Article Text

Download PDFPDF

Deaths due to injury, including violence among married Nepali women of childbearing age: a qualitative analysis of verbal autopsy narratives
  1. Kiely T Houston1,
  2. Pamela J Surkan1,
  3. Joanne Katz1,
  4. Keith P West Jr1,
  5. Steven C LeClerq1,
  6. Parul Christian1,
  7. Lee Wu1,
  8. Sanu M Dali2,
  9. Subarna K Khatry3
  1. 1Department of International Health, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA
  2. 2Sarvanga Swasthya Sadan (Sarvanga Hospital), Lalitpur, Nepal
  3. 3Nepal Nutrition Intervention Project—Sarlahi (NNIPS), Nepal Netra Jyoti Sangh, Nepal Eye Hospital Complex, Kathmandu, Nepal
  1. Correspondence to Dr Pamela J Surkan, Department of International Health, Johns Hopkins Bloomberg School of Public Health, 615 North Wolfe Street, Baltimore, MD 21205, USA; psurkan{at}


Objectives Self-harm and interpersonal violence are important causes of death among women in Nepal. We analysed prospectively collected data to investigate the extent and nature of injury-related deaths among married women aged 15–49 years, recorded through verbal autopsy in rural Sarlahi District, Nepal.

Methods Verbal autopsies were systematically collected on all deaths of married women of reproductive age (15–49 years) over a 3-year period (1994–1997) as part of a randomised community-based trial of maternal vitamin A and β-carotene supplementation. This analysis included a three-way comparison of verbal autopsy data: qualitative free-response narratives, closed-ended responses, and physician-assigned consensus cause of death.

Results We focused on 46 of 559 deaths (8.2%) that were determined to be injury-related. Of the 46, 28% were identified as intentionally self-inflicted, and 11% as intentionally inflicted by another. Inconsistencies were noted between verbal autopsy reports of causes of deaths and physician assessments. Conflicts within the family figured prominently in the narratives. Women with unstable family situations and suffering from mental illness were often described as having experienced violent deaths.

Conclusions Findings highlight that intervention efforts might be necessary especially in situations where there are poor family dynamics or mental health issues in order to prevent potential intrafamily violence and possible death. Results also point to the need for further documentation of violent deaths in rural Nepal.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Globally, the fraction of deaths from intentional injuries from self-harm and interpersonal violence, is increasing.1 Violent deaths have been acknowledged as a public health problem and a leading cause of death in low-income and middle-income countries.2 ,3 However, accounts of these violent deaths are difficult to capture in these settings due to under-reporting and a lack of comprehensive vital registration systems. In place of vital registration systems, verbal autopsy interviews are often used as the standard method for recording data on causes of death.4 However, verbal autopsy methods vary in implementation and have lacked standardisation.5 Verbal autopsy may also face particular challenges when documenting injury-related deaths.6

In Nepal, as in other developing countries, violent deaths among women are of particular interest due to their implications for gender-based violence. Though official statistics on suicide or homicide among women do not exist, a study in eight districts of Nepal conducted in 2008 found that suicide was the leading cause of death among women of reproductive age, accounting for 16% of deaths among women aged 15–49 years.7 Further evidence of self-harm and interpersonal violence as important causes of death is available from the region. In Matlab, Bangladesh, suicide rates among women have also been found to be higher than among men. In more than half these suicides, verbal, physical or emotional violence from another household member was a contributing factor. In the same study, nearly 10% of the study sample were killed by their husbands or in-laws.8

Though violent deaths have not been well documented, violence against women has been shown to be a widespread problem in Nepal.9 ,10 In a recent cross-sectional survey of married women aged 15–24 years in rural Nepal, more than half reported experiencing violence from their husbands.11 A 2006 report found that one in five men in one region of Nepal agreed that wife beating was justifiable, a percentage that is even higher among rural households and those headed by younger men.12 Husband's drinking, polygamy, low household economic status and early marriage are factors shown to be associated with intimate partner violence in Nepal.9

Against a general lack of available data on death related to injury, particularly violent deaths, in developing country settings, we had a unique opportunity to analyzes, prospectively, collected data from a large population-based community trial in rural Nepal. It included standardised verbal autopsy data on deaths of married women of child-bearing age participating in the study in a subdistrict region over a 3-year period.13 Narratives recorded as part of the verbal autopsy interview allowed for insights into deaths beyond that which could be gleaned from the closed-ended questionnaire. For this study, we defined violent death with the Centers for Disease Control definition: ‘a death resulting from the intentional use of physical force or power against oneself, another person, or a group or community’.14

The study aims were to (1) examine the extent to which injury-related deaths are described in the verbal autopsy data and (2) to characterise the causes and nature of injury-related deaths for married Nepali women aged 15–45 years through examination of the free-response narratives.



