Probabilistic Cause-of-death Assignment using Verbal Autopsies

Urban-rural and sex differentials in tuberculosis mortality in Bangladesh: results from a population-based survey

OBJECTIVE

To assess tuberculosis mortality in Bangladesh through a population-based survey using a Verbal Autopsy tool.

METHODS

Nationwide mortality survey employing the WHO-recommended Verbal Autopsy (VA) tool, and using InsilicoVA, a data-driven method, to assign the cause of death. Using a three-stage cluster sampling method, 3997 VA interviews were conducted in both urban and rural areas of Bangladesh. Cause-specific mortality fractions (CSMF) were estimated using Bayesian probabilistic models.

RESULTS

6.8% of total deaths in the population were due to TB [95% CI: (5.1, 8.9)], comprising 12.0% [95% CI: (11.1, 12.8)] and 6.42% [95% CI: (5.4, 7.3)] of total male and female deaths, respectively. This proportion was highest among adults age 15-49 years [12.2%, 95% CI: (9.4, 14.6)]. The urban population is more likely to die from TB, and urban males have highest CSMF [13.6%, 95% CI: (9.1, 16.9)].

CONCLUSION

Our survey results show that TB is the fifth major cause of death in the general population and that sex and place of residence (urban/rural) have a significant effect on TB mortality in Bangladesh. The underlying causes of higher rates of TB-related deaths in urban areas and particularly among urban males, who have better knowledge and higher enrollment in the DOTS Program, need to be explored.

Integrating community-based verbal autopsy into civil registration and vital statistics (CRVS): system-level considerations

Background: Reliable and representative cause of death (COD) statistics are essential to inform public health policy, respond to emerging health needs, and document progress towards Sustainable Development Goals. However, less than one-third of deaths worldwide are assigned a cause. Civil registration and vital statistics (CRVS) systems in low- and lower-middle-income countries are failing to provide timely, complete and accurate vital statistics, and it will still be some time before they can provide physician-certified COD for every death.

Proposals: Verbal autopsy (VA) is a method to ascertain the probable COD and, although imperfect, it is the best alternative in the absence of medical certification. There is extensive experience with VA in research settings but only a few examples of its use on a large scale. Data collection using electronic questionnaires on mobile devices and computer algorithms to analyse responses and estimate probable COD have increased the potential for VA to be routinely applied in CRVS systems. However, a number of CRVS and health system integration issues should be considered in planning, piloting and implementing a system-wide intervention such as VA. These include addressing the multiplicity of stakeholders and sub-systems involved, integration with existing CRVS work processes and information flows, linking VA results to civil registration records, information technology requirements and data quality assurance.

Conclusions: Integrating VA within CRVS systems is not simply a technical undertaking. It will have profound system-wide effects that should be carefully considered when planning for an effective implementation. This paper identifies and discusses the major system-level issues and emerging practices, provides a planning checklist of system-level considerations and proposes an overview for how VA can be integrated into routine CRVS systems.

Performance of InSilicoVA for assigning causes of death to verbal autopsies: multisite validation study using clinical diagnostic gold standards

BACKGROUND

Recently, a new algorithm for automatic computer certification of verbal autopsy data named InSilicoVA was published. The authors presented their algorithm as a statistical method and assessed its performance using a single set of model predictors and one age group.

METHODS

We perform a standard procedure for analyzing the predictive accuracy of verbal autopsy classification methods using the same data and the publicly available implementation of the algorithm released by the authors. We extend the original analysis to include children and neonates, instead of only adults, and test accuracy using different sets of predictors, including the set used in the original paper and a set that matches the released software.

RESULTS

The population-level performance (i.e., predictive accuracy) of the algorithm varied from 2.1 to 37.6% when trained on data preprocessed similarly as in the original study. When trained on data that matched the software default format, the performance ranged from -11.5 to 17.5%. When using the default training data provided, the performance ranged from -59.4 to -38.5%. Overall, the InSilicoVA predictive accuracy was found to be 11.6-8.2 percentage points lower than that of an alternative algorithm. Additionally, the sensitivity for InSilicoVA was consistently lower than that for an alternative diagnostic algorithm (Tariff 2.0), although the specificity was comparable.

CONCLUSIONS

The default format and training data provided by the software lead to results that are at best suboptimal, with poor cause-of-death predictive performance. This method is likely to generate erroneous cause of death predictions and, even if properly configured, is not as accurate as alternative automated diagnostic methods.

