89

How do verbal autopsy models estimate the burden of maternal mortality?. Synthesize the InterVA-M validation evidence with the GBD maternal mortality burden findings, separating method, validation, burden estimates, uncertainty, and limitations. Highlight that probabilistic models can approximate cause-specific mortality fractions but also require careful interpretation because case-by-case agreement and data quality constraints affect confidence.

How verbal autopsy models estimate maternal mortality burden

Verbal autopsy models like InterVA-M estimate maternal mortality burden by assigning likely causes of death probabilistically from interview data, rather than relying on a single physician judgment for each case.[1] In the GBD study, maternal mortality burden is then estimated at population level by cause, age, country, and time using standardized modeling and correction methods applied to incomplete mortality data.[2]

Taken together, these studies show that probabilistic models can approximate cause-specific mortality fractions and produce usable burden estimates, but they still need careful interpretation because individual-case agreement is imperfect and the underlying data may be incomplete or poor quality.[3][4]

Method

  • InterVA-M: a probabilistic verbal autopsy interpretation model developed and refined with adult female verbal autopsy data, then tested against 258 additional verbal autopsy interviews from Burkina Faso by comparing model-assigned causes with physician-assigned causes.[5]
  • GBD 2015 Eastern Mediterranean Region study: a regional burden analysis using Global Burden of Disease methods, with mortality modeled from population-level data and cause-of-death ensemble modeling (CODEm), plus corrections for incompleteness and underestimation.[6]

Validation evidence and burden estimates

For InterVA-M, the main validation signal was that the cause-specific mortality fractions were broadly similar after rationalization between model output and physician interpretation, and case-by-case agreement was about 60% with any reviewing physician, increasing to about 80% when discrepant cases were reviewed by an additional physician.[7]

The GBD paper does not validate a case-level verbal autopsy model; instead, it presents regional burden estimates for maternal mortality and related outcomes, reported with uncertainty intervals for measures such as maternal mortality ratio, deaths, years of life lost, years lived with disability, and disability-adjusted life years.[8]

Uncertainty and limitations

  • InterVA-M: disagreement remained for some causes, especially cardiovascular disease, malaria, and pregnancy-related infections, showing that case-level attribution is not fully stable across reviewers or methods.[9]
  • InterVA-M data limitation: free-text material did not make an appreciable difference to the model’s output in this assessment, suggesting limited added value from that narrative detail in the validation sample.[10]
  • GBD limitation: many countries in the Eastern Mediterranean Region have poor health data and weak vital statistics, so the estimates depend heavily on GBD modeling and garbage-code correction.[11]
  • GBD limitation: the paper also notes limited information on unsafe abortion in the region, which constrains how completely maternal mortality can be characterized.[12]

Bottom line

The shared lesson is that probabilistic models such as InterVA-M can help estimate maternal mortality burden when direct measurement is difficult, and they can approximate cause-specific mortality fractions at scale.[13][14] But confidence should remain calibrated: case-by-case agreement is imperfect, and GBD-style burden estimates are only as strong as the source data and correction assumptions behind them.[15][16]


Related Content From The Pandipedia