Causes and incidence of community-acquired serious infections among young children in south Asia (ANISA): an observational cohort study

1The Millennium Development Goals led to reductions in deaths among children younger than 5 years by 56%, from 12·7 million in 1990 to 5·6 million, in 2016 worldwide. In south Asia the reduction was 64%, from 4·8 million to 1·7 million, but the percentage of neonatal deaths increased from 45% in 1990 to 59% in 2016.

2The main causes of neonatal deaths in low-income and middle-income countries are prematurity, intrapartum-related events, severe infections, and congenital anomalies.,, Severe infections, including sepsis, meningitis, and pneumonia, are estimated to cause about a third of the 2·6 million neonatal deaths globally or often lead to long-term disabilities. Prevention of neonatal infections will be necessary to achieve Sustainable Development Goal 3 (ensure healthy lives and promote wellbeing for all at all ages) and the objectives of the Every Woman Every Child initiative., Although neonatal illness is commonly referred to as possible serious bacterial infection (pSBI), the causes remain largely unknown,, which is an important impediment to the design of effective prevention and management programmes.

3Evidence before this studyThe Millennium Development Goals reinforced the global commitment to improve child survival and have led to mortality in children younger than 5 years being lowered by 56% from 12·7 million in 1990 to 5·6 million in 2016. In south Asian countries, mortality in this age group has been reduced by 64%, from 4·8 million to 1·7 million. Neonatal mortality, however, increased from 45% to 59% between 1990 and 2017. We searched PubMed and Ovid, without date restrictions, with 51 combinations of the keywords “neonatal infections”, “South Asia”, “sepsis”, “young infant”, “etiology”, “low income countries”, “clinical algorithm”, and “surveillance”. This search yielded 317 references that were relevant to our research questions. The main causes of neonatal deaths seem to be prematurity, intrapartum complications, severe infection, and congenital anomalies. Severe infection is addressed by treatment and prevention policies focused on clean delivery at birth, breastfeeding, appropriate antibiotic therapy for ill babies, and vaccination. Nevertheless, the causes of most possible serious bacterial infections (pSBIs) remain elusive. In 2012, an estimated 6·9 million neonates worldwide met the WHO definition for pSBI—ie, referral to hospital and antibiotic treatment—although in studies bacteria have been detected by culture in only 5–10% of samples. Information about viral causes of illness is even scarcer. Treatment remains mostly empirical, particularly in developing countries, where most data on antimicrobial resistance derive from hospital-acquired infections. Given the gap in knowledge about causes of pSBI in neonates and the widespread use of treatment with antibiotics, multidrug resistance is a concern. Few studies have included neonates, and even fewer babies on the day of birth, when most neonatal deaths occur. Moreover, studies in low-income and middle-income countries have mostly been hospital based and enrolled babies with pSBI who were seeking health care. In many parts of south Asia, home births are still common and seeking care in the days after birth is rare, even when referred by a community health worker.Added value of this studyWe designed the ANISA study to assess unanswered questions specifically in developing countries, in a large sample of pregnant women and babies, by testing blood and respiratory samples from those with symptoms and from healthy controls. We generated population-based evidence on the causes of pSBI through community-wide pregnancy surveillance and enrolment of babies born at home and in health facilities. We included areas with high child mortality and those with state-of-the-art diagnostic services. Stringent protocols were put in place to minimise blood-culture contamination and an external expert panel confirmed classifications of blood culture isolates. As well as blood culture, we incorporated molecular testing in almost 2000 healthy babies and used an innovative partially latent class model to attribute causes to pSBI.Implications of all the available evidenceThe list of causal bacteria we identified was long and diverse, making prevention and treatment strategies complex. Respiratory syncytial virus was the leading causal organism in our study, suggesting an opportunity to develop interventions based on maternal immunisation and provision of monoclonal antibodies to neonates. Ureaplasma spp had the second leading causal proportion, but have not previously been recognised as common except in very preterm babies, and are not targeted by current recommended antibiotic regimens. Research into the epidemiology and pathogenesis of this group of bacteria in neonates is needed. Despite testing of blood and respiratory samples, for 72% of pSBI episodes, no cause could be attributed. Use of metagenomics approaches, minimally invasive tissue sampling, or both, to investigate causes, with consideration of factors unrelated to infections might facilitate appropriate decisions and investments to save lives among very young babies, limit unnecessary antibiotic use, and curb the emergence of antimicrobial resistance in the community.

