimester and later pregnancy visits is the minimum of ANCfirst (as processed by the HS team) and ANC4
(GBD covariate also processed by HS team). There is non-zero prevalence of first trimester visits only when ANC4 > ANC1 (such as in Pakistan). We are likely
overestimating the correlation between first trimester ANC and later pregnancy ANC (i.e., the prevalence of a first trimester ANC visit ONLY is likely non-zero
despite this assertion in our model.) We assume that the DHS data used to produce the ANCfirst, ANC1, and ANC4 covariates applies to abortion/miscarriage/ectopic pregnancies
* We use ANC4 as an upper bound on attending ANC throughout pregnancy in Pakistan, though strictly speaking this isn't logically necessary,
but represents our assumption that this proxy measure is an underestimate.
This assumption could be wrong if many people attend 4+ visits all during one phase of pregnancy.
* We assume that the DHS data used to produce the ANCfirst, ANC1, and ANC4 covariates applies to abortion/miscarriage/ectopic pregnancies
as well as pregnancies resulting in live birth or stillbirth.
* We assume that abortion/miscarriage/ectopic pregnancies can only attend ANC in the first trimester (though in fact abortions and miscarriages in particular can happen beyond this point).

.. todo::

Expand Down