Assessing The Mortality Patterns in Rural Villages of Gujarat State of India Through Field Based Study Using Verbal Autopsy Tool

Background: Well-planned verbal autopsies can provide timely, robust, and plausible information on mortality causes and can have positive influence on disease control priorities. The present study was intended to assess probable causes of death in rural areas of Anand district of Gujarat State by implementing “Verbal autopsy tool.” Material and Methods: A Cross-sectional study was conducted where all recorded 227 deaths from six villages of selected areas during January 2017 to December 2018 were considered for the study. Multiple steps were employed in conducting verbal autopsy including staff training, development of study questionnaires, physician’s involvement in assigning causes of death etc. The data analysis was performed using Statistical Package for Social Sciences (SPSS) Software with Version 24. Results: Out of 227 deaths, the maximum deaths were attributed to non-communicable diseases (NCDs). The predominant NCDs among deceased individuals were cardiovascular diseases (66.01%), malignancy (21.56%), chronic pulmonary diseases (9.80%) and Diabetes Mellitus type 2 (2.63%). Among communicable diseases, nearly 48% of deceased individuals had bacterial infections including Tuberculosis followed by viral infections including HIV (31%), parasitic infections (21%). Conclusions: Despite few limitations, lay reporting of deaths supplemented with physician assignment of cause of death for verbal autopsies, remains a practicable strategy to record the various patterns of mortality reliably for unattended deaths.


INTRODUCTION
Evidence of death and disease is essential in providing standards of health care and eventually to understand the health in the society. 1 Out of 60 million estimated deaths globally, 9.6 million deaths are estimated to occur in India annually, which is 16% of total global deaths. 1 More than 75% of deaths in India occur at home and majority of them do not have certified cause. 2 Mortality statistics provides valuable insights into the health status and current and future health needs of population. However, the patterns of mortality as well as associated rates such as crude death rate, age and sex specific death rates, proportional mortality rates still vary greatly around the world. The countries with high mortality rates like South Africa, Rwanda and Republic of Tanzania indicate the pattern of high childhood and adult mortality primarily due to infectious ailments like tuberculosis, HIV/AIDS, and nutritional deficiencies. 1,2 The upsurge in maternal mortality in these nations might be attributed to inadequate healthcare facilities for care of the woman during pregnancy, during delivery and post-natal care. 3 In developed countries like United Kingdom, Australia and United States of America, there is low level of child and adult mortality except in old age, where non communicable diseases form the main cause of death. 1,2 Countries like Russian Federation and Estonia has been shown low childhood and high adult mortality whereas India and Afghanistan report high childhood and adult mortality. 4 Verbal Autopsy is an investigation of chain of events, circumstances, key symptoms leading to death through structured interviews of living family members or associates of deceased. Trained physicians then utilize this information to determine causes of death. It is widely employed for the estimation of cause specific mortality in regions with little or no medical death certification. 5,6 It has also been witnessed that well implemented verbal autopsies can provide timely, robust, and plausible information on mortality causes and may have positive influence on disease control priorities. 7,8 Pertaining to mortality patterns in India, only onefifth (20%) of deaths from rural regions and onethird (33%) of deaths from urban areas take place in hospital settings. 9 Most of the hospital deaths are generally attended by a physician but many deaths which occur outside the health care settings make identifying cause of death very difficult. [5][6][7][8][9] Even the deaths which occur in hospital settings, the use of clinical and laboratory investigation is limited. In conjunction with very infrequent use of post-mortem examinations, the available scarce clinical information makes reliable certification of cause of death very challenging throughout India. [10][11][12][13][14][15][16][17][18][19][20][21] Studies in various states of India have revealed that verbal autopsies can be used effectively among rural popula-tion. 11,12 The supporting evidence also indicates that health workers can be potentially trained to use the verbal autopsy to ascertain the cause of mortality among various population groups. 13 The present study was envisioned to assess mortality patterns by cause, age groups and gender in welldefined rural areas of Anand District of Gujarat by conducting "Verbal autopsy tool" over the period of 24 months. Study Population and Sample Size Estimation: All recorded 274 deaths which occurred in the six villages of the selected area from January 2017 to December 2018 were considered for the study. All deaths were verified with the local community for completeness of death reporting. The houses of deceased individuals in selected villages were approached for conducting interviews of respondents. Of which, 47 houses were identified as locked and revisited the same houses after one week, but the houses were locked. No third attempt was made to visit these 47 houses and excluded from the study. The revised sample size was 227 deaths.

