Scroll Top

Getting reliable mortality and burden of disease estimates – So near but yet so far!!!

Conversation between Dr Chalapathi Rao (CR) and Prof T Sundararaman (TS

Reliable mortality data is essential for measuring the health outcomes of Universal Health Coverage (UHC) and for health planning in general. In this conversation, we discuss the sources of information on mortality and note that while there is ongoing work to improve death reporting and Medical Certification of Cause of Death (MCCD), and this includes digitization of data, we have not arrived at a situation where any of this can be used for guiding policy or action. The only information readily available is the Global Burden of Disease and preventable mortality indicators generated by IHME, but these have considerable limitations, but our focus should be on why it is feasible and desirable to do much better with the data and data systems we already have. Much improvement can happen if the states assisted by the public health community take the lead in this area of public health practice and research.

TS: I’ll get to the larger question of measuring Universal Health Coverage (UHC) a bit later. But I’m going to start with this what do we know about the sources of information on adult mortality? There is mortality data available in the census, the NSSO health rounds, the National Family Health Survey (NFHS), and the Sample Registration System (SRS). Then there is the registration of deaths in the Civil Registration System (CRS) and Medical Certification of Cause of Deaths (MCCD). Would you give me a sense of what are the pros and cons of each source? We are talking of mortality from all causes and not just maternal mortality.

CR: Let’s start with the census. The census does have information on under 5-year-old child survival. There is a series of questions on birth history, from women from 15 to 49 years, asking about how many children were ever born and how many of them are still surviving. To that information, they apply a series of equations, which are known as the Brass methods by which they can calculate the risk of dying between birth and five years. I am not aware of whether the census includes a specific question on maternal mortality or all-cause mortality. I don’t think so, because I haven’t seen a published report on that, but one needs to check.  In some other Asian countries, in the census, they also ask a question about whether there has been a household death in the last year or three years and they use that data to estimate mortality, but in the Indian census, they don’t. Census-based child survival estimates have several disadvantages- related to low periodicity, recall bias especially with regard to still births, and issues on statistical methodology. So, although this data is available in the census, this has not been taken seriously, and even now other sources have better information.

The NSSO health rounds are also not reliable for mortality data. There is a recall bias, and the sample size is not adequate for measuring adult mortality rates, or life expectancy at birth and anything related to the causes of death.

Even the Sample Registration System (SRS) and the NFHS themselves are hopelessly inadequate in terms of sample size, to be able to estimate an adult mortality rate with any degree of precision, which can be put to use for either assessing cross-sectional scenario at a point in time or to measure trends over time.   The confidence intervals are so wide that they will always overlap, and you’ll never be able to interpret differentials. I have the data published on this for SRS1, which has a sample size several times higher than the NFHS. So, if the confidence interval is wide for the SRS, it will be even wider for the NFHS

TS: SRS publishes data on all-cause mortality. NFHS does not. Both SRS and NFHS publish data on child mortality, but they seldom match. So when you compare the SRS and NFHS, which is the more reliable source and why do you think so?

CR: I would actually not rely on either of them, but if you want me to compare them, then we need to actually go a little deeper into the methodology of each source. The sample size in SRS is designed to measure the infant mortality rate (IMR) at the level of the “natural division” within 15% relative standard error. A natural division is a big geographical region somewhat made up of a cluster of districts, so each large state has 2-4 natural divisions. We see these natural divisions used in weather reporting- an area like Cooch Behar in West Bengal, or say Vidarbha in Maharashtra. Now, based on the sample size, SRS is precise enough to measure the IMR. But in that same cluster, which has been selected to measure IMR the SRS is also recording the mortality at all ages. But obviously, because adult mortality is rarer than infant mortality in most populations, the confidence intervals will be much wider than 15% as a relative standard. ​​So, that is the reason why there is no substitute to the Civil Registration System (CRS), in terms of sample size given the vast population and the wide dispersion that we have and the need for precise mortality measurement at least at the natural division of the state level, if not at the district level. For our population size, if we accurately measure mortality through the CRS, we can get very robust estimates even at the district level and we are actually looking into some of those analyses as we speak.

TS: Yes. So this is one big message that we have arrived at. The civil registration system (CRS) cannot be substituted.  But before we leave SRS, could we agree that SRS would be reliable in a larger state, and perhaps the current practice of giving mortality estimates of a five-year period instead of one year will provide more reliability?

