Using social autopsy to understand maternal, newborn, and child mortality in low-resource settings: a systematic review of the literature

ABSTRACT Background: Social, cultural, and behavioral factors are often potent upstream contributors to maternal, neonatal, and child mortality, especially in low- and middle-income countries (LMICs). Social autopsy is one method of identifying the impact of such factors, yet it is unclear how social autopsy methods are being used in LMICs. Objective: This study aimed to identify the most common social autopsy instruments, describe overarching findings across populations and geography, and identify gaps in the existing social autopsy literature. Methods: A systematic search of the peer-reviewed literature from 2005 to 2016 was conducted. Studies were included if they were conducted in an LMIC, focused on maternal/neonatal/infant/child health, reported on the results of original research, and explicitly mentioned the use of a social autopsy tool. Results: Sixteen articles out of 1950 citations were included, representing research conducted in 11 countries. Five different tools were described, with two primary conceptual frameworks used to guide analysis: Pathway to Survival and Three Delays models. Studies varied in methods for identifying deaths, and recall periods for respondents ranged from 6 weeks to 5+ years. Across studies, recognition of danger signs appeared to be high, while subsequent care-seeking was inconsistent. Cost, distance to facility, and transportation issues were frequently cited barriers to care-seeking, however, additional barriers were reported that varied by location. Gaps in the social autopsy literature include the lack of: harmonized tools and analytical methods that allow for cross-study comparisons, discussion of complexity of decision making for care seeking, qualitative narratives that address inconsistencies in responses, and the explicit inclusion of perspectives from husbands and fathers. Conclusion: Despite the nascence of the field, research across 11 countries has included social autopsy methods, using a variety of tools, sampling methods, and analytical frameworks to determine how social factors impact maternal, neonatal, and child health outcomes.


Background
Social, cultural, and behavioral factors are often potent upstream contributors to poor maternal, neonatal, and child health outcomes, especially in low-and middleincome countries (LMICs). While anatomical autopsies and verbal autopsies (typically interviews with family members about the days leading up to death) identify leading clinical causes of death, social autopsy aims to push beyond the medical causes of mortality. Social autopsywhich includes verbal autopsy questions and a host of additional sociocultural and behavioral questionsseeks to discern the social, behavioral, and health system factors that contribute to deaths [1,2]. Such factors could include access and transportation to medical care, care-seeking behaviors, cultural norms surrounding illness, and local health system practiceswhich may not be evident when looking at biomedical causes alone.
Adding social autopsy questions to verbal autopsy tools can create a more robust understanding of the modifiable factors leading to deaths and allow for more expertly tailored interventions [1]. For example, a verbal autopsy might attribute the death of a newborn to neonatal sepsis. While this is valuable information, it would not be clear whether the sepsis was likely a result of maternal sepsis, a non-sterile delivery in a health facility, a non-sterile home delivery, or some other cause. Adding a social autopsy component might reveal that in some locations it is not uncommon to apply non-sterile substances to the umbilical cord, such as shea-nut butter or cow dung, and it is often the grandmothers who insist upon such practices [3]. This information could change the focus of an intervention from facilitybased efforts to community-level outreach to extended family members of new mothers. This paper builds on the 2011 comprehensive review of Kalter et al. [2] to provide an update of the peer-reviewed social autopsy literature and: (1) determine the extent to which Social Autopsy methodology has been used in low-and middle-income CONTACT  countries to understand the contributors to maternal, newborn and child mortality; (2) identify the most common Social Autopsy instruments being used, as well as commonalities and differences in methodologies; (3) describe overarching findings across disparate populations and geographic areas; and (4) describe the quality of the evidence to date and identify potential gaps in the existing research literature.