We analysed data relating to 559 women who died between April 1994 and September 1997. These included all deaths which occurred over 3.5 years among 44 576 women who participated in a cluster-randomised double-blind community trial concerning the effect of Vitamin A and β-carotene supplementation on maternal and infant health and survival (Nepal Nutrition Intervention Project—Sarlahi (NNIPS)—2). Details regarding the study population have been reported previously.13 ,15 ,16 The number of non-maternal deaths was 375, that is, women who enrolled but did not get pregnant and died before the end of the trial. There were 184 deaths among women who enrolled in the trial, became pregnant, and died before the end of the trial.

Participants in our study were married women of childbearing age (15–45 years) living in 30 village development committees (district subdivisions) of Sarlahi district who died during the original study period. This included all women identified through a house-to-house census at the start of the trial, and women living in this same area who married during the period of study. Verbal autopsies were conducted by four trained field supervisors, hired by the study. All interviewers were Nepali males, with a high school education. Interviewers were instructed to conduct the interview with a family member close to the deceased who had knowledge of the events leading up to the death. Interviews were conducted in the home and in some cases additional family members were also present. In order to be sensitive to the family, interviews were conducted at least 1 month after the death occurred. The median number of days between the date of death and the interview was 77 days, or approximately 2.5 months. Interviewers first conducted the structured questionnaire and then asked the open-ended questions. The structured questionnaire included standard questions about the signs and symptoms of common illnesses and pregnancy-related illness and mortality, as well as questions about injury prior to death. Causes of death were assigned following independent review of the structured questionnaire and the narrative by two physicians. If these independently assigned causes were not in agreement, a consensus cause was assigned following discussion. A cause of ‘uncertain’ was assigned to those cases in which there was complete data but the strength of evidence as possible to capture in a verbal autopsy was insufficient to suggest that any one or more event or condition was severe enough to have caused death. The verbal autopsy questionnaire was developed specifically for this study as no standard questionnaire was available at the time of the study.

Verbal autopsy analysis

Secondary data analysis was conducted to analyse verbal autopsies that were collected as part of the trial, particularly focusing on free-response narrative data that characterise injury-related and violent deaths. Our analysis re-examined responses to an open-ended question at the end of a structured verbal autopsy form. These narratives often included informal information gathered from persons other than the primary respondent.

The narratives were elicited through the open-ended question ‘Please tell me how the woman became ill and died.’ Responses varied in length from a one-sentence summary to one to two paragraphs of detailed text. From analysis of the narratives, we categorised whether a death was injury-related or not. This designation was then cross-checked with the physician-assigned cause of death and the response to the question ‘Did the woman have an accident/injury/animal bite before death?’ on the verbal autopsy survey form. As this study was designed to investigate narrative responses, deaths without an injury-related narrative were excluded from further analysis. Injury-related narrative responses were then coded further to identify violent deaths and categorise them by intent. Qualitative data were manually coded separately by two investigators and reconciled through discussion. Coders included a masters-level public health student and a public health faculty member with a doctoral degree in public health.

The original trial received ethical approval from the Nepal Health Research Council in Kathmandu, the Joint Committee on Clinical Investigation at the Johns Hopkins School of Medicine and the Teratology Society in Bethesda, Maryland. A data safety and monitoring committee reviewed and approved the protocol and oversaw the safety of participants in the trial. The verbal autopsy analysis received an exemption from the Johns Hopkins Bloomberg School of Public Health Institutional Review Board.