CR: There is one more drawback we can say of the sample registration system, even for the under-five mortality. Although they may be doing a fair job for the under-five mortality at the state level, stillbirths are under-reported and we need further attention to measuring stillbirths and perinatal mortality rates.

SRS does publish a 5-year aggregate of deaths. The life tables from the sample registration system are grouped for those 5 years. They are useful and I have used some of those reports for my comparative analysis. The most recent of these is the 2015 to 2019 report. But we need to check them.  In principle, aggregating the data would yield a stronger state-level estimate. And we might also conjecture that the mortality may not change so much over a period of five years.  But this is still a lame excuse. In states which have a population of at least 40 to 50 million population, we definitely need indicators at the sub-state level because there are going to be differentials and you can’t apply a uniform rate for such a large population. So, that is something we have to keep in mind when we aggregate data across years. It is, statistically, somewhat like cheating in doing that. Note also that SRS involves an additional supervisor who visits all households in each sample cluster every six months, and independently records vital events that had occurred during this period; which is then compared with the records maintained by the local enumerator.  A reconciled list of events is then prepared, based on the matching, thereby providing more reliable data for the villages in this sample. So the data collection methodology of the SRS is sound, the problem is with the limited sample size.

TS: So let us come back to Civil Registration System (CRS) which is where we will focus the rest of our discussion. In CRS, the basic data is from form 2 which is a death report filed by a non-medical notifier.  If the death is attended by a medical officer they will fill form 4, which is the MCCD form, and if the death is elsewhere or brought dead in an ambulance to a doctor, then Form 4a is filled. Am I correct? Can you fine-tune me on this?

CR: All deaths are to be reported using a form 2. Whether the death happened at home or in a hospital or anywhere, it is to be reported using form 2.  There is a list of notifiers and registrars for local level on who can fill in the form and to whom they must submit it to at district and sub-district level and state level that is issued by the registrar-general of births and deaths. On form 2, the notifier also records the names of the informant from the household and they are supposed to get the signature or thumbprint of the informant because in official terms this data is being reported by the household to the government. [Of the total 36 states and UTs, in about 20 states/UTs it is the panchayats or general administration which notifies and registers in local and the district level. In about 15 states/UTs it is the health department which does so at the local level and in about 10 of these it also has the registrar-function at the district level as well.  Before independence and for many years after, in most states it was the police department which was the registrar, with the village kotwal playing the notifier. Now that arrangement exists in only J &K. In all other states it is the panchayat system and general administration or the health system which is in charge. Earlier, as inherited from the colonial state, registration was a citizen duty and the state bestowed it or denied it as part of the exercise of its powers. Now registration is increasingly perceived as an entitlement- which is a welcome development].

Deaths which have been in a health facility or hospital, are to be issued both form 2 and form 4 by the hospital.  So, when they are filling in the form 2, they get the signature of the person who is taking the body and they give him one copy. They also send one copy of form 2 and one of form 4 which is the Medical Certification of Cause of Death (MCCD) to the Registrar’s office directly.

The third type of deaths are those that were ‘brought dead’ to a physician or which happened outside the health facility, but which was attended to by a physician and for these he/she issues the form 4a. The family is then required to go to the Registrar and report the death, where they will fill in a form 2. And this form 4a is not given to the family. It comes directly to the Registrar’s Department and sent to the Health Department for further processing of all MCCDs.

I’m not really sure about the extent to which form 4 and 4a is implemented across the country. There may be some data from the Registrar General’s report.  I know, that in Tamil Nadu about 30% of deaths are reported with a form 4 and 13% across the state are reported with a form 4a, so an so the total MCCD coverage is 44%, which is quite good. Situation is similar in Goa, Maharashtra, Kerala- but these are high compared to other states. I also know that in Maharashtra and Goa (a much smaller state), the coverage of MCCD is quite high compared to other states, but across the country, there is a lot of shortfall in the availability of this information.

TS: If I remember right, MCCD coverage of deaths would be about 25% across the country?

CR:  Yes. I suspect that there are likely to be a lot of hospitals, where they are actually filling in the MCCD form, but the form is not progressing into the Registrar’s system for some reason or the other. So, if we pay a little more attention to compiling with greater efficiency, whatever forms are already being filled, and we might be able to have a much larger MCCD coverage.  To substantiate this, note that from SRS, nearly 50+ % of deaths in the whole country were hospital-based. Since 2015-16, MCCD reporting has been made mandatory across all hospitals in the country. And Central Bureau of Health Intelligence (CBHI) have conducted a lot of training programmes and distributed materials on MCCD.

