The Symptom Pattern method, proposed in 2007, is an automated method for analyzing verbal autopsy (VA) data to generate cause of death data in regions of the world without vital registration systems. New research from IHME, the Department of Health Services at the University of Washington, and the University of Queensland as part of the Population Health Metrics Research Consortium (PHMRC) shows that a simplified version of Symptom Pattern, termed Simplified Symptom Pattern (SSP), can be used to accurately interpret VAs.
Recent studies show that Symptom Pattern, as originally proposed, can be simplified with enhanced performance. This study was designed to test this simplified version, assessing its performance in predicting causes of death at both the individual and population level. The quality of assessment by the SSP method was tested by validating its performance using deaths for which the causes of death are known, collected as part of the PHMRC gold standard verbal autopsy validation study, and by comparing the SSP method to a widely used method of interpreting VAs, physician-certified verbal autopsy (PCVA). This study is part of ongoing work by IHME to develop the most accurate and efficient methods of predicting causes of death from VAs.
The SSP method was tested both with and without household recall of health care experience. Household recall of health care experience includes any information the caretaker has about the patient's medical treatment, including whether health workers provided documentation for the cause of hospitalization or cause of death.
The results showed that the SSP method performed better than Symptom Pattern method, and that the SSP method was as good as or better than PCVA in adults, neonates, and children, both in correctly determining cause of death at the individual level and in accurately estimating the cause-specific mortality fractions at the population level. The only exception was in neonates with health care experience information included, although direct comparison in neonates is not possible, because PCVA analysis used a shorter cause of death list.
Analysis of the quality of the SSP method’s results for specific causes shows that it may systematically over- or underestimate predictions of causes of death. In many cases, however, the trend is predictable and precise, and therefore predictions in these causes can be adjusted.
In order to develop the SSP method, the authors investigated parameters in Symptom Pattern’s Bayesian framework that allow for optimal performance in assigning individual causes of death and determining cause-specific mortality fractions. The SSP method was developed by combining parameters that produced the best performance. The SSP method was then applied to 12,542 adult, child, and neonatal VAs from the PHMRC gold standard VA validation study (for which gold standard causes of death were established using strict clinical diagnostic criteria) and was compared to the results from PCVA on the same dataset.
A practical limitation of the SSP method, noted by the authors, is that the computational power required to implement the method for a single cause of death is greater than other methods, such as the Tariff or Random Forest methods. However, for analysis of large groups of deaths or for research studies, the required computational power might be a reasonable trade-off, given the reliable results produced by the SSP method.
If the SSP method is implemented on a user-friendly computational platform, it can be a reliable, accurate method for interpreting VA data to estimate individual- and population-level mortality in populations that lack vital registration systems.