Cause-specific mortality estimates for Malaysia in 2013: results from a national sample verification study using medical record review and verbal autopsy

Reliable vital statistics on births and deaths (including causes of deaths) are important parameters for monitoring health and are routinely used as evidence for health planning [1]. Numbers of death and mortality rates by age, sex and cause are essential indicators for population health assessment. However, in 2005, only about a third of the world’s countries had almost 100% completeness of death registration [2]. Furthermore, only 23 countries were classified to have high-quality registration data on causes of death [2]. Malaysia is one of few Asian countries with long standing and functional vital registration systems. However, the quality of data has been inadequate for routine estimation of key mortality indicators. This is principally due to a persistently high proportion of deaths attributed to non-specific causes of death. Even in 2013, approximately 30% of deaths in the registration data were noted to be from such non-specific causes, which limits the utility of the overall dataset for population health assessment [3].

Currently, there are two methods in place for death certification and registration in Malaysia, based on the place of occurrence of the death. Firstly, deaths in hospitals are certified as to cause by attending physicians, and these are termed as medically-certified deaths. These include deaths from injuries for which forensic post mortem investigations are undertaken to determine the causes of death from medico-legal perspectives. Secondly, deaths that occur at home are reported to the local police station by relatives of the deceased, who also provide a ‘lay’ opinion of the cause, and these deaths are termed as ‘non-medically certified’ deaths. All information from both certification methods are reported to the National Registration Department (Jabatan Pendaftaran Negara), where annual mortality datasets are compiled and subsequently transferred to the Department of Statistics, Malaysia (Jabatan Perangkaan, Malaysia) for quality control, coding, and analysis. In the past decade, the proportion of deaths that are medically certified has improved from 39.0% in 2000 to 51.5% in 2014 [3].

However, there are several specific problems with data quality, particularly in regard to the registered causes of death from either source. Analysis of the data for 2013 indicated that 12.5% of medically certified deaths were assigned to conditions listed in the chapter titled ‘Symptoms, signs and ill-defined conditions’ in the International Classification of Diseases and Health Related Problems, Tenth Revision (ICD-10) [4]. In addition, 6% of deaths are coded ‘sepsis’, and other smaller percentages of deaths are coded to non-specific conditions such as ‘cardiac arrest’ and ‘respiratory failure’. For the non-medically certified deaths, about 60% are coded to ‘old age’ [5]. On combining the data from the two certification methods, it was found that nearly 30% of all deaths had been assigned such ill-defined or non-specific causes of death. These findings clearly indicate the limited utility of available data on causes of death from registration for health planning, or epidemiological research.

In view of these limitations in the Malaysian Vital Registration (VR) data, this study was designed to apply Medical Record review (MR) and Verbal Autopsy (VA) methods in a national sample of deaths in 2013. The study was commissioned by the Malaysian Ministry of Health in 2014, to quantify the biases in the VR data and utilise the findings to develop national estimates of deaths by age, sex and cause for Malaysia. At that time, the most recent year for which VR data was available was 2013 and was hence chosen as the reference period for the study.

The total sample for this study was 14,497 deaths, comprising 7487 (51.6%) hospital deaths and 7010 (48.4%) home deaths. Field data collection successfully conducted medical record review for 5041 hospital deaths. VA procedures were successfully completed for 6843 deaths, which included 3724 home deaths and 3119 hospital deaths. The study response rate was 67.0% (ie: at least one of the verification procedures was successfully conducted) but this ranged from 40 to 84% across all the districts.

On matching the cases across the two study arms, it was found that there were 2172 cases for which both MR review and VA procedures were completed. Overall, the losses to follow up were substantially higher for non-medically certified deaths (47%) as compared to medically-certified deaths (32%). We investigated whether the losses to follow-up as shown in the Fig. 1 resulted in any major bias in the final study sample, from two aspects. Firstly, we compared the age-sex proportionate distributions of the final study sample for hospital and home deaths with the proportionate distributions for each sub group from the vital registration data (see Table 1). As can be seen, there was close approximation between the proportionate distributions, with < 3% difference in proportions across all sub groups.