TS: I understand that before 2015, MCCD was by policy limited to medical colleges hospitals and larger hospitals, and even district or much less private hospitals were not required to fill it. But today has MCCD expanded even to primary health centres?

CR: It is expanded to all clinical establishments with an inpatient facility providing treatment where a death can take place. So, unless our primary health centre has a delivery unit or something like that, around the clock, it won’t be implemented. In other words primary health centres, and private outpatient clinics do not fill form 4. But to both public and private hospitals it is a mandatory and a legal requirement.

TS: is there a chance of duplication of data from form 2 and form 4, where a hospital files the death and so does field level functionary?

CR: There may be, but I think that they in the existing system. when this form comes to the local registrar’s office they will check in and ensure that there is no double counting of events. This is unlikely to be a significant problem.

TS: The MCCD records only hospital deaths, which is about 25 percent and there is no clear denominator for these deaths. So, other than the legal requirement of certification of death, what public health value does it have currently?

CR: True. It does not, by itself, have public health value, since one cannot, estimate population-based mortality rates or even the leading causes of death. Those who die in hospitals are not representative of deaths in the general population. If we could combine it with data from form 2 we would do better.

TS: So back to CRS and form 2 as the main source of reliable public health data on mortality.  What would be the amount of completion of form 2 across the country and in Tamil Nadu in particular?

CR: We were getting an annual Civil Registration System report based on this form 2 data. The most recent report as we have seen is for 2020, which was published around mid-2022. Since then, we haven’t had a report because of various administrative reasons related to the covid pandemic. We are expecting the reports to resume in the near future.

In the latest, report, the level of completeness is estimated at 92% across the country and in several states, it is 100%.  But there is a problem in how they arrived at this number.  I have done some analysis for the data for 2019 and previously also, in which we have estimated the completeness at national level and for each state and we have published this along with its methodology on which it is based2. It was about 79% for males, and around 72% for females, and if you put both together, it’s about 75%2. We generally do not publish or discuss mortality statistics for both sexes together, because it doesn’t have any relevance. So, we always calculate it separately. The shortfall is mostly in the counting of infant deaths, even for state like Tamil Nadu where completeness is 98 percent for men and 95 percent for women.  For Andhra Pradesh, the completeness estimate based on form 2, is 94% per males and 85% for females. And the other end for Bihar, it is 48% for males and 35% for females. Karnataka is 96% males and 88% females, and Kerala is 98% and 97 %. And Tamil Nadu is 98 and 95. Gujarat is 89% for females and about 90% for males. In fact, I would say that, except for a few states, the completeness is sufficient for us to start thinking of generating state and district level data with population denominators. (the latter needs the next census which is now many years overdue).

TS: Now I’m going to go into my main question. What we have understood till now is that of all the data sources, form 2 is the most robust and for many states we do have this report with high reliability and completeness. What prevents us from presenting this information annually and putting this data up in the public domain? Not as yet by cause of death, but definitely by age and gender. If from smaller research studies we can have an age and gender-wise cause of death and we could apply this to the CRS age and sex-related mortality rates, and we then would have a fairly robust measurement of important UHC outcome indicators at the district level.  So why don’t we have a district-wise annual report of mortality? Do other countries have that and can we have that?

CR: So, rather than answering the question of why we don’t have that, maybe we can focus on whether and how we can have that? The first enabling factors is this circular3 issued in November 2017 by the office of the Registrar General that empowers the state governments and their registrar generals to calculate the vital rates through data from the CRS in all states, at both the district and at state level beginning 1st January 2018. This authorization of the states means that the public health community can inform and work with states to close this gap. Need not wait for the center.

Secondly, if we look at the civil registration system report of 2019 4. we see the details of number of deaths registered by state, union territory and district and they also have deaths by age and gender (pages 62 through the 69). So clearly the deaths by age and gender for each district is available.

TS: So it is clear that age and sex-wise data is available by district in most districts, and even the other states can catch up. Lack of digitization has been mentioned as a problem, but in my understanding most states have already digitized. This would mean almost real-time availability of this information. The goal is so very near, even for presumptive cause of death. But the digitized information is not used. Form 2, captures data on paper form and then the data is entered into computers. But after the data is entered, the aggregate of that disappears.  There is no output from that path. There is a manual aggregation process that happens in parallel and that seems to be actually be data that is on flow. But even this is not used for state and sub-state information and action.

