Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=zgha20

Download by: [197.152.200.205] Date: 01 February 2017, At: 02:32

Global Health Action

ISSN: 1654-9716 (Print) 1654-9880 (Online) Journal homepage: http://www.tandfonline.com/loi/zgha20

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

Don de Savigny, Ian Riley, Daniel Chandramohan, Frank Odhiambo, ErinNichols, Sam Notzon, Carla AbouZahr, Raj Mitra, Daniel Cobos Muñoz, SonjaFirth, Nicolas Maire, Osman Sankoh, Gay Bronson, Philip Setel, Peter Byass,Robert Jakob, Ties Boerma & Alan D. Lopez

To cite this article: Don de Savigny, Ian Riley, Daniel Chandramohan, Frank Odhiambo, ErinNichols, Sam Notzon, Carla AbouZahr, Raj Mitra, Daniel Cobos Muñoz, Sonja Firth, NicolasMaire, Osman Sankoh, Gay Bronson, Philip Setel, Peter Byass, Robert Jakob, Ties Boerma &Alan D. Lopez (2017) Integrating community-based verbal autopsy into civil registration andvital statistics (CRVS): system-level considerations, Global Health Action, 10:1, 1272882, DOI:10.1080/16549716.2017.1272882

To link to this article: http://dx.doi.org/10.1080/16549716.2017.1272882

© 2017 The Author(s). Published by InformaUK Limited, trading as Taylor & FrancisGroup

Published online: 31 Jan 2017.

Submit your article to this journal

View related articles

View Crossmark data

STUDY DESIGN ARTICLE

Integrating community-based verbal autopsy into civil registration and vitalstatistics (CRVS): system-level considerationsDon de Savigny a,b,c, Ian Rileyc, Daniel Chandramohand, Frank Odhiamboe, Erin Nicholsf, Sam Notzonf,Carla AbouZahrg, Raj Mitrah, Daniel Cobos Muñoza,b, Sonja Firthc, Nicolas Mairea,b, Osman Sankohi,j,Gay Bronsonk, Philip Setelk, Peter Byassl,m, Robert Jakobn, Ties Boerman and Alan D. Lopezc

aDepartment of Epidemiology and Public Health, Swiss Tropical and Public Health Institute, Basel, Switzerland; bUniversity of Basel,Basel, Switzerland; cMelbourne School of Population and Global Health, University of Melbourne, Carlton, Australia; dDepartment ofDisease Control, London School of Hygiene and Tropical Medicine, London, UK; eAfrican Field Epidemiology Network (AFENET), Kisumu,Kenya; fNational Centre for Health Statistics, Centres for Disease Control and Prevention, Hyattsville, MD, USA; gCAZ Consulting, Geneva,Switzerland; hAfrica Centre for Statistics, United Nations Economic Commission for Africa, Addis Ababa, Ethiopia; iINDEPTH Network,Accra, Ghana; jSchool of Public Health, University of Witwatersrand, Johannesburg, South Africa; kVital Strategies, New York, NY, USA;lWHO Collaborating Centre for Verbal Autopsy, Umeå Centre for Global Health Research, Epidemiology and Global Health, Departmentof Public Health and Clinical Medicine, Umeå University, Umeå, Sweden; mMRC-Wits Rural Public Health and Health Transitions Unit(Agincourt), School of Public Health, University of Witwatersrand, Johannesburg, South Africa; nDepartment of Health Statistics andInformation Systems, World Health Organization, Geneva, Switzerland

ABSTRACTBackground: Reliable and representative cause of death (COD) statistics are essential toinform public health policy, respond to emerging health needs, and document progresstowards Sustainable Development Goals. However, less than one-third of deaths worldwideare 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, althoughimperfect, it is the best alternative in the absence of medical certification. There is extensiveexperience 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 algorithmsto analyse responses and estimate probable COD have increased the potential for VA to beroutinely applied in CRVS systems. However, a number of CRVS and health system integrationissues should be considered in planning, piloting and implementing a system-wide interven-tion such as VA. These include addressing the multiplicity of stakeholders and sub-systemsinvolved, integration with existing CRVS work processes and information flows, linking VAresults to civil registration records, information technology requirements and data qualityassurance.Conclusions: Integrating VA within CRVS systems is not simply a technical undertaking. It willhave profound system-wide effects that should be carefully considered when planning for aneffective implementation. This paper identifies and discusses the major system-level issuesand emerging practices, provides a planning checklist of system-level considerations andproposes an overview for how VA can be integrated into routine CRVS systems.

ARTICLE HISTORYReceived 27 September 2016Accepted 12 December 2016

RESPONSIBLE EDITORJennifer Stewart Williams,Umeå University, Sweden

KEYWORDSMortality surveillance; causeof death; systemsintegration; informationtechnology; healthinformation systems; processmapping; internationalclassification of disease;Sustainable DevelopmentGoals

Background

Real-time and accurate statistics on mortality and CODare essential for the development of national health andpopulation policies, and underpin the ability of coun-tries to respond to emerging health threats and epi-demics. The United Nations (UN) and World HealthOrganization (WHO) standards for mortality statisticsrequire the recording in the civil registry of all deaths,by age, gender, date and place of occurrence, along withmedical certification of the cause of death (COD) bytrained physicians according to the InternationalClassification of Diseases (ICD). The informationshould be regularly compiled into vital statistics

through national civil registration and vital statistics(CRVS) systems. Although there has been growingmomentum to strengthen CRVS systems in recentyears [1–4], it will still be some time before the routineCRVS systems currently present in low- and lower-middle-income countries will be able to ensure physi-cian-certified COD for a significant proportion ofdeaths. This is especially so for those that occur inrural or remote areas, owing to the lack of trainedphysicians and the fact that at least two-thirds – andoften more than 80% – of deaths occur at home.Summary COD statistics available in such countriesare often biased since they derive mainly from deaths

CONTACT Don de Savigny d.desavigny@unibas.ch Swiss Tropical and Public Health Institute, Socinstrasse 57, Basel 4002, Switzerland

*PB is Deputy Editor of Global Health Action, but played no part in the editorial process for this paper.

GLOBAL HEALTH ACTION, 2017VOL. 10, 1272882http://dx.doi.org/10.1080/16549716.2017.1272882

© 2017 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis GroupThis is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permitsunrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

in health facilities and in urban settings, which are notrepresentative of the experience of the general popula-tion [5]. As a result, CRVS systems are unable to gen-erate data that are sufficiently reliable andrepresentative for public health policy and planningpurposes [6]. This is of particular concern given thedynamics of the demographic and health transitionsoccurring around the world and the need to documentprogress towards the Sustainable Development Goals[7–9].

Drawing on the UN Sustainable DevelopmentGoals, the WHO targets for universal civil registra-tion of births and deaths, including COD, the WorldBank CRVS targets [10], the UN Economic andSocial Commission for Asia and the Pacific [11] andthe UN Economic Commission for Africa’s regionalprogrammes supported by ministerial level engage-ment [12], there are new goals to achieve significantimprovements in CRVS reporting by 2020, such that:

● 60% of deaths in a given year are continuouslynotified, registered and certified with keycharacteristics

● 80% of deaths in hospitals have COD reliablydetermined and officially certified in real time

● 50% of deaths in communities have probableCOD determined in real time, and collectionsystems designed in a representative way.

