Using Minimally Invasive Tissue Sampling and Determination of Cause of Death to Establish Etiologies of Community Respiratory Deaths Among Zambian Infants and Children

Abstract

In low-to-middle-income countries, acute lower respiratory infection (ALRI) remains the leading infectious cause of death among infants and children under 5 years old. Case-control studies based on upper respiratory sampling have informed current understandings of ALRI etiologies; in contrast, minimally invasive tissue sampling (MITS) offers a method of directly interrogating lower respiratory tract pathogens to establish etiologic distributions. This study performed in the postmortem setting used MITS and a Determination of Cause of Death (DeCoDe) panel to elucidate the causes of fatal pneumonia in the community in Lusaka, Zambia. For deceased infants and children under age 5 years whose next-of-kin provided consent, a verbal autopsy was obtained and 6 lung tissue biopsies from each case were sent for histopathology and multiplex polymerase chain reaction testing. Subsequently, a multi-disciplinary DeCoDe panel met to review each case, determine if the child died of respiratory causes, construct a causal chain of diagnoses directly leading to the death, and determine if the death was preventable (i.e., if an identifiable intervention would have averted the death). Among 106 deaths, 49 were adjudicated as respiratory deaths, with etiologic causes including Klebsiella pneumoniae (13), Streptococcus pneumoniae (5), and Pneumocystis jirovecii (4), among others. Of note, for 21 respiratory deaths, a causative pathogen could not be identified despite clinical and histopathologic evidence of ALRI. A large majority of all deaths were considered preventable (90/106 or 85%). This study demonstrates the impact of certain respiratory pathogens through direct in situ tissue sampling with supportive pathologic data and presents a useful method of studying the etiologic distribution of fatal ALRIs in settings where many deaths occur in the community.

This postmortem study in Lusaka, Zambia used verbal autopsy and minimally invasive lung sampling/testing to determine respiratory causes of death of infants and children who died outside of healthcare settings.

INTRODUCTION

Acute lower respiratory infections (ALRI) remain a leading cause of death among infants and children under 5 in low- and middle-income countries (LMIC) [1]. Establishing the etiologies of fatal ALRI is key to informing interventions to prevent mortality. Yet this task is complicated by challenges related to: (1) non-invasively identifying the pathogens causing disease in the lower respiratory tract and (2) representing the distinct etiologic distribution in the community (where most deaths occur in LMIC) rather than in the hospital setting (where most studies are performed). The Pneumonia Etiology Research in Child Health Study (PERCH) [2] initially intended to incorporate percutaneous pulmonary biopsy into its protocol for determining etiologic causes of pneumonia but the procedural risks were prohibitive. Instead, it used novel Bayesian analytics to infer an etiologic distribution of ALRI causes from case-control nasopharyngeal-oropharyngeal sampling across 9 sites in LMICs in Asia and Africa, finding that a small number of pathogens—mainly viruses, and especially respiratory syncytial virus (RSV)—accounted for most severe pneumonias.

Despite the major progress presented through PERCH’s novel analytical methods, its findings are based on indirect inference of lower respiratory tract pathogens. These methods may be prone to bias, in particular, under-representing bacteria within the etiological distribution. Moreover, the low 30-day mortality among cases in PERCH suggests these findings may not be generalizable to fatal pneumonias. As understanding fatal pneumonia etiologies in the community is critical to informing child survival interventions, we aimed to directly identify lower respiratory tract pathogens within the context of an ongoing postmortem surveillance study. The prevailing question of this present study was: what pathogens are responsible for under-5 respiratory deaths in the community in Lusaka, Zambia?

We addressed this question using minimally invasive tissue sampling (MITS) and a Determination of Causes of Death (DeCoDe) panel adjudication approach modified from the Child Health and Mortality Prevention Surveillance (CHAMPS) network [3, 4]. In analyzing the results, we address 3 specific aims, each of which highlights the powerful potential for MITS and DeCoDe within pneumonia etiology research: (1) what caused the deaths of these children? (2) which deaths were preventable and how? and (3) what are the case features of deaths due to specific notable causes?

METHODS

This study was embedded within 2 larger postmortem surveillance studies conducted in Lusaka, Zambia—the Zambia Pertussis RSV Infant Mortality Estimation Study (ZPRIME) [5] and the Lusaka COVID Impact Study (LCIS) [6]—and drew upon an approach to postmortem research that was refined over time. This project was approved by the ethical review boards at Boston University Medical Center and the University of Zambia.

Study Setting and Data Collection

The postmortem enrollment setting and protocols have been previously described in detail [5]. Enrollment took place at the University Teaching Hospital morgue. This facility issues nearly all burial permits for the city, regardless of the socio-economic status of the decedents, allowing for a highly representative site for enrollment in this study.

