The study was conducted from 29 June 2015 to 1 August 2016 at Chris Hani Baragwanath Academic Hospital (CHBAH). Further details of the study site, including healthcare setting and human immunodeficiency virus (HIV) prevalence, are given in the companion manuscript on neonatal deaths, also published in this supplement [8]. Additionally, in South Africa, the Expanded Programme on Immunization includes pneumococcal conjugate vaccine (PCV) and rotavirus vaccine that have been introduced since 2009, in addition to 8 other routinely administered childhood vaccines [9].

In this prospective, observational study, deaths of children <14 years of age occurring in the medical and emergency department of CHBAH, and those certified as dead upon arrival at the hospital, were identified through daily screening of death registries, excluding from 18 December 2015 to 3 January 2016 (vacation period). We initiated the study by initially only screening and enrolling deaths occurring in the general medical wards. The surveillance and enrollment of cases was expanded to include deaths occurring in the surgical burn unit (31 August 2015), hematology/oncology ward (16 September 2015), surgical wards (30 September 2015), and the nursery ward for care of stable very low-birth-weight infants (kangaroo-care facility; 7 October 2015). Details on the counseling and consenting processes are as detailed elsewhere in the companion paper on neonatal deaths [8].

We aimed to undertake MITS within 24 hours of death and excluded those cases in which the procedure could not be performed within 72 hours. Corpses were kept in the hospital mortuary at 4°C until retrieval for burial.

Trained study staff (medical doctor or professional nurse assisted by research assistants) undertook MITS as detailed elsewhere in this supplement [8]. The testing algorithm for the collected samples was similar to that done in neonatal deaths, with the exception that molecular testing of blood samples using the FastTrack Diagnostics (FTD; Sliema, Malta) neonatal sepsis kit was not undertaken in the age group 1 month to 14 years; and only the FTD Respiratory-33 panel kit was used on lung samples. Methods used for tissue histology processing are also described elsewhere in this supplement [8].

The CoD attribution was based on consensus of an international panel convened in South Africa from 26 March to 5 April 2017 composed of pathologists, pediatricians, epidemiologists, microbiologists, obstetricians, infectious disease specialists, and international coding and certification experts (the Determination of Cause of Death [DeCoDe] panel group is listed in the Acknowledgements). The panel, chaired by either Chris Wilson or Scott Dowell, deliberated on CoD attribution using the International Classification of Diseases, Tenth Revision (ICD-10) [10]. This included attributing a single underlying medical condition that likely initiated the sequence of events culminating in death, and the most proximal event (ie, immediate cause) of death, if applicable. Also evaluated were any antecedent conditions in the casual pathway leading to death, and medical conditions considered to have a contributory role, but not directly involved in the causal pathway. Details on the working methods of the DeCoDe panel are described in the companion manuscript on causes of neonatal deaths [8]. The DeCoDe panel allowed for inclusion of two “immediate” CoD, where concurrent diseases could not be prioritized as the main condition that caused the death; for example, histologically confirmed Pneumocytis jirovecii and respiratory syncytial virus (RSV) pneumonitis being concurrently present (and attributed as the immediate CoD) in a child with underlying HIV infection (attributed as the underlying CoD).

The CoD was generally based on consensus, in the absence of which a majority viewpoint was used, and recorded according to World Health Organization (WHO) guidelines for death certification. In addition, CoD was stratified into first-level United Nations Inter-agency Group on Child Mortality Estimation (UN-IGME) categories [11]: group I (an aggregation of communicable, maternal, perinatal, and nutritional conditions), group II (noncommunicable diseases), and group III (injuries).

We stratified analyses to infants (1–11 months of age), children (12–59 months of age), and older children (60 months to 14 years of age). Descriptive statistics were calculated providing medians with interquartile ranges for continuous variables, and proportions for categorical variables. For select variables, differences between the infants, children, and older children were tested using Kruskal-Wallis or Fisher exact test. Weight-for-age was calculated in Stata software using the UK WHO Preterm Growth Charts (version UKWHOpreterm) and UK WHO Term Growth Charts (version UKWHOterm) for the preterm babies and children aged 0–20 years, respectively. Statistical analysis was done using Stata software version 15 (StataCorp, College Station, Texas).

The University of the Witwatersrand Human Research Ethics Committee (number 150215) and the CHBAH Protocol Review Committee approved the study. Parents/legal guardians provided consent prior to any MITS study procedure.

For solid tissues, most tissue biopsy samples were graded as adequate: 71.6% (n = 91) for left lung, 77.2% (n = 98) for right lung, 81.9% (n = 104) for brain, and 84.2% (n = 107) for liver; or as suboptimal: 8.7% (n = 11) for liver, 11% (n = 14) for brain, 15.7% (n = 20) for right lung, and 19.7% (n = 25) for left lung. No target organ tissue was obtained in 5.5% (n = 7) of liver, 7.1% (n = 9) of right lung and brain, and 8.7% (n = 11) of left lung samples.

Using the findings from the molecular testing, microbiology, and histopathology in addition to the antemortem clinical findings, the DeCoDe panel assigned a CoD in 99% of cases. Details of the certainty of diagnoses for the different underlying CoD and immediate CoD categories are reported in Supplementary Tables 1 and 2. The DeCoDe panel was confident (level 1) for the majority of underlying (81.1%) and immediate CoD (77.2%) diagnoses. Overall, the UN-IGME categories for underlying CoD were 55.1% due to communicable diseases, maternal, perinatal, and nutritional conditions (group 1), 36% due to noncommunicable diseases (group 2), and 8% due to injuries (group 3) (Tables 2 and 3; Figure 2). In group I, these included deaths attributed to underlying HIV/AIDS (22.9%), diarrheal disease (15.7%), acute respiratory infections (14.3%), and complications of prematurity (14.2%).

