RESEARCH

Availability and quality of cause-of-death data for estimating the global burden of injuries

Disponibilité et qualité des données relatives aux causes de décès dans l'estimation du poids mondial des traumatismes

Disponibilidad y calidad de los datos sobre las causas de defunción para calcular la carga mundial de traumatismos

Kavi Bhalla^I,*; James E Harrison^II; Saeid Shahraz^I; Lois A Fingerhut^III; on behalf of the Global Burden of Disease Injury Expert Group

^IDepartment of Global Health and Population, Harvard School of Public Health, 718 Huntington Avenue, Boston, MA, 02115, United States of America (USA)
^IIResearch Centre for Injury Studies, Flinders University, Adelaide, Australia
^IIIInternational Collaborative Effort on Injury Statistics, Washington, USA

ABSTRACT

OBJECTIVE: To assess the availability and quality of global death registration data used for estimating injury mortality.
METHODS: The completeness and coverage of recent national death registration data from the World Health Organization mortality database were assessed. The quality of data on a specific cause of injury death was judged high if fewer than 20% of deaths were attributed to any of several partially specified causes of injury, such as "unspecified unintentional injury".
FINDINGS: Recent death registration data were available for 83 countries, comprising 28% of the global population. They included most high-income countries, most countries in Latin America and several in central Asia and the Caribbean. Categories commonly used for partially specified external causes of injury resulting in death included "undetermined intent," "unspecified mechanism of unintentional injury," "unspecified road injury" and "unspecified mechanism of homicide". Only 20 countries had high-quality data. Nevertheless, because the partially specified categories do contain some information about injury mechanisms, reliable estimates of deaths due to specific external causes of injury, such as road injury, suicide and homicide, could be derived for many more countries.
CONCLUSION: Only 20 countries had high-quality death registration data that could be used for estimating injury mortality because injury deaths were frequently classified using imprecise partially specified categories. Analytical methods that can derive national estimates of injury mortality from alternative data sources are needed for countries without reliable death registration systems.

RÉSUMÉ

OBJECTIF: Évaluer la disponibilité et la qualité des données mondiales d'enregistrement des décès utilisées pour estimer la mortalité d'origine traumatique.
MÉTHODES: Nous avons évalué l'exhaustivité et la couverture des données récentes d'enregistrement des décès nationaux, données provenant de la base de l'OMS sur la mortalité. La qualité des données relatives à une cause spécifique de décès par traumatisme a été jugée élevée si moins de 20% des décès étaient attribuables à l'une des causes de blessure partiellement définies, comme une «blessure non définie et involontaire».
RÉSULTATS: Les données récentes d'enregistrement des décès étaient disponibles pour 83 pays, soit 28% de la population globale. Elles incluaient la plupart des pays à revenu élevé, la majorité des pays d'Amérique latine et plusieurs pays d'Asie centrale et des Caraïbes. Les catégories les plus utilisées pour les causes de traumatismes externes partiellement définies entraînant le décès comprenaient: une «intention indéterminée», un «mécanisme non défini de traumatisme involontaire», des «blessures non définies suite à des accidents de la circulation» et un «mécanisme non défini d'homicide». Seuls 20 pays présentaient une qualité de données élevée. Néanmoins, comme les catégories partiellement définies contiennent des informations sur les mécanismes des traumatismes, il est possible d'obtenir des estimations fiables des décès dus à des causes de traumatismes externes spécifiques pour de nombreux autres pays, notamment les accidents de la circulation, les suicides et les homicides.
CONCLUSION: Seuls 20 pays ont présenté des données d'enregistrement des décès de qualité élevée, pouvant être utilisées pour estimer la mortalité attribuable à un traumatisme car les décès d'origine traumatique ont souvent été classés à l'aide de catégories partiellement définies et imprécises. Les méthodes analytiques qui peuvent fournir des estimations nationales sur la mortalité par traumatisme à partir d'autres sources de données sont nécessaires aux pays qui ne disposent pas de systèmes fiables d'enregistrement des décès.