Selection of Deceased Individuals and Recruitment Process
There steps were implemented in conducting verbal autopsy.
Step-1: Continuous monitoring and recording of vital events in the registers such as deaths, births, pregnancies by Female Health Worker (FHW) under supervision of field supervisors.
Step-2: Verbal autopsy interview process in the field area by field supervisors and investigator.
Step-3: Assigning "cause of death" by two trained physicians using ICD-10 classification.
Survey Instrument: The study used validated structured verbal autopsy (VA) questionnaire as a tool for data collection. The questionnaire was modified version of existing World Health Organization (WHO) standard verbal autopsy (VA) questionnaire. 22 It was translated into local language (Gujarati) and its fea-sibility in data collection was ensured. Four separate translated VA questionnaires were used to collect detailed information each for neonatal, child, adult, and maternal deaths respectively. The questionnaire was translated (Forward translation) and back translated (Backward translation) by panel of three independent translators. The back translated version of questionnaire was further assessed by other two multilingual experts. A pilot study was carried out to ensure the validity of the tool and to assess the VA procedure. The VA questionnaire comprised of demographic information of respondents and deceased, information on the events surrounding the death, the cardinal symptoms and probing details for each cardinal s y m p t o m . When available, laboratory investigation reports, hospital records, and death certificates were photocopied and incorporated in the review process. The face validity, content validity and consensus validity of the survey instrument were ascertained before the commencement of the study.
Training of Field Supervisors: Two full days of training sessions were utilized for field supervisors who were female graduates and willing to perform assigned tasks. All sessions were facilitated by experts and were concluded with constructive feedback process. The classroom and field work-based training were imparted to field supervisors. The training agenda emphasized on skills to conduct a good interview, mock VA interview exercises, orientation about cardinal signs and symptoms related to mortality.
Data Collection: The list of the deceased individuals was prepared according to their place of residence (Faliya) and respondents were contacted by houseto-house visit. An approximate duration of each interview was around 20-25 minutes. Two-third of the data collection work was done by field supervisors, while one-third work was performed by investigators to ensure the data quality and to assess the completeness of the field work. Entire verbal autopsy process was supervised and monitored by investigators as per preformed checklist. Weekly review meetings were organized to evaluate the progress of data collection process till its completion. Figure one illustrates the entire process of assigning cause of death. In case of any disagreements among physicians, they were given 'reconciliation task' to ensure consensus. 'Adjudication Process' was also instigated in case of further disagreement between two physicians by consulting third senior experienced physician to make final decision on ICD code.

Assignment of Causes of Death by Trained Physicians:
Statistical Analysis: The data analysis was performed using Statistical Package for Social Sciences (SPSS) software version 24. The frequency tables were prepared for nominal variables. Interval variables (viz. Age of the respondent, age of deceased) were converted into categorical variables and fre-quency, proportions were calculated. Outcome variables were probable causes of death, quality of narrative, certainty of diagnosis. Outcome variables such as non-communicable diseases, communicable diseases were converted into dichotomous variables category for the purpose of applying chi-square test and logistic regression. P value less than 0.05 was considered as statistically significant.   Verbal autopsy forms assigned to two physician coders All 227 Verbal Autopsy forms were received from field supervisor    Table 4 indicates the distribution of co-morbidity profiles of adult deceased who were above 15 years of age. Out of 218 adult deceased individuals, 110 (50.45%) were smokers and tobacco chewers while 35 (16.05%) were alcoholics. Tobacco chewing (X 2 = 6.37, p=0.01) and alcohol (X 2 = 9.17, p = 0.002) showed statistically significant association with the mortality due to NCDs. More than two-thirds (71.1%) of the deceased were vegetarian and around one-thirds (28.9%) were non-vegetarians. However, the study could not show any statistically positive correlation between other risk factors and NCDs induced mortality. Table 5 explains the relationship between various risk factors and occurrence of noncommunicable diseases among deceased. Tobacco chewing and alcohol were significantly associated with NCDs (p<0.05). The information about variables could not be retrieved for one deceased person.