Search strategy
A systematic search of the peer-reviewed, published literature from 2005 to 2016 was conducted to identify original research using social autopsy methodology. The year 2005 was chosen to ensure sufficient overlap with Kalter et al. (2011) comprehensive literature review [2], given that this systematic review methodology was designed to be more rigorous than a comprehensive review and include every article in the peer-reviewed literature that used specifically designated social autopsy methodology. We chose not to go farther back than 2005 because social autopsy methodology was the focus of this systematic review, and formalized social autopsy methods have only come into common usage within the past 10-15 years. Searches used: Ovid MEDLINE, International Pharmaceutical Abstracts, Journals@Ovid Full Text, Epub ahead of print, Global Health database, PubMed, Popline, and BioMed Central. Initial searches were conducted by two of the authors (CAM, CJ) on 10 August 2016, and repeated on 20 August 2016. Given our explicit focus on research using 'social autopsy' methods (as opposed to the many researchers who explore the social determinants of health or social factors related to death), the following search terms were used in various combinations: social autopsy, psycho-social autopsy, social + autopsy, psychosocial + autopsy (search strategy available upon request). Additional hand searching was conducted by reviewing the references of all retrieved studies.

Study selection and data extraction
Studies were included in the review if they were published in a peer-reviewed journal in English between January 2005 and August 2016, were conducted entirely or in part in a low-or middle-income country, focused on maternal/neonatal/infant/child health, reported on the results of original research, and explicitly mentioned the use of a social autopsy instrument, tool, survey, interview guide, or questionnaire. Review articles were not included, as this review aimed to focus on primary sources. Review articles were used to verify the identification of and inclusion of all articles that met the search criteria.
Study inclusion was determined in a multi-step procedure. First, two of the authors (CAM, CJ) evaluated bibliographic data and abstracts for concordance with formal inclusion rules. Studies that clearly did not meet the remaining inclusion criteria were discarded. The remaining studies were selected for full-text retrieval. Publications that did not present empirical data or meet inclusion criteria were discarded, but not before hand-searching the references. Full-text of studies identified from the references were retrieved as well. In a final step, two of the authors (CAM, CJ) examined the remaining studies in detail to identify the final sample of studies meeting all inclusion criteria.

Assessment of quality
The USA National Institutes of Health 'Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies' was adapted for use in this research [4]. This instrument includes 14 items that assess various components of a research study, from clear presentation of the research question through such factors as whether the assessors were blinded to the exposure status of participants. Eight of the 14 items were deemed to be appropriate for assessing the studies included in this review, although not every item was appropriate for every study (See Table 1.) Two authors (CAM, EK) separately reviewed each article against the quality checklist and then compared their individual ratings. In all but two cases, ratings were identical. For the two articles with disparate scores, the authors discussed their perspectives and reached consensus on a final score. A score between 0 and 1 was then calculated for each article, reflecting the number of 'yes' answers on the checklist relative to the number of items that were determined to be applicable to the study in question (e.g. 6 out of 7 possible yes answers yields a score of 0.86, whereas 7 out of a possible 8 yes answers yields a score of 0.88).

Analysis and synthesis strategy
Given the variety and types of studies included in this systematic reviewincluding descriptive and evaluative studies that ranged from simple bivariate analyses to complex multivariate modelinga metaanalysis was neither possible nor appropriate.
One table was created that synthesized the findings across all selected studies, including the study design, the countries in which the research was conducted, methods for identifying respondents, the number and types of respondents interviewed, the recall period, the study's main findings, the social autopsy instrument used, and the conceptual model used to drive the analysis. Another table was used to juxtapose quantitative findings regarding care-seeking across studies and by neonatal, child, and maternal events. These tables were used to address the main research questions. Figure 1 (adapted from Moher et al. [4]) illustrates the process of identifying, screening, and selecting the articles included in this review. A total of 1,950 citations were identified through a systematic protocol of electronic searches, with an additional 14 records identified through other sources. After removing those articles that clearly did not meet the inclusion criteria upon review of the abstracts, a total of 50 articles were identified for full text review. At this final stage another 34 were removed for such reasons as not including original data (n = 8), not mentioning or citing a social autopsy tool (n = 17), not addressing maternal/infant/child health (n = 2), not being peer reviewed (n = 1), not occurring in LMIC (n = 3), or not being available (n = 3). This left 16 published studies that met all inclusion criteria and for which data were extracted (see Table 1).