CR: Yes, Form 2 is a manual process- but this is essential. Because form 2 is an official paper record, which has to be maintained at the local registrar level as an archive. Then the data from form 2 is entered into the death register which is now becoming electronic.  So they have this dual process of data entry and transmission, but also the paper archives, and totals at the local registrar level. I am aware that in both Tamil Nadu as well as Punjab, all the variables of form 2 (including the presumptive cause of death) are entered in the computer and they are submitted to the state vital statistics unit. However, the cause of death data from form 2 is unavailable in the reports. Till this computerization started, such data was not available for analysis. The information was only on that paper record which is kept at the local registrar level. Now this data is available and adequate for analysis for all age and gender specific mortality rates at state and district level. However, with regard to the data on presumptive cause of death, though its availability now is a huge step forward, it still has a major problem of reliability.  Form 2 is technically data as reported by the family, and the quality of presumptive cause of death reported depends on whether a documents with proper medical diagnosis or prescriptions are available, or it is only the family’s knowledge and interpretation and what it wants to share. Often family statements would be very broad like heart disease or lung disease. I presume that a lot of the data will be unreliable because the purpose of that presumptive cause of death column on form 2 is only to distinguish as to whether the death was from natural causes or was it associated with any medical, or legal, issue.

Neither notifier nor local registrar is trained to understand the technical terms that are present in reports from a hospital and derive a proper diagnosis on that. And we cannot blame them. You know, their function is not epidemiologic. You know, their function is from a social and government perspective that if a person has died, then their death has to be registered and the form has to be given to them for them to carry on with all other things.

TS: But should we not be able to address this gap. We note that the SRS itself is a way of going deeper and putting another layer of verbal autopsy inquiry over the presumptive cause of death reporting and thereby improving the counting of births and deaths. There have been very good pilots run by people like Yogesh Jain, in JSS where they have used interactive voice recording with notifiers to also get a better quality of death reporting. We know that standardizing symptomatic, presumptive reporting is possible and this would add great value to CRS reporting on deaths. There is a WHO standardisation of around 128 diagnoses or even less that was suggested for verbal autopsy backed presumptive diagnosis of cause of death. Even the HMIS under NRHM had way back standardized about 12 diagnosis entities on which till today HMIS reports deaths. These categories are broad like fever related deaths, trauma related deaths, cardiovascular related deaths etc which  in combination with age, sex and geography is enough for quite a lot of district and sub-state health planning and actions. . Even if causes are relatively unknown, one can unpack it with information from surveillance sites. The other way to do it is to, like for IMR, use information from research studies to  look at the nature of deaths in a specific age group and then apply it to the other figures elsewhere.

So why aren’t we able to make these presumptive causes talk more? By standardizing protocols and giving training we should be able to achieve this almost immediately.  I believe the CBHI and the mainstream departments are focused on Form 4 and MCCD, which have inherent limitations even if we get it correct. On the other hand,  information from form 2 is reaching universal coverage and is digitized. Just a little bit more is required.

CR: I completely agree we could do get an adequate quality of mortality report with CRS data if we do this. There have been, to my mind two barriers to doing so: neglect and disconnect. Neglect because nobody paid attention to try and strengthen this cause of death reporting in form 2 through any kind of standardization. But what I want you to focus now is on the disconnect, in that research projects in this area are few and have never really translated into proper action which is integrated with the regular death reporting program under the CRS.  Because there has to a proof of that concept on scale, and only then can it be considered to be scaled up and strengthened across all state populations. We can get reliable data with numerators based on defined denominators even now, but the reasons we have not done so can be attributed to neglect and the disconnect.

TS: Right. So, what is being done by the government to address this challenge in Tamil Nadu?

CR: So as opposed to neglect, the TN government has been paying attention. And as opposed to disconnect, an implementation research project is underway, which connects research on data quality improvement with the routine operations of the CRS system. The project us led by the Director of Public Health, with technical support from the WHO. I am aware that field work is being undertaken in two districts Karur and Krishnagiri to record causes for all the deaths in these districts. The district population is the denominator. MCCD forms (form 4) are assured for all the deaths in facilities, and form 4A for deaths which happen at home and certified by a physician.  And for the rest of the home deaths which is around 55 %, the WHO verbal autopsy questionnaire is being used to investigate the causes of death. A detailed presentation of the project aims, design, methodology and progress till April 2024 was delivered to the Verbal Autopsy Reference Group Community of Practice, on 25th April 2024.