Infants and children who died in the community and were “brought in dead” within 48 hours were eligible for enrollment. During the enrollment period occurring under ZPRIME, from January to September 2020, infants aged 4 days to 6 months at the time of death were enrolled; during the enrollment period occurring under LCIS, from September 2022 to December 2022, infants and children under age 5 years were enrolled. To achieve a rough balance, presumed respiratory and non-respiratory deaths were enrolled in a 1:1 fashion, where deceased infants were considered presumptive “respiratory deaths” if the next-of-kin answered “yes” to 2 or more of 5 respiratory questions on the verbal autopsy tool, and were otherwise considered presumptive “non-respiratory deaths” (see Supplementary Table S1).

Eligible infants and children were identified according to the morgue registry. Experienced field team members approached next-of-kin of the deceased and offered grief counseling irrespective of their interest in participating in the study. For families interested in discussing the study, a screening form was administered and an informed written consent process was conducted. Verbal autopsy was performed using a shortened version of a validated verbal autopsy instrument, culminating in an open narrative of the events preceding the death [7].

Sample Collection and Processing

The body of the deceased was taken to the autopsy theater at the University Teaching Hospital’s morgue for the MITS procedure. The MITS procedure was performed by field team members who had received training from the MITS Alliance (supported by a grant from the Bill & Melinda Gates Foundation that is managed by the Research Triangle Institute International [8]). First, significant physical findings were noted. Then, the thorax of the deceased was cleaned with ethanol and iodine. Sterile 16-guage biopsy needles were used to obtain 2 core needle biopsies from the upper, middle, and lower zone of each lung. From each zone, the first pass biopsy was obtained using a new sterile biopsy needle. Puncture sites were along the axillary lines. Portable ultrasound was introduced later in the study to guide the location of the biopsy needle insertion sites and target areas of pathology. Ultrasound was introduced midway through the study because it was found to be a useful tool in a separate concurrent protocol for diagnosing pediatric pneumonia and was used for the last 30 cases.

From each set of core needle biopsies corresponding to the 6 lung zones, the first biopsy specimen was sent to the molecular laboratory while the other 6 core needle biopsies were placed immediately in neutral buffered formalin and sent to the pathology laboratory. In the pathology laboratory, the specimens were left to fix in formalin for at least 24 hours prior to paraffin embedding, microtomy, and mounting on slides for final hematoxylin and eosin staining and slide review. These slides were read by 2 pathologists (VM and CM) who provided a report of the findings from each lung zone and took representative photographs from the slides. “Pneumonitis” was used to describe a pattern of inflammation that was localized to the interstitium (lymphocytic infiltrate with interstitial expansion/fibrosis). “Pneumonia” was used to indicate histologic bronchopneumonia, with an intra-alveolar/bronchial distribution of acute (neutrophilic) inflammation. Within the molecular laboratory, each of the 6 lung tissue specimens per case was frozen at −80 degrees Celsius for multiplex polymerase chain reaction (PCR) testing with 32-pathogen multiplex quantitative PCR (FTD-33 kit, Fast-Track Diagnostics [9–11], Supplementary Table S2). We ran PCR for 45 cycles and reported a result as “detected” (with its associated cycle threshold) if there was a reactive logarithmic fluorescence curve at any threshold. Study procedures did not include microbiological cultures, specialized stains for pathogen detection, or sampling of other organs.

Adjudication Approach

To adjudicate the causes of death within this cohort, we adapted the DeCoDe process [4] developed by the CHAMPS network to the design of our study. We convened a panel of 7 experienced clinicians based in Lusaka: 2 pathologists and 5 pediatricians, including 1 pediatric neurologist. The panel was tasked with reviewing the information for each case, determining if the death was primarily due to respiratory causes, and providing a cause of death. For respiratory deaths, the panel constructed a causal chain which included an immediate cause of death, with the option of adding 1 or more underlying causes. The requirement for inclusion in this causal chain was based on a counterfactual argument: “the death would likely not have occurred in the absence of this diagnosis.” The panel indicated a subjective degree of certainty (high, moderate, or low) for the respiratory/non-respiratory determination as well as the diagnoses within the causal chain for respiratory deaths. For non-respiratory deaths, the panel could posit a hypothesis for the cause of death based on the information available. Causes of non-respiratory deaths were assumed to be low certainty given that the diagnostics in this study focused on pulmonary pathology.

Finally, the panel determined if the death was preventable and, if so, what interventions could in theory have prevented the death. The panel produced 1 adjudication per case, and these were subsequently ranked with 1 of 3 levels of consensus within the panel: unanimous, majority, or non-consensus (which could include a plurality of panelists without achieving an outright majority).