To address the last goal, over the past decade there hasbeen significant progress in the development of verbalautopsy (VA) methods to identify the probable COD inthe absence of a physician. VA involves an interview –after a culturally acceptable mourning period – with anappropriate next of kin, family member or care-giver of adeceased person to collect information on signs andsymptoms experienced by the deceased prior to death.This information is then analysed, either by a physicianor, more innovatively, by automated computer algo-rithms, to yield a probable COD that can be codedaccording to ICD standards [5,13–20].

There is extensive experience in the use of VA inresearch settings for determining the COD [21–23].This experience has led to advances in the VA question-naire design [19,24,25], in data capture onmobile devices[19,26], and in the use of computer algorithms for deter-mining and coding the probable COD [13,20,27–30].These advances make paper-based and mobile tablet-based VA at community level an increasingly availableand effective substitute for physician-certified COD insettings where a significant proportion of deaths occuroutside hospitals. The methodological developmentshave been accompanied by efforts to significantly reducethe time and cost of VA interviews and ascertainment ofCOD, thus greatly increasing the potential of automatedVA methods to be applied in routine CRVS systems to

cost-effectively monitor, in real time, COD patterns inpopulations [14,19,31,32].

VA is an imperfect tool for ascertaining COD, butit is the only alternative in the absence of medicalcertification [29,33–37]. For this reason, and in lightof the technological advances that make scaled appli-cation feasible, there is growing interest in applyingpaper-based or mobile tablet-based VA beyond theresearch setting, as a routine component of CRVSoperations where physician-certified COD reportingis currently rare or absent [38]. Community VA mayalso prove useful in hospital settings to establish CODfor cases that are ‘dead on arrival’, or shortly there-after, where post-mortem pathological examinationsare unavailable, or that are certified to non-specificcauses such as ‘cardiorespiratory failure’ that are oflittle value for public health decision making [39].

Despite many important technical advances, theintroduction and sustainable integration of VA intoCRVS systems is more than just a technical challenge.A number of systems integration and methodologicalissues will need to be carefully addressed during theplanning, piloting and implementation stages of anysuch endeavour. VA is nascent in its use within routineCRVS settings. This paper describes emerging practicesfrom which we can begin to establish an evidence basefor best practices and recommendations. The goal of thepaper is to describe how VA processes can be integratedinto a CRVS system; as such, it is also important tohighlight implications of VA on the civil registrationsystem. In this paper, we identify and discuss some ofthe more important system-wide issues. These fallbroadly under the domains of governance, design, opera-tions, human resources, financing, infrastructure, logis-tics, information technologies and data quality assurance.

Proposals

Governance issues

Owing to the multiplicity of actors and sub-systemsinvolved in the production of complete and reliablevital statistics and the importance of civil registrationfor policy making, CRVS systems engage an extraor-dinary range of critical stakeholders across multiplesectors. See Figure 1 for a typical example. Such com-plexity presents numerous governance challenges.

National CRVS Policy and Coordinating CommitteeA high-level National CRVS Policy and CoordinatingCommittee often best governs effective CRVS systems,because of the many sectors, government agencies andpartners concerned [40]. These typically include theministries of local government, health, justice and theinterior, as well as specialized agencies or bureaux suchas statistics, national identity (ID), e-government,national health insurance, and others concerned with

2 D. DE SAVIGNY ET AL.

various population registries (e.g. electoral commis-sions). The chair of this high-level committee is a cri-tical choice, and often reports directly to the highestlevel of government above-the-line ministries, such as aplanning commission in either the president’s or primeminister’s office, or is the minister responsible for civilregistration. This will enable the committee to haveconvening authority for those in other ministries andagencies. Where there is a national ID system, with orwithout biometric identifiers, it is beneficial that it isalso represented on the high-level committee. In theabsence of such cross-sectoral involvement, there arerisks that parallel data capture systems and conflictingvital statistics will occur, leading to inefficiencies and(potentially) undermining demand for civil registrationservices. The National CRVS Policy and CoordinatingCommittee could typically be concerned with:

(1) the coordination and development of theCRVS system, including establishing its long-term vision and funding strategy

(2) overseeing efforts to define and harmonizestakeholder functions, systems and datasources, ensuring standards for interoperabil-ity of information flow and databases

(3) keeping political, legal and regulatory environ-ments supportive of the CRVS systemthroughout its evolution, in particular as newtechnologies emerge

(4) monitoring opportunities and new develop-ments to improve the efficiency of the CRVSsystem, such as decisions for incorporating VAas an integral part of CRVS

(5) overseeing any interagency technical workinggroup(s) or sub-committees on CRVS thatundertake day-to-day work of implementingCRVS programmes and plans.

National sub-committee on mortality and cause ofdeathGiven the importance of accurate, real-time mortalitydata for public decision making, the National CRVSCoordinating Committee could convene a National

Sub-committee on Mortality and Cause of Death,responsible for developing and overseeing a strategyfor ensuring that nationally representative mortalitydata are available to, and used by, decision makers forplanning. Such a sub-committee could be under theauspices of the Ministry of Health. Among othertasks, this mortality Sub-committee or equivalentwould advocate for the need to monitor COD anddevelop materials and plans to support this objectivefor both deaths in health facilities and deaths outsidehealth facilities, including deaths due to unnaturalcauses. Advocacy may entail explaining the need tostay abreast of epidemiological change and transitionand to measure the mortality-specific indicators thatcould be monitored in response to SustainableDevelopment Goals and targets for health [7]. TheSub-committee would also advocate for the impor-tance of real-time tracking of mortality in order tocontinuously update national population registers, IDdatabases and electoral rolls.

The Sub-committee would probably establish aTask Force on CRVS-VA or similar mechanism toaddress the particular challenges involved in generat-ing information on COD using VA. Typically, theCRVS-VA Task Force would deal with the legal,design, operational, technical and systems issuesthat could be addressed in preparing plans forCRVS VA integration, including:

(1) commissioning the detailed process mappingof how death and COD are captured at healthfacilities and outside health facilities, and usingthese to conceptualize how VA could beintegrated

(2) addressing the legal, regulatory and ethicalissues that are likely to arise in relation tonon-physician ascertainment of COD

(3) making recommendations to the NationalCRVS Planning and Coordinating Committeevia the Mortality Sub-Committee regardingthe legal framework that is likely to be mostappropriate to support VA implementation inthe country

Figure 1. Typical civil registration and vital statistics (CRVS) stakeholders at national level. NGOs = non-governmentalorganizations; FBOs = faith-based organizations.

GLOBAL HEALTH ACTION 3

(4) guiding the detailed design and piloting ofintegrated VA in CRVS, considering the finan-cing, human resources, logistics, quality con-trol and information technology (IT) needs,including resolving technical issues, e.g. choiceof questionnaire, algorithms and samplingstrategy

(5) developing recommendations to the NationalCRVS Planning and Coordinating Committeevia the Mortality Sub-committee for the planand budget for integrating VA into the CRVSsystem

(6) monitoring and evaluating the implementationand scale-up of VA in CRVS

(7) addressing the question of how VA-ascer-tained COD data from communities will beanalysed along with physician-certified hospi-tal deaths to permit the CRVS to provide themost comprehensive picture of CODs in thecountry.

