A total of 3623 under-5 deaths were reported during the study period, and 2509 were included in the analysis (Fig. 1). Overall, 50.2% (n = 1352) were allocated to no narrative and 49.8% (n = 1341) to have an open narrative, with 95% and 94% of VAs conducted per-protocol in each arm. Balance in the randomisation was achieved for respondent type, socio-economic status, child sex and time since the death. However more open narrative interviews were conducted for neonates (44.5% versus 39.9%, p-value = 0.042) and location of death differed between the two groups (Web-Appendix 2). Of the deaths, 41.9% were neonates, 52.9% were male and 31.8% occurred at home (Web-Appendix 2). Primary respondents were mainly mothers (77.0%, n = 1931), followed by grandparents (10.6%, n = 266) and fathers (8.0%, n = 200). The mean time between death and verbal autopsy was 22.5 weeks (range: 1–52 weeks). We present the quantitative and qualitative results together under the following themes: VA processes, data quality and emotions.
VA Processes and Procedures
Open narratives took a median of 7 minutes (range: 1–113 minutes) and closed questions took 17.5 minutes (range: 6–164 minutes). Overall, interviews which included narratives took longer to complete, with 32.3% taking longer than 30 minutes compared to only 5.2% of those without a narrative (p-value < 0.001), with the accompanying closed questions correspondingly taking longer on average to complete (Table 1).
Table 1
Description of VA duration.
| No Narrative N = 1265 | Narrative N = 1242 | p-value |
Total VA duration (Minutes) | | | |
< 10 | 259 (20.5%) | 33 (2.7%) | |
11–20 | 663 (52.4%) | 381 (30.7) | |
21–30 | 252 (19.9%) | 422 (34.0%) | |
> 30 | 66 (5.2%) | 401 (32.3%) | |
Missing | 25 (2.0%) | 5 (0.4%) | < 0.001 |
Closed question duration (Minutes) | | | |
Min – Max | 6–134 | 6–164 | |
Median (IQR) | 15.0 (9.7) | 19.8 (9.9) | < 0.001 |
Narrative duration (Minutes) | | | |
Min – Max | | 1–113 | |
Median (IQR) | | 7.0 (5.0) | |
From the group discussion, MEOs reported narratives generally taking between 3 and 15 minutes. A key factor in the duration of these was the respondent and whether they were capable and willing to respond. Respondents who were keen to relay their story were reported to do so without prompting, including in interviews randomised to not include a narrative. Conversely respondents who were hard to engage in interviews with a narrative were also reported.
“My experience has been that after getting consent sometimes a respondent starts to recount before you ask, so you don’t interrupt, you just listen. But because your phone has asked you not to take an open history, you don’t take notes on that, you just go straight to the questions.” (MEO 8)
“And you can see that there were some open histories that were very short, maybe just 2 minutes [general agreement]. You just know that the respondent was not ready to give you information. It happened like that.” (MEO 2)
However, in general the MEOs agreed that the main difference in interviews with and without open narratives was the time taken. Not taken into consideration in the quantitative measures of duration was the time to identify appropriate respondents before an interview could start. This could involve multiple visits to a respondent’s household before an appropriate respondent could be located (e.g. mother or father), or needing to gain community trust to access the respondent.
“But some other times it may take even 10 minutes because these people know who you want to talk to you, but they are trying to shield them because they are not very sure at first what you’ve come to do.” (MEO 1)
When asked what they would recommend as the best VA procedure, there was a consensus that both the open narrative and closed questions were important and should be included: “The best way is the one which has the open history, that way you have the full explanation.” (MEO 7)
Data Quality
Based on InterVA analysis of closed questions, 94.3% of deaths had a cause of death assigned; there was no difference between those with and without an open narrative (94.7% versus 93.9%, p-value = 0.404). Comparing the number of positive responses in the closed questions found no differences with a mean of 22.4, 21.4 and 21.8 “yes” answers for neonates, infant and child VA interviews (Table 2). The addition of the open narrative was not associated with respondents expressing a desire to know or suggest a potential cause of death.
Table 2
Description of respondent emotions and VA quality indicators, between interviews with and without open narratives.
| No Narrative n (%) | Narrative n (%) | Total n (%) | p-value |
Respondent displayed visible emotion | | | | |
No | 1042 (82.4%) | 990 (79.7%) | 2032 (81.0%) | |
Yes | 223 (17.6%) | 252 (20.3%) | 475 (19.0%) | 0.089 |
*Type of emotion displayed during interview | | | | |
Crying | 4 (1.8%) | 12 (4.8%) | 18 (3.4%) | |
Long silence | 59 (26.5%) | 68 (27.0%) | 127 (26.7%) | |
Other signs of emotion | 160 (71.8%) | 172 (68.2%) | 332 (69.9%) | 0.191 |
*Interview paused due to respondent becoming too emotional | | | | |
No | 89 (39.9%) | 117 (46.4%) | 206 (43.4%) | |
Yes – Once | 31 (13.9%) | 41 (16.3%) | 72 (15.2%) | |
Yes – More than once | 103 (46.2%) | 94 (37.3%) | 197 (41.5%) | 0.146 |
Respondent expressed desire to know the cause of death | | | | |
No | 1235 (97.6%) | 1216 (97.9%) | 2451 (97.8%) | |
Yes | 30 (2.4%) | 26 (2.1%) | 56 (2.2%) | 0.638 |
Respondent suggested potential cause of death | | | | |
No | 909 (71.9%) | 890 (71.7%) | 1799 (71.8%) | |
Yes | 356 (28.1%) | 352 (28.3%) | 708 (28.2%) | 0.912 |
Inter-VA able to assign cause of death | | | | |
Indeterminate | 77 (6.1%) | 66 (5.3%) | 143 (5.7%) | |
Determinate | 1188 (93.9%) | 1176 (94.7%) | 2364 (94.3%) | 0.404 |
Number of “Yes” responses to closed questions** | Mean (SD) | | | |
Neonates | 22.6 (5.3) | 22.2 (5.3) | 22.4 (5.3) | 0.297 |
Infants | 21.5 (6.8) | 21.3 (7.3) | 21.4 (7.0) | 0.658 |
Child | 22.2 (8.2) | 21.3 (8.1) | 21.8 (8.1) | 0.122 |
*Questions only asked for respondents who had a visible display of emotion (n = 475) |
**Different numbers of questions are asked for different age groups |
There was consensus from the MEOs that data collected was of better quality when they conducted an open narrative. The first reason was that they effectively asked the questions twice, once as the narrative and then a second time in the closed questions, enabling them to cross-check responses. Secondly, MEOs reported respondents being more comfortable narrating a story than responding to “yes/no” questions.