A total of 46 of the 559 (8.2%) deaths were ascribed to possible injury through analysis of the free-response narratives. Intentional injuries, either self or other-inflicted accounted for 18 of the 46 (39%) injury-related deaths, 3% of all deaths. Of the 46 deaths identified through the free response narratives, 31 deaths also had an injury-related physician-assigned cause of death and 32 family-member respondents indicated that the woman had an accident/injury/animal bite before death (table 1). Thus, nearly a third of deaths identified as potentially violent deaths through analysis of the free response narratives were not identified by the physician consensus method. In these 15 cases, 10, 4, and 1 were assigned a cause of death of uncertain, sepsis and tetanus, respectively. Four deaths were attributed to head injury by the physicians, but were excluded from the textual analysis because the narrative attributed the death to epilepsy in three cases and to miscarriage in the fourth case. Those deaths that family members described as due to epilepsy may well have included falls leading to head injury and subsequent death. Twenty-nine percent of injury-related deaths were identified as intentionally self-inflicted, 11% were intentionally afflicted by another (table 2). Of the self-inflicted deaths, 10 were due to hanging and two due to self-immolation. Of the 18 deaths identified as violent deaths in the narratives, 13 were also identified by the physicians with a cause of death of hanging or burns, four were assigned a cause of death of uncertain and one as sepsis. None of the deaths identified as injury-related occurred while the woman was pregnant.

Table 1

Identification of injury-related death by method

Table 2

Categories of violent deaths or injuries reported in narrative data from Nepali women

We identified narratives that suggested an injury-related death may have been intentional, and categorised these responses. Additionally, we found two main recurring themes describing the context of injury-related deaths: family relationships and mental illness/mental disturbance.

Family relationships

In the verbal autopsies, stressful events in the family or relationships with family members were often described in relation to violent deaths. Examples include the death of a spouse or child, conflict with husband or mother-in-law, childbirth, polygamy and elopement of a daughter. Often, the respondent for the verbal autopsy was implicated in the death, and the interviewer included pertinent information from others. Some examples: 1) The woman had no illness. She died by hanging. Note: According to the dead woman’s mother-in-law the woman hanged herself. In spite of what the mother-in-law says, what actually happened is that the mother-in-law killed her with a single blow from a piece of firewood. Respondent: Husband's family, Physician-assigned cause of death: Hanging 2) Because of an altercation between the husband and wife, the wife was burned (by her husband presumably) and died. Respondent: Natal family member, Physician-assigned cause of death: Burns 3) According to the dead woman's family members she had had a stomach illness (‘gastric’) for a long time. The dead woman's parents' home is in India. Because the woman died ten or fifteen days after going there, it was not possible to find out what happened to her, the events leading up to her death, or where, if anywhere she was taken for medical treatment. News of her death arrived 25 or 26 days after her death. Note: After completing this form it was learned from other villagers that this woman's death was not natural. It is suspected that in order to bring in another wife, the husband arranged to say he was taking her to her parents and then had her killed by ruffians along the way. The result of all this is that a court case is now underway. Respondent: Husband, Physician-assigned cause of death: Sepsis 4) In the two years before her death, the woman had an illness that caused her chest pain. At the time she was treated by a private doctor but still did not get better. According to a local person, the woman's husband and family members collaborated to hang her to death. Respondent: Husband's family, Physician-assigned cause of death: Uncertain

Mental illness

Many of the narratives also suggested mental illness or disturbance. Though some narratives presented a detailed account of the woman's illness, linking the mental disturbance to life events or a general history of mental illness, most cited mental disturbance without additional detail. Narratives that describe mental illness also commonly mentioned treatment by traditional and religious healers. Some narratives mention mental illness in conjunction with marital or family discord, usually with husbands or mothers-in-law. Some examples: 1) Since Magh of 2043 (January/February 1987) she was mentally disturbed. She used to have temper tantrums every now and then. She was treated with medicines and by shamans but in the end she committed suicide by hanging herself in a room with a sari. Respondent: Husband, Physician-assigned cause of death: Hanging 2) From a young age this woman was affected by ghosts. Her mind was also not quite normal. Because of this situation her husband kept her with him where he was staying in [the district capital]. One night all of a sudden her body caught fire and she became unconscious. She was also taken to [the district] hospital for treatment but died while being treated. Respondent: Husband's family, Physician-assigned cause of death: Burns 3) The woman hanged herself at a time when neither her husband nor any other members of her family were present. The woman was mentally disturbed. Respondent: Husband's family, Physician-assigned cause of death: Hanging.