Results
To address the first objective of this review (to determine the extent to which Social Autopsy methodology has been used in low-and middle-income countries to understand the contributors to maternal, newborn, and child mortality), Table 2 illustrates that the 16 articles included in this systematic review represent social autopsy research conducted in 11 different countries. India, Niger, and Uganda were the sites of several research studies, whereas Bangladesh, Cameroon, Ghana, Indonesia, Kenya, Malawi, Senegal, and South Africa were represented by single studies. Most of the studies included a focus on the neonatal period (N = 8), with six studies including a focus on children under age 5, five studies focusing on maternal mortality, and three studies focusing on infants under the age of 1.   • Delay at home occurred in 62.5% of cases of neonates.
• Out of 296 newborns taken to health facility, 27.7% reported difficulties in transit.
• For post-neonatal deaths, delay at home occurred in 42.3%. In more than half (58.7%) of these, delay was due to waiting for home remedy to take effect. Delay and difficulties in transit (46%), were primarily due to transport problems and lack of social support for carrying the sick child. • About 113 of the caregivers stated that distance, lack of transport, and cost were the biggest constraints to seeking care at a health facility.
• 96.2% of caregivers recognized that their child had a severe symptom. Median length from illness onset to care-seeking was 1 day.   To address the second objective of this review (to identify the most common Social Autopsy instruments being used, as well as commonalities and differences in methodologies). Table 2 illustrates that the instruments used to conduct social autopsies varied across the 16 published manuscripts, and Table 3 describes the instruments in more detail based upon what was reported in the literature, which did not typically include the instruments themselves, detailed descriptions of the domains, or details on the structure or complexity of the tool. Five sources for the instruments were described, sometimes distinguishing between verbal autopsy and social autopsy, and other times combining the two. Six articles cited the Child Health Epidemiology Research Group's social autopsy tool [22]; six cited the World Health Organization's verbal autopsy tool [24]; five cited the INDEPTH Network's verbal and social autopsy tool [23]; four cited the Population Health Metrics Research Consortium's verbal autopsy tool [25]; and one cited the International Clinical Epidemiology Network (INCLEN) verbal autopsy tool [7]. Four of the 16 manuscripts did not describe the social autopsy tool in sufficient detail to be classified. Overall, the two most commonly used instruments for social autopsy assessment were the Child Health Epidemiology Research Group's social autopsy tool and the INDEPTH network's combined verbal and social autopsy tool.
Methodologically, studies also varied in how deaths were identified and the length of the recall period permitted for identified deaths. Many studies (N = 6) relied upon national or large-scale mortality surveys, with a smaller number relying upon data from demographic health surveillance sites (N = 4), household surveys (N = 3) or village-based informants (N = 3). Four studies had a recall period of up to five years for each identified death, and two studies included deaths older than 5 years. One study focused on deaths within the previous year, and five studies focused on deaths within the previous six weeks. Four studies did not report the recall period.
Analytically, the studies typically used one of two frameworks to guide the analysis. Six articles used some variant of the Pathway to Survival Framework [2], six articles used the Three Delays Model [26], and 5 articles were not clear about whether they used a conceptual model to guide their analysis (see Table 2). The Pathway to Survival Framework focuses on the essential steps both inside the home and in the community to prevent child illness and includes recognition of illness, care seeking, and quality of care provided [2]. The Three Delays Model focuses on delays in deciding to seek care, delays in getting to a healthcare facility, and delays in obtaining high-quality care once at the facility [26]. Note that the definitions of delays across the six studies that focused on the Three Delays Model were not consistent (See Table 4.) Pursuant to the third objective of this review (to describe overarching findings across disparate populations and geographic areas), findings across the studies indicate that in general, recognition of severe symptoms for both mothers and babies is relatively high. Of the 16 manuscripts, 8 indicated high recognition of danger signs. However, subsequent action in terms of care seeking is not necessarily commensurate with the seriousness of the problem (see Table 5). Of the 8 manuscripts in which danger signs were recognized, only 3 reported respondents seeking prompt care. For example, in Uganda [11], 96% of respondents recognized severe symptoms in their children under the age of 5 before death, but a third were treated at home before going to a facility. Similarly, in Niger [12], while 95.8% of pregnant women's caregivers reported a serious or severe symptom prior to death, 60.3% of women received no care for their illness. In Malawi, 97.8% of caretakers recognized a severe or possibly severe symptom, but only 61.1% sought or tried to seek care [14]. Table 5 illustrates that care-seeking patterns differed for neonatal, child, and maternal events. For neonates, across studies, the percentage of families who sought care for the fatal illness ranged from a low of 28.0% to a high of 76%. For children 1-59 months of age, between 78.3 and 88% of families sought care for the fatal illness. For mothers, care-seeking patterns were less reliably reported, but between 4.6 and 59.6% of mothers died at home. Across all types of events, the strongest barrier to care-seeking reported was cost, with 59.4% of respondents in one study citing cost as a primary care-seeking barrier, and 87% of respondents in another study citing cost as a reason for noncompliance with referrals.
