Maternal disorders: GBD 2023, MNCNH

Note

This page is adapted from the previously developed GBD 2021 maternal disorders cause model, which was updated from the GBD 2019 maternal disorders cause model. In contrast to previous versions of the model, in the MNCNH Portfolio project we are modeling several of the maternal disorders subcauses (see Modeled Subcauses) rather than modeling a single overarching cause. This page documents the strategy for capturing the burden of the remaining maternal disorders subcauses that are not explicitly modeled, as well as the strategy for dealing with interactions between the different subcauses.

This page was originally written to be compatible with GBD 2021 and was updated to reflect the maternal disorders cause hierarchy for GBD 2023 in October of 2025. This was the only change relevant for this model between GBD rounds.

Modeled Subcauses 

The following maternal disorders subcauses will be modeled individually, in the indicated model-building wave:

Todo

The intrapartum component (which contains this model) says in the concept model that abortion/miscarriage/ectopic pregnancies skip it altogether. That contradicts this page, in which we apply abortion/miscarriage/ectopic pregnancy maternal disorders. This should be resolved by moving that cause model to a module in the pregnancy component.

Note that the postpartum depression cause model is outside of the GBD maternal disorders hierarchy (implemented using custom rather than GBD data) and also differs from the remaining maternal disorders causes in that only simulants who do not die due to any maternal disorder cause are eligible to become “infected” with postpartum depression.

Additional maternal disorders causes to be included in a future version of the model:

Maternal hypertensive disorders

The remainder of this document describes maternal disorders overall, describes the strategy for capturing the burden of the maternal disorders subcauses that are not explicitly modeled above, and explains how to incorporate multiple modeled subcauses into the same simulation.

Disease Overview 

GBD 2023 Modeling Strategy 

Cause Hierarchy 

All causes (c_294) [level 0]
- Communicable, maternal, neonatal, and nutritional diseases (c_295)
  - Maternal and neonatal disorders (c_962)
    - Maternal disorders (c_366) [level 3]
      - Maternal hemorrhage (c_367)
        
        Maternal hemorrhage with less than 1 liter blood loss (s_180)
        
        Maternal hemorrhage with greater than 1 liter blood loss (s_181)
        
        Mild anemia due to maternal hemorrhage (s_182)
        
        Moderate anemia due to maternal hemorrhage (s_183)
        
        Severe anemia due to maternal hemorrhage (s_184)
      - Maternal sepsis and other maternal infections (c_368)
        
        Puerperal sepsis without anemia (s_937)
        
        Other maternal infections (s_938)
        
        Infertility due to puerperal sepsis (s_675)
        
        Puerperal sepsis with mild anemia (s_23488)
        
        Puerperal sepsis with moderate anemia (s_23489)
        
        Puerperal sepsis with severe anemia (s_23490)
      - Maternal hypertensive disorders (c_369)
        
        Severe pre-eclampsia (s_185)
        
        Eclampsia (s_186)
        
        Other hypertensive disorders of pregnancy (s_676)
        
        Long term sequelae of severe pre-eclampsia (s_187)
        
        Long term sequelae of eclampsia (s_677)
      - Maternal obstructed labor and uterine rupture (c_370)
        
        Obstructed labor, acute event (s_188)
        
        Rectovaginal fistula (s_189)
        
        Vesicovaginal fistula (s_190)
      - Gestational diabetes (c_1118)
        
        Gestational diabetes mellitus (s_23468)
      - Peripartum cardiomyopathy (c_1119)
        
        Mild heart failure due to peripartum cardiomyopathy (s_23464)
        
        Moderate heart failure due to peripartum cardiomyopathy (s_23465)
        
        Severe heart failure due to peripartum cardiomyopathy (s_23466)
        
        Controlled, medicalled managed heart failure due to peripartum cardiomyopathy (s_23467)
      - Ectopic pregnancy (c_374)
        
        Ectopic pregnancy (s_5165)
      - Maternal abortion and miscarriage (c_995)
        
        Maternal abortive outcome (s_191)
      - Other direct maternal disorders, internal (c_379)
        
        Other maternal disorders (s_192)
      - Indirect maternal deaths (c_375)
      - Late maternal deaths (c_376)
      - Maternal deaths aggravated by HIV/AIDs (c_741)

Note

Anemia-specific sepsis sequelae are new to GBD 2023.

