Ian Douglas Riley

Objectives: Gold standard cause of death data is critically important to improve verbal autopsy (VA) methods in diagnosing cause of death where civil and vital registration systems are inadequate or poor. As part of a three-country research study—Improving Methods to Measure Comparable Mortality by Cause (IMMCMC) study—data were collected on clinicopathological criteria-based gold standard cause of death from hospital record reviews with matched VAs. The purpose of this data note is to make accessible a de-identified format of these gold standard VAs for interested researchers to improve the diagnostic accuracy of VA methods. Data description: The study was conducted between 2011 and 2014 in the Philippines, Bangladesh, and Papua New Guinea. Gold standard diagnoses of underlying causes of death for deaths occurring in hospital were matched to VAs conducted using a standardized VA questionnaire developed by the Population Health Metrics Consortium. 3512 deaths were collected in total, comprised of 2491 adults (12 years and older), 320 children (28 days to 12 years), and 702 neonates (0–27 days).

We recently published in BMC Medicine an evaluation of the comparative diagnostic performance of InSilicoVA, a software to map the underlying causes of death from verbal autopsy interviews. The developers of this software claim to have failed to replicate our results and appear to have also failed to locate our replication archive for this work. In this Correspondence, we provide feedback on how this might have been done more usefully and offer some suggestions to improve future attempts at reproducible research. We also offer an alternative interpretation of the results presented by Li et al., namely that, out of 100 verbal autopsy interviews, InSilicoVA will, at best, correctly identify the underlying cause of death in 40 cases and incorrectly in 60 – a markedly inferior performance to alternative existing approaches.

Background: Accurate and timely cause of death (COD) data are essential for informed public health policymaking. Medical certification of COD generally provides the majority of COD data in a population and is an essential component of civil registration and vital statistics (CRVS) systems. Accurate completion of the medical certificate of cause of death (MCCOD) should be a relatively straightforward procedure for physicians, but mistakes are common. Here, we present three training strategies implemented in five countries supported by the Bloomberg Philanthropies Data for Health (D4H) Initiative at the University of Melbourne (UoM) and evaluate the impact on the quality of certification. Methods: The three training strategies evaluated were (1) training of trainers (TOT) in the Philippines, Myanmar, and Sri Lanka; (2) direct training of physicians by the UoM D4H in Papua New Guinea (PNG); and (3) the implementation of an online and basic training strategy in Peru. The evaluation involved an assessment of MCCODs before and after training using an assessment tool developed by the University of Melbourne. Results: The TOT strategy led to reductions in incorrectly completed certificates of between 28% in Sri Lanka and 40% in the Philippines. Following direct training of physicians in PNG, the reduction in incorrectly completed certificates was 30%. In Peru, the reduction in incorrect certificates was 30% after implementation and training on an online system only and 43% after training on both the online system and basic medical certification principles. Conclusions: The results of this study indicate that a variety of training strategies can produce benefits in the quality of certification, but further improvements are possible. The experiences of D4H suggest several aspects of the strategies that should be further developed to improve outcomes, particularly key stakeholder engagement from early in the intervention and local committees to oversee activities and support an improved culture in hospitals to support better diagnostic skills and practices.

Background: Verbal autopsy (VA) is increasingly being considered as a cost-effective method to improve cause of death information in countries with low quality vital registration. VA algorithms that use empirical data have an advantage over expert derived algorithms in that they use responses to the VA instrument as a reference instead of physician opinion. It is unclear how stable these data driven algorithms, such as the Tariff 2.0 method, are to cultural and epidemiological variations in populations where they might be employed. Methods: VAs were conducted in three sites as part of the Improving Methods to Measure Comparable Mortality by Cause (IMMCMC) study: Bohol, Philippines; Chandpur and Comila Districts, Bangladesh; and Central and Eastern Highlands Provinces, Papua New Guinea. Similar diagnostic criteria and cause lists as the Population Health Metrics Research Consortium (PHMRC) study were used to identify gold standard (GS) deaths. We assessed changes in Tariffs by examining the proportion of Tariffs that changed significantly after the addition of the IMMCMC dataset to the PHMRC dataset. Results: The IMMCMC study added 3512 deaths to the GS VA database (2491 adults, 320 children, and 701 neonates). Chance-corrected cause specific mortality fractions for Tariff improved with the addition of the IMMCMC dataset for adults (+ 5.0%), children (+ 5.8%), and neonates (+ 1.5%). 97.2% of Tariffs did not change significantly after the addition of the IMMCMC dataset. Conclusions: Tariffs generally remained consistent after adding the IMMCMC dataset. Population level performance of the Tariff method for diagnosing VAs improved marginally for all age groups in the combined dataset. These findings suggest that cause-symptom relationships of Tariff 2.0 might well be robust across different population settings in developing countries. Increasing the total number of GS deaths improves the validity of Tariff and provides a foundation for the validation of other empirical algorithms.