CATEGORY	Age group	MALES		FEMALES		TOTAL
		Study sample	Vital registration	Study sample	Vital registration	Study sample	Vital registration
Hospital deaths	0–14	5.0%	5.5%	5.4%	6.6%	5.2%	5.9%
	15–59	44.0%	46.5%	35.9%	33.8%	40.9%	41.7%
	60+	50.9%	48.0%	58.7%	59.5%	54.0%	52.3%
TOTAL DEATHS		3078	46,205	1963	27,898	5041	74,103
Home deaths	0–14	1.1%	1.4%	1.6%	1.2%	1.3%	1.3%
	15–59	24.3%	26.0%	14.3%	15.6%	19.6%	21.1%
	60+	74.4%	72.5%	84.1%	83.2%	79.1%	77.6%
TOTAL DEATHS		1948	35,736	1776	32,363	3724	68,099

We also analysed the potential impact of losses to follow up on the distribution of causes of death in the final study sample. Figure 2 below shows a scatter plot with each dot representing the proportion of a specific cause in the vital registration data on the x-axis, and its corresponding proportion as per registration diagnosis in the final recruited sample for the field study on the y-axis. It can be seen that for the majority of causes, the proportions for VR diagnosis in the national data and in the final recruited sample are very similar (along the 45° line). For ill-defined conditions, the proportion in the VR data was 29.7%, as compared to 32.8% in the final study sample Overall, these findings suggest that the losses to follow up in this study are non-differential according to age, sex and cause distributions.

As described in the methods section, the study findings on reclassification of causes of death from the medically certified and non-medically certified deaths were used to develop adjusted cause-specific mortality estimates for Malaysia. Tables 2 and 3 show the estimated leading causes of death for males and females for 2013 from the study. Ischaemic heart disease emerged as the leading cause of death among males in Malaysia, estimated to cause 12,656 (15.4%) of all male deaths. Cerebrovascular disease and chronic obstructive pulmonary disease are second and third on the list respectively, causing 13.7 and 8.5% of all male deaths. For females, cerebrovascular disease was the leading condition causing 11,057 (18.3%) deaths, followed by Ischaemic heart disease and lower respiratory infections causing 12.7 and 11.5% of deaths respectively.

^a Based on WHO Global Burden of Disease categories

COPD Chronic Obstructive Pulmonary Disease

^a Based on WHO Global Burden of Disease categories

COPD Chronic Obstructive Pulmonary Disease

The changes between the original registration data and the final study estimates are also shown in Tables 2, 3 and 4. These demonstrate the impact and importance of the field research study. Ill-defined causes are no longer the leading cause of death for both males and females. Overall, marked changes in both rank order and magnitude of proportionate mortality that have resulted from the field research are indicative of the problems with the quality of data on causes of death from the vital registration system. In particular the three-fold increase in proportionate mortality from chronic obstructive pulmonary disease in males, and the four-fold increase from diabetes in both males and females have now highlighted the importance of these causes, and the urgent need for interventions to reduce the premature mortality burden from them. Marked increases in rank order were also observed for lung and colorectal cancers in males.

^a Based on WHO Global Burden of Disease categories

COPD Chronic Obstructive Pulmonary Disease

In view of the limitations in the quality of data on causes of death from the vital registration system in Malaysia, this study was designed to conduct a thorough verification of registered causes in a national sample of deaths that occurred in 2013. The results from the verification were used to derive corrected estimates of cause-specific mortality in Malaysia. The mortality estimates for 2013 from this study are clearly more useable than registration data, for health policy, monitoring and research. Ill-defined causes are no longer the leading cause of death, and the verification procedures from the field research have resulted in a large-scale empirical reassignment of these deaths to more specific causes. Furthermore, the reassignment has resulted in major changes to the leading causes of death among males and females. Not only are we more accurately able to quantify the burden of the leading causes of death, but also the changes seen in diseases such as diabetes and chronic obstructive lung disease would clearly highlight the importance of prioritizing, managing and preventing these diseases at all levels.

A similar study to verify a sample of registered causes of death and utilize the findings to develop national cause-specific mortality estimates was conducted in Thailand for 2005 [18, 19]. The Thai study identified that verbal autopsy was found useful to understanding cause specific mortality at population level. Over the past decade, VA had been used as a method to determine cause of deaths in places where majority of deaths occur without medical supervision [20]. It has been used for deaths in various age groups [21, 22] and also across a range of causes of death [23–25]. In Brazil, verbal autopsy has been successfully implemented to improve the accuracy of deaths originally classified to ill-defined causes in the vital registration system [26]. The experience from this study indicates that it is feasible to use VA to improve cause of death reporting in Malaysia and would help increase the accuracy of the mortality statistics for home deaths reported to the national vital registration system. A recent qualitative study also demonstrated a high level of community support in Malaysia for verbal autopsies [27]. From another perspective, the study findings also indicate weaknesses in the quality of medical certification of causes for hospital deaths. Several international studies have also indicated the importance of medical record review to improve the quality of cause of hospital deaths in developing countries [28–31]. Hence, death certification and coding for hospital deaths in Malaysia should also be strengthened in order to increase the overall accuracy of national mortality statistics from vital registration.