RESUMEN

OBJETIVO: Evaluar la disponibilidad y la calidad de los datos de los registros mundiales de defunción utilizados para calcular la mortalidad por traumatismos.
MÉTODOS: Se han evaluado la integridad y el alcance de los datos recientes de los registros nacionales de defunción procedentes de la base de datos de mortalidad de la Organización Mundial de la Salud. Se consideró que la calidad de los datos en un caso específico de defunción por traumatismos era elevada si menos del 20% de las muertes se atribuyeron a cualquiera de las causas de traumatismos especificadas parcialmente, como « traumatismos no intencionados y no especificados».
RESULTADOS: Se dispuso de los datos recientes de los registros de defunciones de 83 países, lo que representa el 28% de la población mundial, entre los que se encontraban la mayoría de los países de ingresos elevados, la mayor parte de los países de América Latina y algunos de Asia Central y el Caribe. Las categorías empleadas generalmente para las causas externas parcialmente especificadas de los traumatismos con resultado de muerte fueron: «intención indeterminada», «mecanismo no especificado de traumatismo no intencional», « traumatismos por accidente de tránsito no especificados» y «mecanismo no especificado del homicidio». Únicamente 20 países poseían datos de alta calidad. No obstante, debido a que las categorías especificadas parcialmente contienen cierta información sobre los mecanismos de los traumatismos, en muchos países se pudieron extrapolar estimaciones fiables de las defunciones por traumatismos externos específicos, tales como los traumatismos por accidentes de tránsito, los suicidios y los homicidios.
CONCLUSIÓN: Sólo 20 países disponían de datos de alta calidad de registro de defunciones que se pudieran utilizar para calcular la mortalidad por traumatismos, debido a que las defunciones por traumatismos se clasificaron con frecuencia en categorías imprecisas y sólo parcialmente especificadas. Es necesario aplicar métodos analíticos que permitan obtener las estimaciones nacionales de mortalidad por traumatismos a partir de fuentes de datos alternativos para aquellos países que carezcan de sistemas fiables de registro de defunciones.

Introduction

Reliable estimates of the burden of death and disability due to injury are essential for shaping national and global health priorities. Although the quality of the information available in developing countries is relatively poor, past efforts at quantifying the global burden of disease^1-3 have convincingly established that injuries contribute approximately 10% to global mortality and 12% to global morbidity. A study into the global burden of disease was commissioned by The World Bank in 1991 and a new study is currently under way.⁴ Substantial collaborative efforts by the global injury research community will ensure that the best available evidence is incorporated into new estimates of the global burden of injuries.⁵ Thus, this is an opportune time to undertake an evaluation of the global data sources used for estimating the global burden of injuries. This paper focuses on the availability and quality of global mortality data reported by national death registration systems to the World Health Organization (WHO) mortality database.⁶

The WHO mortality database is the largest single repository of international data on causes of death. Our study builds on past work by Mathers et al.,⁷ who examined the quality of cause-of-death data in this database. They assessed quality by examining the proportion of deaths assigned to ill-defined cause-of-death codes, or "dump" codes. Unfortunately, some of the most important dump codes for injury were not included in their quality assessment. For instance, there was no assessment of code X59 of the International classification of diseases and related health problems, tenth revision (ICD-10), namely "accidental exposure to other and unspecified factors," which is extensively used in death registers to classify injury deaths. Moreover, the WHO mortality database has grown over the past 5 years and a reassessment of data quality is now due.

Following the lead of Mathers et al.,⁷ our aim was to assess the availability of recent mortality data, the completeness and coverage of regional death registration, and the quality of data derived from injury dump codes. In addition, we discuss the effect of data quality on the reliability of estimates of deaths due to road traffic injury, suicide and homicide.