The adjudication process began with a 2-day panel orientation, including practice adjudications of cases drawn from a published MITS study performed in Argentina [12, 13]. For each case, a dossier of data was reviewed ahead of time by the panelists (Supplementary Table S3). Adjudication sessions required the participation of at least 1 pathologist and 3 pediatricians; otherwise, the session was rescheduled. Adjudications were recorded in a secure online form by a notetaker using REDCap (Research Electronic Data Capture) hosted at Boston University [14, 15]. The form contained both structured questions using International Classification of Diseases-11 codes to assign diagnoses within the causal chain and free text fields for notes on the discussion and interventions that could have prevented the death, if any (Supplementary Table S4). Deaths were determined to be preventable if the panel identified an intervention they believed would have averted the death.

Statistical and Text Analysis

This analysis used descriptive statistics to characterize demographic, verbal autopsy structured responses, and pathology and microbiologic features of the various causes of death. Statistical analysis was performed in RStudio. We analyzed both the immediate causes of death (1 diagnosis per respiratory death) and the full causal chain (at least 1 diagnosis per respiratory death). For non-respiratory deaths where the panel hypothesized multiple contributors to the death, we present the immediate proposed etiologic cause of death when provided (e.g., “acute gastroenteritis” instead of “dehydration due to acute gastroenteritis”). Free text regarding interventions that could have prevented the death were reviewed by 2 researchers (AW and MC) who jointly coded the interventions into categories.

RESULTS

Description of the Cohort

Field team members enrolled 121 deceased infants and children from January 2020 to December 2022 during the study windows discussed above, with next-of-kin consenting to MITS procedures. Of note, enrollments were paused from April 1 to May 6, 2020 due to suspension of all human subjects research in Zambia during the initial wave of the coronavirus diseases 2019 (COVID-19) pandemic. For 15 cases, we had incomplete molecular data. This yielded 106 cases with complete histopathology and molecular data that proceeded to adjudication and comprise the analytic set for this paper. Most enrollments (87/106, 82%) occurred from June to December; few enrollments (7/106, 7%) occurred during March—April, comprising part of respiratory virus season, and in particular respiratory syncytial virus (RSV) season in Lusaka [16, 17]. The median age at the time of death was 3.8 months, ranging from < 1 month to 48 months, with an interquartile range (IQR) of 6.4 months. The deceased included 60 females (57%) and 46 males (43%). Deceased infants and children lived throughout the city, with the greatest number (30/106, 28%) coming from the poor and densely populated Chawama and Kanyama townships. Among those with this information provided, 11/44 (25%) had a reported history of malnutrition and 22/106 (21%) had been previously admitted to the hospital for any reason.

Any respiratory symptom (Supplementary Table S1) was reported by next-of-kin as part of the illness preceding death for 68/106 (64%) children, while 38/106 (46%) children had no respiratory symptoms. At least 2 respiratory symptoms were reported for 58/106 (55%) deceased children. Fever was noted for 51/73 (70%) and diarrhea for 9/73 (12%) when asked via structured verbal autopsy questions. The reported duration of illness leading to death was a median of 1 day (range 0–27 days, IQR 2 days) (reported for 69 children).

Adjudicated Causes of Death

Case dossiers were adjudicated over the course of 17 meetings, with 4–7 panelists each, from February to July 2023.

Following review and discussion, 49/106 (46%) deaths were determined to be primarily due to respiratory causes and 57/106 (54%) non-respiratory, with high or moderate certainty in the respiratory designation for 98 of the 106 (93%) cases (Figure 1a). Based on the totality of evidence adjudicated by the panel, 17 of the 58 suspected respiratory death enrollments based on symptoms were adjudicated as non-respiratory, and conversely, 8 of the suspected non-respiratory enrollments were adjudicated as respiratory deaths. The median age of respiratory deaths was 2.7 months (IQR 5.5 months) compared to 4 months for non-respiratory deaths (IQR 7 months). Descriptive data pertaining to respiratory and non-respiratory deaths are presented in Table 1.

Figure 1.

Causes of respiratory deaths. a. Determination of respiratory vs. non-respiratory death following determination of cause of death panel adjudication. b. “Other” includes polymicrobial bacterial pneumonia (1) and sepsis due to unspecified pathogen (due to respiratory source) (1). c. “Other” includes polymicrobial bacterial pneumonia, sepsis due to unspecified pathogen, cleft palate, congenital heart disease, pneumonia due to Bordetella pertussis, pneumonia due to respiratory syncytial virus, and hyaline membrane disease (each accounting for 1).

Table 1.