Business caseCountries may also want to commission the prepara-tion of a business case to justify using VA as amethod to ascertain the most likely underlyingCOD in selected populations. Such a business casemay point out the substantial emerging advantages ofdigital data collection for automated VA on mobiletablet devices instead of relying on paper for inter-views at household level. Electronic data capture athousehold level: (1) vastly facilitates the work of theVA interviewer owing to automated skip logic navi-gation through the complex questionnaire; (2)reduces error rates in data collection owing to controlof valid data entry for respective variables; (3) reducesthe cost of manual double data entry from paper; and(4) eliminates transcription errors when transcribingdata from paper to computer databases. There arealso significant savings in both time and cost bytaking advantage of computer coding in place ofphysician coding of the VA results to determine theunderlying COD [15,41,42]. The business case willneed to consider higher level benefits from integrat-ing VA in CRVS, such as strengthening communitydeath notification as part of CRVS, increasing thecompleteness of both fact of death and COD data inCRVS, and the use of community-based mortalitystatistics to better target programme implementationand funds.

Design and operational issues

Integrating processesEmergent experience is showing that preparingdetailed process maps is a valuable early step in

designing the integration of VA into CRVS systems.1

These maps depict all major steps, processes andactivities related to notifying, registering and certify-ing deaths in the CRVS system. Process mappingdescribes the major actors or agencies involved,from the family through the local authorities, localhealth system, civil registry system, eventually tonational databases and vital statistics. The processmaps also help to establish efficient business func-tions, reporting forms, standard operating proce-dures, requirements, rules and information flowsconcerned with capturing the mortality event. These‘blueprints’ of the CRVS architecture allow all stake-holders to have a common understanding of thecurrent system ‘as-is’. These maps are helpful toenable effective participation in designing how theCRVS and VA steps (declaration, notification, regis-tration, certification, VA interview, VA analysis, link-ing COD and fact of death, validation, qualitycontrol, data transfer, production and disseminationof vital statistics and reports, etc.) are to be integratedinto the CRVS system. In addition, process maps canbe useful in supporting the digitization of CRVS(http://www.crvs-dgb.org/en/) and in ensuring theinteroperability of data systems, for example, whenintegrating VA information flows into both the CRVSsystem and digital health information systems such asDHIS2 (https://www.dhis2.org/). The process mapsfacilitate the identification of pathways to enable thetransition from ‘as-is’ CRVS processes design to ‘as-desired’ processes that can accommodate the additionof routine VA and are key to undertaking a mean-ingful legal review, which is an important step inplanning the addition of VA to CRVS.

SamplingIdeally, VA could be conducted to estimate the CODin all out-of-facility deaths where there is no physi-cian certification. Costs savings can be achieved byconducting VA on a representative sample of alldeaths or in a representative selection of registrationadministrative areas [43–45]. Some countries withlarge populations conduct a VA only on samples ofdeaths in communities by Sample RegistrationSystems (SRS) or Sample Vital Registration withVerbal Autopsy (SAVVY) to determine statisticallyuseful cause-specific mortality fractions at populationlevel (e.g. China, India, Tanzania, Indonesia) or inhealth facilities to improve the quality of data fromhealth facility deaths (e.g. Brazil, Thailand). At pre-sent, the SRS and SAVVY sampling units (e.g. India,Tanzania) are census enumeration areas. However,for VA to be better integrated into CRVS, the sam-pling units could be the registration administrativeunits. When sampling is done, sample size and

1Unpublished observations from the Bloomberg Data for Health Initiative in 15 countries.

4 D. DE SAVIGNY ET AL.

designs need to be considered carefully to avoid poli-tical concerns, ensure representativeness, and provideaffordable and cost-effective scenarios. Other chal-lenges faced by sample VA approaches include ensur-ing that the deaths and associated CODs are welllinked in the CRVS, and that the sampling schemedoes not limit provision of sub-national data onCOD, which is important in countries that havelarge sub-national differences.

Community operational considerationsThe following is a brief overview of how VA couldbe integrated as part of a typical CRVS system fordeaths in the community (Figures 2–5). One key

principle proposed here is that the notification ofthe fact of death to the civil registration systemcould occur before the VA is scheduled and con-ducted. The VA could then be conducted on a deaththat already has been assigned an official notifica-tion, registry number or unique ID. This reducespotential problems in later data linkage and therisk of double counting. It also enables data linkageto be conducted to estimate possible bias and gaps innotification and registration. The processes fordeclaring or notifying deaths, particularly those inthe community, become the starting point for thenew VA processes. This is likely to require reconsi-deration of how vital event notification is conducted

Figure 2. Potential verbal autopsy (VA) processes in a civil registration and vital statistics (CRVS) system.

Figure 3. Example of a monitoring and evaluation cycle for verbal autopsy in civil registration and vital statistics (CRVS).VA = verbal autopsy.

GLOBAL HEALTH ACTION 5

at community level. This step is currently a passivestep in most CRVS systems and it will need tobecome an active step.

The critical challenges for the design and opera-tionalization of VA in CRVS are therefore: (1) theimmediate identification of death events as they

0

20

40

60

80

100

Percent of total

deaths

Deaths notified by

health facilities,

burial authorities,

police, families

Deaths officially

registered by

local/national

registry

Deaths assigned a

cause of death

Deaths with cause

of death included in

national vital

statistics system

Figure 4. Potential data loss from notification of deaths to production of mortality statistics. Hypothetical scenario common tolow-income countries.

Figure 5. Generic processes for integrating community verbal autopsy in a hypothetical civil registration and vital statistics(CRVS) setting for community deaths. VA = verbal autopsy; COD = cause of death; CR = civil registry; CHW = community healthworker; MoH = Ministry of Health; GPRS = General Packet Radio Service.

6 D. DE SAVIGNY ET AL.

occur in the population (usually in disadvantaged,remote rural communities); and (2) ensuring thatthe death, with its associated fact of death details[ID (where relevant), name, age, gender, address,location, date of death, etc.] are already registered inthe CRVS system before the VA is conducted (usuallynot before 30 days or later than 3 months from thedate of death, although up to 12 months may still beacceptable) [46].

NotificationMost CRVS systems are passive, relying on families tocome forward to register vital events and placing theonus on the family to report events, which results inlow notification and registration rates. Notifications ofdeaths to the CRVS system can be increased through avariety of active case-finding methods, including villageauthorities or chiefs, community key informants, villagevolunteers, community outreach workers, or surveys orsurveillance systems. Such methods are, however, likelyto be costly and difficult to sustain in a routine system. Insome countries, short message services (SMS) on mobilephones to notify events from the community to theregistration authority or health facilities have been tried[47,48]. Hybrid approaches that use a combination ofreporting through community-based entities such ashealth facilities, burial authorities or local chiefs, alongwith verification and official registration by family mem-bers, may be more cost-effective. Implementing such amodel will be likely to require changes to the existingCRVS system. Where reporting of all deaths is requiredand enforced, however, these notifications can also beused to trigger scheduling of the VA and official registra-tion of the death. Events could also be notified by proxyprocedures such as reviewing the issuance of burial per-mits by local authorities where this has high coverage andenforcement. The VA questionnaire, either on paper oron a mobile tablet computing device, can then be pre-populated with the minimum necessary notified fact ofdeath information, such as name, age, gender, date ofdeath, location, and notification serial number or civilregister or personal ID number for the deceased, beforethe trained VA interviewer is dispatched to the decea-sed’s household.