The field methods for home deaths are quite straightforward. The health inspectors who are positioned in each PHC get the list of deaths for which there is no form 4. They get that list from the local registrar. The have been trained in conducting VA interviews. The health inspector goes to the identified household with a tablet and does the verbal autopsy interview and this gets uploaded to the server. As of now, much progress has been made for deaths in 2023, And recently, training was provided to a team of doctors to review these questionnaires and assign the cause of death based on the information from the questionnaire. The training team includes staff from the Madras Medical College, Nation Institute of Epidemiology and myself. Academics from medical colleges in Krishnagiri and Karur are also involved in the project to help build local capacity, support field supervision, guide local monitoring & evaluation, and other technical support. The cause of death from VA will be ascertained by a local PHC medical officer. It is anticipated that by the end of the year, a detailed analysis of mortality by age, sex and cause for these 2 districts for 2023 could be available.

TS: Are you using this process to upgrade form 2 to form 4?

CR: Yes. Rather, form 2 will stay as the official death report, but will now be supplemented by a VA-derived cause of death to be used for health statistics, instead of the presumptive cause noted on Form 2. But this VA COD will not be called a form 4. Because Form 4 is legally the medical certification by the physician who has observed the death. Here, the physician is basing his opinion on a derived diagnosis in retrospect. So, we are calling it, verbal autopsy cause of death, but the format and the structure is similar to the form 4 of a hospital.

TS: How many diagnostic categories do you have in the verbal autopsy cause of death? What is the level of standardization of this?

CR: Diagnosis can be of different levels of granularity, as it is based on information that is provided by the household in the form of a medical reports and this varies. Broadly it conforms to the International Classification of Diseases – and we go that level of granularity which the information provided allows us. We could for example say a 5-year- old female child died of pneumonia if we have such a diagnosis or we could state that the 5-year-old female child developed fever acutely and died at home if only that is available. If other the other hand we have information that child had fever, cough, breathlessness and chest pain, they will diagnose it as pneumonia. They will not list these four symptoms. The WHO VA standards include a list of about 65 or 70 cause categories of public health importance, which could be used by physicians to assign the COD.

TS: Now onto the big question and this should have been the starting point. Now given all the problems we’ve discussed, should we place our greater reliance on the Global Burden of Disease estimates and the UHC indicators like deaths in the 30 to 70 age group generated for all countries including India by the Institute for Health Metrics and Evaluation (IHME)? Are these reasonable alternatives to reports based on our own mortality data given the problems we are facing?

CR: The Global Burden of Disease estimate is based on statistical models where all available information from whatever sources of data are available for India are said to be used. This information is used and estimates are provided at both the national and at the state level.

The convergence between local data based on projections and Global burden of disease estimates have been studied and published for all-cause mortality as well as cause specific mortality and for a range of indicators. In this publication4 we show that the global burden of disease overestimates life expectancy at birth at the state level. We note that the GBD estimates do not use CRS data, but use only SRS data and for under-five mortality they use the NFHS data. And they claim some other methods, which I am not going into, but which I do not find convincing. When you overestimate life expectancy, you are underestimating mortality.  When you underestimate mortality, your health programs will not be sufficiently powered to make decisions and study trends, and even the resource allocation to and within the health sector may get undermined.  This overestimation of life expectancy is true for both males and females.

I had previously collaborated with the Institute of Public Health in Malaysia5 for cause-specific mortality, using similar methods as being implemented now in Tamil Nadu. In that study, the age-standardised mortality rates from the study, based on country’s mortality data, were compared with those from the GBD for the 10 leading causes of death in males and female. And we found that in 7 out of 10 of the leading causes in males and females, the causes were statistically significantly different.  So GBD is either under-estimating or over-estimating. In many smaller countries, GBD and preventable mortality estimates are based on mortality data from other countries. These are of little use for either planning or measuring progress. Some governments accept the GBD data, but at times, as with Covid 19 deaths, they are not seen to be acceptable.