Methods

Availability of mortality data

We obtained death registration data for a range of countries from the publicly available WHO mortality database (21 April 2009 update).⁶ The database includes details of deaths registered by national civil registration systems in which the underlying cause of death is coded by the relevant national authority in accordance with ICD rules.⁸ The analysis included only WHO member countries (listed at: http://www.who.int/countries) for which data were available from after 2000. For most countries, the latest data available were coded using ICD-10. For some, however, the latest available data were classified using the ICD-9 basic tabulation list, which is not adequate for estimating injury mortality, as subsequently discussed. We also included data classified using detailed ICD-9 codes from countries that made such data available to WHO: Guyana, Kiribati, the Netherlands (for Aruba), Saint Vincent and the Grenadines, Singapore, Slovenia and Thailand.

Completeness of registration

The completeness of national death registration was quantified by comparing the number of deaths recorded by death registration data for each country with estimates of projected mortality from the United Nations Population Division.⁹ Clearly such estimates of completeness are only crude indicators and should not be used for deriving incidence rates for cause-specific mortality. Nevertheless, they do provide some indication of the level of completeness of death registration in each country. In our study, we judged completeness to be high when it was greater than 80% of the expected value, medium when between 60% and 80%, and low otherwise.

Data quality

We classified all injury deaths using 48 categories of specified external causes of death, which constitute the reporting categories recommended by the injury expert group of the 2005 Global Burden of Diseases, Injuries and Risk Factors Study (i.e. the GBD Injury Expert Group),¹⁰ and 21 categories of partially specified external causes (Table 1, avail able at: http://www.who.int/bulletin/volumes/88/11/09-068809). Several countries coded the cause of death using condensed versions of ICD-9 and ICD-10, which do not contain sufficient detail to classify injury deaths according to the full list of external causes of injury shown in Table 1. In addition, these condensed versions group together dump codes with imprecise definitions. For example, the ICD-10 basic tabulation list includes the code X59, for unspecified unintentional injury, in the category 1103 (i.e. all other external causes). The impossibility of separating dump codes means that deaths classified using these codes cannot be reassigned to specified causes and, consequently, the incidence of these causes cannot be estimated. Thus, we excluded countries that used basic tabulation lists from further quality assessment.

Data quality was assessed by determining the proportion of deaths that were classified as belonging to various partially specified categories: the quality of the data improves as the proportion of deaths assigned to partially specified categories decreases. Moreover, the partially specified categories may form a hierarchy in terms of their information content. This hierarchy is related to the specificity of the definition of each category. For example, the death of a car occupant who was killed in a road traffic accident may be coded using any of the following categories in a hierarchy of partially specified categories whose definition shows less specificity towards the end of the list:

To determine data quality, we computed the proportion of deaths in each partially specified category relative to the corresponding total number of deaths at that level of specificity: e.g. the proportion of road injuries recorded as an unspecified unintentional road injury. Data quality was rated as high when this proportion was smaller than 20%.

Finally, we examined the distribution of deaths classified as belonging to the partially specified categories to determine the circumstances in which the data can, nevertheless, be used to derive reliable estimates of mortality for various specific external causes of death. The countries for which deaths due to road injury, homicide and suicide can be reliably estimated are given as an illustrative example.

Results

Availability of mortality data

Table 2 (available at: http://www.who.int/bulletin/volumes/88/11/09-068809) summarizes the availability of death registration data in the WHO mortality database for countries in the 21 different regions of the world defined by the 2005 Global Burden of Diseases, Injuries and Risk Factors Study.¹⁰ For each country, the most recent year for which data are available and the number of years of data availability are listed. In all, 83 countries met the inclusion criteria. They accounted for 28% of the global population. Recent data were available from most countries in high-income regions except for a few notable exceptions: Switzerland used a basic tabulation list and no recent data were available from Belgium. The coverage of death registration data in low- and middle-income regions was more irregular. Three of the four Asian regions (i.e. South, South-East and East Asia) were severely underrepresented, with less than 15% of the regional population covered. The continent of Africa was even less well covered. Data from sub-Saharan Africa were available for only one country (i.e. South Africa). However, low- and middle-income countries in Latin America and the Caribbean were well represented, with over 80% of the population covered. The availability of injury data from eastern Europe, central Europe and central Asia was limited, primarily because of the use of basic tabulation lists.