VA interviewThe paper- or mobile tablet-based VA questionnaire isusually conducted in a place, at or near the household ofthe deceased, where those family members most famil-iar with the circumstances of the death are available andcomfortable to respond freely with proper considera-tion to locally appropriate protection of confidentiality.The interview responses are best captured digitally in amobile tablet computer to ensure quality, increase theefficiency of data collection and reduce the cost of dataentry. The interviewer should not be tasked with inter-preting, providing or counselling on the COD. The

family could be referred to the appropriate local healthfacility where they can be provided at a later date withan interpreted COD from the VA on an opt-in basis, atleast to broad-cause level, with counselling. However,the VA interviewer will need to be well trained in anapproach to managing the interview process that isethically and culturally appropriate, including familiar-ity with principles of grief counselling for self-care andfor managing the emotionally sensitive interview pro-cess [49,50]. At the end of the interview, the contents ofthe completed VA interview questionnaire should beautomatically encrypted and transmitted securely fromthe VA tablet to a dedicated server at central level orotherwise linked to the CRVS database.

VA interpretationExternal to the mobile tablet-based VA questionnaireat the data server level, the probable underlying COD(and where feasible the associated ICD-10 code) canbe imputed through an appropriate validated algo-rithm. This is likely to be the most feasible methodfor routinely estimating population COD structure,which can be more accurately specified than an indi-vidual diagnosis owing to compensating diagnosticerrors standards [5,13–20].

To comply with legal requirements in certifying theindividual COD, physician certification may still berequired in some countries for CODs that are addedto CRVS civil registers [33]. In such scenarios, a physi-cian could be asked to read and consider the VA inter-view results and sign off, or modify, the computer-generated cause. Even where a physician is not requiredby law to certify the COD, the potential implications ofautomated VA results applied to individual records inthe CRVS, for example with regard to insurance claimsor inheritance, could be considered. Gouda et al. discussthese and other ethical issues around VA [51]. Thus,legal and regulatory review may be required as an earlystep in considering VA integration into CRVS. Thecause, with its ICD-10 code, could then be linked andadded to the CRVS civil record according to the registrynumber and transmitted to both the vital statisticsauthority and the Ministry of Health as required. Acheck boxwith other detail could be added to the recordto indicate that the cause was derived by a VA algorithmthat is clearly specified by name and version. SeeFigure 2 for an overview of the approach, and Figure 5for a more detailed generic flow diagram of VA integra-tion into CRVS.

Human resource issues

It is important to have a profile of the existing CRVShuman resources and needs. Introducing communityVA will require new functions to be added to existinghuman resource cadres or new cadres to be recruited,trained, supplied, supervised and possibly

GLOBAL HEALTH ACTION 7

remunerated. Those at the front line who identifydeaths in the community can be community volun-teers (key informants) who would not need financialcompensation, or paid community outreach workerswhose job descriptions can be expanded to includenotification of births and deaths. Including births inthe notification process is important since high-mor-tality rural populations have relatively high neonatalmortality rates that are often underestimated becausepregnancies and births go unrecorded andunregistered.

Job descriptions, training plans and training mate-rials need to be developed for the new functions orcadres. It is expected that the addition of mobile com-munity-based VA will greatly increase the number ofdeaths, and CODs, that enter the CRVS system, andthis will increase the workload of the existing CRVSsystem at higher levels. This consideration needs to beestimated and planned for to ensure that the increasedworkload can be absorbed and managed.

VA interviewers can be selected from local com-munity members who have secondary school qualifi-cations and could be working in some officialcapacity within the country’s government, as theyare facilitating an official function of the country’sCRVS system. Owing to the relatively infrequent nat-ure of mortality in geographic areas accessible toindividual community interviewers, VA interviewwork is not generally a full-time occupation, andthese tasks can be added to position descriptions ofother community outreach workers who may beemployed by local authorities. While some healthexpertise background may be helpful for VA inter-viewers, it is not necessary that highly trained health-care providers (e.g. physicians, clinical/medical offi-cers or nurses) serve as VA interviewers, as trainedlay interviewers perform this work adequately inresearch settings [52].

Countries that already host research systems con-ducting VA (e.g. Health and DemographicSurveillance Systems) will have an advantage bybeing able to tap into these skills to help developthe human resource plan, assist in training and qual-ity control [53], and participate in respective commit-tees (see ‘Governance issues’, above).

In Table 1 we outline an approximation of thehuman resource cadres and numbers required. Thisassumes that a country has a national crude deathrate of 6–7 per 1000, but that the community VAwould likely to be conducted for rural deaths in moredisadvantaged populations where the crude death ratemay be 9 or more per 1000. It assumes that VAinterviewers will be community based and that thefrequency of deaths in their catchment area will berelatively low (approximately one death per week perinterviewer), so that their work will not be full time.An alternative would be to have fewer VA

interviewers with larger catchment areas. Thiswould need to be traded off against the higher costsand greater difficulty for such workers to find andreach households, and the need to provide them withtransport. There is the additional trade-off of havinga sufficient frequency of VA work to maintain theirskills in conducting VA, while not overloading andcrowding out their other functions.

Financing issues

Because there is limited experience in integrating com-munity VA into CRVS, it is difficult to estimate the costsof routine CRVSVA. In sample registration systems withVA, such as in India [54], Indonesia [55], Tanzania [43]and Ethiopia [56], costs per VA conducted are estimatedbetween 5 and 15 USD per case. In some cultures, con-dolence fees or gifts may be expected when visitinghouseholds after a death, and this may need to be con-sidered. Formal costing studies in these CRVS-VA pio-neering countries will be valuable, as other countries willwant to know the financial and economic costs of addingVA to their CRVS system. As countries gather moreexperience at scale, more costing studies are requiredand a CRVS costing tool needs to be developed toguide country decision making. Countries will want toknow the total system cost versus the added value ofinformation gained. We estimate that the main costdrivers will be training and retraining, IT infrastructure,remuneration of VA interviewers and supervising VAinterviewers.

Infrastructure and logistics issues

The existing infrastructure and its gaps should be wellknown and mapped. The process mapping of CRVSalone, and CRVS plus VA, developed at the design

Table 1. Suggestions for human resource cadres needed tooperate verbal autopsy (VA) in a civil registration and vitalstatistics (CRVS) system.

Cadre

Number/1,000,000population (for 9000 deaths

per year) Level

Community keyinformants fornotification

1000 Unpaidvolunteer,part time

or communityoutreach or healthworkers

1000 Paid bygovernment,part time

VA interviewers 250 Paid, part timeor per event

VA regionalsupervisors

25 Paid, part time

VA physician codersor signers –optional

6 Paid per event,part time

VA IT, logistics andhelp desk

2 per country Paid, full time

VA analyst 1 per country Paid, full timeVA nationalcoordinator

1 per country Paid, part time

8 D. DE SAVIGNY ET AL.

stage is valuable for understanding these gaps ininfrastructure, logistics and IT. This is particularlyso when adding new cadres to operate the VA system,since they will have office and infrastructure needsadditional to the current CRVS infrastructure.