The WHO 2016 verbal autopsy instrument: An international standard suitable for automated analysis by InterVA, InSilicoVA, and Tariff 2.0

BACKGROUND

Verbal autopsy (VA) is a practical method for determining probable causes of death at the population level in places where systems for medical certification of cause of death are weak. VA methods suitable for use in routine settings, such as civil registration and vital statistics (CRVS) systems, have developed rapidly in the last decade. These developments have been part of a growing global momentum to strengthen CRVS systems in low-income countries. With this momentum have come pressure for continued research and development of VA methods and the need for a single standard VA instrument on which multiple automated diagnostic methods can be developed.

METHODS AND FINDINGS

In 2016, partners harmonized a WHO VA standard instrument that fully incorporates the indicators necessary to run currently available automated diagnostic algorithms. The WHO 2016 VA instrument, together with validated approaches to analyzing VA data, offers countries solutions to improving information about patterns of cause-specific mortality. This VA instrument offers the opportunity to harmonize the automated diagnostic algorithms in the future.

CONCLUSIONS

Despite all improvements in design and technology, VA is only recommended where medical certification of cause of death is not possible. The method can nevertheless provide sufficient information to guide public health priorities in communities in which physician certification of deaths is largely unavailable. The WHO 2016 VA instrument, together with validated approaches to analyzing VA data, offers countries solutions to improving information about patterns of cause-specific mortality.

Tuberculosis mortality and the male survival deficit in rural South Africa: An observational community cohort study

BACKGROUND

Women live on average five years longer than men, and the sex difference in longevity is typically lower in populations with high mortality. South Africa-a high mortality population with a large sex disparity-is an exception, but the causes of death that contribute to this difference are not well understood.

METHODS

Using data from a demographic surveillance system in rural KwaZulu-Natal (2000-2014), we estimate differences between male and female adult life expectancy by HIV status. The contribution of causes of death to these life expectancy differences are computed with demographic decomposition techniques. Cause of death information comes from verbal autopsy interviews that are interpreted with the InSilicoVA tool.

RESULTS

Adult women lived an average of 10.4 years (95% confidence Interval 9.0-11.6) longer than men. Sex differences in adult life expectancy were even larger when disaggregated by HIV status: 13.1 (95% confidence interval 10.7-15.3) and 11.2 (95% confidence interval 7.5-14.8) years among known HIV negatives and positives, respectively. Elevated male mortality from pulmonary tuberculosis (TB) and external injuries were responsible for 43% and 31% of the sex difference in life expectancy among the HIV negative population, and 81% and 16% of the difference among people living with HIV.

CONCLUSIONS

The sex differences in adult life expectancy in rural KwaZulu-Natal are exceptionally large, atypical for an African population, and largely driven by high male mortality from pulmonary TB and injuries. This is the case for both HIV positive and HIV negative men and women, signalling a need to improve the engagement of men with health services, irrespective of their HIV status.

Measuring mortality due to HIV-associated tuberculosis among adults in South Africa: Comparing verbal autopsy, minimally-invasive autopsy, and research data

BACKGROUND

The World Health Organization (WHO) aims to reduce tuberculosis (TB) deaths by 95% by 2035; tracking progress requires accurate measurement of TB mortality. International Classification of Diseases (ICD) codes do not differentiate between HIV-associated TB and HIV more generally. Verbal autopsy (VA) is used to estimate cause of death (CoD) patterns but has mostly been validated against a suboptimal gold standard for HIV and TB. This study, conducted among HIV-positive adults, aimed to estimate the accuracy of VA in ascertaining TB and HIV CoD when compared to a reference standard derived from a variety of clinical sources including, in some, minimally-invasive autopsy (MIA).

METHODS AND FINDINGS

Decedents were enrolled into a trial of empirical TB treatment or a cohort exploring diagnostic algorithms for TB in South Africa. The WHO 2012 instrument was used; VA CoD were assigned using physician-certified VA (PCVA), InterVA-4, and SmartVA-Analyze. Reference CoD were assigned using MIA, research, and health facility data, as available. 259 VAs were completed: 147 (57%) decedents were female; median age was 39 (interquartile range [IQR] 33-47) years and CD4 count 51 (IQR 22-102) cells/μL. Compared to reference CoD that included MIA (n = 34), VA underestimated mortality due to HIV/AIDS (94% reference, 74% PCVA, 47% InterVA-4, and 41% SmartVA-Analyze; chance-corrected concordance [CCC] 0.71, 0.42, and 0.31, respectively) and HIV-associated TB (41% reference, 32% PCVA; CCC 0.23). For individual decedents, all VA methods agreed poorly with reference CoD that did not include MIA (n = 259; overall CCC 0.14, 0.06, and 0.15 for PCVA, InterVA-4, and SmartVA-Analyze); agreement was better at population level (cause-specific mortality fraction accuracy 0.78, 0.61, and 0.57, for the three methods, respectively).

