International Journal of Health System and Disaster Management

ORIGINAL ARTICLE
Year
: 2013  |  Volume : 1  |  Issue : 3  |  Page : 155--162

Reporting systems on cause of deaths in India: An assessment


Dinesh Kumar, Bharathi Balaiah 
 Health Division, Planning Commission of India, New Delhi, India

Correspondence Address:
Bharathi Balaiah
24/1, Dhayalammal Street, Padmanabha Nagar, Chennai, Tamil Nadu, 600094
India

Abstract

Introduction: Realization of such causes on the mortality data is very critical and can aid at informed decision making and at devising evidence based definite and timely intervention strategies. The paper attempts to review the existing systems that report causes of death in India. To assess the quality of existing reporting systems on causes of death in India in terms of methodology followed, coverage, completeness, reliability, and pattern of cause of deaths. Materials and Methods: The study design followed was a cross-sectional exploratory one. Secondary data available under Medical Certification of Cause of Death (MCCD) 2007, Report on Cause of Death (RCD) 2001-2003, Global Burden of Disease (GBD) 2004 (updated 2008), and Sample Registration System (SRS) 2008 were reviewed to assess the quality of existing reporting systems on causes of death in India in terms of methodology followed, coverage, completeness, reliability, and pattern of cause of deaths. A correlation analysis was also done to understand the similarity between the reports. The data was analysed using Excel and Statistical Package for Social Sciences (SPSS) version 20. Results: Wide discrepancies observed between the existing reporting systems on cause of death in India. RCD was found to be more appropriate in terms of coverage, completeness, quality check, reliability, and representative character. RCD 2001-03 also showed high correlation with MCCD 2007 and GBD 2007. Conclusion: Having multiple reporting systems with the same objective could pose unnecessary duplication of efforts and wastage of various resources. Having observed discrepancies in the reporting systems, it is emphasized that the existing systems are reorganized/collated to provide quality, accurate, and uniform data on cause of death.



How to cite this article:
Kumar D, Balaiah B. Reporting systems on cause of deaths in India: An assessment.Int J Health Syst Disaster Manage 2013;1:155-162


How to cite this URL:
Kumar D, Balaiah B. Reporting systems on cause of deaths in India: An assessment. Int J Health Syst Disaster Manage [serial online] 2013 [cited 2024 Mar 29 ];1:155-162
Available from: https://www.ijhsdm.org/text.asp?2013/1/3/155/129141


Full Text

 Introduction



Around the world, data related to mortality are the most widely and commonly used ones to frame the policies and programs to combat the various health problems. Mortality data have distinct advantages over other sources of health data. Because mortality data are based on death registration systems, they have the advantage of virtually complete coverage. Consequently, they can be used for the analysis and comparison of small geographic areas and population subgroups. Almost all the countries have legislation that establishes vital registration systems to collect the statistics on mortality and cause of mortality (Mathers et al., 2005).

About 46 million deaths out of 60 million deaths occur in developing countries, but there is dearth of reliable data on mortality in these countries.

In India there are various systems which measures mortality, [Table 1] gives a glimpse of various systems that provide mortality data.{Table 1}

Few of the major sources, as enlisted in the above table, are Civil Registration System (CRS), which is unreliable due to gross under-registration (Jha et al., 2005); Medically Certification of Cause of Death (MCCD), which is majorly confined to the urban settings; and Sample Registration System (SRS) which is giving number of deaths only. Global Burden of Disease (GBD) reports also made efforts in same direction by providing the numbers with cause specific data of mortality. National Family Health Survey (NFHS) and Census also provide some information on mortality indicators. Recently; Office of the Registrar General of India (ORGI), New Delhi has come out with Report on Cause of Death (RCD) 2001-2003 In India, which is prepared by ORGI with the help of Centre for Global Health Research (CGHR), Canada.

Ideally, the mortality measures should be carried out routinely, in a reproducible way, classified according to the internationally accepted criteria, for example, ICD and be made available in a reproducible and accountable manner in an open forum to help in evidence manner.

However, this does not happen in India in real sense. It has been stated in earlier studies that India needs better epidemiological evidence about the relevance of physical, behavioral, and biological measurements to the development of disease in individuals or disease rates in individuals (Jha et al., 2005).