National Identification agencies in low-incomecountries are currently building physical and IT infra-structure. CRVS agencies need to be cognizant of thesedevelopments and seek synergies, particularly for theplanned introduction of VA into CRVS systems.

Infrastructure for CRVS and VA in particular willincreasingly depend on mobile data and wireless net-work capacity. These also need to be mapped in orderto forecast what will be needed to support currentand developing infrastructure.

Data and information technology systems

Increasingly, VA for CRVS will be a digital rather thanpaper-based undertaking from the point of data captureusing tablet or smart-phone devices. VA questionnaireinstruments need to be adapted and translated in a waythat captures the correct medical concepts in locallyunderstood terminology, yet does not compromise theinstrument’s established performance. Detailed gui-dance for interviewers and translators is currentlybeing developed by the WHO [57]. Cognitive testingto ensure that respondents understand the questions asthey are intended is also recommended [58].

Some of the instrumentation challenges includethe fact that tablet interview script language and thelanguage in which the interviewer interacts withrespondents may not be the same, or the fact thatsome countries use different calendars.

Devices will need to link to servers, either locally orcentrally, and the flow of data needs to be designed andmanaged from an enterprise architecture (EA) perspec-tive. EA is the industry standard for designing interoper-ability in information and communication applications.TheUNEconomicCommission forAfrica has developedguidance specifically for how to employ EA for CRVSdigitization [59]. Open-source software such as OpenData Kit (ODK) [60] is ideally suited to developing thenecessary software solutions for data collection and datatransmission. However, e-governance, data security, con-fidentiality and data encryption issues should beaddressed. Some countries may want to feed results intoor via their digital health information infrastructure, suchas DHIS2 [61]. The IT systems could have dedicatedcustomized dashboards with controlled level access forvarious managers and implementers. The IT arrange-ments and specifications for such servers and commu-nications are available [62]. Aggregate results will need tobe managed in a timely fashion. Skills for these functionsare not often available, hence the inclusion of dedicatedIT and logisticians in human resources. In locationswhere mobile and wireless networks are not sufficiently

reliable, an off-linemethod, involving theVA interviewerbringing their tablet to a health centre and downloadingthe VA manually through a USB connection to a com-puter, is a less IT-intensive solution.

Access to dataThere are important legal and ethical dimensions relatedto COD data in CRVS systems [51]. VA will be anadditional source of such data and should be consideredsensitive, personalmedical data, accessible by the smallestnecessary number of agencies and their agents for clearlydefined purposes. For this reason, a legal and regulatoryreview (See Table 2) is an important early step in plan-ning and designing theVA integration. Aswithmedically

Table 2. Planning checklist of system-level considerations forplanning verbal autopsy (VA) integration in civil registrationand vital statistics (CRVS) systems.Ↄ Ensure that a high-level National CRVS Policy and CoordinationCommittee is in operation;

Ↄ Ensure that the relevant authorities, agencies or ministries for civilregistration, statistics, local government and health and areengaged collectively for CRVS;

Ↄ Ensure that a Comprehensive CRVS Assessment has been conductedin the past 4 years and has been used to develop a national CRVSvision and strategy or is being planned;

Ↄ Set up a National Sub-committee on Mortality and Cause of Death;Ↄ Establish a task force for VA implementation reporting to theNational Sub-committee on Mortality and Cause of Death;

Ↄ Ensure that detailed process mapping of CRVS processes forregistration of death in health facilities and death in communitieshas been done as part of the comprehensive assessment, and if not,prepare such process maps;

Ↄ With all relevant stakeholders, use these process maps ofnotification and registration processes of death in the community asa base to develop the plan of implementation for how VA would beintegrated into a modified set of processes;

Ↄ Prepare an investment case to justify using VA as a method toincrease notification and registration of deaths and ascertainunderlying cause of death;

Ↄ Consider a legal and regulatory review of the implications of VA inCRVS as an early step in the plan;

Ↄ Apply the enterprise architecture Digital CRVS Guidebook to assessthe additional IT needs (http://www.crvs-dgb.org/en/);

Ↄ Map the existing CRVS and DHIS2 IT infrastructure and its gaps;Ↄ Seek synergies with existing IT for population registration efforts(i.e. National Identification agencies);

Ↄ Determine how mobile tablets will be supported, maintained andsecurely transmit/receive data (wireless, General Packet RadioService, etc.);

Ↄ Design data flow and quality assurance mechanisms;Ↄ Ensure that e-governance, interoperability, data security,confidentiality and data encryption issues addressed;

Ↄ Decide how VA-coded deaths will be distinguished from medicallycertified deaths in aggregate databases;

Ↄ Decide on scale (sample system or full coverage) and phasedintroduction;

Ↄ Use a VA costing tool to develop the start-up and annualizedbudgets;

Ↄ Prepare a profile of the existing CRVS human resources and needs;Ↄ Develop job descriptions, training plans and training materials fornew and revised positions;

Ↄ Plan for an increase in the workload for existing staff;Ↄ Consider adding VA functions to existing position descriptions ofcommunity workers;

Ↄ Develop a training programme for Master Trainers, Trainer ofTrainers, and training of VA supervisors, interviewers and analysts;

Ↄ Prepare a monitoring and evaluation plan for the new VA processes,including the use of VA costing tools to document costs and anindependent quality assurance mechanism;

Ↄ Work with stakeholders to develop a learning platform for phasedintroduction and assemble necessary funding.

GLOBAL HEALTH ACTION 9

certified CODs from health facilities being captured inCRVS systems, this review could specify levels of accessto VA COD data on a need-to-know basis, including theneed for encryption of data during transmission to theservers, and server security and access protocols. Toprotect confidentiality, data can be anonymized, mini-mizing access to personal identifying information andlinking the COD via unique identifiers only at levels thatneed to know. Population-level estimation of COD doesnot require personal identifying information, which canbe stripped before COD data are made available to thoseagencies conducting population-level analyses. The algo-rithms that estimate the VACOD are not resident on theVA interviewer’s tablet computer and so the cause cannotbe known or communicated to the family at that point.But in some instances, the health system may be obli-gated to inform next of kin of the likely COD from theVA on an opt-in basis to satisfy family demands. If so,provisions should be made for ensuring secure transmis-sion of the identity-linked result from the VA analyticserver back to the appropriate health-facility/health-worker level, where confidential counselling can be pro-vided. Although challenging, it is important to makeprovision to maximize the chances that the informationis managed securely.

Quality assurance

Monitoring and evaluationIntegration of VA into CRVS is a complex undertak-ing. While the potential benefits are great, the successof such a venture relies on a variety of stakeholders,each with their own incentives and impediments toperforming their particular role in the process. It isalso reliant on the communication mechanisms, ITprocesses and infrastructure used to facilitate theinformation flow and eventual use of the COD infor-mation for population health statistics. Critical to thesustainability and effectiveness of such a system is theneed to introduce changes in a phased manner, withrobust monitoring and evaluation processes leading toadaptations and continuous improvements. Usually,that would involve testing during a small-scale pilotand a larger scale implementation before planning fora roll-out across the country. Figure 3 describes anexample of a monitoring and evaluation cycle forintegrating VA into CRVS.

