The model
InterVA: Bayesian interpretation of verbal autopsies
InterVA is a family of probabilistic models that interpret verbal autopsy data and assign likely causes of death. Developed under the leadership of Professor Peter Byass — originally at Umeå University — it became one of the most widely used automated VA methods in the world.
How the model works
InterVA applies Bayes' theorem to verbal autopsy questionnaires. At its core is a matrix of symptom–cause probabilities: for each cause of death in its cause list, the model encodes how likely each interview indicator (a reported sign, symptom, or circumstance) is to be present. Given the set of indicators recorded in an individual VA interview, the model computes the likelihood of each cause and reports the most probable one(s), along with likelihood values that can be aggregated into cause-specific mortality fractions for a population.
Because the logic is fully deterministic and the symptom–cause matrix is published, InterVA outputs are transparent and reproducible: the same input data always yield the same result, and researchers can inspect why a particular cause was assigned. This transparency was a deliberate design goal — a standardised, repeatable alternative to the variability of physician coding.
A short history
- Origins. The InterVA approach grew out of work on standardising VA interpretation in the early 2000s, with earlier versions distributed as a desktop application and documented in the VA methods literature of that period. From roughly 2006 to 2019, this domain — interva.net — served as the project's public entry point, redirecting visitors to the InterVA pages hosted first at Umeå University and later at peterbyass.com.
- InterVA-4 (2012). Released together with the 2012 WHO verbal autopsy instrument, InterVA-4 aligned the model's indicators and cause list with the WHO standard. It is described in Byass et al., 2012, Global Health Action.
- InterVA-5 (2019). A substantial update aligned with the 2016 WHO VA instrument, adding integrated processing of WHO-format questionnaire data and extended outputs. It is described in Byass et al., 2019, BMC Medicine.
- Today. InterVA-5 remains actively referenced, and open-source implementations of the models are maintained as R packages on CRAN within the openVA ecosystem (see Tools).
Official resources
This site does not distribute InterVA software. For the model itself, use the official sources:
- peterbyass.com/interva — the InterVA project pages maintained by the model's authors.
- InterVA4 on CRAN and InterVA5 on CRAN — open-source R implementations of the models, part of the openVA toolkit.
- The original publications listed in Research.
Note. interva.net is an independent informational resource and is not the official website of the InterVA project. References to the InterVA model are nominative and descriptive only. See About.
Frequently asked questions
What is the InterVA model?
InterVA is a family of Bayesian models that interpret verbal autopsy data and assign likely causes of death. Developed under the leadership of Professor Peter Byass, it became one of the most widely used automated VA methods in the world.
How does InterVA assign a cause of death?
The model holds a published matrix of symptom–cause probabilities. For the set of indicators recorded in an interview, it applies Bayes' theorem to compute the likelihood of each cause in its cause list and reports the most probable one(s). Because the matrix and logic are fixed, results are fully transparent and reproducible.
What is the difference between InterVA-4 and InterVA-5?
InterVA-4 (2012) is aligned with the 2012 WHO verbal autopsy instrument; InterVA-5 (2019) is a substantial update aligned with the 2016 WHO instrument, adding integrated processing of WHO-format questionnaire data and extended outputs. Both are described in the papers listed in Research.
Where can I download InterVA software?
Not from this site — interva.net does not distribute software. Use the official sources: the open-source R packages InterVA4 and InterVA5 on CRAN, and the project pages at peterbyass.com/interva. See Tools for an overview.
Is interva.net the official InterVA website?
No. From roughly 2006 to 2019 this domain was the project's public entry point, but today it is an independent informational resource, not affiliated with the InterVA team. The current official home of the project is peterbyass.com/interva.
See also: verbal autopsy methodology · tools & software