“I have that feeling that, without the open history, the quality is compromised. Because it’s like the recall system, the set-up of the brain of the respondent, is disturbed by question time and again. Unlike when he or she is free to express everything from her memory, it happens to be good quality data […] I think that open history gives a respondent a feeling that you are really concerned, because you take a lot of time to listen to him or her.” (MEO 8)
While only 28.2% of respondents were recorded as providing a cause of death (Table 2), the MEOs noted that caregivers would often give a reason for their child’s death – especially if they had sought care. However, they also noted that cause of death was not limited to medical reasons:
“In their narrations, they will tell you the cause, ‘yes this baby was suffering from malaria, but we think this baby died because they delayed in referring us to a health centre’. Maybe in the most remote areas there was no ambulance, they were told to come to the [town] but the ambulance was not available. They were told to look for their own transport to the [town]. So they will tell you those ones as reasons, not the actual sickness of the baby.” (MEO 4)
Emotion
In the majority of interviews, respondents did not display visible signs of emotion (81%), with similar proportions between those with and without an open narrative (79.7% versus 82.4%, p-value = 0.089). Of those who were recorded as showing signs of emotions, 3.4% cried, 26.7% had a long silence and 69.9% displayed other signs of emotion – over half of these interviews needed to be paused once or more (Table 2).
Table 3 shows the logistic regression for respondent emotion. While having an open narrative was associated with 20% (aOR: 1.20; 95% CI: 0.98, 1.47) higher likelihood of the respondent becoming emotional during the interview; this was not statistically significant but may be pragmatically relevant. Factors associated with lower odds of becoming emotional during the VA interview included: non-parental respondents and increased time between the death and interview (2% lower odds for each week passed). Factors associated with increased odds of visible displays of emotion include: deaths amongst infants compared to neonates (aOR: 1.42; 95% CI: 1.09, 1.85); the death occurring at a health centre (aOR: 1.36; 95% CI: 1.04, 1.77) or en-route to hospital (aOR: 1.49; 95% CI: 1.00, 2.22); and being in the middle (aOR 1.52; 95% CI: 1.17, 1.97) or highest wealth tercile (aOR: 1.49; 95% CI: 1.13, 1.95).
Table 3
Logistic regression exploring associations between respondent and child characteristics and emotions during VA.
Visible emotion due to open narrative |
Descriptors | | aOR* (95% CI) | p-value |
Open narrative | No | 1.00 | |
| Yes | 1.20 (0.98, 1.47) | 0.084 |
Respondent | Mother | 1.00 | |
| Father | 0.72 (0.49, 1.07) | 0.102 |
| Grandparent | 0.23 (0.13, 0.39) | < 0.001 |
| Others | 0.04 (0.01, 0.28) | 0.001 |
Child’s age | Neonate | 1.00 | |
| Infant | 1.42 (1.09, 1.85) | 0.010 |
| Child under-5 | 1.21 (0.86, 1.69) | 0.274 |
Child’s sex | Male | 1.00 | |
| Female | 0.99 (0.80, 1.22) | 0.920 |
Location of death | Home | 1.00 | |
| Health centre | 1.36 (1.04, 1.77) | 0.023 |
| MDH | 0.96 (0.72, 1.27) | 0.753 |
| En route to hospital | 1.49 (1.00, 2.22) | 0.049 |
| Other | 0.38 (0.23, 0.64) | < 0.001 |
Socio-economic status by tercile | Tercile 1 (Lowest) | 1.00 | |
| Tercile 2 (Middle) | 1.52 (1.17, 1.97) | 0.002 |
| Tercile 3 (Highest) | 1.49 (1.13, 1.95) | 0.004 |
Delay between death & VA (Weeks) | | 0.98 (0.98, 0.99) | 0.002 |
*All variables presented were included in the adjusted analysis |
The emotion of respondents was not directly raised by the MEOs during the discussion; however, they noted a key challenge in conducting the VAs as being unable to help respondents or feeling hopeless when respondents related their stories. They raised specific examples around HIV positive respondents seeking advice or requests for referrals of malnourished children to NGO programmes.
“A challenge, in a nut shell, was not being able assist where questions were raised. You have raised questions to them. In the end they raise questions to you, that need action, for you to not be able to do anything. That was a big challenge and a let-down.” (MEO 4)
The MEOs raised the fact that the VA process is emotional from the interviewer’s perspective, as well as the respondent, with many of the MEOs also having families and young children which can relate to the narrative.
“The verbal autopsies are not easy to be carried as they involve or concern somebody who has lost life, so it’s always emotional between the interviewer and the interviewee” (MEO 2)