This figure illustrates a number of stages in the VAcycle, from the effectiveness of data collection (‘areVAs being conducted on all notified deaths in thepopulation or in the designated sample?’) to the pub-lication and dissemination of information arising fromVA (‘is the information from VA guiding policy andhealth programmes?’). It will be important to monitoreach of these stages and periodically evaluate, tounderstand where any problems may be arising. Asmall number of key indicators will need to be

identified at each stage, and these indicators willneed to be monitored over time and problem solvingemployed to understand the root cause of any issues.Critically, strategies to alleviate any identified issuesneed to be implemented and continuously monitoredto ensure that the strategy is working. Some stagesrelate more to processes (‘is something happening asit should be?’). Others, including data quality assur-ance, described in more detail in the following section,are more concerned with the quality of the activity (‘iswhat is happening valid?’). The different stages are alsohighly dependent on each other. Therefore, a periodicreview of the appropriateness of the entire monitoringcycle, with required changes to activities or indicatorsbeing monitored, could be performed. This may hap-pen between the pilot and the first phase testing, andbetween each significant scale-up in the process. Thebringing together of all elements of the monitoringand evaluation cycle will be important, given thatresponsibilities for monitoring are likely to be acrossa number of different stakeholders. This broader eva-luation could be conducted under the auspices of theNational Sub-committee on Mortality and Cause ofDeath and the Task Force on CRVS-VA althoughother options are possible. VA is a diagnostic tooland therefore its performance will vary based onlocal epidemiology and the prevalence of specificcauses in the mortality cause mix. Locally applicablevalidation studies could help to determine the externalvalidity of VA results; however, in the absence oflocally relevant gold-standard data sets, practicalmethods for doing this are still in need of development[6,63].

Data quality assessment and assuranceStrategies for monitoring data quality will need to bereassessed at appropriate stages in the CRVS VA pro-cess maps. It is often unclear in CRVS systems whereand when data monitoring and quality assuranceoccur, as information and data move through thesystem to become vital statistics. For independence,this may require the engagement of independent sup-port from academia and research. As a CRVS VAsystem will rely on a strong death notification processand will have implications on the vital statisticsreported, the CRVS VA system will need audit trailsthat allow quality assurance of individual and aggre-gate data.

After a death has been notified and officially regis-tered, the information should be processed and for-warded to the agency responsible for compiling data(microdata) from the individual records and for analys-ing the statistics (aggregated data). This is usually theresponsibility of the National Statistics Office. Dataquality assurance is needed at each stage. At the initialdata collection stage, when a death is notified and VA isconducted, the focus of quality assurance is on

10 D. DE SAVIGNY ET AL.

microdata; that is, the accuracy and completeness of theinformation contained in each individual record.During the later stages of data compilation, analysisand packaging through to dissemination, reportingand use, quality-assessment processes will primarilyfocus instead on the accuracy, consistency and plausi-bility of the aggregated data; that is, the total counts ofdeaths by age, gender and COD.

A frequently observed problem is the decay of infor-mation coverage at various stages along the continuumfrom the occurrence of an event to its inclusion in thenational vital statistics system (Figure 4). Because regis-tration and the production of vital statistics are often theresponsibilities of different government agencies, thereis a potential for inconsistencies to arise. Any failure toensure that all registered vital events are reflected invital statistics is a major concern in terms of the overallquality of CRVS systems.

Different types of errors and inconsistencies canoccur between the notification of a death, its officialregistration and its inclusion in vital statistics. Errorscan arise from both under- and over-reporting. Over-reporting can occur when a vital event is noted as adelayed registration without checks being carried outto ensure that it has not already been previouslyregistered in another district. In other cases,re-registration occurs because the initial registrationpapers cannot be traced, possibly because registrationoccurred in another area and there is no centralizedstorage of records, or because of archiving problems.Avoiding duplication of registration requires a well-managed computerized central register and effectivearchiving, checking and retrieval procedures.

Under-reporting of deaths is far more commonthan over-reporting. This will be reflected in datashowing that the expected annual number of registra-tions is higher than the figure presented in vitalstatistics reports. Under-reporting can result because:

● deaths are not notified to the health authorities,burial authorities or other officials, and there islack of enforcement of public health measuresfor the safe disposal of bodies

● deaths are notified but not officially registered bythe family owing to distance, financial or socio-cultural barriers; in some countries, death regis-tration is not free which, in itself, is a barrier

● there is delayed or incomplete reporting ortransfer of the information from civil registra-tion agencies to the vital statistics system

● delayed registrations are treated differently fromtimely registrations, and not transferred to thevital statistics system

● there is a failure to transmit information oncoronial cases or police investigations oncethey have been resolved and returned to thecivil registration system.

Under-reporting can be reduced by introducing anaudit system clearly outlining responsibilities for log-ging all events sent to the vital statistics system, and forrecording what is received. Particular challenges arise inrelation to the registration of deaths due to accidents orsuspicious COD. In most countries, special proceduresare in place for registering and certifying accidental ornon-natural deaths such as homicides or suicides.Doctors or police called to attend these deaths usuallyreport them to a coroner or judicial authority, and aninvestigation is carried out to determine the preciseCOD. In practice, such deaths are frequently under-reported. Stillbirths and neonatal deaths also pose aparticular challenge and are chronically under-reportedeven though they are particularly important outcomesto document. Modern VA questionnaires include thetechnical possibility of processing information on still-births but, in contrast to neonatal deaths, there is noclear guidance about identifying and including them invital statistics.

Conclusions

Reliable information on the number of deaths by age,gender and cause is the cornerstone of an effectivehealth information system and national statistics sys-tem. However, less than one-third of deaths worldwideare assigned a cause, with the most impoverishednations having the least information while arguablyneeding it the most. The conventional standard fordetermining COD is medical certification by a physi-cian familiar with the case and trained in the certifica-tion rules and procedures of the ICD. Many deaths inlow- and lower-middle-income countries occur in set-tings without health professionals present or easilyaccessible to certify the COD. VA is an imperfect toolfor ascertaining COD, but the best alternative in theabsence of medical certification.

To date, VA has been primarily used in researchenvironments. There are only a few examples whereVA has been implemented in a routine manner on alarge scale (e.g. Brazil). Recent methodological develop-ments suggest that VA is ready for wider nationalapplication in routine CRVS systems. Integrating VAwithin the CRVS system to capture representative dataon a continuous basis is not a trivial undertaking, andthere are many unknowns. The successful integration ofVA will need to consider the potentially far-reachingsystem-wide effects in areas such as legal frameworks,medicolegal–ethics issues, human resources, financing,logistics and information systems. Understanding therequirements and potential (positive and negative) sys-tem effects of integrating VA is an essential element foran effective implementation. In these early stages and aspart of designing any pilot phase, countries need toconsider what integrating VA in their CRVS systementails and how the resulting mortality data will be

GLOBAL HEALTH ACTION 11

used. Questions about the type of VA method, changesin legislation or regulations, acceptability and feasibilityof different approaches and sustainability of the systemin the long run could be addressed in early stages of theplanning process.

This paper provides an important system-level per-spective and overview for how VA can be integratedinto routine CRVS systems. It also provides a practicalsummary checklist of the main points and systemsissues to keep in mind when developing the CRVS-VA implementation plan. The checklist identifies onepotential set of actions for emerging practices.