Characteristics of Adjudicated Respiratory and Non-Respiratory Deaths

	All cases N = 106	Respiratory Deaths N = 49	Non-Respiratory Deaths N = 57
Demographics/social information
Age—no. (%)
<1 mo	12 (11)	8 (16)	4 (7)
1–<6 mos	62 (58)	28 (57)	34 (60)
6–<12 mos	14 (13)	5 (10)	9 (16)
12–59 mos	18 (17)	8 (16)	10 (18)
Sex—no. (%)
Female	60 (57)	24 (49)	36 (63)
Time of year—no. (%)
January–June	33 (31)	21 (43)	12 (21)
July–December	73 (69)	28 (57)	45 (79)
Time of death—no. (%)a
Day (6:00–<18:00)	66 (62)	36 (73)	30 (53)
Night (18:00–<6:00)	40 (38)	13 (27)	27 (47)
Number of co-habitants in home—no. (%)
1–4	42 (40)	18 (37)	24 (42)
5–8	51 (48)	26 (53)	25 (44)
9+	13 (12)	5 (10)	8 (14)
History
Length of illness—no. (%)
≤1 day	38 (36)	12 (24)	26 (46)
2–7 days	26 (25)	17 (35)	9 (16)
>7 days	5 (5)	4 (8)	1 (2)
Missing	37 (35)	16 (33)	21 (37)
Cough—no. (%)	44 (42)	38 (78)	6 (22)
Severe cough—no. (%)	19 (18)	18 (37)	1 (2)
Fast breathing, difficulty breathing, or grunting—no. (%)	64 (60)	42 (86)	22 (39)
Frequent loose or liquid stoolsb—no. (%)
Yes	9 (8)	2 (4)	7 (12)
No	64 (60)	33 (67)	31 (54)
Not elicited	33 (31)	14 (29)	19 (33)
Fever—no. (%)b
Yes	51 (48)	29 (59)	22 (39)
No	22 (21)	6 (12)	16 (28)
Not elicited	33 (31)	14 (29)	19 (33)
Prior healthcare encounterc—no. (%)
Clinic sick visit	25 (24)	14 (29)	11 (19)
Hospitalization	22 (21)	11 (22)	11 (19)
None	59 (56)	24 (49)	35 (61)
Histopathology impression —no. (%)
Normal in all regions	5 (5)	0 (0)	5 (9)
Bronchiolitis in at least 1 region	5 (5)	4 (8)	1 (2)
Pneumonitis in at least 1 region	31 (29)	21 (43)	10 (18)
Pneumonia in at least 1 region	15 (14)	15 (31)	0 (0)
Other histopathology abnormalities in at least 1 region	104 (98)	48 (98)	56 (98)
32-pathogen multiplex PCR results
Number of pathogens per child across all biopsies—median (IQR)	4 (3)	4 (4)	4 (3)
Number with any biopsy detecting—no. (%)
K. pneumoniae	54 (51)	28 (57)	26 (46)
Adenovirus	53 (50)	22 (45)	31 (54)
IAV H1N1	49 (46)	21 (43)	28 (49)
Bordetella spp.	45 (42)	23 (47)	22 (39)
S. pneumoniae	42 (40)	18 (37)	24 (42)
P. jirovecii	39 (37)	16 (33)	23 (40)
H. influenzae	29 (27)	15 (31)	14 (25)
Number of detecting biopsies (of 6) within each cased—median (IQR)
K. pneumoniae	3 (3)	3 (3)	2 (3)
Adenovirus	1 (1)	1 (0)	1(1)
IAV H1N1	1 (1)	1 (1)	1 (1)
Bordetella spp.	1 (1)	1 (1)	1 (1)
S. pneumoniae	2 (3)	4 (2)	2 (2)
P. jirovecii	2 (2)	3 (3)	2 (2)
H. influenzae	2 (3)	3 (3)	2 (2)

^aDay refers to 06:00–17:59 and night refers to 18:00–05:59.

^bPrior symptoms as reported by next-of-kin in verbal autopsy.

^cCoded from narrative verbal autopsy. Refers to any prior hospitalization or sick visit mentioned in the next-of-kin’s description of the events preceding the death.

^dSummary statistics for those cases where the pathogen was detected on at least 1 biopsy—the most common pathogens are highlighted here. See Supplementary Materials for more detailed histopathology and PCR results.

For respiratory deaths, 33 cases had 1 diagnosis in the causal chain, 15 cases had 2 diagnoses, and 1 case had 3 diagnoses. The most common immediate causes of respiratory deaths were pneumonia due to an unspecified organism (21/49, 43%), pneumonia due Klebsiella pneumoniae (13/49, 27%), pneumonia due to Streptococcus pneumoniae (5/49, 10%), and pneumonia due to Pneumocystis jirovecii (4/49, 8%) (Figure 1b). Analyzing all diagnoses within the respiratory causal chains resulted in the same most common diagnoses (Figure 1c). For diagnoses within the respiratory causal chain where a pathogen was identified, 16 were gram-negative bacteria, 5 were gram-positive bacteria, 5 were fungal (all Pneumocystis jirovecii), 2 were due to Mycobacterium tuberculosis, and 1 was viral (RSV); up to 6 of the pathogens included in the causal chains were potentially vaccine preventable (serotype data were not available for Streptococcus pneumoniae isolates; none of the Haemophilus influenzae detected were reported as type B by FTD-33).