While historical death registration data were available for over a decade for many high-income countries, the most recent data for many of these countries dated from before 2005 (Table 3), which suggests that there were delays in reporting to WHO. Although low- and middle- income countries had been reporting data for a shorter time, 5 years of recent data were available for many. It should be noted that all of the ICD-9 data in the WHO mortality database were coded using the ICD-9 basic tabulation list, which limits the length of the historical record in these cases.

Completeness of registration

Of the 83 countries analysed, completeness was high (i.e. > 80%) in 62, medium (i.e. 60-80%) in 9, low (< 60%) in 5, and could not be assessed in the 7 for which no estimate of all-cause deaths was available from the United Nations Population Division. Table 3 lists the completeness of national death registration and the proportion of deaths assigned to various partially specified causes for selected countries. The figures for the remaining countries are listed in Appendix A (available at: http://www.globalburdenofinjuries.org/gimd/Quality_Global_Injury_Mortality_Data.pdf).

Data quality

Use of the broadest unspecified cause-of-death category, i.e. unknown cause of death (ICD-10 codes: R95-R99), was relatively rare. Only Haiti assigned more than 20% of all-cause deaths to this category. However, the use of codes for undetermined intent (ICD-10 codes: Y10-Y34, Y87.2) was common, with 18 countries, including one high-income country (i.e. Singapore), classifying more than 20% of injury deaths as due to undetermined intent. Two of the three countries from North Africa and the Middle East (i.e. Bahrain and Egypt) placed over one-third of all injury deaths in this category. The other countries with an exceptionally high number of deaths in this category included: Azerbaijan (83%), the Dominican Republic (45%), Egypt (42%), Guatemala (37%), Maldives (98%), Suriname (35%) and South Africa (66%). In contrast, the number of deaths classified as being due to an unspecified injury mechanism with undetermined intent (ICD-10 code: Y89.9) was negligible in all countries.

With regard to unintentional injuries, many countries coded a large number of deaths using the broadest unspecified mechanism category: unspecified unintentional injury (ICD-10 code: X59). In 15 countries, over half of which were in Western Europe, over 20% of unintentional injury deaths were allocated this code. The proportion of deaths coded to the unspecified non-transport injury subcategory of unspecified unintentional non-transport injury (ICD-10 code: Y86) was negligible in all countries, with the notable exception of Cuba.

With regard to transport deaths, only Georgia (46%) and Serbia (22%) allocated more than 20% of deaths to the broadest unspecified category: unspecified unintentional transport injury (ICD-10 codes: V99, Y85.9). However, many countries coded a large number of deaths due to road injury using the broadest unspecified category: unspecified unintentional road injury (ICD-10 codes: V89, Y85.0). In total, 40 countries used this code for more than 20% of road deaths. Many were high-income countries, including two in North America (i.e. Canada and the USA) and six in Western Europe, two of which (i.e. France and Portugal) allocated over 80% of road deaths to this category. The other partly specified subcategory of road injury, unspecified road injury not including a pedestrian or bicyclist (ICD-10 codes: V87-V88), was used less often. Only seven countries used this category for more than 20% of vehicle occupant deaths. Three of these (i.e. Greece, Ireland and San Marino) used ICD-9, which does not differentiate between different types of vehicle occupant.

With regard to suicide, the mechanism of death was usually specified. With the exception of Georgia, Haiti and Saint Lucia, no country attributed more than 20% of suicide deaths to an unspecified mechanism. However, the mechanism of homicide deaths was much less likely to be specified. In 13 countries, the mechanism was not specified for more than 20% of homicides. These countries included Portugal (27%) and Spain (22%) in Western Europe and, notably, Israel (43%). The mechanism of deaths classified as being due to a legal intervention was always specified. In no country, were more than 3% of these deaths attributed to an unspecified mechanism.