Acknowledgments

This paper is based on initial discussions among 15 of theauthors at an international workshop on verbal autopsy meth-ods held at the Swiss Tropical and Public Health Institute,Basel, Switzerland, 16–19 November 2015, sponsored by theBloomberg Philanthropies’ Data for Health Initiative.

Author contributions

DDS conceptualized the paper and prepared the first draftbased on workshop discussions with IR, DCM, FO, EN,SN, CAZ, DCM, SF, NM, GB, PS, PB, RJ and ADL. Allauthors contributed towards editing the manuscript andhave read and approved the final version.

Disclosure statement

No potential conflict of interest was reported by theauthors.

Ethics and consent

None.

Funding information

None.

Paper context

More than two-thirds of deaths in low- and lower-middle-income countries occur outside health facilities and are neitherregistered in national civil registration and vital statistics(CRVS) systems nor assigned a physician-certified COD. Therates and patterns of cause-specific mortality in such countriesare changing rapidly owing to epidemiological and healthtransitions. There is renewed interest globally, regionally andnationally to strengthen CRVS systems to provide such essen-tial evidence to steer national health policies. The means tonotify and determine the COD at community level in low-income countries has only been available in research settingscapable of implementing population surveillance with verbalautopsy (VA). However, recent developments and innovationsin VA methods with the use of tablet computers, mobilecommunications and diagnostic algorithms obviating theneed for physicians to attend every death, open the possibilityto integrate VA in routine CRVS systems in such countries.

This paper identifies and discusses the systems integrationconsiderations needed to take such decisions and proposesthe systemic approaches necessary to conceptualize, design,implement and monitor the introduction of VA in CRVS.

ORCID

Don de Savigny http://orcid.org/0000-0002-6903-4722

References

[1] AbouZahr C, de Savigny D, Mikkelsen L, et al.Towards universal civil registration and vital statisticssystems: the time is now. Lancet. 2015;386:1407–1418.

[2] AbouZahr C, de Savigny D, Mikkelsen L, et al. Civilregistration and vital statistics: progress in the datarevolution for counting and accountability. Lancet.2015;386:1373–1385.

[3] Mikkelsen L, Phillips DE, AbouZahr C, et al. A globalassessment of civil registration and vital statistics sys-tems: monitoring data quality and progress. Lancet.2015;386:1395–1406.

[4] Phillips DE, AbouZahr C, Lopez AD, et al. Are wellfunctioning civil registration and vital statistics sys-tems associated with better health outcomes? Lancet.2015;386:1386–1394.

[5] Murray CJ, Lopez AD, Barofsky JT, et al. Estimatingpopulation cause-specific mortality fractions from in-hospital mortality: validation of a new method. PLoSMed. 2007;4:e326.

[6] Joubert J, Bradshaw D, Kabudula C, et al. Record-linkage comparison of verbal autopsy and routinecivil registration death certification in rural north-east South Africa: 2006–09. Int J Epidemiol.2014;43:1945–1958.

[7] World Health Organization. Towards a monitoringframework with targets and indicators for the healthgoals of the post-2015 sustainable development goals.Geneva: World Health Organization; 2015.

[8] Sankoh O, Byass P. Cause-specific mortality atINDEPTH health and demographic surveillance sys-tem sites in Africa and Asia: concluding synthesis.Glob Health Action. 2014;7:25590.

[9] Byass P, de Savigny D, Lopez AD. Essential evidencefor guiding health system priorities and policies:anticipating epidemiological transition in Africa.Glob Health Action. 2014;7:23359.

[10] World Bank, World Health Organization. Global civilregistration and vital statistics scaling up investmentplan 2015–2024. Washington (DC): World Bank;2014.

[11] ESCAP Statistics Division. Regional strategic plan forthe improvement of civil registration and vital statisticsin Asia and the Pacific (Draft). Bangkok: ESCAP; 2012.

[12] Economic Commission for Africa. Reforming andimproving civil registration and vital statistics systemsin Africa. Regional Medium-Term Plan: 2010-2015.Durban: Republic of South Africa: UNECA; 2012.

[13] McCormick TH, Li Z, Calvert C, et al. Probabilisticcause-of-death assignment using verbal autopsies.ArXiv. 2015;1:1411.304.

[14] Byass P, Chandramohan D, Clark SJ, et al.Strengthening standardised interpretation of verbalautopsy data: the new InterVA-4 tool. Glob HealthAction. 2012;5:19281.

12 D. DE SAVIGNY ET AL.

[15] Murray CJL, Lozano R, Flaxman AD, et al. Usingverbal autopsy to measure causes of death: the com-parative performance of existing methods. BMC Med.2014;5:12.

[16] Garenne M. Prospects for automated diagnosis ofverbal autopsies. BMC Med. 2014;12:18.

[17] AbouZahr C. Verbal autopsy: who needs it? PopulHealth Metr. 2011;9:19.

[18] Oti SO, Kyobutungi C. Verbal autopsy interpretation:a comparative analysis of the InterVA model versusphysician review in determining causes of death in theNairobi DSS. Popul Health Metr. 2010;8:21.

[19] Serina P, Riley I, Stewart A, et al. A shortened verbalautopsy instrument for use in routine mortality sur-veillance systems. BMC Med. 2015;13:302.

[20] Serina P, Riley I, Stewart A, et al. Improving perfor-mance of the tariff method for assigning causes ofdeath to verbal autopsies. BMC Med. 2015;13:291.

[21] King G, Lu Y, Shibuya K. Designing verbal autopsystudies. Popul Health Metr. 2010;8:19.

[22] Sankoh O, Byass P. The INDEPTH network: fillingvital gaps in global epidemiology. Int J Epidemiol.2012;41:579–588.

[23] Streatfield PK, Khan WA, Bhuiya A, et al. Cause-specific mortality in Africa and Asia: evidence fromINDEPTH health and demographic surveillance sys-tem sites. Glob Health Action. 2014;7:25362.

[24] Leitao J, Chandramohan D, Byass P, et al. Revising theWHO verbal autopsy instrument to facilitate routinecause-of-death monitoring. Glob Health Action.2013;6:21518.

[25] World Health Organization. Verbal autopsy stan-dards: the 2014 WHO verbal autopsy instrument.Geneva: WHO; 2015.

[26] King C, Zamawe C, Banda M, et al. The quality anddiagnostic value of open narratives in verbal autopsy:a mixed-methods analysis of partnered interviewsfrom Malawi. BMC Med Res Methodol. 2016;16:13.

[27. Liu L, Li M, Cummings S, et al. Deriving causes of childmortality by re-analyzing national verbal autopsy dataapplying a standardized computer algorithm in Uganda,Rwanda and Ghana. J Glob Health. 2015;5:010414.

[28] Byass P, Herbst K, Fottrell E, et al. Comparing verbalautopsy cause of death findings as determined byphysician coding and probabilistic modelling: a publichealth analysis of 54 000 deaths in Africa and Asia. JGlob Health. 2015;5:65–73.

[29] Desai N, Aleksandrowicz L, Miasnikof P, et al.Performance of four computer-coded verbal autopsymethods for cause of death assignment compared withphysician coding on 24,000 deaths in low- and mid-dle-income countries. BMC Med. 2014;12:20.