Reconciliation and Adjudication of Deaths:
The present study also assessed the process of reconciliation and adjudication of deaths. Out of 227 deaths, major disagreements were observed between two physicians in 43 (18.9%) of deaths. Both the physicians were agreed in diagnosis of 81.1% of deaths  NCD -Non communicable diseases P<0.05 = statistically significant before reconciliation, which was raised to 88.5% after reconciliation. Out of 227 deaths, 26 (11.5%) deaths required adjudication by third senior physician.   (Table 6). Overall, the agreement between both the physicians in ascertaining diagnosis was very strong (k=0.76, p <0.001). The level of agreement between causes of death according to respondents and causes of death given by ICD-10 was with moderately strong agreement (k=0.59, p <0.001).
Age Group Patterns: Table 7 depicts distribution of the mortality patterns due to infectious diseases. Except gender, for none of the variables, there was a statistically significant difference (Table 7). However, the detailed information of 5 deceased individuals could not be obtained. The study also detected trends of deaths among different age groups. The deaths due to NCDs revealed rising trends among young and middle-aged individuals followed by decline between age group of 45-55 years and again there was a sharp upsurge among elder groups. The mortality due to communicable diseases was high among individuals of 15-45 years age and with more than 60 years of age. Injuries were more common among deceased of 15-45 years of age.