In general there was considerable heterogeneity between countries in the use of unspecified codes. While Australia and the United Kingdom of Great Britain and Northern Ireland classified a large proportion (i.e. 18% and 24%, respectively) of injury deaths as due to unspecified unintentional injury (ICD-10 code: X59), far fewer deaths were coded in this way in the USA (5%). On the other hand, 12% of all injury deaths were coded as due to undetermined intent (ICD-10 codes: Y10-Y34, Y87.2) in the United Kingdom, while the proportions were much smaller in Australia and the USA (1% and 3%, respectively).

Overall, only 20 countries did not allocate more than 20% of deaths to any partially specified category (Table 4). However, since the use of most partially specified categories influences data on only certain external causes of death, the number of countries whose data can be used to provide reliable estimates of deaths due to a particular external cause may be much larger than 20. For example, Table 4 lists 47 countries for which reliable estimates of road injury deaths were available and 60 for which reliable estimates of death due to suicide or homicide were available.

Discussion

Previous work on national and global mortality patterns has not paid sufficient attention to injuries. We undertook this analysis because the only other assessment of the quality of global cause-of-death data⁷ did not consider the use of the most common injury dump codes. Furthermore, the external causes of injuries are often poorly categorized in many administrative data systems. For instance, even in high-income countries with an extensive history of disease surveillance, hospital administration records commonly omit the external cause of an injury.^11-13 Similarly, in an unpublished analysis of mortality surveillance data based on verbal autopsy, which is the only source of mortality data in many information-poor settings, we found that often only the nature of the injury was reported for injury deaths. This is a serious shortcoming that will hamper attempts to develop injury prevention strategies since reliable estimates of the incidence of external causes of injuries are needed.^14,15

The inadequate classification of injuries by ICD basic tabulation lists is another reason why injury deaths have not been fully considered. These summary lists do not provide codes for the mechanism of a suicide or homicide. Thus, the incidence of many important mechanisms of injuries (e.g. firearm injury, poisoning and burn injury) cannot be determined. In addition, the summary lists pose a more substantial problem for assessing mortality due to specific causes of injury. Usually the key injury dump codes identified in our analysis (e.g. ICD-10 code: X59) are grouped together with other specified mechanisms. Without access to data on these injury dump codes, the quality of the data overall cannot be assessed and injury mortality cannot be reliably estimated.

Ultimately, the purpose of this analysis was to identify those countries where the incidence of death due to injury can be reliably estimated. In the absence of empirical evidence to support a more nuanced characterization of quality, we adopted a maximum of 20% for each partially specified category for the cause of death to define high-quality data. While this threshold is to some extent arbitrary, it is based on the understanding that, when a large number of deaths have been allocated to partially specified categories, reapportioning deaths to specified categories can introduce substantial biases. Nevertheless, we showed that these categories contain a hierarchy of information content (i.e. they are partially specified) that should be harnessed fully to derive estimates of the incidence of death due to injury. Thus, although only 20 countries had high-quality data on all quality indicators, many more had high-quality data for specific external causes such as road injury, suicide and homicide.

The considerable heterogeneity between countries in the use of partially specified codes shows that they differ in coding practices. An understanding of those differences may enable us to identify biases arising from the dump codes used in specific countries and to derive appropriate reclassification rules for these deaths. There may also be commonalities in coding practices across countries that could help explain, for instance, why death registers in most countries tend to report the mechanism of suicides but not the mechanism of homicides.

In many countries, vital registration is the only comprehensive source of data for estimating mortality due to specific causes. Consequently, it has been proposed that national data should be validated and data quality should be improved, and this has already been done in several settings.^16-18 Although we focused on injury deaths, the standardized method for assessing data quality we used could be implemented by national agencies as part of their routine quality assurance practices. Temporal changes in data quality could be evaluated and one country's data could be compared with those from other countries.

