Tyler McCormick- Univ. of Washington
TITLE:Assessing cause of death using verbal autopsies
ABSTRACT: Worldwide, fewer than 1/3 of deaths occur in a setting where there is a medically assigned cause. In populations like this, where most deaths happen outside of hospitals, verbal autopsy (VA) is a commonly used tool to assess cause of death and estimate cause-specific mortality rates and the distribution of deaths by cause. VA uses an interview with caregivers of the decedent to elicit data describing the signs and symptoms leading up to the death. This talk describes recent methodological and implementation focused work on verbal autopsies. First, we present a probabilistic approach to classifying cause of death using verbal autopsies. Our approach uses a latent Gaussian graphical model to characterize dependence between symptoms, lifting a conditional independence assumption present in currently available cause classification methods. Our model also accommodates informative prior information about marginal relationships between causes and symptoms, making it well-suited for settings where we combine inputs from many data sources. In the second part of the talk, we discuss the our ongoing efforts to integrate verbal autopsies in settings with partial coverage vital registration systems. We describe an open source software platform designed to integrate with existing verbal autopsy & vital registration data infrastructures, and provide insights based on a pilot study currently underway in 10 countries.