Blog

Accurate information on causes of death is essential to estimate the burden of disease, informing health policies and tracking progress to achieve global targets, such as the Sustainable Development Goals. However, current estimates for the causes of death are inadequate at a global level, so the health community is trying to resolve this.

In this blog, we find out about the unique study and the first assessment of its kind carried out by the team from the Institute for Global Health of Barcelona (ISGlobal), Spain. To improve on the lack of precise information on causes of death in low income and resource-restricted settings, the team compared the commonly used Interpreting Verbal Autopsy method, where verbal autopsies based on structured interviews with the witnesses of the death are interpreted and coded by a computer, with Complete Diagnostic Autopsy, the current gold standard methodology.

From the group, Clara Menéndez and Llorenç Quintó discuss their analysis and their Research Article, of which they’ve recently published a Version 2 and is available on Gates Open Research.

Within the ISGlobal program, the evaluation of methods to determine the causes of death in vulnerable populations in low- and middle-income countries has been a priority area of research. Our current study is part of a prospective observational post mortem evaluation (the CaDMIA study) to validate the minimally invasive autopsy method for cause of death (CoD) determination in different age groups using the complete diagnostic autopsies (CDA), as the gold standard technique. Since all deaths in this study were hospital deaths, the information feeding the verbal autopsies (VAs) in this analysis was extracted from the clinical record of the deceased individual and from the obstetric record in perinatal deaths.

Validating a model

This was unified into the WHO 2012 VA standard format, converted into the 245 input indicators of the VA, and processed using the InterVA4 package to determine the CoD. We also made a rigorous review of the methodology developed by other research groups in order to apply standardized methods that made our results comparable with previous studies of concordance in CoD determination, although in those studies the CDA was not used as the reference method.

A main characteristic of the methodology applied in this study is that 500 cause compositions based on resampling techniques were generated for each study group to validate the VA model across a variety of CoD distributions, i.e. not only for the distribution particularly observed in the study sample.

The performance of the VA at both individual and population level was estimated comparing the CoD established by the CDA with the CoD provided by the VA, calculating several statistical measures of agreement corrected by chance. A complete explanation of the methodology applied can be found in our manuscript and in the extended data.

Clinical errors and inaccurate data on mortality rates

Accurate knowledge of the CoD is crucial for the design of appropriate and effective health policies and strategies. Determining the CoD is complex and is probably the data recorded with the least precision and reliability, especially in low-income countries where many deaths occur at home or at poorly resourced health facilities. VAs are the only source of data for CoD in such settings, and their practice and improvement should therefore be encouraged. However, rigorous studies are needed to verify the validity of VAs in estimating CoD at both the individual and population levels.

Several “gold standards” have been used over the years in so-called “validation” studies of VAs, that used clinical diagnoses made by physicians despite being prone to frequent misclassification errors to estimate the CoD. Comparison of VAs with the CDA is the most appropriate approach, being the main limitation that CDAs are rarely performed in general and specially in low and middle-income countries. In this study we were able to access the information necessary to perform this unprecedented and unique validation analysis.

Diseases and deaths misclassified

The study shows an overall poor agreement both at the individual and at the population level of the InterVA model against the gold standard. In general, the sensitivity of the InterVA model was low in identifying CoD assigned by the CDA. In maternal deaths, the probability that a death diagnosed as eclampsia by the verbal autopsy procedure was low, where out of 26 deaths attributed to eclampsia, only four patients actually died from the condition.

At the population level the chance-corrected accuracy of the InterVA model in predicting CoD was overall low, best in children, but performed worst of all in stillbirths. Frequent misclassification of the type of infection were observed among CoD for all categories of infectious diseases, including tuberculosis, malaria and HIV.  

What these results tell us, is that the reliability of current estimates for common causes of death is generally poor or very poor. Therefore, programs to prevent or manage these diseases are prone to be ineffective or at worst will fail.

Improving quality and performance

We need to identify the indicators of the VA responsible for the misclassification especially for infectious diseases. This information should be used to adapt the existing VA interview to include more specific indicators in certain signs/symptoms to improve the identification of infections as CoD. Also, InterVA estimates can be improved by predictive models for misclassification using a small number of important indicators as covariates.

We expect that improved reliability of VA methods would translate into more precise estimates of relevant causes of death in the population, guiding targeted health programs and facilitating health monitoring and evaluation.

The pillar of responsible research

One of our main reasons for choosing to publish on Gates Open Research is because we think that the open peer review on this open access platform, is a much more transparent and constructive process than anonymous review. Openness forces referees to think more carefully about the scientific issues and to write more thoughtful reviews, which may also help to expose possible conflicts of interest in some cases.

Sharing information freely is the essence of scientific activity and the principle that makes the advancement of knowledge possible. It is the pillar of responsible research, as it is concerned with making scientific research more transparent, collaborative and efficient. It is not only the research itself that needs to be open to society to improve access to the final results, but all the intermediate elements need to be freely and easily available so that they are communicable and potentially useful for other researchers. Initiatives such as open access scientific journals are necessary, but we feel that they are currently inadequate and more needs to be done encourage a more open research culture. 

“This is a very useful contribution to the field,” says Clara Calvert from the London School of Hygiene & Tropical Medicine, UK, in her review of the study. Read the full article and the peer review reports here.