An important finding of our analysis is that reliable national death registration data were available for less than 30% of the global population, which means that alternative data sources must be used for estimating global injury mortality. The two most populous countries in the world, India and China, do not have reliable national death registration systems though they do have sample registration systems¹⁹ that may be useful for deriving estimates of death due to injury. Similarly, at present there are no national death registration systems in most of Africa and such systems are unlikely to become a reliable source of data for decades. Nevertheless, even in these information-poor settings there are several alternative data sources that can be used to estimate injury mortality. Notably these include demographic surveillance sites,²⁰ mortuaries,^21,22 national censuses containing information on the cause of death,²³ and national health surveys that report the details of sibling mortality.²⁴ Analytical methods that can derive national estimates of injury mortality from multiple data sources are urgently needed.

Our evaluation of the availability and quality of global death registration data has several limitations. First, we considered only the data available from the WHO mortality database, supplemented by additional data from a few countries. While this data set is the single largest global death registration repository, there are many countries that collect death registration data but do not report to WHO. It is also likely that more recent data are available from the national vital statistics agencies of many countries included in the WHO database. Second, our judgment of data quality was based on the proportion of deaths that was assigned to partially specified categories. However, the misclassification of deaths was not considered. For instance, it is likely that some deaths due to suicide were coded as due to an unintentional or undetermined cause because of the stigma associated with suicide, and other deaths may have been similarly coded because of medico-legal considerations associated with intentional death. Moreover, these misclassifications may vary substantially between countries. Misclassification could have significant impact on the accuracy of injury mortality estimates derived from death registration data.

Despite these shortcomings, this analysis is a step towards making global injury data comparable across countries. The next step in this process is to develop methods for processing death registration data sets such that reasonable estimates of cause-specific injury mortality can be derived. This would lead to the construction of an international database of injury deaths that could provide useful insights into the structural causes of variations in the incidence of injury mortality between countries. With access to such a database, the international injury research community could learn from the experiences of different countries and could identify the social, political and environmental prerequisites of safe and sustainable living conditions.

Funding: No member of the GBD Injury Expert Group derived salary support from the Global Burden of Disease project. However, several members have ongoing funded research in closely associated areas. Kavi Bhalla and Saeid Shahraz are supported by a grant from the World Bank Global Road Safety Facility, but this funder played no role in the study design, data analysis, decision to publish or preparation of this manuscript.

Competing interests: None declared.

References

1. Murray CJL, Lopez AD. The global burden of disease: a comprehensive assessment of mortality and disability from diseases, injuries, and risk factors in 1990 and projected to 2020. Cambridge: Harvard University Press; 1996.         

2. Mathers CD, Loncar D. Projections of global mortality and burden of disease from 2002 to 2030. PLoS Med 2006;3:e442. doi:10.1371/journal.pmed.0030442 PMID:17132052        

3. World Health Organization. The global burden of disease: 2004 update. Geneva: World Health Organization; 2008.         

4. Murray CJ, Lopez AD, Black R, Mathers CD, Shibuya K, Ezzati M et al. Global burden of disease 2005: call for collaborators. Lancet 2007;370:109-10. doi:10.1016/S0140-6736(07)61064-2 PMID:17630021        

5. Bhalla K, Harrison J, Abraham J, Borse NN, Lyons R, Boufous S et al.; Global Burden of Disease Injury Expert Group. Data sources for improving estimates of the global burden of injuries: call for contributors. PLoS Med 2009;6:e1. doi:10.1371/journal.pmed.1000001PMID:19166263        

6. WHO mortality database. Geneva: World Health Organization; 2008. Available from: http://www.who.int/healthinfo/statistics/mortdata [accessed 4 June 2010]          .

7. Mathers CD, Fat DM, Inoue M, Rao C, Lopez AD. Counting the dead and what they died from: an assessment of the global status of cause of death data. Bull World Health Organ 2005;83:171-7. PMID:15798840        

8. World Health Organization. International statistical classification of diseases and related health problems. Geneva: World Health Organization; 1992.         