4Evidence before this study

5The Millennium Development Goals reinforced the global commitment to improve child survival and have led to mortality in children younger than 5 years being lowered by 56% from 12·7 million in 1990 to 5·6 million in 2016. In south Asian countries, mortality in this age group has been reduced by 64%, from 4·8 million to 1·7 million. Neonatal mortality, however, increased from 45% to 59% between 1990 and 2017. We searched PubMed and Ovid, without date restrictions, with 51 combinations of the keywords “neonatal infections”, “South Asia”, “sepsis”, “young infant”, “etiology”, “low income countries”, “clinical algorithm”, and “surveillance”. This search yielded 317 references that were relevant to our research questions. The main causes of neonatal deaths seem to be prematurity, intrapartum complications, severe infection, and congenital anomalies. Severe infection is addressed by treatment and prevention policies focused on clean delivery at birth, breastfeeding, appropriate antibiotic therapy for ill babies, and vaccination. Nevertheless, the causes of most possible serious bacterial infections (pSBIs) remain elusive. In 2012, an estimated 6·9 million neonates worldwide met the WHO definition for pSBI—ie, referral to hospital and antibiotic treatment—although in studies bacteria have been detected by culture in only 5–10% of samples. Information about viral causes of illness is even scarcer. Treatment remains mostly empirical, particularly in developing countries, where most data on antimicrobial resistance derive from hospital-acquired infections. Given the gap in knowledge about causes of pSBI in neonates and the widespread use of treatment with antibiotics, multidrug resistance is a concern. Few studies have included neonates, and even fewer babies on the day of birth, when most neonatal deaths occur. Moreover, studies in low-income and middle-income countries have mostly been hospital based and enrolled babies with pSBI who were seeking health care. In many parts of south Asia, home births are still common and seeking care in the days after birth is rare, even when referred by a community health worker.

6Added value of this study

7We designed the ANISA study to assess unanswered questions specifically in developing countries, in a large sample of pregnant women and babies, by testing blood and respiratory samples from those with symptoms and from healthy controls. We generated population-based evidence on the causes of pSBI through community-wide pregnancy surveillance and enrolment of babies born at home and in health facilities. We included areas with high child mortality and those with state-of-the-art diagnostic services. Stringent protocols were put in place to minimise blood-culture contamination and an external expert panel confirmed classifications of blood culture isolates. As well as blood culture, we incorporated molecular testing in almost 2000 healthy babies and used an innovative partially latent class model to attribute causes to pSBI.

8Implications of all the available evidence

9The list of causal bacteria we identified was long and diverse, making prevention and treatment strategies complex. Respiratory syncytial virus was the leading causal organism in our study, suggesting an opportunity to develop interventions based on maternal immunisation and provision of monoclonal antibodies to neonates. Ureaplasma spp had the second leading causal proportion, but have not previously been recognised as common except in very preterm babies, and are not targeted by current recommended antibiotic regimens. Research into the epidemiology and pathogenesis of this group of bacteria in neonates is needed. Despite testing of blood and respiratory samples, for 72% of pSBI episodes, no cause could be attributed. Use of metagenomics approaches, minimally invasive tissue sampling, or both, to investigate causes, with consideration of factors unrelated to infections might facilitate appropriate decisions and investments to save lives among very young babies, limit unnecessary antibiotic use, and curb the emergence of antimicrobial resistance in the community.

10Few data are available on the annual number of pSBI episodes in south Asia, despite this region having the highest number worldwide (3·5 million)., Determination of the causes of pSBI in neonates has depended on isolation of bacteria by blood culture, which is deemed to be the gold standard test despite poor sensitivity. Two outpatient-based WHO studies done in multiple countries and a population-based study done in low-income and middle-income countries yielded bacterial isolates from less than 10% of babies with pSBI.,, A study in India showed 10% positivity by blood culture, with organisms being predominantly Gram-negative (69%). A study in Bangladesh examined 62 respiratory samples and found a high burden of viral infections (48%).

11Ascertainment of the causes of pSBI has been hindered by several factors: difficulty in case finding (ie, many babies born at home die soon after birth without entering the formal health system); insufficient availability of diagnostic tools to detect bacterial and viral causes; collection of blood samples with inadequate volume, frequent contamination of blood cultures, or both; lack of parallel samples from healthy controls and assessments by sensitive molecular methods to provide information on background colonisation by pSBI pathogens; and absence of appropriate analytical approaches to use information from multiple types of tests and samples to attribute cause. To address these gaps in knowledge, we did the Aetiology of Neonatal Infections in South Asia (ANISA) study in three countries (Bangladesh, India, and Pakistan) to investigate the incidence of pSBI episodes in the first 2 months of life and estimate the proportions of bacterial and viral causes.