For non-respiratory deaths, hypotheses regarding the cause of death were provided for all deceased children. The most common hypothesized causes were acute gastroenteritis/colitis (17/57, 30%), unspecified sepsis (8/57, 14%), and sudden unexpected infant death (8/57, 14%) (Figure 2).

Figure 2.

Diagnoses causing non-respiratory deaths. Note that non-respiratory diagnoses were considered hypotheses and assumed to have low certainty given that MITS testing for histopathology and molecular diagnostics was limited to the lungs. Each diagnosis accounting for 1 death were grouped into “Other” (these included anemia, congenital infection, hepatitis, intestinal obstruction, myocarditis, and vaccine reaction).

Opportunities to Prevent Avoidable Deaths

Most deaths (90/106, 85%) across all age categories were deemed by the DeCoDe panel to have been preventable and all 22 deaths of infants and children 6 months or older were considered preventable. Of the respiratory deaths, 44/49 (90%) were preventable, compared to 46/57 (81%) of the non-respiratory deaths. When analyzed by immediate cause of death, all 21 deaths due to acute gastroenteritis/colitis or and/or dehydration were preventable (Figure 3). Deaths due to unspecified pneumonia and bacterial pneumonia were nearly all preventable (95% and 85%, respectively). The specific interventions that could have prevented the death are listed in Table 2. Earlier access to care/healthcare-seeking, avoiding inappropriate hospital or clinic discharge, and timely referral to a higher level of care were the most common interventions proposed by the panel. In many cases with abrupt evolution (symptom onset of less than 1 day), earlier access to care would have been on the order of hours, not days.

Figure 3.

Immediate causes of preventable and non-preventable deaths. In this figure, dehydration as a cause of death was grouped with gastroenteritis/colitis.

Table 2.

Interventions to Prevent Death

	Number of Deaths That Could Have Been Prevented Through This Interventiona—No. (%)b
Earlier presentation to care	54 (51)
Avoid inappropriate discharge home	21 (20)
Timely referral to higher level of care	19 (18)
Counseling on danger signs for return to care	11 (10)
Appropriate diagnostic investigation	9 (8)
Appropriate patient triage	6 (6)
Sudden infant death syndrome prevention measures	5 (5)
Inpatient infection control measures	4 (4)
Prior diagnosis and management of malnutrition	3 (3)
Timely antibiotic administration	3 (3)
Other†	12 (11)

^aNinety deaths were deemed to have been preventable through at least 1 potential intervention. For 43 deaths there were 2 or more potential interventions that could have prevented the death. Thus, the sum exceeds the number of deceased children in the cohort.

^bPercentages reflect the percent of deceased children who could have been affected by the listed intervention. Percentages do not add to 100.

^†Other: Oral rehydration therapy (2), appropriate perinatal HIV care (2), avoid exposure of newborn to crowded areas (1), prompt oxygen administration (1), choking resuscitation training for parents (1), adequate management of cleft palate (1), appropriate medication management of preexisting condition (1), appropriate surgical management (1), frequent monitoring of infant at home (1), and safe feeding practices (1).

Characterizing Notable Immediate Causes of Death

Pneumonia Due to Unspecified Pathogen. 

The median age of the 21 children who died of unspecified pneumonia was slightly younger than the remainder of the cohort, at 2.5 months (IQR 3.9 months). Six of these cases had histologic pneumonia identified in 1 or more biopsy and 8 had pneumonitis identified. When observed, pneumonia was isolated to 1 biopsy in 4 out of 6 cases; pneumonitis was seen on at least 2 biopsies in 6 cases and over half of the biopsies in 3 cases. All 21 cases had other histopathology abnormalities noted on at least 1 biopsy, ranging from vessel congestion to interstitial expansion. None of the cases with completely normal histopathology and none of those with bronchiolitis were classified as deaths due to unspecified pneumonia. The number of pathogens detected on multiplex PCR ranged from 0–9, with a median of 4 (IQR 3), similar to other cases. The case was determined to be of unspecified etiology due to having no pathogen of clear invasive disease potential (including 1 case where all 6 32-pathogen PCR tests were negative) or because the causal pathogen could not be distinguished amongst several pathogens that were detected (e.g., 1 case where Staphylococcus aureus, Haemophilus influenzae, and Streptococcus pneumoniae were each detected on at least 3 biopsies, with low cycle thresholds for each).

Pneumonia due to Klebsiella pneumoniae. 

