GBD 2023 Maternal Sepsis Risk Effects
Risk Overview
This page describes the Vivarium modeling strategy for risk effects. For a description of Vivarium modeling strategy for risk exposure (in this case a cause model document), see the maternal sepsis page.
GBD 2023 Modeling Strategy
Todo
Add link to 2023 anemia impairment page once it is created and describes the new anemia causal attribution process for GBD 2023.
GBD does not explicitly model maternal (also referred to as puerperal) sepsis as a risk factor. However, as of GBD 2023, it models a hemoglobin shift associated with puerperal sepsis in the anemia causal attribution process. Newly for GBD 2023, this hemoglobin shift is time-dependent (by days relative to delivery) and is informed by an observational analysis of US MarketScan data. The graph below is from the nonfatal methods appendix. Note that it compares postpartum people with sepsis to postpartum people without sepsis, so the hemoglobin shift values are in addition to the hemoglobin changes associated with pregnancy itself.
Vivarium Modeling Strategy
Outcome |
Outcome type |
Outcome ID |
Affected measure |
Note |
|---|---|---|---|---|
Hemoglobin concentration during the first six weeks after the end of pregnancy |
Risk exposure |
376 |
Hemoglobin concentration |
During the first six weeks after the end of pregnancy only |
Hemoglobin concentration between 6 weeks and 39 weeks after the end of pregnancy |
Modelable entity |
27596 |
Hemoglobin concentration |
During the period from 6 weeks to 39 weeks after the end of pregnancy only; this is the “non-pregnant” hemoglobin distribution |
Hemoglobin effects after the end of pregnancy
For simulants who experience an incident case of puerperal sepsis as determined in the maternal sepsis page,
we will decrease the simulant’s hemoglobin concentration during two distinct time periods: the first six weeks after the end of pregnancy and the period from 6 weeks to 39 weeks after the end of pregnancy.
We chose the 39 week cutoff for consistency with postpartum hemorrhage effects.
The magnitude of the decrease will be determined by the maternal sepsis hemoglobin effect estimated in GBD 2023,
which is time-dependent by days relative to delivery.
For each time period, we take an average of the continuous hemoglobin shift curve across that time period to determine the hemoglobin shift value applied during that time period.
The curve was provided to us by the anemia modelers at /mnt/team/anemia/pub/emotive/mat_sep/, specifically in the pred_data.csv file in that directory.
We will copy this file to our simulation repository for tracking purposes.
Validation and Verification Criteria
The hemoglobin concentration stratified by puerperal sepsis incidence (also stratified by anemia status in pregnancy to avoid confounding by this factor) should differ by the magnitude of the puerperal sepsis hemoglobin effect, in each post-pregnancy time step.
Assumptions and Limitations
This modeling strategy does not consider that the impact of maternal sepsis is already reflected in the pregnancy adjustment factor used for the hemoglobin model and therefore we may slightly underestimate hemoglobin concentration (and therefore overestimate anemia prevalence) on average during the pregnancy and lactation period by applying an additional negative hemoglobin shift associated with maternal sepsis.
This modeling strategy applies an average hemoglobin shift for all incident maternal sepsis cases during the first six weeks after the end of pregnancy and an average hemoglobin shift for all incident maternal sepsis cases during the period from 6 weeks to 39 weeks after the end of pregnancy, which does not capture the continuous nature of hemoglobin changes over time. The continuous curves could result in simulants moving between several anemia categories over the course of the postpartum period, which is not captured here. This also means that the (relatively arbitrary) choices of durations of these periods are impactful.
The GBD shift is derived from USA MarketScan data, which may not be generalizable to other locations. In particular, treatment in the USA may be more effective at mitigating the hemoglobin shift associated with maternal sepsis than treatment in other locations, which could lead to an underestimation of the hemoglobin shift in locations with less effective treatment. Additionally, the USA has more access to antibiotic treatment than our modeled locations which could lead to better prevention.
Todo
Consider a modeling strategy that calibrates the pregnancy-specific hemoglobin exposure to the baseline level of maternal sepsis in the population