Verbal autopsy tool version history

SmartVA-Analyze Application

Version 2.2.0 (current, released January 9, 2023)

This release updates the app to work with a new questionnaire:

  • Add support for WHO 2022 Questionnaire
  • Fix coding of missing values in Pregnancy Duration question in WHO Questionnaires

Version 2.1.0 (released January 18, 2021)

This release includes minor changes to reduce the fraction of child deaths that are mapped to an undetermined cause:

  • Reduce the threshold for child undetermined causes of death. The proposed threshold brings the undetermined fraction in various child test datasets closer to the undetermined fraction that is seen for adult datasets.
  • Remove the majority of child sepsis spurious symptoms. Inadvertently, too many symptoms were listed as spurious (i.e., these symptoms were removed as predictors for child sepsis) such that child sepsis was not predicted in any dataset. Removing these spurious symptoms results in an increase in sepsis predictions that were mostly previously classified as "undetermined".
  • Amend the child homicide rule so that an additional condition is reporting the decedent was a victim of violence.

Version 2.0 (released December 10, 2018)

This release adds support for the WHO 2016 version 1.5.1 Questionnaire. The differences between this and the previous version are:

  • Refactored implementation of Tariff 2.0 Algorithm for faster analysis
  • Updated graphical user interface removes pop-up regarding warnings
  • Addition and reorganization of output files (SmartVA Outputs Interpretation)
  • Report.txt instead of warnings.txt in folder “4-monitoring-and-quality”
  • Sex-specific CSMF plots only include sex-specific causes
  • “Renal Failure” renamed to “Chronic Kidney Disease” and “Violent death” renamed to “Homicide”
  • Adjusts the child thresholds used to determine if an observation has enough information to generate a valid prediction, which reduces the proportion of child Undetermined deaths
  • Women who were reported to have gone through menopause cannot be assigned a maternal cause
  • Adds stricter criteria to predict child cardiovascular diseases

Version 1.2.1 (released June 30, 2017)

This is a minor update to Smart-VA-Analyze 1.2 released May 2017. The differences between this and the previous version are:

  • Stillbirth is no longer predicted in neonates who live at least one day
  • Child AIDS is no longer deterministically predicted in cases where the mother was told by a health worker that she had HIV.
  • Maternal causes can now be predicted in ages 12 through 14
  • Observations with missing ages now use a default age which allows age-restricted causes, like cancer, to be predicted.

Version 1.2 (released May 15, 2017)

This release provides additional features, more robust handling input data, and methodological updates to increase the predictive performance.

Release notes

  • Updates weights used to reallocate undetermined predictions at the population-level with data from GBD 2015 and Tariff performance using the shortened questionnaire.
  • Adds an option to remove AIDS from the list of predictions if the data comes from a country with low HIV prevalence.
  • Adds progress bar and estimated time remaining for each step.
  • Adds an option to turn off generation of figures.
  • Fixes issues with properly classifying observations into the correct age-specific module if the data is incomplete.
  • Changes the aggregation of adult causes reported in the final prediction files.
  • Fixes errors in the data cleaning which previously prevented the analysis from completing when the input data was missing columns or had unexpected values.
  • Adds rules to predict injury or stillbirth if a key symptom is endorsed.
  • Adjusts the thresholds used to determine if an observation has enough information to generate a valid prediction.
  • Removes tariff scores for implausible cause-symptom pairs which resulted from small sample size in the training data. 
  • Enhances Tariff 2.0 algorithm with expert rule-based algorithms which provide simple rule-out criteria or necessary symptoms from certain diagnoses.

Version 1.1.1 (released June 17, 2016)

This release is a minor update to SmartVA-Analyze version 1.1, released December 2015. The differences between this and the previous version are:

  • Fixes a spurious correlation issue when a sign or symptom gets a high tariff, but probably should not have. An example is “pox”, which appears in the free-text responses of the PHMRC validation dataset in a very unexpected way.
  • Fixes a critical bug in interpreting the ODK version of the PHMRC Shortened Questionnaire, which affects how the free-text response in child VAIs is used.
  • Fixes an issue which prevented analysis from completing when the input data was downloaded from some instances of ODK Aggregate.

Version 1.1 (released December 17, 2015)

This release provides support for the Tariff Method 2.0 and the PHMRC Shortened Questionnaire.

Release notes

  • Implements the Tariff 2.0 Method computer certification of verbal autopsy (VA). While designed and validated with the Population Health Metrics Research Consortium (PHMRC) Gold Standard VA database, this method may be applied to any VA survey. See documentation for more details.
  • Includes some minor bug fixes.

Version 1.0 (released October 14, 2013)