Additionally, peripartum cardiomyopathy (c_1119) and gestational diabetes (c_1118) are new causes to GBD 2023. However, they are only present in the computation hierarchy (cause_set_id==2) and not the reporting hierarchy (cause_set_id==3). Cause ID #1160 (other direct maternal disorders, inclusive of gestational diabetes and peripartum cardiomyopathy) is the cause used in the reporting hierarchy that is analogous cause to cause ID #379 that is used in the computation hiearchy (and excludes burden due to peripartum cardiomyopathy and gestational diabetes).

Subcause case definitions 

Maternal disorders are direct obstetric complications of pregnancy, childbirth, and the postpartum period:

Abortion is defined as elective or medically indicated termination of pregnancy at any gestational age, regardless of symptoms or complications (abortion), or spontaneous loss of pregnancy before 24 weeks of gestation (miscarriage) with complications requiring medical care. Miscarriages that do not require medical care are not included in this cause.
Ectopic pregnancy is defined as a pregnancy that implants outside of the uterine cavity, generally presenting with abdominal pain and vaginal bleeding.
Obstructed labour and uterine rupture.
1. Acute event includes failure to progress (no advance of the presenting part of the fetus despite strong uterine contractions), cephalopelvic disproportion (fetal size that is too large for maternal pelvic dimensions), non-vertex fetal positioning during labour (any fetal position besides head down during labour; excludes non-vertex positioning during antepartum period), and uterine rupture during labour (non-surgical breakdown of uterine wall during labour and delivery). Perineal lacerations without any of the above conditions are excluded from the case definition. (Estimation of the incidence and short- term disability due to these conditions is described in this appendix section.)
2. Obstetric fistula is defined as an abnormal opening between the vagina and the bladder or rectum with involuntary escape of urine, flatus, and/or faeces following childbirth. (The non-fatal burden of fistulas is included in the non-fatal burden of obstructed labour in reporting, but estimation is described in a separate appendix section on “Fistula – impairment.”)
Maternal haemorrhage includes heavier than expected postpartum bleeding (>500 ml following vaginal delivery or >1000 ml after cesarean delivery) or antepartum bleeding (vaginal bleeding from any cause at or beyond 20 weeks of gestation and prior to onset of labour), or placental disorders with haemorrhage regardless of blood volume lost or timing of bleeding event. Placental disorders without haemorrhage are included with other [direct] maternal disorders.
Maternal sepsis and other maternal infections encompasses maternal sepsis during or following labour and delivery (defined as temperature <36°C or >38°C and clinical signs of shock [systolic blood pressure <90 mmHg and tachycardia >120 bpm]) as well as other maternal infections believed to have a close epidemiological relationship with pregnancy, including genitourinary tract infections (excluding sexually transmitted diseases), obstetrical wound infections, and breast infections related to childbirth and lactation.
Hypertensive disorders of pregnancy includes several subcategories:
1. Gestational hypertension is the new onset of hypertension in a pregnant person after 20 weeks’ gestation, as defined by having a blood pressure measured >140/90 mmHg on more than one occasion.
2. Preeclampsia is defined by hypertension [>140/90 mmHg] and proteinuria [≥0.3 g/l], with or without signs of end-organ damage.
3. Severe preeclampsia is defined by preeclampsia with severe hypertension [>160/100 mmHg] or other signs of end organ damage [liver: low platelets, elevated liver enzymes, coagulation issues; kidney: elevated creatinine; central nervous system: headaches or visual disturbances], syndrome of hypertension, elevated liver enzymes, low platelets [HELLP syndrome]).
4. Eclampsia is defined as hypertension and seizures, with or without proteinuria. This definition excludes chronic hypertension in a pregnant person (hypertension present prior to 20 weeks’ gestation) unless superimposed preeclampsia develops.
Other [direct] maternal disorders includes a variety of different obstetric complications. The most common of these in ICD-10-coded vital registration sources in terms of number of deaths include O88 (obstetric embolism), O26 (maternal care for other conditions predominantly related to pregnancy), O90 (complications of the puerperium, not elsewhere classified), O75 (other complications of labour and delivery, not elsewhere classified), C58 (malignant neoplasm of placenta), and O36 (maternal care for other fetal problems).