Rationale

India is a large country with a diverse population. It is estimated that 9.5 million deaths occur in India every year (Jha et al., 2005). Over three-quarter of these deaths occur at home and more than half of these deaths do not have a certified cause. Any disease which is common in one part of the country may not be common in another part of the country and the reasons for which are diverse and at many times remain unknown. Realization of such causes on the mortality data is very critical and can aid at informed decision making and at devising evidence based definite and timely intervention strategies.

Though there are many systems and schemes that report the number of deaths and the causes of deaths in India, lots of asymmetries have been observed between them. Having said this background, this paper attempts to understand the different reporting systems on causes of death in India, to assess their quality in terms of methodology followed, coverage, completeness, reliability, and pattern of cause of deaths and to provide the suitable recommendations for desired improvements in the reporting systems on causes of deaths in India.

 Materials and Methods



The study design followed was a cross-sectional exploratory one. Secondary data available under MCCD 2007, RCD 2001-2003, GBD 2004 (updated 2008), and SRS 2008 were reviewed to assess the quality of existing reporting systems on causes of death in India in terms of methodology followed, coverage, completeness, reliability, and pattern of cause of deaths.

As SRS is considered to be a reliable source of vital statistics in India, the reporting systems on causes of death in India, MCCD 2007, RCD 2001-2003, and GBD 2004 (updated 2008) were compared with SRS 2007 in terms of completeness.

The three systems that report causes of death in India were then compared among themselves in terms of methodology followed, coverage, reliability, and pattern of cause of deaths to assess their quality. A correlation analysis was also done to understand the similarity between the reports. The data was analyzed using Excel and Statistical Package for Social Sciences (SPSS) version 20.

 Results



Understanding the different reporting systems on causes of death in India

Realizing the dearth of quality data on cause of deaths in India, a scheme on MCCD was initiated under the domains of Registration of Births and Deaths Act, 1969 and thereafter reports on cause of death across the country are being collected annually. Though it has been operationally functional in almost all of the states/union territories (UTs), levels of efficiency across states/UTs vary considerably. It has been observed that only deaths occurring in urban medical institutions and urban areas are being captured, rendering the efficiency of the scheme in capturing the mortality data across states/UTs a debatable one. To fulfill the void in data from the rural areas, Survey of Cause of Death was initiated, which later on was merged with SRS from 1999, thus encompassing both rural and urban areas. Since then, a system of verbal autopsy under the domain of SRS has been in operation.

The ORGI in collaboration with the CGHR did a Special Survey on Cause of Death to cover all the deaths that occurred in the years 2001-2003 during 2004-05 in the SRS. The RCDs based on the survey was published in the year 2009.

Globally, the MCCD database for urban India, the Annual Survey of Causes of Death (SCD) for rural areas, information from World Health Organization (WHO) technical programs and UNAIDS are used with few mathematical calculations in arriving at the death rate and in assigning the cause of death.

Methodology followed

Though all of the three systems that report cause of death in India used International Classification of Disease, tenth revision (ICD-10) for assigning cause of death, considerable variations were found in the methodology followed. The methodologies followed in each of the three reports are schematically presented in the flow chart below [Figure 1].{Figure 1}

Under MCCD 2007, the institutional and non-institutional deaths are certified by the medical professionals using different formats of MCCD as evolved by WHO. The certification is carried according to the national list of causes of death based on ICD-10. Data thus obtained are then compiled, tabulated, and sent by the Registrar (births and deaths) of the respective districts to the Chief Registrars of Births and Deaths of the respective states/UTs. These are then sent in the form of statistical table for consolidation at national level to the ORGI for consolidation at national level.

Under the RCD (2001-2003), the Special Survey of Deaths for the period 2001-2003 was conducted by the ORGI in collaboration with the CGHR during 2004-2005. Details of signs and symptoms that can help in determining the underlying cause of death were collected retrospectively by the SRS supervisors using a special form of verbal autopsy called "RHIME" (Representative, Reproducible, Routine Household Interview of Mortality using medical Evaluation) from each of the households in the SRS sample units where a death had occurred. Later, the field reports were examined by two independent trained physicians and the cause of death were assigned using the three-digit code of the ICD-10. On instances of disagreements between the physicians, forms were anonymously reshuffled between them for reconciliation. On continuing disagreement, the decision of a senior third physician is sought to assign the cause of the death.