Reducing maternal mortality is a key focus of development strategies and one of the indicators used to measure progress towards achieving the Sustainable Development Goals. In the absence of medical certification of the cause of deaths that occur in the community, verbal autopsy (VA) methods are the only available means to assess levels and trends of maternal deaths that occur outside health facilities. The 2016 World Health Organization VA Instrument facilitates the identification of eight specific causes of maternal death, yet maternal deaths are often unsupervised, leading to sparse and generally poor symptom reporting to inform a reliable diagnosis using VAs. There is little research evidence to support the reliable identification of specific causes of maternal death in the context of routine VAs. We recommend that routine VAs are only used to capture the event of a maternal death and that more detailed follow-up interviews are used to identify the specific causes.

Almost all countries without complete vital registration systems have data on deaths collected by hospitals. However, these data have not been widely used to estimate cause of death (COD) patterns in populations because only a non-representative fraction of people in these countries die in health facilities. Methods that can exploit hospital mortality statistics to reliably estimate community COD patterns are required to strengthen the evidence base for disease and injury control programs. We propose a method that weights hospital-certified causes by the probability of death to estimate population cause-specific mortality fractions (CSMFs). We used an established verbal autopsy instrument (VAI) to collect data from hospital catchment areas in Chandpur and Comilla Districts, Bangladesh, and Bohol province, the Philippines, between 2011 and 2014, along with demographic covariates for each death. Hospital medical certificates of cause of death (death certificates) were collected and mapped to the corresponding cause categories of the VAI. Tariff 2.0 was used to assign a COD for community deaths. Logistic regression models were created for broad causes in each country to calculate the probability of in-hospital death, given a set of covariate values. The reweighted CSMFs for deaths in the hospital catchment population, represented by each hospital death, were calculated from the corresponding regression models. We collected data on 4228 adult deaths in the Philippines and 3725 deaths in Bangladesh. Short time to hospital and education were consistently associated with in-hospital death in the Philippines and absence of a disability was consistently associated with in-hospital death in Bangladesh. Non-communicable diseases (excluding stroke) and stroke were the leading causes of death in both the Philippines (33.9%, 19.1%) and Bangladesh (46.1%, 21.1%) according to the reweighted method. The reweighted method generally estimated CSMFs that fell between those derived from hospitals and those diagnosed by Tariff 2.0. Statistical methods can be used to derive estimates of cause-specific probability of death in-hospital for Bangladesh and the Philippines to generate population CSMFs. In regions where hospital death certification is of reasonable quality and routine verbal autopsy is not applied, these estimates could be applied to generate cost-effective and robust CSMFs for the population.

# Background This study investigated healthcare decision-making surrounding terminal illnesses of adults and elderly persons (aged 15 years or more), who later died at home, hospitals, or in-transit, in a rural, low-income area of Bangladesh where out-of-pocket health expenditure is very high. # Methods There were 1330 adult and elderly deaths in Matlab Health and Demographic Surveillance site in 2013. To record how decisions were made and implemented when caring for terminal illness, and how treatment-related costs were managed, the main caregivers of 69 randomly selected persons who died at home, 74 who died in hospitals, and 11 who died in-transit were interviewed with a social autopsy questionnaire in 2014. Differences between groups were tested by χ^2^ for significance at *P* \<0.05. # Results Those who died at home were sick for longer periods than those who died in hospitals or in-transit. During terminal illness, 33% of the home deaths were admitted, but discharged from hospitals prior to death for no chance of cure (52%) or financial constraints (17%). The reasons for 67% of home deaths not being admitted to a hospital were no chance of cure (28%), misjudgment of illness severity (21%), too short duration of illness (20%), or financial constraints (17%). For hospital deaths, final decisions were mostly made by family members (95%) as opposed to physicians (5%) for cure of serious illness (93%). Hospitals were chosen considering quality of care (59%), distance (36%), known doctors (35%) or referral (29%). After hospitals were chosen, 42% of hospitalizations were delayed due to lack of money (33%), misjudgment of illness severity (31%), or lack of someone to accompany the patient (31%). Payment of hospital costs included family savings (87%), borrowing from relatives (46%) or borrowing from moneylenders (6%). # Conclusions Healthcare decisions for terminally ill adults and elderly persons were influenced by the caregivers' assessment of chances of cure, judgment of illness severity and financial constraints. Improving understanding of illness severity and lowering direct out-of-pocket expenditures may help patients who have a reasonable chance of cure receive care in a hospital.