CONCLUSIONS

Current VA methods underestimate mortality due to HIV-associated TB. ICD and VA methods need modifications that allow for more specific evaluation of HIV-related deaths and direct estimation of mortality due to HIV-associated TB.

Trends in the burden of HIV mortality after roll-out of antiretroviral therapy in KwaZulu-Natal, South Africa: an observational community cohort study

BACKGROUND

Antiretroviral therapy (ART) substantially decreases morbidity and mortality in people living with HIV. In this study, we describe population-level trends in the adult life expectancy and trends in the residual burden of HIV mortality after the roll-out of a public sector ART programme in KwaZulu-Natal, South Africa, one of the populations with the most severe HIV epidemics in the world.

METHODS

Data come from the Africa Centre Demographic Information System (ACDIS), an observational community cohort study in the uMkhanyakude district in northern KwaZulu-Natal, South Africa. We used non-parametric survival analysis methods to estimate gains in the population-wide life expectancy at age 15 years since the introduction of ART, and the shortfall of the population-wide adult life expectancy compared with that of the HIV-negative population (ie, the life expectancy deficit). Life expectancy gains and deficits were further disaggregated by age and cause of death with demographic decomposition methods.

FINDINGS

Covering the calendar years 2001 through to 2014, we obtained information on 93 903 adults who jointly contribute 535 42 8 person-years of observation to the analyses and 9992 deaths. Since the roll-out of ART in 2004, adult life expectancy increased by 15·2 years for men (95% CI 12·4-17·8) and 17·2 years for women (14·5-20·2). Reductions in pulmonary tuberculosis and HIV-related mortality account for 79·7% of the total life expectancy gains in men (8·4 adult life-years), and 90·7% in women (12·8 adult life-years). For men, 9·5% is the result of a decline in external injuries. By 2014, the life expectancy deficit had decreased to 1·2 years for men (-2·9 to 5·8) and to 5·3 years for women (2·6-7·8). In 2011-14, pulmonary tuberculosis and HIV were responsible for 84·9% of the life expectancy deficit in men and 80·8% in women.

INTERPRETATION

The burden of HIV on adult mortality in this population is rapidly shrinking, but remains large for women, despite their better engagement with HIV-care services. Gains in adult life-years lived as well as the present life expectancy deficit are almost exclusively due to differences in mortality attributed to HIV and pulmonary tuberculosis.

FUNDING

Wellcome Trust, the Bill & Melinda Gates Foundation, and the National Institutes of Health.

Naive Bayes classifiers for verbal autopsies: comparison to physician-based classification for 21,000 child and adult deaths

BACKGROUND

Verbal autopsies (VA) are increasingly used in low- and middle-income countries where most causes of death (COD) occur at home without medical attention, and home deaths differ substantially from hospital deaths. Hence, there is no plausible "standard" against which VAs for home deaths may be validated. Previous studies have shown contradictory performance of automated methods compared to physician-based classification of CODs. We sought to compare the performance of the classic naive Bayes classifier (NBC) versus existing automated classifiers, using physician-based classification as the reference.

METHODS

We compared the performance of NBC, an open-source Tariff Method (OTM), and InterVA-4 on three datasets covering about 21,000 child and adult deaths: the ongoing Million Death Study in India, and health and demographic surveillance sites in Agincourt, South Africa and Matlab, Bangladesh. We applied several training and testing splits of the data to quantify the sensitivity and specificity compared to physician coding for individual CODs and to test the cause-specific mortality fractions at the population level.

