The apparent burden of unexplained sudden infant deaths in Lusaka, Zambia: Findings from analysis of verbal autopsies

Study setting

Data collection for this study occurred in Lusaka, the capital city of Zambia. Lusaka has a predominantly urban population (84%) of 1.7 million. It is surrounded by unofficial peri-urban compounds/towns where most of the city’s poor reside bringing the total population of Lusaka to approximately 2.4 million individuals, or roughly 1/8th of the total population of Zambia itself. A dozen primary health facilities provide health care to the population with the University Teaching Hospital (UTH) serving as the main referral facility⁶. UTH is the largest hospital in Zambia with 1655 beds and serves as the main training institution for doctors, nurses and other clinical officers^7,8.

Data was collected as part of the Zambian Pertussis/RSV Infant Mortality Estimation (ZPRIME) project. The ZPRIME project, a Bill and Melinda Gates sponsored project, was a post-mortem prevalence study designed to identify the proportion of deaths aged four days to less than six months that were attributable to Bordetella pertussis and respiratory syncytial virus (RSV) in Lusaka, Zambia⁹. ZPRIME enrolled deceased subjects aged four days to less than six months who died in UTH or in the community and presented at the UTH morgue⁹. Ethical approval for ZPRIME was provided by the institutional review board at the University of Zambia (Ref. No. 2017-May-053, Approval date: 07/21/2017) and Boston University Medical Center (IRB Number: H-36469, Approval date: 06/06/2017).

Study population

For this present study, we focus on infant deaths that occurred in the community. Verbal autopsies (VA) were conducted with families and/or caregivers of 809 decedents aged four days to six months who presented at the University Teaching Hospital Morgue (UTH) as BIDs (Brought in dead).

An infant was eligible for a verbal autopsy if the infant had died:

Data collection

Data collectors trained in grief counselling conducted verbal autopsies with families of eligible infants using an abbreviated verbal autopsy tool with close ended questions about the symptoms immediately preceding the infant’s death. An open response narrative question encouraged respondents to describe, in as much detail as possible, the circumstances leading to the infant’s death. The open response field was prompted by the question: “Now, using your own words, please describe the events leading up to your child’s death. Please take as much time as you need and be as detailed as you can.”⁶ The VA tool modified for use in this study was the IHME-modified version of the verbal autopsy tool created and validated by the Population Health Metrics Research Consortium (PHMRC) (PHMRC: Shortened verbal autopsy instrument)^9,10.

Data collectors were notified if an eligible BID decedent presented at the UTH Morgue. Informed consent was sought from respondents and those who gave consent were interviewed. In addition to the VA tool, demographic information such as maternal and paternal education, occupation, and household census was collected. We also collected data on the closest clinic where the infant usually received care as a proxy for the location of the residence. In total, 809 verbal autopsies (VAs) were conducted between August 2017 and August 2020 as part of ZPRIME. Please see the underlying data¹¹.

Coding and statistical analysis

To estimate the proportion of infants who died suddenly and unexpectedly, we focused our analysis on the open response narrative in the VA tool. We did this in two steps. Firstly, we qualitatively coded the free text narratives in Microsoft Excel 2021 (RRID:SCR_016137) and classified the responses into symptomatic and asymptomatic (suspected SUIDs) deaths based on reports of danger signs of ill health within the week immediately preceding death. Any narrative with reports of fever, difficulty breathing, cough, hospital admissions or other symptoms within the week immediately preceding death were classified as symptomatic deaths and assigned a score of 0. Any narrative that described an infant who was otherwise healthy with no antecedent illness prior to death or reported to have no symptoms or hospital admissions in the week prior to death and reported to be found dead in bed after a sleep episode were classified as asymptomatic deaths or suspected SUID and assigned a score of 1.

