Merge pull request #1877 from ihmeuw/limitations

Document a few more MNCNH limitations

Author: zmbc

docs/source/models/concept_models/vivarium_mncnh_portfolio/maternal_disorders_module/module_document.rst

@@ -103,6 +103,7 @@ Incidence, mortality, YLDs, and YLLs due to cause-specific maternal disorders.
 * We assume that all deaths due to maternal disorders occur during the intrapartum period prior to the postpartum period although especially for causes of deaths such as "late maternal deaths," this may not be the case.
 * We assume that YLDs due to subcauses in the :ref:`Residual maternal disorders <2021_cause_residual_maternal_disorders_mncnh>` cause is spread evenly across all live/stillbirth pregnanices rather than modeling YLDs-per-case among incident cases specifically.
 * We assume that abortion/miscarriage/ectopic pregnancies may not experience any maternal disorders subcauses except for those included in the :ref:`Abortion/miscarriage/ectopic pregnancy maternal disorders <2021_cause_abortion_miscarriage_ectopic_pregnancy_causes_mncnh>` cause model
+* We assume there is no direct correlation between maternal disorders and LBWSG exposure apart from that induced through the iron interventions that affect maternal disorders via hemoglobin and also act on LBWSG exposure. However, maternal complications are likely to vary in frequency with gestational age. Obstructed labor incidence is more common among term than preterm births, while hemorrhage is more common among preterm and post-term than term births, for instance.
 
 4.0 Verification and Validation Criteria
 +++++++++++++++++++++++++++++++++++++++++

docs/source/models/intervention_models/mncnh_pregnancy/anemia_screening.rst

@@ -61,15 +61,21 @@ Ferritin is a protein that stores iron within the body and low blood ferritin le
 screen will also be screened for ferritin levels via a minimally invasive screening (finger prick). Pregnancies that have a hemoglobin level <100 g/L and a blood ferritin level below 15 ug/L 
 (anemic AND iron deficient) are eligible for IV iron.
 
-Notably, the GBD does not have any estimates related to ferritin exposure. However, the GBD assigns specific causes to all cases of anemia. Some of these causes of anemia are considered "iron 
+Notably, the GBD does not have any estimates related to ferritin exposure or ferritin screening. However, the GBD assigns specific causes to all cases of anemia. Some of these causes of anemia are considered "iron 
 responsive," indicating that they are iron deficiency anemia. An example of an iron deficiency anemia is anemia caused by maternal hemorrhage (caused by blood loss, decreasing systemic levels 
 of both hemoglobin and iron). An example of a non-iron responsive anemia is sickle cell trait (low hemoglobin is due to a defect in hemoglobin protein rather than low iron levels). Notably, it 
 is possible for non-iron-responsive anemias to also have low iron levels. See the :ref:`anemia impairment document <2019_anemia_impairment>` for a list of iron responsive and non iron responsive 
 causes of anemia in the GBD.
 
-Therefore, in our model we will use the severity-specific fraction of iron responsive anemia among all causes of anemia in GBD as a proxy measure for the fraction of anemia cases with low ferritin. 
-This approach is limited in that we may slightly underestimate total eligibility by not considering the proportion of the population who has low hemoglobin due to an iron-non-responsive cause and 
+Unlike hemoglobin screening, we do not explicitly model any test inaccuracy with e.g. a sensitivity and specificity.
+We assign simulants only a *tested* ferritin value, not a true underlying value.
+
+In our model we will use the severity-specific fraction of iron-responsive anemia among all causes of anemia
+in GBD as a proxy measure for the fraction of anemia cases (of that severity) who receive a low ferritin test result when screened. 
+This approach is limited in that we may slightly underestimate low ferritin by not considering the proportion of the population who has low hemoglobin due to an iron-non-responsive cause and 
 also coincidentally has low ferritin.
+It is also limited in that we may overstate the differences between anemia severities with respect to ferritin *test* results
+due to testing inaccuracies (false positives and false negatives).
 