The 13 infants and children who died of pneumonia due to Klebsiella pneumoniae ranged in age from 0.5 to 20 months in age (median 3.2 months, IQR 6 months). Whereas a minority of the deceased in the overall cohort had a prior healthcare encounter (clinic sick visit or hospitalization), the majority who died of Klebsiella pneumoniae had a prior healthcare encounter described within the verbal autopsy narrative: 5 (38%) had previously been hospitalized and 4 (31%) had a prior clinic sick visit. The remaining 4 (31%) had no reported prior healthcare encounter. Of the 11 (85%) Klebsiella deaths determined to be preventable, the panel identified 3 that could have been averted through inpatient infection control measures. Four children had pneumonia identified on at least 1 histopathology section and 5 had pneumonitis identified on at least 1 histopathology section; no children had normal histopathology on sections from all 6 lung zones. 12 out of 13 cases had at least 1 other pathogen detected on the 6 lung biopsies, with a median of 2 other pathogens detected (IQR 2). Among these 13 cases, 7 had Klebsiella pneumoniae detected on 5 or 6 out of 6 biopsies (Supplementary Figure S1).

Pneumonia due to Streptococcus pneumoniae. 

The median age of the 5 infants and children who died from Streptococcus pneumoniae was 8 months (IQR 13 months). On histopathology, 2 cases had pneumonia (both in only 1 of 6 biopsies), 1 had pneumonitis (observed on 2 biopsies), and 3 had bronchiolitis on any lung section (seen on 1 biopsy in 1 case and 2 biopsies in 2 cases; this was interpreted as evidence for unspecified viral infection preceding bacterial pneumonia). Four of the 5 deaths were determined to be preventable, all through earlier access to healthcare services.

Pneumonia due to Pneumocystis jirovecii. 

The 4 infants who died of Pneumocystis jirovecii pneumonia were all under 6 months (median 4.4 months, IQR 1.7 months). Two cases had at least 1 area of pneumonitis (1 of which had diffuse pneumonitis seen on all 6 biopsies and the other with pneumonitis observed on one biopsy), 1 case had 1 out of 6 biopsies showing a focus of bronchiolitis, and 3 cases had other histopathology abnormalities. For 1 infant, maternal HIV status was positive but not confirmed in the infant, and for the others, this information was not collected or was not known. Three out of 4 of the deaths attributed to this cause were thought to have been preventable.

Respiratory Tuberculosis. 

One infant and 1 child were diagnosed as having died of tuberculosis. In 1 case, all 6 lung zones revealed granulomatous inflammation and pulmonary ultrasound revealed thickened pleura. In the other case, 4 of 6 lung biopsies revealed granulomatous inflammation with other areas showing infiltrates and interstitial expansion with lymphocytes and macrophages. Both infants had been evaluated in a hospital or clinic, administered unidentified medicines, and discharged prior to death.

DISCUSSION

In this study, we employed MITS and DeCoDe to determine and characterize the most likely etiologic causes for community respiratory deaths occurring among infants and children in a low-resource setting. Within the 106 cases adjudicated, we found 49 respiratory deaths due to pathogens including Klebsiella pneumoniae, Streptococcus pneumoniae, and Pneumocystis jirovecii; of note, in 21 respiratory deaths, the causative pathogen could not be identified amidst the pathogens detected by multiplex PCR. A large majority of all deaths were preventable, as was the case for each of the major causes of respiratory deaths. Proposed interventions to prevent each death were often multiple and spanned diverse settings and phases of care. Analyzing notable immediate causes of death revealed that the detection of multiple pathogens on multiplex PCR was more the rule than the exception. In some cases, histopathology was diffuse, but in others, only 1 or 2 of 6 biopsies revealed pneumonia or pneumonitis that was used diagnostically.

In contrast to prior landmark studies of pneumonia etiologies in LMIC settings [2, 18, 19], we relied on direct lower respiratory tract sampling and the clinical reasoning of our panel to elucidate the causes of ALRI, as opposed to indirect inference from statistical models. This method allowed us to interrogate the worst outcomes in a population that is often invisible to research and public health study—infants and children who die outside of a formal healthcare setting. Though this method bypasses the potential bias imposed by inferring causal deep respiratory pathogens from upper respiratory flora, other challenges are introduced, including the unavoidable subjectivity of adjudicating cases with incomplete clinical information and the uncertainty of ascribing infection to 1 pathogen when several are detected. Through this subjective process, the panel was able to weigh each piece of evidence distinctly depending on the strength and limitations of each finding (e.g., aspects of the narrative history, the number of biopsies detecting a given pathogen, the cycle thresholds, and severity of associated histopathologic findings, etc). By way of example, an infant who had had a prolonged neonatal intensive care unit hospitalization developed signs of respiratory distress at home shortly before death. Pneumonitis was observed in some areas of histopathology and 1 biopsy detected Klebsiella pneumonia at a cycle threshold of 27. Although bronchopneumonia was not detected on any biopsy, the panel concluded there was moderate evidence that the cause of death was pneumonia due to Klebsiella pneumoniae based on other subjective and objective findings and that the absence of supporting histopathology may have been due to sampling error.