We further classified the asymptomatic deaths into unexplained (possible SIDS) and explained deaths (possible suffocation/smothering (ASSB)). A SUID death was explained if the narrative suggested suffocation or smothering as the likely cause of death based on:

In the second step, the coded data was then uploaded into SAS software v9.4 (SAS Institute Inc., Cary, NC, USA) (RRID:SCR_008567) and quantitatively analyzed (The analysis can also be performed using R Statistical Software (v4.2.1; R Core Team 2022) (RRID:SCR_001905)). We calculated frequencies and percentages for dichotomous and categorical data and estimated mean and standard deviation for continuous data. We used logistic regression for univariate and multivariate analysis to test statistical differences between symptomatic and asymptomatic deaths on key infant, maternal and other demographic risk factors of SUIDs. All statistical analysis were conducted at a 0.05 significance level. We calculated odds ratios (OR), and 95% confidence intervals (95% CI) to show statistical differences in univariate and multivariate analysis. Included are sample narratives for SUID cases to show our coding decisions.

Population characteristics

There were slightly more females than males in the sample (44% vs 43.8%). Most of the BIDs occurred in infants younger than two months old with a mean age of 2.3 months (SD = 1.7 months). The majority had normal birth weight (mean = 2,629 kg, std = 643 kg) and lived in households with siblings (79.4%, 642/809). The mothers of these infants were mostly unemployed (78.4%, 634/809) and seldom had education beyond secondary school. Only 2.9% (24/809) of mothers reported completing post-secondary school. Educational attainment for fathers was also low. A total of 6.1% (49/809) of fathers had some or completed post-secondary school. However, the majority of fathers were reported to be self-employed or salaried employees (79.4%, 642/809). Almost all the infants lived with their mother at the time of death (99.4%). Nearly a quarter of infants did not live with both parents with fathers present in 77.1% (624/809) of households. Household sizes tended to be larger with a mean of 5.6 persons and standard deviation 2.4 persons. The majority of these households had greater than four children in the household (34.2%, 277/809). Demographic characteristics of these BIDs are presented in Table 1.

Cause of death

Most of the BID infants, 92.6% (749/809), presented with symptoms prior to death with 38.2% (309/809) presenting with respiratory symptoms and 54.4% (440/809) with non-respiratory or other symptoms. The proportions of each cause of death identified in the narratives are shown in Table 2. However, 7.4% (60/809) presented with no preceding symptoms and were classified as suspected SUIDs. Of these, 27% (16/60) had narratives that were suggestive of accidental suffocation or strangulation in bed (ASSB) and 73% (44/60) were classified as suspected SIDS. Table 3 and Table 4 contain a verbatim list of representative narratives for possible SIDS and ASSB to show the coding decisions made in distinguishing between these two causes of SUIDs. A full list of narratives has been included as supplementary tables.

Population characteristics of asymptomatic (suspected SUID) and symptomatic cases

More than half, 51.7% (31/60) of the suspected SUID cases were female with a mean age of 2.1 months (SD = 1.6 months). The majority lived in households with a sibling, 80% (48/60). Educational attainment of SUID mothers was seldom beyond secondary school. Only 1.7% (1/60) were reported to have completed post-secondary school. Most of these mothers were also unemployed, 86.7% (52/60), and lived in larger households with a mean household size of 6.1 persons with more than four children in the household (40%, 24/60). All the suspected SUID cases lived with their mother in the household at the time of death with fathers present in 78.3% (47/60) of these households. Nearly 22% (13/60) of mothers were single mothers. Symptomatic cases were likely to be aged 2.3 months (SD = 1.7months), male (44.6%, 334/749), and with normal birthweight (mean = 2,646 grams, std = 642 grams). Mothers of symptomatic cases were also likely to have had some or completed secondary education (55.3%, 414/809), and were unemployed (77.7%, 582/749). Household sizes of symptomatic cases tended to be smaller (mean = 5.6, std = 2.3), and almost always included the mother (99%, 744/749) or father (77%, 577/809). Demographic characteristics of asymptomatic and symptomatic cases are shown in Table 5.

Suspected SUID cases

Peak age of SUID. Figure 1 summarizes the distribution of suspected SUID cases by age, time of death, month, and season. The mean age of suspected SUID cases was 2.1 months. Overall, deaths were concentrated in infants younger than two months with a peak age of 1–2 months. More importantly, we found that a quarter, 25% (15/60) of SUID deaths occurred in infants within the first month of life although it has generally been considered that SUIDs are rare in the neonatal period.