12ANISA was an observational cohort study done in five sites in three countries. Four sites were rural (Sylhet, Bangladesh; Matiari, Pakistan; and Vellore and Odisha, India) and one was urban (Karachi, Pakistan). These sites covered a total population of around 1·6 million people, including roughly 250 000 married girls and women of reproductive age (13–49 years) and around 30 000 births per year. Eligible participants were married women of reproductive age and live babies that could be enrolled up to age 7 days.

13We obtained informed verbal consent from pregnant girls and women when they were registered in the study and written informed consent when samples were collected from babies. The study was approved by the ethics committees or internal review boards of all participating organisations.

14We recruited community health workers (CHWs) from the local area through open advertisements. CHWs established active community-based surveillance for pregnancy, pregnancy outcomes, and signs of illness in neonates. Briefly, they followed up married girls and women of reproductive age by home visits to identify pregnancies. They recorded dates of last menstrual period, facilitated or encouraged accessing of antenatal care, built rapport, and arranged visits to assess neonates immediately after birth. Live neonates visited by CHWs within 7 days of birth were registered in the study and scheduled for up to ten home visits up to age 59 days, three week 1 and one per week thereafter. CHWs also collected socioeconomic, household, and maternal information at enrolment and information on labour and delivery at the first visit after birth.

15Babies were assessed at every visit for signs of illness, defined as respiratory rate 60 breaths per min or more, severe chest in-drawing, hypothermia (temperature <35·5°C), hyperthermia (temperature ≥38·0°C), movement only with stimulation, convulsions, and poor feeding as reported by the mother and confirmed by CHW assessment. Babies with one or more signs of pSBI were referred by the CHWs to study-designated health-care facilities or visited at home by a study physician (only in Matiari, Pakistan). To capture illness episodes that occurred between visits, family members were encouraged to seek assessment at a study health facility or via the outreach physician (in Matiari). Ill babies in Bangladesh and Pakistan were examined and managed at primary care facilities or, if they had severe or complicated illness, were referred to tertiary-level hospitals. All ill babies in India were assessed and managed at tertiary-level hospitals.

16Sample collection has previously been described in detail. Briefly, trained phlebotomists followed standardised protocols to collect up to 3 mL blood from babies with pSBI episodes for culture and molecular testing. 1 mL blood samples were taken from healthy controls for molecular testing. Strict procedures to minimise contamination during phlebotomy were followed: use of sterile hand gloves and disposible dignity sheets for babies; regular cleaning of the phlebotomy room with disinfectant; and systematic use of antiseptic (alcohol and iodine). Physicians used flocked swabs (Copan Diagnostics, Brescia, Italy) to collect nasopharyngeal and oropharyngeal samples at the health centre from symptomatic babies who met the criteria for pSBI and from healthy babies when scheduled by the study team. To characterise the background presence of potential pSBI pathogens in the community, we used an automated algorithm triggered by CHWs at the first postnatal visit to select randomly registered healthy babies for collection of samples. These babies were assigned an age at which to undergo clinical assessment and sample collection, with assignment probability based on the age distribution of babies with pSBI episodes enrolled to that point.

17We used automated blood culture (culture media BacTec, BD, Franklin Lakes, NJ, USA or BacT/Alert, Biomérieux, Basinstoke, UK) to isolate bacterial pathogens. Molecular testing was done by PCR with Custom TaqMan Array Cards (Thermo Fisher, Bartlesville, OK, USA) and LifeTech Viia 7 software version 1.2.4 to detect 15 bacterial and 13 viral pathogens., Bacterial strains from the cultures were identified with a conventional biochemical test and the API 20E strip test (bioMérieux-Vitek, Hazelwood, MO, USA), and were confirmed at the Centers for Disease Control and Prevention by matrix-assisted laser desorption/ionisation-time of flight mass spectrometry. Antimicrobial susceptibility testing was done by disc diffusion, in accordance with Clinical and Laboratory Standards Institute guidelines. We used a MagNA Pure LC total nucleic acid isolation kit (Roche, Basel, Switzerland) to extract nucleic acid from blood samples and nasopharyngeal and oropharyngeal swabs for assessment.