DISCUSSION
The present study used local female graduates since it is highly expensive to recruit professionally trained persons to conduct verbal autopsy interviews and related tasks. Generally, it is challenging to get female workers willing to perform to do field work but in the present study, all field supervisors (interviewers) were females and were extremely motivated to handle assigned responsibilities.
The VA tool for adult mortality is an open narrative format which includes the check list of signs and symptoms with filters to retrieve additional data about events or circumstances associated with death. This tool is highly sensitive and the validity of it was influenced by the intensive training provided to the field supervisors (interviewers) on instant random scrutiny of 5% of interview data and reviewing VA reports centrally by two physicians to ascertain probable underlying cause of mortality which is found to be more effective than opinion-based algorithms. 11 However, in India, there is sparse literature about validity of cause of mortality among adults through verbal autopsy.
In the current study, overall, the deaths from unspecified and unknown reasons reduced from 51% to 19% (p<0.05) in study area. Two full day's comprehensive training to write VA reports followed by sustained supervision of submitted reports led to identify probable underlying cause of mortality for most of the deaths. The specific causes of mortality arrived through VA reports may be utilized to compute death rates for Anand District of Gujarat State.
One of the significant findings of the present study was that physicians were accurate in appropriately certifying causes of mortality among the target population. In this study, during reconciliation process, only among 18.9% of deaths, there was major disagreement between two physicians; otherwise, they diagnosed causes of deaths accurately in the majority (81.9%) of deaths. This might be attributed to high level of agreement between two physicians in first round of coding only and good quality of verbal autopsy narrative. Physicians are precisely trained to assess pathological processes and, in principle at least, to suitably apply the directions and procedures of the ICD to certify the cause of death. Reporting of deaths paired with physician coding assignment of verbal autopsies, despite some challenges, is more practicable technique to document mortality patterns than automated algorithms. 16 These findings were supported by the present study.
In our study, the leading causes of mortality, in descending order, were non-communicable (NCDs) diseases, infectious diseases and injuries. Prospective study of one million deaths conducted by Jha P et al. 16 reported NCDs as cause of death among 42% of deceased individuals. A MINErVA study carried out by AIIMS New Delhi 17 mentioned NCD as major cause of death among 55% of study population.
Comparing to these studies, the mortality rate due to NCDs was higher in the current study. The difference in findings could be due to enhancing trend of NCD induced mortality over last decade due to changes in life-style patterns as well as social and economic transitions in the society.
In the current study, the majority (96%) of deceased individuals were adults and remaining 4% were neonates and children. Two-third (64.8%) of deceased were males and one-third (35.2%) of them were females. Almost similar findings were also noted in studies of Jha P et al 18 and Khademi H et al 19 .
A key element in the reliability of data regarding cause of death through verbal autopsy (VA) is the recall period. The existing literature shows varying optimal recall time to attain maximum validity of a VA from as soon as the mourning occurs up to 12 months and beyond. 20 World Health Organization (WHO) recommends that, after a period of death, the verbal autopsy should be implemented as soon as possible and recalls exceeding one year must be interpreted with caution. [20][21][22] In the present study, around 80% of interviews were conducted within 12 months and recall time exceeded one year for 20% of interviews which were derived up to maximum 24 months. Overall, in this study, the recall time during verbal autopsy was consistent with WHO recommendation.
The existing study indicated the mean crude death rate (CDR) of all the deaths occurred in six villages was 9.2 which was comparable to national average CDR of rural population of India. 23 By comparing CDR of study site (9.2) with CDR of rural residence of Gujarat State (7.3), it was seen that, the current study had higher than state average. This disparity might be possibly due to variation in demography across the other parts of Gujarat state of India. The present study also reported comparative findings of gender specific CDR in males (11.1) and females (7.3) with national CDR of 8.3 and 6.8 in males and females respectively. 23 These differences could be attributed to differences in the geographical and demographic patterns across other regions of country.
This study showed that information on probable causes of death can be effectively obtained by this alternative tool (planned verbal autopsy) and the study can also be expanded to cover remaining 21 villages of study area. In the current study, it was also observed that, even though, the performance of field supervisors in conducting interviews was satisfactory, the study revealed few lacunae in their functioning. Peripheral health workers such as female health workers (FHWs), field supervisors should be trained in conducting verbal autopsy and refresher training should be taken at regular intervals. Resident doctors can also be involved in capacity enhancement as well as monitoring of these health workers.
The present study recommends few reforms in government health system. Undoubtedly, to ensure the authenticity and validity of verbal autopsy method, it should be implemented by government as a routine data collection tool at Primary Health Centres (PHCs) and grass root level staff such as Multipurpose Workers (MPHWs), Accredited Social Health Activists (ASHAs), Anganwadi Workers (AWWs) should be trained in conducting verbal autopsy. Panchayat (local government) system in the village should be linked with health care systems such as sub centres and PHCs. Deaths registered in the Panchayat should be regularly informed to PHC Medical officers and accordingly verbal autopsy can be carried out by concerned authorities. Similarly, death certification system may be implemented at the PHCs and other government health centres.

LIMITATIONS
An important potential limitation of the present study was the small sample size and restriction of sample only from rural settings. Therefore, the emerging findings cannot be generalized. The reliability of the information on symptoms at the time of deaths given by the respondents was based on cooperation and reliable description on the part of the respondents. Additionally, extra medical information of deceased regarding investigations, diagnosis, treatment received was not available with some of the family members. These things might have affected the quality of narrative and probable causes of death. In the study, maternal deaths had not been documented so maternal autopsy could not be performed and conclusions for maternal deaths could not be drawn. Neonatal and child deaths were also less in number, so valid conclusions about their mortality causes could not be made.

CONCLUSION
Lay reporting of unattended mortalities with cause of death assignment' by physician can be widely practicable. High prevalence of NCDs induced deaths among population of age groups less than 45 years and geriatrics calls for well-planned and an effective life style modification measure. The upsurge in mortality due to infectious diseases like tuberculosis and HIV among young population reiterates the need for targeted interventions to control similar diseases. The outcomes of this study can be beneficial to the local health authority in establishing health and research priorities. However, multicentric studies with large sample size are highly warranted to generalize the study findings.