Under GBD 2004 (updated 2008), the MCCD for urban India (1996), the Annual SCD for rural areas of India for 1996-1998, information from WHO technical programs, and UNAIDS were used to calculate the GBD 2002 cause distributions for India. The GBD 2002 cause distributions for India were adjusted to the 2004 all-cause envelope, and the resulting cause-specific estimates were further adjusted with information for 2004 from WHO technical programs and UNAIDS.

Coverage

The total absolute number of deaths derived as per the crude death rate of 7.4 of SRS 2007 was found to be 8,398,192. The total number of deaths reported under the different reporting systems on cause of death was compared with this derived total number of deaths to comment on the coverage.

As per the report on MCCD 2007, the total registered deaths that occurred across the country was 5,804,922 and in the MCCD states/UTs was 4,219,770. Out of which, only 798,546 deaths (13.76% of the total registered deaths across the whole country and 18.9% of the total registered deaths in MCCD reporting states/UTs) were medically certified.

As mentioned above, the Special Survey of all Deaths within the SRS for the calendar years 2001-2003 was conducted in 2004-2005. It covered over 6,645 sample units and 6,452,000 populations in all the States/UTs. A total of 140,000 deaths at all ages were investigated. However, due to out migration and change of households, 12% of deaths could not be surveyed. And 9% of the deaths were not included in the analysis due to poor quality of image and non-legible or incorrect language code. The total deaths that were assigned cause of death and analyzed were 113,692.

As mentioned above, the cause distributions for India in GBD 2004 was calculated using the MCCD for urban India (1996), the Annual SCD for rural areas of India for 1996-1998, information from WHO technical programs and UNAIDS. The total number of deaths recorded under GBD in the year 2007 were more than 9,800,000, which was considerably higher than the total number of deaths registered in same year in SRS.

Completeness

Completeness of Death Registration (C) is defined as the proportion registered of actual deaths, which is same as the ratio of registered deaths to actual deaths (Mahapatra, 2010).

As per the definition described above, the completeness of the reports MCCD 2007, RCD 2001-2003, and GBD 2007 were found to be 0.19, 0.81, and 1.17 respectively.

As the difference between the completeness of the MCCD 2007 and GBD 2007 were found to be too high, the data available in the two sources were further compared with the crude death rate in SRS 2007.

[Table 2] gives the difference in the total registered deaths in absolute numbers and the crude death rate between the report on MCCD 2007 and SRS 2007 database.{Table 2}

Crude death rate is the total number of deaths to residents in a specified geographic area (country, state, county, etc.) divided by the total population for the same geographic area (for a specified time period, usually a calendar year) and multiplied by 100,000 [Table 3].{Table 3}

[INLINE:1]

Wide discrepancies were observed in the absolute number of deaths registered and crude death rate in India for the year 2007 among the databases of Report on MCCD 2007 and the SRS 2007. This discrepancy was significantly higher when the data on absolute number of deaths and CDR were analyzed for individual states as evident in the table given above.

A correlation analysis was done between the two reports and it was found to be 0.081, which is quite insignificant. [Figure 2] depicts the graphical representation of the correlation between the two reports.{Figure 2}

Due to unavailability of individual state-wise details, a similar exercise was not possible with GBD 2007.

Comparison of CDR of MCCD 2007 and GBD 2007 with that of SRS

Crude death rate mentioned in SRS for the year 2007 = 7.4Crude death rate for MCCD reporting states/UTs for the year 2007 (by using total deaths) =5,805,130/1,131,043,000 × 1,000 = 5.1

Crude death rate for MCCD reporting states/UTs for the year 2007 (by using only MCCD deaths) =798,456/866,428,000 × 1,000 = 0.9Crude death rate for GBD for the year 2004 = 9,894,700/1,181,412,000 × 1,000 = 8.3

The crude death rate arrived for the whole country using the database of RCD 2007 (5.1) was found to be lower than that of SRS 2007 (7.4). When the same was calculated using only the medically certified deaths, it became negligible (0.9). In contrast, the CDR arrived using GBD 2007 was found to be higher (8.3).