Despite the positive outcomes in terms of more reliable estimates, as well as the strengths of the analytical approach used in this study, there are certain limitations. Firstly, retrospective data collection is susceptible to losses to follow-up, as observed in this study. In many of the instances of failure to conduct VA interviews, the primary caregiver of the deceased had moved residence to a different location or could not be contacted despite three attempts. Further, there were cases where the deceased had been living alone, with no surviving relative. Consequently, the VA interview could not be conducted. Finally, there were a few instances of incomplete recording of the address of the deceased in the registration data. In the study sample for hospital deaths, MR review could not be completed in three hospitals, owing to non-availability of manpower to complete this task during the study period. Fortunately, additional analysis, as described in Table 1 and Fig. 2, suggest that the losses to follow up in this study were largely non-differential in regard to age, sex and cause distributions, hence minimising any potential biases in the study findings from this aspect.

Other potential sources of bias include data collection errors during verbal autopsy interviews, arising from inaccurate information provided by family members who might not have been the primary caregiver of the deceased during the final illness. In addition, verbal autopsy is also prone to recall bias as a result of poor memory on the part of the respondent. Difficulty in recalling the symptoms and illness prior to death by relatives increases with the length of time since death. Since the VA interviews were done for deaths that had occurred about 12 months before the date of interview, the large gap between time of death and the VA interview potentially introduces recall bias. Finally, bias could also occur on account of a deliberate pattern of mis-reporting of symptoms by the interviewee, on account of administrative pressures or social stigma, notably in the event of injury causes with medico-legal implications, or from diseases such as HIV/AIDS or tuberculosis. As mentioned earlier, while the study implementation, quality control measures, and data analysis have included provisions to ensure accuracy of study findings, these limitations should be borne in mind.

From a policy perspective, the marked differences between the estimated mortality rates from this study and similar estimates for Malaysia in 2013 from the IHME GBD 2013 Study require careful attention. The estimates derived from this study are based on field research, which involved rigorous processes for data collection, management and analysis, as described in this article. These field research methods create confidence in the resultant mortality estimates, and a sense of political relevance and ownership which enhance their utilisation for national health policy and research. In contrast, the IHME GBD estimates for several causes are based on statistical models constructed from a dataset of mortality and cause of death patterns from a wide range of countries across the world [32] and may not always be applicable to the Malaysian context. Also, the estimation methods apply model-based reassignment strategies for deaths originally assigned ill-defined and non-specific codes in national datasets, as opposed to the empirical reassignment strategies adopted for our study estimates [33]. The IHME GBD model-based estimates for Malaysia could be biased from these aspects. Future mortality estimation exercises for Malaysia could take into account the findings from the research reported here as local inputs for the statistical models which have been extensively used in more recent analyses reported by the GBD Study Collaborators [34–38].

Inaccurate mortality statistics from vital registration have an extensive and often unquantifiable negative implications for national health care systems. While this study provides the first ever Malaysian national mortality estimates derived from empirical local data, Malaysia needs to pay close attention to strengthening the performance of its civil registration and vital statistics system. The research experience from this study, along with the critical mass of human and technical resources developed in hospitals, health clinics, district health offices and the Institute of Public Health should be utilised in devising interventions to improve the quality of registered causes of death.

This study highlights that there is room for and a need to strengthen the mortality statistics system in Malaysia, and this should be made a priority. The verbal autopsy and medical record review methods applied in this study and its findings have improved our understanding of mortality patterns in the country. Verbal autopsy methods should be built into routine practice and scaled up across the country, to improve registration data. Training on medical certification of causes of death should be regularly provided in all hospitals. Studies similar to this one should also be conducted on a periodic basis to improve the empirical basis for cause of death estimation in Malaysia and enable a better understanding of the epidemiological transition that occurs at national and sub national level across the country, and better plan the future health needs of the nation.