Surprisingly, a viral pathogen (RSV) featured in the causal chain for only 1 case, in contrast to other studies [2, 18, 19] where viruses have comprised the majority of severe ALRI causes among children in LMIC. We note that our enrollment periods coincided with the peak of the COVID-19 pandemic, during which time other respiratory virus cases declined globally [20], and enrollment was low during the height of Lusaka’s RSV season [5, 17]. Other explanations for observing more bacterial than viral pneumonias may pertain to the postmortem context. Our study provides a snapshot of the latest phase of a clinical trajectory that may include post-viral superimposed bacterial pneumonia and could miss the causal contribution of pathogens with relatively brief periods of viral shedding such as RSV and influenza [21]. As such, postmortem surveillance with MITS/DeCoDe methods may be complementary to traditional case-control epidemiologic studies of pneumonia etiology in order to give a full picture of pathogens along the causal chain.

Another notable finding was the large number of respiratory deaths that could not be attributed to a specific pathogen—in large part due to the detection of multiple pathogens across the different lung zones. The frequent finding of multiple pathogens is consistent with other MITS studies focused on the lung [22, 23]. Interestingly, detection of multiple pathogens was not restricted to respiratory deaths, highlighting the importance of experienced panelists to discriminate between causative and non-causative pathogens. Each pathogen detected by PCR could reflect numerous scenarios, e.g., the cause of a fatal invasive infection, the cause of a mild infection present at the time of death, residual DNA/RNA from a prior infection, a lower respiratory tract colonizer, or a contaminant with potential for overgrowth in the postmortem setting [21]. For example, adenovirus was frequently detected, but usually only on 1 of 6 biopsies, and at a higher cycle threshold, so was attributed to prolonged shedding rather than acute infection in all cases. Further complicating the picture, detecting a pathogen in 1 lung zone had to be interpreted within the context of the sampling bias inherent to MITS, which has been shown to perform better in cases of diffuse infection [24]. Within our study, fatal pneumonias were not always diffuse by molecular methods or histopathology. Therefore, future pneumonia etiology studies using MITS should be cautioned against inferring pneumonia etiologies using fewer lung samples. In theory, some of this sampling error could be reduced through the use of ultrasound guidance, a hypothesis our group is investigating further.

Klebsiella pneumoniae was unexpectedly the most common pathogen-level cause of fatal pneumonia identified in this community-based case series. This stands in contrast to earlier landmark childhood pneumonia etiology studies where this pathogen has been largely absent or not systematically studied [2, 18]. More recent studies have revealed Klebsiella pneumoniae as an important pathogen in serious infections among children [3, 25, 26] and specifically in pneumonia [27]. In a recent study from the CHAMPS network, Mahtab et al have found K. pneumoniae to be among the leading pathogenic causes of fatal pneumonia in infants and children (78/455 or 26% pneumonia deaths) [28]. Potential explanations for this finding include a changing etiologic distribution of bacterial ALRI due to uptake of the pneumococcal vaccine and/or effects of broad antimicrobial use and misuse. Of note, we observed Klebsiella pneumoniae deaths among children with and without a history of likely nosocomial exposure. Our study adds urgency to the need to investigate and design interventions to address this pathogen, especially considering its quickly evolving global epidemiology featuring more hypervirulence factors and antimicrobial resistance genes [29–31].

Our study also shows numerous opportunities to prevent respiratory and non-respiratory deaths. As previously described in the ZPRIME study, delays in seeking care, arriving to care, and receiving care were prevalent [32]. In some cases, discharges home from a medical setting occurred shortly before death; we were unable to formally assess medical services provided prior to death but this is an area of needed study. Quality of healthcare can play a major role in differential mortality globally [33], as has been seen in CHAMPS [34]. Post-discharge mortality is poorly understood and can exceed inpatient mortality in some LMIC settings [35]. Finally, most causal respiratory pathogens in our study were not vaccine preventable, highlighting areas of potential for future vaccine development to address a shifting etiologic landscape (Supplementary Tables S5 and S6).

