Detailed characteristics of the study population, study site, the MITS procedure, laboratory assays, and CoD attribution are provided in the Supplementary Materials, and briefly described here.

Of 236 neonatal deaths eligible for enrollment and whose parents were approached for study participation, 153 (65.7%) consented. Additional results are provided in the Supplementary Materials. There were 47 ENDs and 106 LNDs (Supplementary Figure 1). Thirty-five percent (n = 51) of neonates who died were born to HIV-infected women, including 41.5% (17/41) of ENDs and 32.5% (34/105) of LNDs (Table 1). Only 2 (1.3%) of the HIV-exposed neonates acquired HIV from the mother, diagnosed by HIV PCR. Eighty percent of cases weighed <2500 g at birth, and 78.6% were born prematurely (Table 1). The median ages on admission were 1 day and 9 days for END and LND, respectively. All of the END and 90 of the LND cases had never been discharged from hospital since birth. The median duration of hospitalization before death of the remaining 16 LNDs was 11 days (range, 3–26 days). The median time between death and undertaking the MITS was 23.4 hours (IQR, 14–37 hours) (Table 1). The majority (76.8%) of neonatal deaths with MITS had adequate core samples for histological examination, with 8.4% suboptimal samples and only 0.6% autolyzed (Supplementary Table 3).

This study demonstrated that MITS, coupled with medical record review, is a robust method to determine the underlying and highly specific immediate causes of deaths occurring in neonates in a LMIC setting. The acceptability of MITS in our study, where two-thirds of bereaved parents agreed to study participation, was similar to the hypothetical acceptability thereof in a multicenter study [8] and affirms its ability to bridge the data gap arising from inter alia cultural and religious barriers in undertaking CDA in LMICs [8].

The data generated from this pilot study to the CHAMPS program provide proof of principle that MITS is feasible with informative findings. These findings set the stage for multicenter mortality surveillance, which could be instrumental in prioritizing strategies and interventions to reduce neonatal mortality [6]. This could include identifying and addressing diseases, which could be prevented, or treated in neonatal deaths attributed to underlying conditions such as “LBW/preterm complications,” which need not be life-threatening. An overwhelming finding from our study was the dominant but underemphasized role of hospital-acquired infections to in-facility neonatal deaths, especially in LNDs (60.4%). Notably, the dominant pathogen, A. baumannii, was resistant to all classes of antibiotics except for colistin, which is not licensed for use in neonates in South Africa. The observations on the contribution of hospital-acquired infections as a cause of neonatal deaths in this study have underpinned a reevaluation and introduction of strategies aimed to improve infection control practices at CHBAH, the success of which will be evaluated through the ongoing CHAMPS program in Soweto.

Our study also demonstrates the need to evaluate cause-specific mortality fraction beyond analysis of the underlying CoD as is currently the focus of estimates on causes of neonatal deaths [2]. Notably, infections that could potentially be prevented or treated were the immediate CoD among 70.4% and 45.0% of cases with “LBW/prematurity complications” and “congenital abnormalities,” respectively, which were among the 3 leading underlying CoD. Furthermore, 79.8% of LNDs were infection related—also almost exclusively due to hospital-acquired infections. In contrast, among ENDs with “LBW/prematurity complications” as an underlying CoD, the immediate CoD were predominantly due to histologically confirmed hyaline membrane disease (50%) and other common complications of prematurity.

The value of postmortem bacterial culture in attributing a casual association to death is controversial due to multiple potential sources of sample contamination, including postmortem aspiration of gastrointestinal and upper airway colonizing bacteria, or translocation of enteric bacteria into the bloodstream. Nevertheless, implicating bacterial infection as the CoD in this study was done only after review of each individual case, and considering multiple corroborating lines of evidence available for each death. For example, pneumonia-attributed deaths were corroborated by lung pathology, showing histological evidence of pneumonia, coupled with immunohistochemical evidence of infection, in addition to PCR detection and/or culture positivity. Also, sepsis was typically diagnosed based on culture and/or PCR positivity from >1 site (postmortem) if premortem culture was negative or PCR positivity from at least 1 site (postmortem) coupled with premortem culture showing the same organism, and often with histopathological evidence of sepsis from multiple organs. Furthermore, in support that MITS sampling for identifying infectious-related deaths was not attributable to environmental contamination was the difference in frequency of identification and spectrum of pathogens implicated as the cause of infection between END and LND cases. The dominant bacteria implicated as causing infection-related END were those commonly colonizing the maternal rectovaginal tract in our setting [12], which could cause invasive disease in the newborn following bacterial infection in utero or during birth. Nevertheless, the possibility of postmortem contamination by these organisms cannot be excluded.