9. World population prospects: the 2002 revision. New York: United Nations Population Division; 2003. Available from: http://esa.un.org/unpp [accessed 4 June 2010]          .

10. Global Burden of Disease Injury Expert Group [Internet site]. Boston: Global Burden of Disease Injury Expert Group; 2009. Available from: http://sites.google.com/site/gbdinjuryexpertgroup [accessed 4 June 2010]          .

11. Finkelstein EA, Corso PS, Miller TR. The incidence and economic burden of injuries in the United States. New York: Oxford University Press; 2006.         

12. Conner KR, Langley J, Tomaszewski KJ, Conwell Y. Injury hospitalization and risks for subsequent self-injury and suicide: a national study from New Zealand. Am J Public Health 2003;93:1128-31. doi:10.2105/AJPH.93.7.1128 PMID:12835197        

13. Bhalla K, Shahraz S, Naghavi M, Lozano R, Murray C. Estimating the distribution of external causes in hospital data from injury diagnosis. Accid Anal Prev 2008;40:1822-9. doi:10.1016/j.aap.2008.07.002 PMID:19068282        

14. Sniezek JE, Finklea JF, Graitcer PL. Injury coding and hospital discharge data. JAMA 1989;262:2270-2. doi:10.1001/jama.262.16.2270 PMID:2795809        

15. Lawrence BA, Miller TR, Weiss HB, Spicer RS. Issues in using state hospital discharge data in injury control research and surveillance. Accid Anal Prev 2007;39:319-25. doi:10.1016/j.aap.2006.08.001 PMID:17026946        

16. Rosenberg HM. Improving cause-of-death statistics. Am J Public Health 1989;79:563-4. doi:10.2105/AJPH.79.5.563 PMID:2705587        

17. Sirken MG, Rosenberg HM, Chevarley FM, Curtin LR. The quality of cause-of-death statistics. Am J Public Health 1987;77:137-9. doi:10.2105/AJPH.77.2.137 PMID:3799853        

18. McKenzie K, Chen L, Walker SM. Correlates of undefined cause of injury coded mortality data in Australia. HIM J 2009;38:8-14. PMID:19293431        

19. Setel PW, Macfarlane SB, Szreter S, Mikkelsen L, Jha P, Stout S et al.; Monitoring of Vital Events. A scandal of invisibility: making everyone count by counting everyone. Lancet 2007;370:1569-77. doi:10.1016/S0140-6736(07)61307-5 PMID:17992727        

20. Adjuik M, Smith T, Clark S, Todd J, Garrib A, Kinfu Y et al. Cause-specific mortality rates in sub-Saharan Africa and Bangladesh. Bull World Health Organ 2006;84:181-8. doi:10.2471/BLT.05.026492 PMID:16583076        

21. Butchart A, Peden M, Matzopoulos R, Phillips R, Burrows S, Bhagwandin N et al. The South African National Non-Natural Mortality Surveillance System-rationale, pilot results and evaluation. S Afr Med J 2001;91:408-17. PMID:11455806        

22. Wolde A, Abdella K, Ahmed E, Tsegaye F, Babaniyi OA, Kobusingye O et al. Pattern of injuries in Addis Ababa, Ethiopia: a one-year descriptive study. East Centr Afr J Surg 2008;13:14-22.         

23. Setel PW, Sankoh O, Rao C, Velkoff VA, Mathers C, Gonghuan Y et al. Sample registration of vital events with verbal autopsy: a renewed commitment to measuring and monitoring vital statistics. Bull World Health Organ 2005;83:611-7. PMID:16184280        

24. Gakidou E, King G. Death by survey: estimating adult mortality without selection bias from sibling survival data. Demography 2006;43:569-85. doi:10.1353/dem.2006.0024 PMID:17051828        

(Submitted: 17 June 2009 - Revised version received: 4 May 2010 - Accepted: 6 May 2010 - Published online: 22 June 2010)

* Correspondence to Kavi Bhalla (e-mail: kavi_bhalla@harvard.edu).