Limitations

The DeCoDe process for adjudicating cause of death, and by extension, determining whether the death was preventable or not, is subjective, imperfect, and prone to potential biases. There is no ground truth against which to judge the determinations of the DeCoDe panelists in our study. The process reflects the individual experience of the panelists alongside their collective clinical reasoning, calling into question the reproducibility of this method. Nonetheless, the process is a significant improvement over the status quo, which is even more subjective, less transparent, and completely unstructured. On a spectrum of data quality for cause of death analysis, we go from (1) no formal process for adjudication at all (sadly true in most settings, and largely the case in Zambia to this point); to (2) verbal autopsy-driven adjudications (which can only provide syndromic cause of death assessments); to (3) MITS with pathology and molecular evidence interpreted by a DeCoDe panel (who follow a process that seeks to bring a further step up in terms of specificity); and to (4) formal open autopsy with extensive histopathological/molecular/other testing with a systematic adjudication process. The latter is the gold standard—ideal but often unachievable in LMICs, especially at scale. Thus, MITS/DeCoDe represents a substantial advance because it is feasible and relatively noninvasive. Similar limitations apply to the determination of whether the death was preventable or not, which was also tasked to the DeCoDe panel, and in many cases, rests on the correct causal adjudication.

Another limitation of our study was that the costs and complexities of MITS sampling meant that we lacked the resources to sample all infants from the ZPRIME and LCIS studies using MITS. The deceased infants and children who had MITS performed and are included in the present study were not purposefully sampled due to logistical and administrative challenges. For this reason, we refrain from typical statistical analyses performed for case-control studies. Therefore, we are unable to determine population-attributable fractions for the causes of death that were identified and it is not suitable to draw inferences about secular disease transmission, such as with RSV. In particular, we note that lower numbers of cases were enrolled during peak RSV and influenza season and likely contributed to the low number of deaths we observed due to these key pathogens. This likely occurred due to the timing of the enrollment windows under the 2 studies and interruptions in enrollment activity in April-May 2020, as well as general effects of the COVID-19 pandemic on the respiratory virus season during the year 2020. Our hope is that insights from our analysis can inform systematic, multi-site epidemiologic study of pneumonia etiology in areas of high burden.

Additionally, our conclusions are limited by the narrow scope of the clinical information that could be collected and the contents of the FTD-33 multiplex panel. Specifically, our methods omitted several key causes of pneumonia in young infants (among them, herpes simplex virus and Chlamydia trachomatis), evaluation for malaria, and anthropometric data on nutritional status for most children. The multiplex notably excludes Mycobacterium tuberculosis and it is possible that relying solely on histopathology for diagnosis of tuberculosis led to under-reporting. Incorporating immunohistochemistry and histochemical stains into the MITS methods would have enabled differentiating co-infections from precedent infections, thereby strengthening the analysis. Finally, variations in the time elapsed and storage conditions from time of death outside the hospital to the MITS procedure pose an important limitation, and could have unknown impacts on the ability to isolate viruses on subsequent molecular testing.

CONCLUSIONS

This study of respiratory causes of death using MITS and DeCoDe methods presents a novel and important approach to studying pneumonia etiologies in LMIC, which is key to reducing global child mortality. Its strengths are complementary to traditional case-control studies of pneumonia etiologies. Future studies using representative sampling across multiple sites could inform population-attributable fractions for specific pathogens.

Notes

Author contributions . AW: methodology, investigation, analysis, Writing—original draft, Writing—editing, MC: methodology, investigation, writing—editing, project administration, data curation, NN: investigation, writing—editing, EJM: investigation, writing—editing, MK: investigation, writing—editing, CM: investigation, writing—editing, BM: investigation, writing—editing, LN: investigation, writing—editing, RP: project administration, writing—editing, funding acquisition, VM: investigation, writing—editing, CC: investigation, data curation, BN: investigation, CC: investigation, SC: investigation, LP: investigation, DN: investigation, JL: investigation, data curation, LF: data curation, WM: methodology, data curation, CM [2]: project administration, supervision, SWS: methodology, investigation, writing—editing, CG: conceptualization, methodology, Writing—editing, supervision, funding acquisition.

Acknowledgments. We would like to acknowledge Dianna Blau and Mauricio Caballero for sharing their expertise related to the DeCoDe process as we designed our protocol. This work could not have been done without the tireless, thoughtful, and compassionate efforts of the field team. They created and refined a grief counseling intervention which they offered to all families in a setting where grief counseling is not routinely provided and guided family members through the study process with great sensitivity. Most importantly, we thank the families of the deceased infants and children, who considered participation in this study during a time of unthinkable loss.

Financial support. Funded by the Bill & Melinda Gates Foundation. Supported by Clinical & Translational Science Institute 1UL1TR001430. AW acknowledges grant funding from National Institutes of Health training grant T32HL125232-07.

Potential conflicts of interest. Conflicts of interest: The authors have no conflicts of interest to report. The sponsors played no role in the development of this manuscript. CG joined the Bill & Melinda Gates Foundation following completion of data collection for this project.

Data availability. Data from this study and access to the biorepository are available upon reasonable request to the corresponding author.