This analysis of verbal autopsy data sheds light on injury-related causes of death among Nepalese women, which includes intentional injuries, self-inflicted and other inflicted. Although occurring more than a decade ago, we have no reason to believe that the situation for rural Nepali women has changed such that these findings no longer apply. Our findings support other research that also suggests that deteriorating relationships with family members, especially with intimate partners or those suffering from mental health problems, may increase the risk of suicide in Nepal.17 Importantly, this investigation also highlights the continuing need for a better understanding of violent death, especially suicide.

Mental illness or mental disturbance featured prominently in the narratives we examined. In Nepal, mental illness remains a stigmatised condition. In general, there is no concept of biological causes of mental illness; rather, mental illness is usually associated with malevolent spirits or with stressful life events, often including conflict within the family.18 It is also possible that a label of mental illness may be used in some cases to marginalise women who break social norms or otherwise behave differently than expected. Formal health services for mental illness are almost entirely located in Kathmandu18 and community-based mental health services have not been widely implemented.19 More research is needed on mental illness in rural Nepal, and more broadly in rural South Asia, to understand the relationship between mental health status and mortality risk.

An in-depth analysis of the narrative portions of the injury-related verbal autopsies also revealed some challenges in using standard verbal autopsy procedures as a method for investigating violent deaths. Current WHO guidelines recommend that respondents should be the ‘primary caregiver (usually a family member) who was with the deceased in the period leading to death.’20 However, in the case of violent deaths, it is possible that this respondent is unable or unwilling to give an accurate account of the death, either due to their involvement in the death or due to stigma surrounding a suicide or mental illness. As illustrated in many of the narratives, especially those that described conflict within the family, information from the official respondent (often a husband) did not match that of other sources. This may have led to a misclassification of the cause of death (injury vs non-injury related) or to an incomplete picture of the nature of the death (self vs other-inflicted or intentional vs unintentional). Although there is no guarantee that information from other sources will be objective especially when volunteered on an ad hoc basis, and it may be difficult for a physician reviewer to reconcile conflicting accounts recorded by a lay interviewer, interviewing multiple respondents may provide important information when a violent death is suspected. In rural India, for example, deaths with conflicting accounts from multiple family or community members were successfully resolved through a second round of interviews.21 Though standardisation of verbal autopsy procedures naturally leads to a preference for closed-ended responses and algorithm-based cause of death designation,20 in this case, the inclusion of an open-ended response allowed for sub-categorisation of injury-related causes by intent and provided a deeper understanding of the distribution of deaths in this population.

In our analysis, there were two types of discrepancies between physician-assigned cause of death and designation as injury-related death. In the majority of cases (9 of 15) the physician-assigned cause of death of was ‘uncertain’, meaning, the physicians either agreed there was not enough information to assign a cause of death, or the information provided was unclear. Among these, some narratives included information from someone other than the primary respondent (discussed below). In others, the narrative indicated a possible injury but did not yield enough information to further classify by intent. In the 6 of 15 discrepant cases where a cause of death was assigned by physicians (eg, sepsis), the injury described was likely a more distal cause of death, but still warranted analysis of the narrative.

The clear limitations of mortality data derived from verbal autopsies not specifically designed to probe intimate and social causes of death has led to the development of methods that probe beyond the biologic and proximate causes of death. This ‘social autopsy’ methodology has been primarily used to investigate child mortality and maternal mortality.22 Social autopsies in these realms have focused on care-seeking and healthcare usage, but similar methods could be employed to capture care-seeking for mental illness and, perhaps, identify means for supporting victims of gender-based violence. In the case of violent deaths, where social and environmental issues are likely to play a large role in prevention efforts, more extensive verbal autopsy methods may be a necessary supplement to standard procedures.

One limitation of this study is the lack of demographic information collected with the verbal autopsies. The original data were collected in order to assess maternal causes of mortality, and in-depth demographic data was not recorded for those women who did not become pregnant during the trial. This prevented analysis of the data on the basis of sociodemographic characteristics of those who died due to injury or violence. A second limitation of this study is the high proportion of ambiguous responses, those without a cause of death able to be determined either through physician review or through the narratives. This may have resulted in an underestimate of the count of injury-related deaths. Additionally, some narratives, often those describing suicide or referencing mental illness, were very short. It is possible that this brevity reflects stigma associated with suicide and mental illness, or inadequate training of interviewing staff to probe such causes. Lastly, the coding of the narratives by non-medical personnel may have resulted in more attention to distal causes of death rather than proximal medical causes, accounting for some of the discrepancies with the physician-assigned cause of death.