Cost, distance to facility, and lack of transportation were cited as barriers to care seeking in 14 of the 16 manuscripts. Twelve of the 16 studies provided glimpses into some of the other reasons families may not be seeking prompt attention. These included perceived low quality of care at the facility [6,9,17,21], attributing illness to spiritual and other non-medical causes [6,8,18,19], thinking that the baby would die anyway or was too sick to travel [12,14], thinking that the baby was not sick enough to seek care [14], not having a family member to accompany the woman to the facility [12,20], and needing husband's permission to seek care [8]. Specific examples of how such barriers influence care include that in Bangladesh, 18% of mothers said they did not bring the baby for care because they thought it would die anyway [18]. In India among families experience a post-neonatal death, 104 out of 177 cases with delays at home (58.7%) reported waiting to go to the health facility  • Did not receive any home remedy and was not taken to any health facility • Subjectively judged as a delay if the reviewers deemed that the delay seemed to have partially contributed to the death or if the delay was 'avoidable' by some action by either the caregiver or the health professional • Any newborn baby who died at home or where it took more than 12 hours to seek outside care • Newborn babies whose caregivers expressed problems with getting transport • Delay in receiving or failure to receive quality care at the health facility (as judged by the audit doctor) to see if home remedies would to take effect [7]. In South Africa, one constraint on care-seeking commonly reported through qualitative inquiries was that the cause of the illness was attributed to witchcraft, for which western health facilities are ineffective [19]. The final objective of this review was to describe the quality of the evidence to date and to identify potential gaps in the existing literature. Table 2 shows the quality ratings of the articles, which ranged from a low of 0.67 to 1.0. Overall, the articles scored well on objective measures of quality, with the most consistent deficit being a lack of reporting on the participation rate of eligible subjects. Social autopsy research relies upon identifying and recruiting family members of individuals who have passed away, often in settings without vital registration systems or other ways to definitively determine the sampling frame. While some studies addressed this challenge by using national mortality surveys to generate a sampling frame, four of the articles included in the review did not report how many of the deaths they identified were included in their data collection efforts [8][9][10]21]. This limits the ability to determine potential sampling bias.
With regard to the second half of this final objective, this review identified several notable gaps in the existing social autopsy literature. The first gap is the lack of harmonization across social autopsy tools, including the domains included in the various instruments and the methodologies employed to analyze the findings. This lack of harmonization and methodological consistency makes synthesis across studies difficult. For example, some studies focused exclusively on care-seeking behaviors, while others included background sociodemographic factors such as access to toilet facilities, clean water, and other environmental factors that may impact illness exposure. Similarly, some studies reported the percentage who sought care at a facility for the fatal illness, while others reported the percentage who died at homemaking it difficult to directly compare care-seeking behaviors. This is seen even among studies that relied upon the same conceptual framework, the Three Delays Model. As described, the Three Delays were operationalized differently across the different studies, making comparisons difficult. Future research designed to facilitate consistency in methodology and reporting across social autopsy research is warranted.
The second gap isgiven social autopsy's inclusion of questions about care-seeking decision-makingthe lack of meaningful discussion of the process of decision making with regard to seeking care outside the home during critical and fatal illnesses for mothers and newborns. While some studies describe a delay in making a decision, or attribute delays to waiting for the husband to return home for the women to get permission to leave, no study addressed the potential complexity of decision making in a setting where decisions are often made as a group rather than by any one individual. Further research that attempts to clarify the cognitive, sociocultural, and communication-related processes associated with decision-making is needed.