Reliability

As discussed above, the cause of the deaths in the scheme MCCD 2007 being certified by the medical professionals in person, they could be considered high reliability. However, no specific quality measure being carried out, being added by the limited coverage, and varying levels of efficiency across the country puts the reliability of the scheme doubtful. In contrast, the intense quality assurance procedures carried out in the RCD 2001-2003 along with its relatively better coverage, completeness and representative character renders the data available under the scheme more reliable. The analysis done above clearly indicates a possibility of duplication of data in the GBD 2007.

Pattern of cause of deaths

The grouping for the cause of death of all the reports were reorganized as per the World Health Organization (WHO) guidelines into three categories: (i) communicable, maternal, perinatal, and nutritional conditions; (ii) noncommunicable diseases; (iii) injuries to facilitate comparison among the reports. A fourth category, others, was added where it was difficult to categorize the causes.

[Table 4] provides a snapshot on the pattern of cause of deaths observed in the three reporting systems.{Table 4}

As per all the above three reporting systems, the noncommunicable diseases form the prime disease burden in the Indian healthcare, followed by communicable diseases, maternal, perinatal, and nutritional conditions, and injuries and poisoning. Though in overall view, it appears similar with respect to disease burden, significant discrepancies were observed among the reports. In absolute terms, the discrepancy observed was much higher and a high proportion of deaths were classified under others. But the critical review of deaths under both the reports was done and they were reclassified based on the WHO guidelines. The discrepancy came down, but still a considerable proportion of deaths were classified under others in MCCD 2007 (13.8%) and RCD 2001-2003 (9.9%).

The percentage of communicable diseases, maternal, perinatal, and nutritional conditions was found to be much lower in MCCD 2007 (26.2%) than the observed values in RCD 2001-2003 (38.2%) and GBD 2007 (37.1%). Similarly, the percentage of noncommunicable diseases was found to be much lower in RCD 2001-2003 (42.4%) as compared to MCCD 2007 (50.8%) and GBD 2007 (53%). The percentage of injuries and poisoning remained more or less similar in all the three reports. A considerable proportion of deaths were classified under others in MCCD 2007 (13.8%) and RCD 2001-2003 (9.9%).

Having observed the discrepancies under broad categories, the cause of deaths were grouped in narrow categories. Vast discrepancies were observed among various subcategories. [Table 5] provides a snapshot on the subcategories on pattern of cause of deaths observed in the three reporting systems.

A correlation exercise was carried out using the data in [Table 5] among the three reporting systems to understand the similarities among them. The results of correlation exercise between the three reports can be seen in [Table 6].{Table 5}{Table 6}

High correlation ranging from 0.8 to 0.9 was observed between the reports. However, RCD 2001-2003 had high correlation with both MCCD 2007 and GBD 2007.

 Discussion



Methodology followed

Under MCCD 2007, as the deaths are certified by the medical professionals in person, the cause of death thus assigned shall be considered of high reliability. However, no specific quality measure appears to be followed in the procedure to validate the data gathered.

Under the RCD (2001-2003) though the assigning of the cause of death is not done by the physician in person and is rather done by retrospectively examining the reports collected by field supervisors, intense quality measures are carried to ensure the quality of the data. The assigning of the death is being carried by two physicians to reduce the inter person variation. The reports examined are randomly assigned between two of the 120 physicians to increase the comparability across the states. When disagreement is there among the physicians, the forms are randomly shuffled and redistributed among the physicians using a web-based system. Moreover, a 5-10% of the sample units were resurveyed and 5-10% of the supervisor's visits were resampled to ensure quality, completeness, accuracy, to analyze the training input, and correction of method.

Under GBD 2004 (updated 2008), though the data available in various databases are being pooled to arrive at the cause of death in India under this database, the possibility of duplication of the reports amidst of increasing reliability could not be ignored.

Coverage

Though the MCCD has been operational since 1969, the scheme has not been taken up in the states and the UTs in a uniform manner. The percentage of the medically certified deaths observed among the total registered deaths across the whole country and among the total registered deaths in MCCDs reporting states and UTs appears quite low. Moreover, the medically certified deaths were collected only from 25 states/UTs against 35 states/UTs. Different level of efficiency was also observed among the different states/UTs. The percentage of medical certification carried and the type of the hospitals covered varied among the states/UTs. The scheme mostly covers the deaths that occur in medical institutions located in urban areas. Owing to the low coverage and varying efficiency observed, the report as much may lack a representative character.