Although the burden of hospital infections and spectra of implicated pathogens might be site-specific [13], the importance of hospital infection as a cause of neonatal mortality has also been observed in other LMIC settings on antemortem sampling. In Zambia, high case fatality ratios (29%–47%) were reported in hospitalized neonates with suspected sepsis, among whom extended-spectrum β-lactamase K. pneumoniae was the dominant (75%) pathogen and primarily hospital-acquired [14]. Gram-negative bacteria, particularly Acinetobacter species (22%) and Klebsiella species (17%), were also the dominant cause of neonatal sepsis in India, with high rates of multidrug resistance observed for Acinetobacter species (82%) and Klebsiella species (54%). Notably, one-quarter of neonatal deaths in the Indian study were attributable to sepsis based on antemortem sampling [15], which could be an underestimate due to low-to-moderate sensitivity of blood culture in detecting bacterial infections [13].

The proportional distribution of underlying CoD in neonates in our study differed from that estimated nationally for South Africa in 2015 [2]. These differences included a higher percentage of deaths in our study being attributed to “LBW/prematurity complications” (52.9% vs 35%) and “congenital abnormalities” (9.2% vs 3.1%), and a lower percentage to “complications of intrapartum events” (15.0% vs 21.4%), and “infections” (9.8% vs 20.6%). The national estimates are, however, based on passive reporting from vital registration databases and limited verbal autopsy reports, which could explain the discordance. A limitation of our study, and the comparison to the national estimates, was that it was a single facility-based study; hence, our findings may not be generalizable to the broader community or other settings with different levels of healthcare access. Although a strength of the study setting was the presence of a single public referral hospital in Soweto, we could have missed community deaths not brought to the hospital.

Another study limitation was that the commercially available multiplex PCR assay was not customized for this study, including lacking targets to evaluate for the most common bacteria (ie, A. baumannii and K. pneumoniae) causing hospital-acquired infections.

Notably, the distribution of timing of neonatal deaths represented in our study differs from the global trend, where approximately 50% of neonatal deaths are estimated to occur within 72 hours of birth [16], compared to 30.7% of cases sampled in our study. This could be explained by the majority (>95%) of deliveries in our setting occurring within healthcare facilities, as well as reasonable access to curative healthcare including intensive care facilities, and hence a better chance of neonatal survival within 72 hours of birth. Furthermore, the exclusion of newborns with birth weight of <750 g, who are highly likely to die within a few hours of birth in the absence of full access to intensive care interventions, including in our setting, might have also contributed to these differences.

The aspiration of providing “universal healthcare” as envisioned in the United Nations Sustainable Development Goal 3.8 [17], could possibly lead to epidemiological shifts in the timing of neonatal deaths globally. This includes potentially a greater role of hospital-acquired infections as an immediate cause of neonatal deaths, as the number of deliveries attended by skilled healthcare workers has steadily increased from 60% in 2000 to 80% by 2015 globally [18], and many babies born preterm are now more likely to stay in hospital after delivery. Consequently, greater focus may be required now to mitigate the risk of hospital infections, particularly in prematurely born neonates, as alluded to in our study.

In conclusion, in this proof-of-concept study to the CHAMPS program, we demonstrate the potential role of MITS interpreted with other clinical information in addressing the knowledge gap on causes of neonatal mortality. Future surveillance using MITS in the multicenter CHAMPS program will contribute to inform decision making in the prioritization of interventions and research that need to be undertaken to reduce neonatal deaths.

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.