Overview of EAVA

Introduction

Causes of death (CODs) are not known for about two-thirds of the global population (Cejudo et al. 2023). Verbal Autopsies (VAs) are routinely conducted questionnaires which capture community or population level causes of death in settings where civil registration and health systems are weak (World Health Organization 2016). Historically, causes were ascertained by physicians, who interpreted signs and symptoms reported by a decedent’s caregiver and assigned a cause. In recent years, computer-coded verbal autopsy (CCVA) algorithms have been developed to conserve physician time in resource-limited settings (Byass et al. 2012; McCormick et al. 2016; Kalter, Perin, and Black 2016). This package makes available the Expert Algorithm Verbal Autopsy (EAVA) CCVA. The EAVA package consists of two functions which parallel data processing in the CrossVA and the openVA packages.

odk2EAVA

Publicly available data from the Countrywide Mortality Surveillance for Action project website https://comsamozambique.org/data-access contains WHO 2016 Verbal Autopsy Questionnaire data, and demonstrates the odk2EAVA() function. The resulting output consists of a unique identifier and 378 signs and symptoms for evaluation and COD assignment by codEAVA().

codEAVA

The output of the odk2EAVA() function becomes the input of the codEAVA() function. codEAVA() assigns causes based on the presence and absence of symptoms. Each record is evaluated and either meets or does not meet diagnostic criteria for each cause. Records are assigned a single cause based on a deterministic hierarchy of all causes, so that each record is assigned the first cause for which it meets diagnostic criteria in the hierarchy. If no cause criteria are met, the record will have an “unspecified” COD.

Example

Taken together, data are first converted from the WHO 2016 Verbal Autopsy Questionnaire format using odk2EAVA() and then that output file is used to generate causes of death based on signs and symptoms and a hierarchy of causes.

# load EAVA package
library(EAVA)
library(stringi)
library(stringr)

# load embedded example data
data <- as.data.frame(data_public)

# run odk2EAVA() using example data
output <- odk2EAVA(data, id_col  = "comsa_id")

# run codEAVA for neonates and children 1-to-59 months of age
EAVA_neonate <- codEAVA(output, "neonate")
EAVA_child <- codEAVA(output, "child")

# check results
head(EAVA_neonate)
#>      ID       cause
#> 1  2021 Intrapartum
#> 5  3169 Unspecified
#> 9 16965 Unspecified
head(EAVA_child)
#>      ID                   cause
#> 2 17485                 Preterm
#> 3 12102             Unspecified
#> 4  9502      Diarrhea/Dysentery
#> 6   434        Other infections
#> 7  4168               Pneumonia
#> 8  5983 Meningitis/Encephalitis

References

Byass, Peter, Daniel Chandramohan, Samuel J Clark, Lucia D’Ambruoso, Edward Fottrell, Wendy J Graham, Abraham J Herbst, et al. 2012. “Strengthening Standardised Interpretation of Verbal Autopsy Data: The New InterVA-4 Tool.” Global Health Action 5 (1). https://doi.org/10.3402/gha.v5i0.19281.

Cejudo, Ander, Arantza Casillas, Alicia Pérez, Maite Oronoz, and Daniel Cobos. 2023. “Cause of Death Estimation from Verbal Autopsies: Is the Open Response Redundant or Synergistic?” Artificial Intelligence in Medicine 143 (102622). https://doi.org/10.1016/j.artmed.2023.102622.

Kalter, Henry, Jamie Perin, and Robert E Black. 2016. “Validating Hierarchical Verbal Autopsy Expert Algorithms in a Large Data Set with Known Causes of Death.” Journal of Global Health 6 (1). https://doi.org/10.7189/jogh.06.010601.

McCormick, Tyler H, Zehang Richard Li, Clara Calvert, Amelia C Crampin, Kathleen Kahn, and Samuel J Clark. 2016. “Probabilistic Cause-of-Death Assignment Using Verbal Autopsies.” Journal of the American Statistical Association 111 (515): 1036–49. https://doi.org/10.1080/01621459.2016.1152191.

World Health Organization. 2016. “Verbal Autopsy Standards: The 2016 WHO Verbal Autopsy Instrument.”