18Three infectious disease specialists reviewed and classified each isolate as a pathogen or as not clinically relevant, based on the clinical features of the patient, drug susceptibility, illness management, and outcome. 10% of molecular samples were retested at the Centers for Disease Control and Prevention to provide quality control and assurance for site results (appendix). As an additional assessment, we used real-time PCR for speciation of Bordetella spp and Ureaplasma spp and confirmation of Escherichia coli positivity.

19We defined confirmed pSBI episodes as the presence of one sign of illness assessed by the CHWs, no hospital admission in the preceding 7 days (except for postnatal hospital stays for babies born in health facilities), and no registered pSBI episode in the preceding 7 days. We excluded babies with increased respiratory rate as the only sign because, during the study period, independent evidence showed low specificity for this sign alone and samples were not collected in several sites. Test results for samples from healthy babies were included if the criteria for pSBI were not met on the day of assessment or in the previous 7 days before and after sample collection.

20We aimed to enrol 400 healthy babies in each site in Bangladesh and Pakistan and 200 in each Indian site. The sample size was constrained by ethical and community acceptability, but we calculated that the target sample size would be sufficient to provide initial estimation of background colonisation with set false-positive rates (appendix) and a maximum SE of 3%.

21The primary objective was to estimate the proportion of pSBI episodes attributed to each pathogen assessed. Because the study design included two types of sample (blood and respiratory), up to three tests per pathogen (blood culture, blood molecular assay, and respiratory molecular assay), and because each of these tests had its own sensitivity and specificity for detection of the true cause of the pSBI episode, we used a partially latent class model, which was originally developed for a similarly designed study of the causes of pneumonia and adapted to meet the ANISA study specifications (appendix). For all babies with pSBI episodes and healthy babies with at least one laboratory test result, we included blood culture final determination data (for those with pSBI episodes) and observed binary molecular assay results (positive or negative) in the model input dataset. We attributed causal proportions to 28 target pathogens and two additional classes: other blood culture (all organisms isolated from blood cultures that did not have an associated pathogen-specific molecular assay test) and other or none (episodes that could not be attributed to any pathogens).

22The primary output of the model was mean pathogen proportions at the population level with 95% credible intervals (CrIs), from which individual-level proportions could be derived (appendix). Covariates were study site, age at onset, date of enrolment, and outcome status (died but had a sample available from the previous 7 days vs alive, except for Vellore, where there were too few deaths to stratify by outcome). We ran the model under various input dataset conditions, for which performance was assessed by simulation studies (appendix). Model convergence was assessed through trace and other diagnosis plots. We assessed model assumptions with interim model outputs by stopping the Gibbs sampler at random cycles. Model fit to input data was evaluated by comparing the fitted and observed numbers of positives for each blood molecular assay and blood culture. We did programming and computation with R (version 3.2.5), SAS (version 9.3), and Stata (version 13).

23The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had access to all data in the study and had responsibility for the decision to submit for publication.

24During surveillance, pregnancy outcome data were collected for 74 145 (87%) of 84 971 pregnancies. 71 361 were livebirths, 2324 were stillbirths, and 1283 were miscarriages. 63 114 (88%) liveborn babies were registered and followed up for 59 days, and the remaining 8247 babies could not be reached within 7 days, due to death, migration out of the area, or refusal to participate in the study (figure 1, appendix). 56 960 (90%) liveborn neonates were recruited within 72 h. Population characteristics, in particular socioeconomic status, home delivery, pregnancy outcomes, and early death among babies, varied by site (table 1). 43 665 (69%) registered babies received all three visits in the first week, and 48 415 (77%) received at least eight of ten scheduled visits by CHWs. These visits resulted in 12 836 referrals for clinical assessment by physicians, of which 10 809 (84%) were attended. Caregivers made 3683 additional self-referrals.

25Physicians identified 6022 pSBI episodes, including 2498 early-onset (<3 days) and 3524 late-onset episodes (3–59 days). 289 (5%) babies had more than one episode (figure 1, table 2, appendix). The overall pSBI incidence was 95·4 episodes per 1000 livebirths (range 78·3 per 1000 in Vellore, India, to 112·6 per 1000 in Karachi, Pakistan) and was notably higher in the first 3 days of life than in the remaining follow-up period (55·8 per 1000 livebirths vs 39·6 per 1000 livebirths), which is equivalent to an increase in incidence density rate of roughly 13·5 times.