Restrictions 

Vivarium Modeling Strategy 

Scope 

Assumptions and Limitations 

For each of our maternal disorders subcauses, we have taken a strategy of calculating the amount of YLDs per incident case and assigning that number of YLDs to each incident case regardless of whether or not that simulant dies in the simulation (see the individual subcause model documents for specific instructions on how to model YLDs). In reality, while simulants who die due to a maternal disorder will accumulate YLDs that occur during pregnancy and acute YLDs during labor and the immediate postpartum period, they will not accumulate YLDs associated with long-term sequelae of maternal disorders subcauses. By not accounting for the difference in YLDs between incident cases of maternal disorders subcauses among simulants who survive and those who do not, we will not capture the phenomenon of averting maternal deaths (averting YLLs) in an intervention scenario leading to increases in maternal disorders YLDs. This will cause us to slightly overestimate the impact of interventions on total DALYs averted due to maternal disorders for interventions that reduce maternal disorders mortality.

Cause Model Diagram 

Data Tables 

Calculating Burden 

Years of life lost 

Years lived with disability 

Modeling multiple maternal disorders together 

Since the MNCNH Portfolio simulation uses Vivarium timesteps in a nonstandard way, we need to do more work to specify how different simulation components interact and in what order decisions should be made. This has two implications for modeling multiple maternal disorders subcauses together:

We need to decide in which order to make decisions about the different maternal disorders subcauses, which will be important if there are causal interactions between them.
If a simulant experiences multiple maternal disorders simultaneously, we need to specify how to determine whether the simulant dies from one of the subcauses, and from which one.

To deal with the above two issues, we will

Split incidence and mortality into separate timesteps
Have a separate incidence timestep for each of the modeled maternal disorders subcauses
Have a single timestep that handles mortality from all the maternal disorders subcauses together

More details are in the following two subsections.

Subcause ordering 

We anticipate that there are correlations and perhaps causal relationships between various maternal disorders subcauses. For now, we are ignoring such interactions and treating the different subcauses as independent. However, to be able to handle such interactions in future waves, the simulation should make decisions about incidence of the different subcauses in the order of the suspected causal relationships. The specified order is:

Abortion/miscarriage/ectopic pregnancy maternal disorders
Maternal hypertensive disorders
Obstructed labor and uterine rupture
Maternal hemorrhage
Maternal sepsis and other maternal infections
Residual maternal disorders
Postpartum depression

The current plan is to have a separate “incidence timestep” for each of the modeled subcauses, ordered as above, and the simulation will decide which simulants experience each subcause on the corresponding timestep.

Note that the residual maternal disorders cause includes a subcause of “late maternal deaths.” Therefore, it is possible in reality that an individual could experience postpartum depression prior to dying of a late maternal death. However, we do not allow for that possibility in our model.

Mortality component 

We will have a single simulation timestep that handles mortality from all the maternal disorders subcauses together. The mortality timestep should happen after the incidence timesteps of all the maternal disorders subcauses. The mortality timestep will work similarly to the mortality component in a standard Vivarium simulation.

On the mortality timestep, first we will determine whether the simulant dies of any of the maternal disorders subcauses. Then, if the simulant dies, we will decide which maternal disorder caused the death. Suppose after the incidence timesteps for all the maternal disorders subcauses, a simulant simultaneously has cases of $k$ different subcauses, say $c_1, \dotsc, c_k$. We will assume that the probability that the simulant dies of one of these subcauses is

(1)\[P(\text{simulant dies of one of $c_1,\dotsc, c_k$} \mid \text{simulant has $c_1,\dotsc, c_k$ only}) = \sum_{i=1}^k \operatorname{cfr}_i,\]

where $\operatorname{cfr}_i$ is the case fatality risk (a.k.a. case fatality rate) of the $i^\mathrm{th}$ subcause, $c_i$. If the simulant dies, we then specify that the probability that they die of $c_i$ should be

(2)\[P\left(\text{simulant dies of $c_i$} \:\middle|\: \genfrac{}{}{0pt}{} {\text{simulant has $c_1,\dotsc, c_k$ only}}{\text{and dies of one of them}} \right) = \frac{\operatorname{cfr}_i}{\sum_{i=1}^k \operatorname{cfr}_i}.\]

Clearly these conditional probabilities sum to 1, so every simulant who dies will be assigned a cause of death. Moreover, we claim that using the above conditional probabilities will lead to the correct case fatality risk for all the maternal disorders subcauses.