Though the deaths investigated and analyzed in the Special Survey of all Deaths were low as compared to the absolute deaths occurred in the years 2001-2003, as the sample units were dispersed across the country with representation from both urban/rural areas, they appear quite comprehensive and representative.

Though the coverage appeared to be quite good in absolute terms in GBD 2004, it did emphasize the chances of duplication in data.

Completeness

The completeness observed in the study clearly indicates that MCCD has extremely poor completeness, which indirectly indicates that the report by itself lacks representative character. RCD 2001-2003 was found to have a considerably reliable completeness. The high level of completeness observed with GBD 2007 could possibly because of the above mentioned duplication of data.

The reasons for the low CDR observed using the database of RCD 2007 as compared to that of SRS 2007 could possibly be errors in the mentioned total registered deaths, varying levels of efficiency across states, MCCD scheme not fully functional, capturing mostly urban data, and neglecting rural areas. The reason for the high CDR observed using the database of GBD 2007 could possibly be the pooling of data from multiple sources resulting in duplication.

Reliability

As discussed above, the cause of the deaths in the scheme MCCD 2007 being certified by the medical professionals in person, they could be considered highly reliable. However, no specific quality measure being carried out, being added by the limited coverage and varying levels of efficiency across the country puts the reliability of the scheme doubtful. In contrast, the intense quality assurance procedures carried out in the RCD 2001-2003 along with its relatively better coverage, completeness, and representative character renders the data available under the scheme more reliable. The analysis done above clearly indicates a possibility of duplication of data in the GBD 2007.

Pattern of cause of deaths

The decrease in the communicable diseases, maternal, perinatal, and nutritional conditions in the MCCD 2007 could possibly be because of poor almost nil rural representation in the medically certified deaths. Similarly, the low percentage of NCDs in the RCD 2001-2003 could possibly be because of the increased rural representation of the sample units. However, these explanations might not justify the differences observed in the two reporting systems when compared with GBD 2007. As one of the disease groups are almost avoidable/preventable and require immediate interventions, and as the other group of disease group requires a well-planned and sustained intervention, even a minimal change in their absolute numbers can make a significant impact in the policy decision, thereby on the healthcare and on the economic status of the nation. If a similar difference exists between each of the systems and decisions are taken relying on only one of the two reports, MCCD 2007 and RCD 2001-2003, there is high possibility that one group of disease is neglected.

 Conclusion



The completeness of the reports MCCD 2007, RCD 2001-2003, and GBD 2007 were found to be 0.19, 0.81, and 1.17, respectively. Wide discrepancies were also observed among the three databases, in terms of the absolute number of deaths registered, crude death rate and the pattern of cause of death in India for the year 2007. This discrepancy was significantly higher when the same was performed for individual states/UTs and considering only the medically certified deaths.

High level of correlation ranging from 0.8 to 0.9 was observed between the reports when compared with respect to individual causes of death. However, RCD 2001-2003 had high correlation with both MCCD 2007 and GBD 2007.

Though the reliability of assigning cause of death appears to be significantly higher in the scheme of MCCD, as the deaths are being certified by the medical professionals in person, the system lacks representative character owing to the different level of efficiencies observed among the different states/UTs, poor coverage, poor completeness errors in data, and lack of validity measures.

Though the assigning of the cause of death is not done by the physician in person and is rather done by retrospectively examining the reports collected by field supervisors under the RCD, the intense quality measures carried to ensure the quality of the data, good coverage, representative character of the sample units, accuracy, reliability, and good comparability with other reporting systems makes RCD, a more commendable one.

As data are been pooled in GBD to arrive at the cause distribution for India, the possibility of duplication of the reports amidst of increasing reliability could not be ignored, which is also been clearly proven in the above exercise.

As one of the disease groups are almost avoidable/preventable and require immediate interventions and as the other group of disease group requires a well-planned and sustained intervention, even a minimal change in their absolute numbers can make a significant impact in the policy decision, thereby on the healthcare and on the economic status of the nation. The asymmetry observed among the reporting systems in providing such valuable information on the disease burden could considerably influence the decision making.

Having multiple reporting systems with the same objective could also pose unnecessary duplication of efforts and wastage of various resources. This strongly emphasizes the need for reorganizing/collating the existing systems of collecting cause of death and bringing down quality measures in such a way that a quality, accurate, and uniform data is available on cause of death which can aid the policy maker in evidence-based decision making.