Figure 1. Distribution of suspected sudden unexpected infant death (SUID) cases by age, time of day, month, and season of death.

A. Age distribution of suspected SUID cases. B. Distribution of suspected SUID cases by time of day. C. Monthly distribution of suspected SUID cases. D. Seasonal* distribution of suspected SUID cases. * Hot and Rainy: Nov, Dec, Jan, Feb, Mar, Apr; Cold and Dry: May, Jun, Jul, Aug; Hot and Dry: Sept, Oct.

Time of day of SUID. All the suspected SUID cases occurred during sleep with peak incidence in the night, 37% (22/60), and early morning, 38% (23/60). Very few occurred in the evening, 15% (9/60) with even fewer in the afternoon, 10% (6/60).

Month and season of SUID. The majority of the suspected SUID cases occurred in the cold months of May, June, July, and August with peak incidence in August, 15% (9/60). By season, this corresponds to the cold and dry seasons of Lusaka. Half of the suspected SUID deaths occurred in the cold and dry season with almost 40% (24/60) occurring during the rainy season. Fewer deaths occurred in the hot and dry season of September and October.

Univariate and multivariate analysis of factors associated with SUID. In univariate analysis, age of death in months and time of death were found to be statistically associated with an asymptomatic presentation. Compared to infants aged less than one month, infants aged between one and two months had 2.6 times increased odds of suspected SUIDs, and this risk was statistically significant (OR: 2.6, 95% CI: 1.31–5.27). The odds of SUID occurring in the night was also 2.64 times higher compared to the afternoon (OR: 2.64, 95% CI: 1.04–6.63). There was an increased odds of SUIDs for female infants, mothers with low education or no employment, larger families with four or more children and the cold/dry season. However, these were not significant with 95% CI which contained the null value as shown in Table 6. In multivariate analysis, infants aged between one and two months had 2.84 increased odds of suspected SUIDs compared to infants in the first month of life (AOR = 2.84, 95% CI: 1.31, 6.16). Unemployed mothers also had 2.49 increased odds of suspected SUID compared to salaried or self-employed mothers (AOR: 2.49, 95% CI: 1.02, 6.08)

Policy implications

This is the first study to report on the prevalence of SUIDs in Zambia and the first study to describe the burden of SUID in Africa outside of South Africa. Our findings suggest SUID could be accounting for a significant proportion of infant deaths, but this cause of infant mortality is going unrecognized. The risk factors of SUIDs are modifiable with simple interventions. This study shows that there is a large enough burden of disease to merit the implementation of specific interventions or programs targeted at reducing sudden infant death in this population. A sudden infant death can be stressful to the family especially young mothers. Some of the narratives were heart breaking. A common theme in these narratives was a need to understand what caused the death of their child. The response to a SIDS/SUID death has been one of blame and sometimes mothers have been criminalized on suspicion that they intentionally caused the death of their child. We want to shed a light on this important cause of infant mortality and make a case for increased research as well as the implementation of targeted campaigns and programs on SIDS in Zambia.

Limitations

A key limitation of this study is that we relied on VA narratives to assign a cause of death. We did not conduct a death scene investigation or formal autopsy. For that reason, we consider these to be ‘presumptive’ SUID cases since they lack the gold standard evaluation. With that said, the apparent burden revealed in our analysis is very similar to what has been described in South Africa using that gold standard evaluation and is also consistent with what was described in high-income settings prior to the introduction of SIDS interventions. While there could be some misclassification, the most parsimonious explanation is that most of these are indeed examples of SUID. Another limitation is that we did not collect any information on other key risk factors of SIDS such as maternal smoking or alcohol use and infant sleep practices from the mothers of the deceased infants. However, in a separate analysis recently accepted for publication, we report a high prevalence of SUID risk factors and behaviors from mothers of infants in the same communities as in the current cohort. These included low rates of sleeping in a prone position as recommended, high frequency of bed sharing, bundling of infants with duvets and blankets, and exposure to indoor air pollution and alcohol use²².