To see this, first observe that multiplying the above two equations yields

(3)\[P(\text{simulant dies of $c_i$} \mid \text{simulant has $c_1,\dotsc, c_k$ only}) = \operatorname{cfr}_i, \text{ if } c_i \in \{c_1,\dotsc, c_k\}.\]

In particular, (3) holds for any set of causes $c_1,\dotsc, c_k$, as long as $c_i$ is one of them. Therefore, when we compute the overall case fatality risk of $c_i$ by averaging over the entire population of simulants who have the cause $c_i$, we get

(4)\[\begin{split}\begin{aligned} P(\text{simulant dies of $c_i$} \mid \text{simulant has $c_i$}) \hspace{-7.5cm}& \\ &= \sum_k \sum_{\substack{c_1,\dotsc, c_k\in \text{causes} \\ c_i \in \{c_1,\dotsc, c_k\}}} P(\text{dies of $c_i$} \mid \text{has $c_1,\dotsc, c_k$ only}) \cdot P(\text{has $c_1,\dotsc, c_k$ only} \mid \text{has $c_i$})\\ &= \sum_k \sum_{\substack{c_1,\dotsc, c_k\in \text{causes} \\ c_i \in \{c_1,\dotsc, c_k\}}} \operatorname{cfr}_i \cdot P(\text{has $c_1,\dotsc, c_k$ only} \mid \text{has $c_i$})\\ &= \operatorname{cfr}_i \cdot \sum_k \sum_{\substack{c_1,\dotsc, c_k\in \text{causes} \\ c_i \in \{c_1,\dotsc, c_k\}}} P(\text{has $c_1,\dotsc, c_k$ only} \mid \text{has $c_i$})\\ &= \operatorname{cfr}_i \cdot 1. \end{aligned}\end{split}\]

The first step and last step hold because the union of the disjoint events $\{\text{has $c_1,\dotsc, c_k$ only}\}$ over all subsets $\{c_1,\dotsc, c_k\}$ containing the cause $c_i$ equals the event $\{\text{has $c_i$}\}$. Since $c_i$ was arbitrary, we get

\[P(\text{simulant dies of $c_i$} \mid \text{simulant has $c_i$}) = \operatorname{cfr}_i\]

for all maternal disorders subcauses $c_i$ as claimed.

Warning

If the case fatality risks of some of the maternal disorders subcauses are too large, the “probability” calculated in (1) may be greater than 1, which would lead the model to underestimate the true case fatality risks. This problem could be fixed, for example, by scaling down the probability in (1) to be less than 1, and scaling up the probability $P(\text{simulant dies of $c_i$} \mid \text{simulant has $c_i$ only})$ so that the average of the probabilities computed in (4) still works out to the correct overall case fatality risk $\operatorname{cfr}_i$. Note that the probabilities in (4) depend on the joint distribution of all the causes, so in order to solve for the correct probability, we would either need some information about the joint distribution, or we would have to calibrate the model empirically.

The potential for probabilities greater than 1 illustrates an inherent deficiency in the above methodology. Specifically, the the two assumptions (1) and (2) lead to the equation (3), which is equivalent to the assumption that the case fatality risk of $c_i$ is conditionally independent of whatever other causes you have, given that you have $c_i$. It is not difficult to see that this assumption cannot hold in general. For example, if you have a particularly deadly form of cancer, your case fatality risk for that cancer might be quite high in general. However, if you are also falling off a cliff, the case fatality risk of the cancer will be 0, since you will die of the fall with probability 1 before the cancer kills you.

That is, in general, we expect the probability of dying from $c_i$ to depend on what other causes you have concurrently, because there could be crowd-out effects (e.g., the fall crowds out the cancer in the example above) or possibly “reinforcement” effects between the different causes. However, we suspect that this won’t be an issue for the maternal disorders we are modeling, and the above calculations will likely be good enough approximations for our purposes.