The data used in this study, and the data on suicides among women in a 2008 study7 were, incidentally, collected in the context of investigating pregnancy-related mortality in Nepal. Although both studies recorded information on all deaths, analysis and discussion centred on medical causes of death, particularly maternal deaths. While valuable information on intentional deaths emerged from these studies, this altered focus may have resulted in instruments that were not best suited to fully capturing the nature or prevalence of violent deaths. A focus on maternal mortality, which by definition explicitly excludes injury-related deaths, may contribute to the lack of emphasis on injury-related deaths among women of reproductive age in Nepal and other developing countries generally and in the pregnancy and postpartum periods. However, evidence from several settings suggests that domestic violence plays a significant role in contributing to mortality during the pregnancy period.23 Physical intimate partner violence has been shown to be associated with increased likelihood of premature labour or delivery,24 low birth weight,25 and higher levels of depression during and after pregnancy.26 In an international study, intimate partner violence during pregnancy was found to be more common than pre-eclampsia and gestational diabetes, conditions that are regularly included in antenatal screening.27 Further research is warranted to understand injury-related mortality risk, including risk due to mental illness, during and after pregnancy.

Despite these limitations, this study is unique in its examination of a large body of data from a population-based open cohort of over 40 000 women in Sarlahi district over a period of several years. Data on cause-specific mortality remains scarce in this setting. Our finding of 8.2% of deaths attributed to injury is comparable to the 9.1% deaths attributed to injury in a recent analysis of deaths in a large Bangladeshi cohort of women of reproductive age.28 An additional strength of the study was the inclusion of information from other family and community members. Though not available in all cases, this supplementary data allowed for a more nuanced understanding of each death and allowed for sub-categorisation by intent and perpetrator. Though there exist many barriers to collecting accurate statistics on suicide and homicide among women, verbal autopsies that include a range of respondents (friends, family and other community members) and some elements of social autopsy techniques, may be one way of gathering the in-depth information necessary for prevention efforts. From this, and more recent data on violent deaths in rural Nepal,17 it is clear that any suicide prevention efforts will need to take into account the prominent role of familial relationships and mental illness in the causal pathway to mortality.

What is already known on the subject

  • Suicide is a significant cause of death among women in South Asia.

  • Verbal autopsies are most often used to assign cause of death in settings without vital registration systems.

  • Documenting violent deaths remains challenging in low-resource settings.

What this study adds

  • In rural Nepal, violent deaths among married women were often related to tension in the household and/or mental illness.

  • Verbal autopsies where the respondent is likely to be involved in a violent death may need to be supplemented by additional data collection.



  • Contributors JK, KPW, SCL, PC, LW, SMD and SKK were integral to the original study design and data acquisition. KTH, PJS and LW contributed to secondary data analysis. KTH drafted the manuscript. All authors reviewed the draft manuscript and contributed to revisions. All authors approve of the final version to be published.

  • Funding The NNIPS­2 trial was conducted by the Center for Human Nutrition in the Department of International Health, at the Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA and the Nepal Nutrition Intervention Project—Sarlahi (NNIPS), whose local in-country partner organisation is the National Society for Comprehensive Eye Care (Nepal Netra Jyoti Sangh), Kathmandu, Nepal under Cooperative Agreement No HRN-A-00-97-00015-00 with the Office of Health, Nutrition and Infectious Disease, US Agency for International Development (USAID), Washington, DC, USA. The Sight and Life Research Institute, Baltimore, Maryland, USA, UNICEF, Kathmandu, Nepal and the Sushil Kedia Seva Mandir (Foundation), Hariaun, Sarlahi, Nepal provided additional support for the trial. The secondary analysis was supported, in part, by Grant #614 from the Bill and Melinda Gates Foundation, Seattle, Washington, USA.

  • Competing interests None.

  • Ethics approval Nepal Health Research Council in Kathmandu, the Joint Committee on Clinical Investigation at the Johns Hopkins School of Medicine and the Teratology Society in Bethesda, Maryland.

  • Provenance and peer review Not commissioned; externally peer reviewed.