Another gap in the social autopsy literature based upon this review are qualitative narratives that can provide deeper context to the events described through social autopsy. Of the 16 articles examined in this review, four included a qualitative component in addition to the quantitative data [6,8,17,19], and only one focused solely on qualitative results [9]. Existing social autopsy tools include an open-ended qualitative portion, yet these data are rarely included in published manuscripts. Future analyses would benefit from the explicit use of qualitative methodology to explore inconsistencies identified through the social autopsy interviews, such as why care was not sought even when danger signs were identified.
The final gap in the social autopsy literature is the absence of the explicit seeking and inclusion of perspectives of husbands and fathers surrounding these events. While invariably some of the caregivers interviewed were men, the role of husbands and fathers is not well understood. To date no published research has compared their responses to the responses of mothers or other relatives involved in the care of mothers and their babies.

Discussion
This systematic review of the peer-reviewed literature addressing original research on maternal, neonatal, and child deaths in low-resource settings using social autopsy methods identified 16 articles representing research in 11 countries. Despite the relative nascence of the field of social autopsy research, these articles represent a diversity of geographic areas, target populations, recruitment strategies, and data collection tools. Nonetheless, there were two key findings that appeared across the studies, including: (1) high rates of recognition of serious symptoms do not always translate to high rates of care-seeking; and (2) cost, distance to facilities, and transportation are common barriers to care-seeking, yet different settings include different barriers to care-seeking, such as needing to wait for a home remedy to take effect before leaving for a western facility.
This systematic review builds upon a comprehensive review undertaken by Kalter et al. [2] that looked at articles published between 1989 and 2010 and focused on the development of social autopsy methods and the utilization of the Pathway to Survival Conceptual Framework [2]. Their review (which included 14 peer-reviewed articles pursuant to child deaths and eight related to maternal deaths) focused heavily on the elements of care-seeking that were examined in each study, as well as how the introduction of their Pathway to Survival model influenced the number of elements of care-seeking that were assessed. Since the publication of the Kalter  This study highlights the breadth and depth of existing social autopsy research, and followed rigorous systematic literature review methodology to ensure inclusion of all relevant articles. Starting from nearly 2000 citations, we identified 16 that included primary research using a designated social autopsy tool to explore the social and cultural factors that impact maternal, newborn, infant, or child mortality in developing country settings. Another strength of this study is that we examined not only the tools being used, but also the analytic frameworks employed. It is clear from our findings that researchers are typically choosing to use either the Pathway to Survival model or the Three Delays Model to guide their analysis, with only one published paper using both [11]. However, it is worth nothing that those articles using the Three Delays Model did not operationalize the variables in the same way, making comparison across studies difficult. While both the Pathway to Survival model and the Three Delays Model have yielded important findings, the reliance on these models above all others suggests that the research literature might benefit from the development and testing of alternative frameworks.
Despite the strengths, this review was limited to English-language articles published in indexed journals, and it is possible that research conducted in non-English-speaking countries may not be adequately represented. This review is also limited by the lack of detail included in many published manuscripts regarding the content, duration, domains, and individual items included in their social autopsy tools. It is also difficult to determine how the instruments have been revised over time and/or how they have been adapted to different contexts.
In sum, this review suggests a need for harmonized social autopsy tools and methodologies, as well as a need to better understand the complexities of decision making around care seeking for serious illnesses among mothers and babies. Given the study findings that care-seeking is more common for older children than for newborns, it also highlights the need for deeper inquiry using qualitative methodology to unpack inconsistencies in care seeking, even in the face of recognition of the severity of symptoms.
The studies included in this review indicate that cost, distance, and transportationwhile common barriers to care seekingare not the only barriers for women and families experiencing life-threatening complications. Until such barriers are understood and addressed through locally appropriate means, delayed or lack of care-seeking will continue to serve as a barrier to improving maternal, neonatal, infant, and child outcomes. This study also illustrates the importance of including, articulating, and expanding upon the potentially unique perspectives of men in influencing care-seeking for mothers and their children.