[30] James SL, Flaxman AD, Murray CJ, et al. Performanceof the Tariff Method: validation of a simple additivealgorithm for analysis of verbal autopsies. PopulHealth Metr. 2011;9:31.

[31. Setel PW. Verbal autopsy and global mortality statistics:if not now, then when? Popul Health Metr. 2011;9:20.

[32] D’Ambruoso L, Boerma T, Byass P, et al. The case forverbal autopsy in health systems strengthening. LancetGlob Health. 2016;5:1.

[33] Lozano R, Lopez AD, Atkinson C, et al. Performanceof physician-certified verbal autopsies: multisite vali-dation study using clinical diagnostic gold standards.Popul Health Metr. 2011;9:32.

34] Miasnikof P, Giannakeas V, Gomes M, et al. Naive Bayesclassifiers for verbal autopsies: comparison to physician-

based classification for 21,000 child and adult deaths.BMC Med. 2015;13:286.

[35] Leitao J, Desai N, Aleksandrowicz L, et al. Comparisonof physician-certified verbal autopsy with computer-coded verbal autopsy for cause of death assignment inhospitalized patients in low- and middle-income coun-tries: systematic review. BMC Med. 2014;12:22.

[36] Garenne M, Fauveau V. Potential and limits of verbalautopsies. Bull World Health Organ. 2006;84:164.

[37] Soleman N, Chandramohan D, Shibuya K. Verbalautopsy: current practices and challenges. Bull WorldHealth Organ. 2006;84:239–245.

[38] Polprasert W, Rao C, Adair T, et al. Cause-of-deathascertainment for deaths that occur outside hospitalsin Thailand: application of verbal autopsy methods.Popul Health Metr. 2010;8:13.

[39] Ohemeng-Dapaah S, Pronyk P, Akosa E, et al.Combining vital events registration, verbal autopsyand electronic medical records in rural Ghana forimproved health services delivery. Stud HealthTechnol Inform. 2010;160:416–420.

[40] United Nations Statistical Division. Principles andrecommendations for a vital statistics system,Revision 3. New York: United Nations StatisticalDivision; 2014.

[41] Giduthuri JG, Maire N, Joseph S, et al. Developingand validating a tablet version of an illness explana-tory model interview for a public health survey inPune, India. PLoS One. 2014;9:e107374.

[42] Bird J, Byass P, Kahn K, et al. A matter of life and death:practical and ethical constraints in the development of amobile verbal autopsy tool. CHI. Paris; 2013 Apr. p. 1489.

[43] Kabadi GS, Geubbels E, Lyatuu I, et al. Data resourceprofile: the sentinel panel of districts: Tanzania’snational platform for health impact evaluation. Int JEpidemiol. 2015;44:79–86.

[44] Ndila C, Bauni E, Nyirongo V, et al. Verbal autopsy asa tool for identifying children dying of sickle celldisease: a validation study conducted in Kilifi district,Kenya. BMC Med. 2014;12:65.

[45] Mudenda S, Kamocha S, Mswia R, et al. Feasibility ofusing a World Health Organization-standard metho-dology for Sample Vital Registration with VerbalAutopsy (SAVVY) to report leading causes of deathin Zambia: results of a pilot in four provinces, 2010.Popul Health Metr. 2011;9:40.

[46] Hussain-Alkhateeb L, Petzold M, Collinson M, et al.Effects of recall time on cause-of-death findings usingverbal autopsy: empirical evidence from rural SouthAfrica. Emerg Themes Epidemiol. 2016;13:10.

[47] Kabadi G, Mwanyika H, de Savigny D. Mobile phoneSMS support to improve coverage of birth and deathregistration: a scalable solution. Queensland (Australia):Health Information Systems Knowledge Hub, School ofPopulation Health, The University of Queensland; 2013.

[48] World health Organization. MOVE-IT: report onmonitoring of vital events using information technol-ogy: WHO; 2013. [cited 6 Dec 2016]. Available from:http://www.who.int/healthinfo/civil_registration/crvs_report_it_2013.pdf?ua=1

[49] Aborigo RA, Allotey P, Tindana P, et al. Culturalimperatives and the ethics of verbal autopsies inrural Ghana. Glob Health Action. 2013;6:18570.

50] Gouda HN, Kelly-Hanku A, Wilson L, et al. “Wheneverthey cry, I cry with them”: reciprocal relationships and therole of ethics in a verbal autopsy study in Papua NewGuinea. Soc Sci Med. 2016;163:1–9.

GLOBAL HEALTH ACTION 13

[51] Gouda HNF A, Joshi R, Riley I, et al. New challenges forverbal autopsy: considering the ethical and social impli-cations of verbal autopsy methods in routine healthinformation systems. Soc Sci Med. Submitted 2016.

[52] INDEPTH. INDEPTH resource kit for demographicsurveillance systems. Accra. 2008. [cited 2016 Dec 6].Available from: www.indepth-network.org

[53] Ye Y, Wamukoya M, Ezeh A, et al. Health and demo-graphic surveillance systems: a step towards full civilregistration and vital statistics system in sub-SaharaAfrica? BMC Public Health. 2012;12:741.

[54] Joshi R, Praveen D, Jan S, et al. How much does averbal autopsy based mortality surveillance systemcost in rural India? PLoS One. 2015;10:e0126410.

[55] Pratiwi ED, Kosen S. Development of an Indonesiansample registration system: a longitudinal study.Lancet. 2013;381:S118.

[56] Godefay H, Abrha A, Kinsman J, et al. Undertakingcause-specific mortality measurement in an unregis-tered population: an example from Tigray Region,Ethiopia. Glob Health Action. 2014;7:25264.

[57] WHO. Verbal autopsy standards: ascertaining andattributing causes of death Geneva. 2014. [cited 6Dec 2016]. Available from: http://www.who.int/healthinfo/statistics/verbalautopsystandards/en/

[58] Scanlon P, Nichols EK. Results of the cognitive inter-viewing study of the 2012 WHO verbal autopsy

instrument in Nyanza Province, Kenya. Hyattsville:National Center for Health Statistics, Centers forDisease Control and Prevention; 2013.

[59] UNECA. CRVS digitisation guidebook: a step-by-step guide to digitising civil registration and vitalstatistics processes in low resource settings Africaprogramme for accelerated improvement of civilregistration and vital statistics; 2015. [cited 6December 2016]. Available from: http://www.crvs-dgb.org/en/

[60. Hartung C, Anokwa Y, Brunnette W, et al., editors.Open data kit: tools to build information services fordeveloping regions. In: ICTD ‘10 Proceedings of the 4thACM/IEEE International Conference on Informationand Communication Technologies and Development;2010 Dec. London (UK): ICTD; 2010.

[61] DHIS2. DHIS2. [cited 6 Dec 2016]. Available from:https://www.dhis2.org/

[62] DHIS2. DHIS2 implementation guide. editor. Oslo(Norway). [cited 6 Dec 2016]. Available from:https://ci.dhis2.org/docs/2.24/en/implementer/dhis2_implementation_guide.pdf

[63] Garenne M, Collinson MA, Kabudula CW, et al.Completeness of birth and death registration in arural area of South Africa: the Agincourt health anddemographic surveillance, 1992–2014. Glob HealthAction. 2016;9:32795.

14 D. DE SAVIGNY ET AL.