Recommendations

It is clearly evident from the above exercise that discrepancies exist between various systems that report cause of death. And none of the system completely encompasses the cause of death on mortality for the entire section of the country. However, the data collected through the SCD was found to be more countable in terms of its representative character, reliability, coverage, quality, and completeness. Moreover, it can't be denied that a real time certification of death by a qualified medical professional will be comparatively superior than a retrospective death certification by physicians using signs and symptoms collected by field staff with quality checks. Moreover, both the RCD and Medical Certification of Death come under the domain of the ORGI, which indicates a clear duplication of efforts with varied outcomes. Having said this, it is recommended that the MCCD shall be emphasized and enforced in all the states to ensure that every death that occurs is been certified. Clear amendments to facilitate the same shall be brought in place. The details of the death shall be continued to be collected as a part of the vital statistics, which in fact can supplement the MCCD process. An annual or periodical retrospective survey could be done to validate or adjust the data that has been collected through the MCCD.

Study limitations

The only major limitation of the study was that all of the databases that have been compared in this study did not belong to exact time frames. However, an attempt has been made by the researchers to compare the databases of similar time period. The databases for the systems MCCD and Sample Registration Survey were taken for the year 2007. The database for RCD (2001-2003), though it has been carried for the years 2001-2003, was published in 2008. However, the database for GBD was comparatively older one which has been updated for the year 2008. Though the databases used do not belong to exact time frames, the exercise carried out are sufficient enough to throw insight on the different reporting systems on cause of death in India.

 Acknowledgment



I would like to thank Dr Rakesh Sarwal, Adviser, Health and Family Welfare Division, Planning Commission, New Delhi for his kind and continuous support. He supported all the way right from setting of objectives to recommendations. Also, I would like to thank Dr. Amandeep Singh, Consultant, Health Division, Planning Commission, New Delhi for his guidance at every step.

Last but not the least I would like to thank my family and all my friends for their unconditional support.[15]

References

1Mathers 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;83:3:171-7.
2Office of registrar General, India. Report on Medical Certification of Cause of Death, 2007. Ministry of Home Affairs, 2013; New Delhi.
3Murray CJ, Lopez AD. Global mortality, disability, and the contribution of risk factors: Global Burden of Disease Study. Lancet 1997;349:1436-42.
4Office of registrar General, India. Report on Causes of Death in India 2001-2003. Ministry of Home Affairs, 2008 New Delhi.
5Murray CJ, Lopez AD. Alternative projections of mortality and disability by cause 1990-2020: Global Burden of Disease Study. Lancet 1997;349:1498-504.
6World Health Organisation. Global Burden. The Global Burden of Disease 2004 update. Available from: http://www.who.int/healthinfo/global_burden_disease/GBD_report_2004update_full.pdf [Last accessed on 8 th May 2013].
7Registrar General, India. (2008). Sample Registration System. SRS Bulletin. New Delhi. India 2008;43:1.
8Anand, K, Chowdhury D, Singh KB, Pandav CS, Kapoor SK. Estimation of mortality and morbidity due to strokes in India. Neuroepidemiology 2001;20:208-11.
9Jha P, Gajalakshmi V, Gupta PC, Kumar R, Mony P, Dhingra N, et al. RGI-CGHR Prospective Study Collaborators. Prospective study of one million deaths in India: Rationale, design, and validation results. PLoS Med 2005;3:e 18.
10Srinath Reddy K, Shah B, Varghese C, Ramadoss A. Responding to the threat of chronic diseases in India. Lancet 2005;366:1744-9.
11Lopez AD, Murray CC. The global burden of disease, 1990-2020. Nat Med 1998;4:1241-3.
12Mahapatra P. An overview of the sample registration system in India. In Prince Mahidol Award Conference and Global Health Information Forum; 2010.
13Reddy KS, Yusuf S. Emerging epidemic of cardiovascular disease in developing countries. Circulation 1998;97:596-601.
14Krug EG, Sharma GK, Lozano R. The global burden of injuries. Am J Public Health 2000;90:523-6.
15Mathers CD, Boerma T, Ma Fat D. Global and regional causes of death. Br Med Bull 2009;92:7-32.