My point of view is just pointing fingers at GBD is not going to get us anywhere. We have to pull up our socks and produce our own reliable mortality data and compare it with the GBD and see which one is more plausible. If we just take GBD estimates at face value or if we outsource our analyses by giving them whatever data is available without trying to do anything ourselves, then we might end up with an estimate which is not reflecting the truth and which we ourselves do not have confidence in.

So in conclusion, what I’d like to say is.  Ensuring reliable CRS mortality data is feasible in the short term.  If you look at what is happening in Tamil Nadu’s they have made a lot of groundbreaking progress with data on mortality in the two study districts, covering a population of approximately 4 million (about 28,000 deaths in 2023). There is also an intention to use available resources and the critical capacity from these two districts to be able to scale up in other districts next year. The methodology is not rocket science. And my view is, given that UHC is related to the Sustainable Development Goals that need to be achieved by 2030, there is enough time for India to build up the required capacity over the next 3-5 years and implement a sufficient number of activities covering larger numbers of states to be able to have their own data and be able to draw inferences and conclusions based on local data and analysis, rather than depending on external agencies.

If you remember, I met you in 2013 in NHSRC when you were then its executive director, and we had discussed this topic. Even now it is the NHSRC for example, with its current leadership, and its many state avatars which can take the lead in building up the capacity required for this across the states. Given the fact that the Registrar General’s November 2017, order has empowered state governments to generate their health outcome indicators from CRS it is an urgent need to draw the attention of public health researchers and practitioners to help the government close this gap.

TS: Would you like to comment on the million-death study?

CR: That is a topic in itself. And quite controversial. We should learn from it. But Ill like to take a pass on that. Perhaps its best left for another conversation.

TS: Just another minor detail to conclude with. In the measure of UHC, other than certain disease-specific outcome measures, the indicator that you have talked a lot about is preventable mortality in the 30 to 70 age group. You have a number of papers on that. This is one reliable indicator which we would be able to generate at the district level from an improved CRS to measure progress towards UHC? Could you help us understand this indicator better?

CR: In 2014, WHO developed an indicator for non-communicable disease (NCD) mortality as an important indicator for measuring UHC progress. This indicator for non-communicable disease mortality was defined as the unconditional probability of dying between the ages of 30 and 70 from diabetes, cardiovascular disease, cancers and chronic obstructive lung diseases. The rationale for this age group was that normally deaths from these causes would not happen below the age of 30. And, after the age of 70, there are challenges in accurately measuring what was the actual underlying cause of death because there can be multiple morbidities and age has a leading contribution to make. But we don’t have data on causes of death to be able to filter out how many have died from these four causes as compared to others. So, in my studies, total all-cause mortality in between age 30 and 70 is taken as a proxy.6 The bulk of it would be from NCDs, but even other common causes like road traffic accidents, or injuries or alcohol, or tuberculosis HIV and hepatitis-B are preventable.   The term amenable or avoidable mortality has been used to refer to those who could be saved because of healthcare and preventable includes those saved by healthcare as well as prevented by larger public health interventions.  So “preventable mortality” term can be used as the unconditional probability of dying between the age of 30 and 70 and the estimated population in this age group is the denominator. But I would caution that such definitions and terms are still evolving.

TS: Thanks for this interview. I know that we have kept to the overview and not gone into the much greater depths that you are capable of.  Much of this information is basic, but just impossible to access elsewhere. So, thank you. One reason why we have chosen this topic, is also as an appeal to all our schools of public health and departments of preventive and social medicine, to get engaged in providing support to state governments to generate reliable all-cause mortality data at the state and district level, which would immensely benefit public health action by the government and civil society and enhance their own efforts in public health research. This also makes for data sovereignty and federalism. Where the countries and states have the capacity to generate, own and use their own high-quality public health data and not become dependent on block-boxed estimates which are difficult to verify or replicate.

Dr Chalapati Rao is a research academic at the Australian National University. His key interests lie in the collection and analysis of cause-specific mortality statistics for health policy and research. His experience covers field projects for implementing or strengthening mortality data collection & validation, as well as secondary data analysis, for countries in the Indo-Pacific region. His work in India includes previous research on causes of death in Andhra Pradesh, verbal autopsy validation in Delhi and surrounding areas, CRS system analysis/evaluation, and subnational mortality estimation. His current work focuses on the design and implementation of activities to strengthen cause-specific mortality measurement at the district and state levels.

Reference List

Leave a comment