Medical certificates of cause of death (MCCOD) issued by hospital physicians are a key input to vital registration systems. Deaths certified by hospital physicians have been implicitly considered to be of high quality, but recent evidence suggests otherwise. We conducted a medical record review (MRR) of hospital MCCOD in the Philippines and compared the cause of death concordance with certificates coded by the Philippines Statistics Authority (PSA). MCCOD for adult deaths in Bohol Regional Hospital (BRH) in 2007–2008 and 2011 were collected and reviewed by a team of study physicians. Corresponding MCCOD coded by the PSA were linked by a hospital identifier. The study physicians wrote a new MCCOD using the patient medical record, noted the quality of the medical record to produce a cause of death, and indicated whether it was necessary to change the underlying cause of death (UCOD). Chance-corrected concordance, cause-specific mortality fraction (CSMF) accuracy, and chance-corrected CSMF were used to examine the concordance between the MRR and PSA. A total of 1052 adult deaths were linked between the MRR and PSA. Median chance-corrected concordance was 0.73, CSMF accuracy was 0.85, and chance-corrected CSMF accuracy was 0.58. 74.8% of medical records were deemed to be of high enough quality to assign a cause of death, yet study physicians indicated that it was necessary to change the UCOD in 41% of deaths, 82% of which required addition of a new UCOD. Medical records were generally of sufficient quality to assign a cause of death and concordance between the PSA and MRR was reasonably high, suggesting that routine mortality statistics data are reasonably accurate for describing population level causes of death in Bohol. While overall agreement between the PSA and MRR in major cause groups was sufficient for public health purposes, improvements in death certification practices are recommended to help physicians differentiate between treatable (immediate) COD and COD that are important for public health surveillance.

Deaths in developing countries often occur outside health facilities, making it extremely difficult to gather reliable cause of death (COD) information. Automated COD assignment using a verbal autopsy instrument (VAI) has been proposed as a reliable and cost-effective alternative to traditional physician-certified verbal autopsy, but its performance is still being evaluated. The purpose of this study was to compare the similarity of diagnosis by Medical Assistants (MA) in the Matlab Health and Demographic Surveillance System (HDSS) with the SmartVA Analyze 1.2 (Tariff 2.0) diagnosis. This study took place between January 2011 and April 2014 in Matlab, Bangladesh. MA with 3 years of medical training assigned COD to Matlab residents by reviewing the information collected using the Population Health Metrics Research Consortium (PHMRC) long-form VAI. Smart VA Analyze 1.2 automatically assigned COD using the same questionnaire. COD agreement and cause-specific mortality fractions (CSMFs) were compared for MA and Tariff. Of the 4969 verbal autopsy cases reviewed, 4328 were adults, 296 were children, and 345 were neonates. Cohen’s kappa was 0.38 (0.36, 0.40) for adults, 0.43 (0.38, 0.49) for children, and 0.27 (0.22, 0.33) for neonates. For adults, the top two COD for MA were stroke (29.6%) and ischemic heart diseases (IHD) (14.2%) and for Tariff these were stroke (32.0%) and IHD (14.0%). For children, the top two COD for MA were drowning (33.5%) and pneumonia (13.2%) and for Tariff these were also drowning (36.8%) and pneumonia (12.4%). For neonates, the top two COD for MA were birth asphyxia (41.2%) and meningitis/sepsis (22.3%) and for Tariff these were birth asphyxia (37.0%) and preterm delivery (30.9%). The CSMFs for Tariff and MA showed very close agreement across all age categories but some differences were observed for neonate preterm delivery and meningitis/sepsis. Given the known advantages of automated methods over physician certified verbal autopsy, the SmartVA software, incorporating the shortened VAI questionnaire and Tariff 2.0, could serve as a cost-effective alternative to Matlab MA to routinely collect and analyze verbal autopsy data in a HDSS to generate essential population level COD data for planning.

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.