 .. note::
 
@@ -189,8 +195,11 @@ Assumptions and Limitations
   in order to match the desired sensitivity and specificity of the screening test, we would need to solve for the uncertainty 
   distribution, likely via optimization, at the location-specific level (as it will depend on the underlying population 
   hemoglobin exposure distribution).
-- We use the severity-specific fraction of iron responsive anemia among all causes of anemia in GBD as a proxy measure for the fraction of anemia cases with low ferritin. This approach is limited in that we may slightly underestimate total eligibility by not considering the proportion of the population who has low hemoglobin due to an iron-non-responsive cause and also coincidentally has low ferritin.
-- In the absence of data to directly inform otherwise, we assume that the population without anemia has half the rate of low ferritin exposure as the population with mild anemia. We made this assumption given that the population without anemia is expected to have a low ferritin exposure level that is greater than zero but less than that of the population with mild anemia.
+- We use the severity-specific fraction of iron responsive anemia among all causes of anemia in GBD as a proxy measure for the fraction of anemia cases with low ferritin test results.
+  This approach is limited in that we may slightly underestimate total eligibility by not considering the proportion of the population who has low hemoglobin due to an iron-non-responsive cause and also coincidentally has low ferritin.
+  It is also limited in that we may overstate the differences between anemia severities with respect to ferritin *test* results
+  due to testing inaccuracies (false positives and false negatives).
+- In the absence of data to directly inform otherwise, we assume that the population without anemia has half the rate of low ferritin test results as the population with mild anemia. We made this assumption given that the population without anemia is expected to have a low ferritin exposure level that is greater than zero but less than that of the population with mild anemia.
 
 .. todo:: 
 

docs/source/models/intervention_models/mncnh_pregnancy/oral_iron_antenatal/oral_iron_antenatal.rst

@@ -183,6 +183,7 @@ Assumptions and Limitations
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 - We assume no effect modification by baseline hemoglobin level. In reality, the individual hemoglobin shifts are likely greater among those who are anemic at baseline.
+- We assume no effect modification by when in pregnancy oral iron is received. In reality, the effect on hemoglobin (particularly at the end of pregnancy) is likely greater for those who have taken oral iron for longer.
 - Our baseline calibration preserves the population mean value of hemoglobin concentration, but only approximates the overall exposure distribution.
 
 Verification and validation criteria
@@ -313,6 +314,8 @@ Assumptions and Limitations
 
 - We do not consider effect modification by maternal anemia status.
 
+- We assume no effect modification by when in pregnancy oral iron is received. In reality, the effect on birthweight is likely greater for those who have taken oral iron for longer.
+
 - For the :ref:`MNCNH portfolio simulation <2024_concept_model_vivarium_mncnh_portfolio>` that uses the baseline coverage value of women that took any antenatal iron: We assume that taking any iron supplement is equally as effective as taking daily a iron supplement in the baseline scenario. If it is in fact less effective, we will overestimate the impact of the baseline IFA coverage and therefore underestimate the impact of the MMS interventions.
 
 - Our baseline calibration preserves the population mean value of birthweight, but only approximates the overall exposure distribution.
@@ -376,6 +379,8 @@ Assumptions and limitations
 
 - Our calculation of the IFA and MMS gestational age shifts does not take into account the correlation between ANC attendance (and therefore the population eligible to receive these interventions) and gestational age exposure that is induced through the correlated propensities used in the :ref:`facility choice model <2024_facility_model_vivarium_mncnh_portfolio>` in the MNCNH portfolio simulation. Given that there is a positive modeled correlation between these variables, there will be slightly less preterm birth among the population eligible for these interventions than among the population overall in our simulation. Therefore, our estimates of the gestational age shifts are likely less than they would be if they were calculated with consideration to this underyling correlation. Additionally, the modeled correlation in our simulation not considered in the calculation of the shifts may cause us to not exactly replicate the observed intervention RR on preterm birth that our modeling strategy aims to replicate.
 
+- We assume no effect modification by when in pregnancy oral iron is received. In reality, the effect on gestational age is likely greater for those who have taken oral iron for longer.
+
 - Our baseline calibration preserves the population mean value of gestational age at birth, but only approximates the overall exposure distribution.
 
 - In the case of MMS, although we have improved the assumption of a single shift applied to the entire distribution through our "dual shift" strategy, it is still limited in that the true shift is likely more of a continuous function with baseline gestational age rather than two conditional values. In particular, a limitation of this approach is the illogical implication that the effect of treatment on a birth that would have been 31.9 weeks without treatment leads to a longer gestation than the effect of the same treatment on a birth that would have been 32.1 weeks without treatment.
@@ -529,6 +534,8 @@ Assumptions and limitations
 
 - There is an observed association between severe anemia during pregnancy and stillbirth, as shown in [Young-et-al-2019]_. Therefore, there should hypothetically be some effect of IFA on stillbirth given that it improves hemoglobin concentration during pregnancy; however, we do not consider this effect due to lack of evidence that shows a direct effect of IFA on stillbirth outcomes.
 
+- We assume no effect modification by when in pregnancy oral iron is received. In reality, the effect on stillbirth is likely greater for those who have taken oral iron for longer.
+
 References
 ------------