RESULTS

The NBC achieved comparable sensitivity (median 0.51, range 0.48-0.58) to OTM (median 0.50, range 0.41-0.51), with InterVA-4 having lower sensitivity (median 0.43, range 0.36-0.47) in all three datasets, across all CODs. Consistency of CODs was comparable for NBC and InterVA-4 but lower for OTM. NBC and OTM achieved better performance when using a local rather than a non-local training dataset. At the population level, NBC scored the highest cause-specific mortality fraction accuracy across the datasets (median 0.88, range 0.87-0.93), followed by InterVA-4 (median 0.66, range 0.62-0.73) and OTM (median 0.57, range 0.42-0.58).

CONCLUSIONS

NBC outperforms current similar COD classifiers at the population level. Nevertheless, no current automated classifier adequately replicates physician classification for individual CODs. There is a need for further research on automated classifiers using local training and test data in diverse settings prior to recommending any replacement of physician-based classification of verbal autopsies.

Methodological challenges in measuring vaccine effectiveness using population cohorts in low resource settings

•

We discuss methodological challenges for evaluating vaccine effectiveness using cohorts.

•

No single set of definitions or analytical approach can address all possible biases.

•

Careful consideration of denominator, exposure and outcome definitions is needed.

•

Sensitivity analyses are crucial to examine assumptions and explore subtle relationships.

Post-licensure real world evaluation of vaccine implementation is important for establishing evidence of vaccine effectiveness (VE) and programme impact, including indirect effects. Large cohort studies offer an important epidemiological approach for evaluating VE, but have inherent methodological challenges.

Since March 2012, we have conducted an open prospective cohort study in two sites in rural Malawi to evaluate the post-introduction effectiveness of 13-valent pneumococcal conjugate vaccine (PCV13) against all-cause post-neonatal infant mortality and monovalent rotavirus vaccine (RV1) against diarrhoea-related post-neonatal infant mortality. Our study sites cover a population of 500,000, with a baseline post-neonatal infant mortality of 25 per 1000 live births.

We conducted a methodological review of cohort studies for vaccine effectiveness in a developing country setting, applied to our study context. Based on published literature, we outline key considerations when defining the denominator (study population), exposure (vaccination status) and outcome ascertainment (mortality and cause of death) of such studies. We assess various definitions in these three domains, in terms of their impact on power, effect size and potential biases and their direction, using our cohort study for illustration. Based on this iterative process, we discuss the pros and cons of our final per-protocol analysis plan. Since no single set of definitions or analytical approach accounts for all possible biases, we propose sensitivity analyses to interrogate our assumptions and methodological decisions.

In the poorest regions of the world where routine vital birth and death surveillance are frequently unavailable and the burden of disease and death is greatest We conclude that provided the balance between definitions and their overall assumed impact on estimated VE are acknowledged, such large scale real-world cohort studies can provide crucial information to policymakers by providing robust and compelling evidence of total benefits of newly introduced vaccines on reducing child mortality.

The impacts of maternal mortality and cause of death on children's risk of dying in rural South Africa: evidence from a population based surveillance study (1992-2013)

BACKGROUND

Maternal mortality, the HIV/AIDS pandemic, and child survival are closely linked. This study contributes evidence on the impact of maternal death on children's risk of dying in an HIV-endemic population in rural South Africa.

METHODS

We used data for children younger than 10 years from the Agincourt health and socio-demographic surveillance system (1992 - 2013). We used discrete time event history analysis to estimate children's risk of dying when they experienced a maternal death compared to children whose mother survived (N=3,740,992 child months). We also examined variation in risk due to cause of maternal death. We defined mother's survival status as early maternal death (during pregnancy, childbirth, or within 42 days of most recent childbirth or identified cause of death), late maternal death (within 43-365 days of most recent childbirth), any other death, and mothers who survived.

RESULTS

Children who experienced an early maternal death were at 15 times the risk of dying (RRR 15.2; 95% CI 8.3-27.9) compared to children whose mother survived. Children under 1 month whose mother died an early (p=0.002) maternal death were at increased risk of dying compared to older children. Children whose mothers died of an HIV/AIDS or TB-related early maternal death were at 29 times the risk of dying compared to children with surviving mothers (RRR 29.2; 95% CI 11.7-73.1). The risk of these children dying was significantly higher than those children whose mother died of a HIV/AIDS or TB-related non-maternal death (p=0.017).

CONCLUSIONS

This study contributes further evidence on the impact of a mother's death on child survival in a poor, rural setting with high HIV prevalence. The intersecting epidemics of maternal mortality and HIV/AIDS - especially in sub-Saharan Africa - have profound implications for maternal and child health and well-being. Such evidence can help guide public and primary health care practice and interventions.