Among cases with HIV as the underlying CoD (n = 16), infection was the immediate CoD in 15 (93.8%) cases, mainly from pneumonia and sepsis (n = 7 [43.8%] and n = 4 [25.0%], respectively; Figure 3; Table 2). Similarly, infections were also the immediate CoD in 83.3% of 18 deaths attributed to complications of prematurity as the underlying cause, also mainly due to pneumonia (n = 9 [50.0%]) or sepsis (n = 5 [27.8%]) (Figure 3; Table 2).

Among the 10 cases with acute respiratory infections as the underlying CoD, 8 deaths were due to the pneumonia caused by RSV (n = 2), Streptococcus pneumoniae (n = 2), Haemophilus influenzae (n = 1), Staphylococcus aureus (n = 1), P. jirovecii (n = 1), and influenza A virus (n = 1). Similarly, 10 of 11 (90.9%) deaths attributed to diarrhea as the underlying cause died as a consequence thereof, whereas 1 child died of community-acquired Escherichia coli sepsis. The diarrhea deaths were due to enteroinvasive E. coli/Shigella (n = 4) and norovirus (n = 1), while in 6 cases no pathogen was identified. Details of other group 1 underlying CoD are outlined in Table 3 and included 3 (2.4%) deaths due to tuberculosis, 2 (1.6%) due to hepatitis A, and 1 each from varicella and congenital rubella syndrome.

The majority of UN-IGME group II (n = 46) category deaths were attributed to congenital malformations (n = 24 [18.9% of overall deaths]) of which 70.8% were in infants and 25.0% in children (Table 3). The most frequent congenital abnormalities were gastrointestinal malformations (37.5%), Trisomy-21 (25%), and cardiovascular system malformations (25%) (Table 3). Infections were also the dominant immediate CoD among cases with underlying congenital abnormalities, including 45.8% (n = 11) pneumonia, 20.8% (n = 5) sepsis, and 4.2% (n = 1) meningitis (Figure 3; Table 2). Among these infection-related deaths (n = 17), 58.8% (n = 10) were community-associated, and 41.2% (n = 7) were hospital-acquired. Malignancies were diagnosed as an underlying CoD in 7.1% of case overall, and contributed to 40.9% (19/43) of UN-IGME group II deaths (Table 2).

The majority of deaths classified as UN-IGME group III (n = 10) occurred in children (60%) and older children (30%). These included deaths from burns (n = 4 [40.0%]), poisoning (n = 3 [30.0%]), foreign body asphyxiation (n = 1 [10.0%]), drowning (n = 1 [10.0%]), and motor vehicle accident (n = 1 [10.0%]) (Table 3). Among the group III deaths, hospital-acquired infection was the immediate CoD in 40.0% of cases, and community-acquired infection in 10.0% (Figure 3; Table 2).

Overall, infections were attributed as an underlying or immediate cause in 78% (n = 99) of deaths. This included 48 of 127 (37.8%) cases where infection was an underlying CoD, and 51 cases where the underlying CoD was a noninfectious medical condition (64.6% of 79).

Pneumonia was the commonest infection-related CoD (44% [n = 44/99]), 73% (n = 32/44) of which were community-associated and 27% (n = 12/44) hospital-acquired (Table 4). Among community-associated pneumonia cases (n = 32), RSV was the most commonly implicated pathogen (21.9% [n = 7]) and occurred exclusively among infants, including 4 cases in children with a noninfectious underlying CoD. Other pathogens causing community-acquired pneumonia deaths were P. jirovecii (18.8% [n = 6]), including 3 in HIV-infected children and 3 in infants without HIV infection, 1 of whom had underlying cerebral palsy, another was HIV exposed; the third was HIV unexposed and nutritional status was unknown. Other organisms implicated in community-acquired pneumonia deaths included K. pneumoniae (15.6% [n = 5]), cytomegalovirus (CMV; 15.6% [n = 5]), influenza virus (12.5% [n = 4]), and S. pneumoniae (12.5% [n = 4]). Forty percent (n = 2/5) of CMV pneumonitis– and 25% (n = 1/4) of pneumococcal pneumonia–attributed deaths were in HIV-infected children.

Klebsiella pneumoniae was the commonest (66.7% [n = 8/12]) pathogen causing hospital-acquired pneumonia deaths, and all cases had a noninfectious underlying CoD. Seventy-one percent (n = 17/24) of the sepsis-related deaths were hospital-acquired, most commonly due to Acinetobacter baumannii (47.1% [n = 8/17]), K. pneumoniae (35.3% [n = 6/17]), and S. aureus (17.6% [n = 3/17]). Among the 7 community-acquired sepsis deaths, E. coli accounted for 57.1% (n = 4/7) of cases and S. aureus for 14.3% (n = 1/7); no pathogen was identified in 2 cases (Table 4).

Of the 6 deaths due to meningitis as the immediate cause, 4 were community-acquired (S. pneumoniae; Neisseria meningitidis and S. pneumoniae coinfection; E. coli; and Mycobacterium tuberculosis); and 2 hospital-acquired cases were due to A. baumannii and Candida albicans coinfection (n = 1) and E. coli (n = 1).