26Blood samples were collected from 4859 (81%) of 6022 babies during pSBI episodes, of which all were tested by blood culture and 4216 (70%) were tested by molecular assay (figure 1). 602 (10%) babies had received antibiotics before blood samples were collected, mainly in Matiari, Pakistan (439 [73%]). Molecular test results for respiratory samples were available for 5209 (87%) of 6022 babies with pSBI episodes. We also collected 1726 blood samples (results available for 1717) and 1893 respiratory samples from 1895 healthy babies for molecular testing (figure 1). Healthy babies were older than sick children at the time of sample collection (median age 13·4 days, IQR 2·8–33·8 vs 5·7 days, 1·4–27·4) and cumulative enrolment of controls did not always track pSBI episodes (appendix).

27Of 4859 pSBI blood cultures, 338 (7%) yielded bacteria, of which 206 were judged not to be clinically relevant based on predefined criteria, 30 not to be clinically relevant by an independent review panel, and 102 to be clinically relevant. The incidence of culture-confirmed infection was 1·6 per 1000 livebirths. Causal organisms in blood cultures were predominately E coli, Klebsiella spp, Staphylococcus aureus, and group A streptococcus. More than half of causal bacteria were Gram negative, led by Klebsiella spp and E coli (table 3), and incidence of Gram-negative organisms was higher among hospital-born babies than in community-born babies (1·3 per 1000 livebirths vs 0·7 per 1000 livebirths). Most isolates were susceptible to penicillin, ampicillin, gentamicin, or a combination of these drugs (table 3). Isolates were more susceptible to gentamicin (75 [74%]) than ampicillin or penicillin (27 [41%]).

28Among pSBI blood samples tested by molecular assay, 487 (12%) of 4216 tested positive for at least one target, led by enterovirus or rhinovirus, Salmonella spp, Klebsiella spp, and E coli, and 57 (1%) were positive for multiple targets. Among 5209 respiratory samples 3907 (75%) were positive for at least one target and 1302 (47%) were positive for multiple targets (table 4). The numbers of positive results were similar for blood and respiratory samples from healthy babies, with some variation by site, age, and outcome (table 4, appendix). Pathogens were more prevalent among babies with pSBI than among healthy babies, including for respiratory syncytial virus and Ureaplasma spp (table 4). Among 566 samples containing Ureaplasma spp, 311 (55%) were U parvum, 255 (45%) were U urealyticum, and the remainder contained both species. Bordetella spp were detected in similar proportions of babies with pSBI and healthy babies, with most samples containing these species seen in Karachi, Pakistan (table 4). Speciation of Bordetella spp revealed that 84 (36%) of 234 samples from babies with pSBI and 30 (41%) of 73 from healthy babies contained B pertussis.

29The estimated test-specific true-positive and false-positive rates underlying the model-based proportions of causal organisms and the estimated incidence are shown in the appendix. Overall, 16% and 12% of pSBI episodes were attributed to bacteria and viruses, respectively, and the remaining 72% could not be attributed to any of the tested pathogens. The leading pathogen was respiratory syncytial virus, seen in 6·5% of cases (95% CrI 5·8–7·6), followed by Ureaplasma spp, seen in 2·8% (1·9–3·8; figure 2). Klebsiella pneumoniae, E coli, enterovirus or rhinovirus, Salmonella spp, Streptococcus pneumoniae, group B streptococcus, and S aureus were each attributed as the causal agent in more than 1·0% of pSBI episodes (figure 2). Attribution of cause to a specific pathogen was less likely among early-onset than among late-onset episodes (22·5%, 95% CrI 19·2–26·0 vs 33·1%, 29·8–36·8). Furthermore, among early-onset episodes, bacterial attributions were much more frequent than viral (18·4% vs 4·1%; figure 2). Ureaplasma spp were the most frequent causal organisms, with an estimated pathogen proportion of 3·1% (95% CrI 1·9–4·4). E coli, K pneumoniae, group B streptococcus, Salmonella spp, S aureus, and respiratory syncytial virus all had proportion point estimates greater than 1·0% (figure 2). Among late-onset episodes, the proportion attributed to viruses was 16·5%, similar to that attributed to bacteria (16·6%) and were predominantly respiratory syncytial virus (9·1%, 95% CrI 7·7–10·9) and Ureaplasma spp (2·5%, 1·6–3·4; figure 2). Attribution of bacterial causes differed by site: the leading bacterial causes were S pneumoniae in Sylhet (2·6% of episodes, 95% CrI 1·4–4·7), Ureaplasma spp in Matiari (5·3%, 3·0–7·4), and Klebsiella spp in Odisha (7·2%, 4·4–7·2; figure 2). Respiratory syncytial virus was the leading viral cause in all sites (figure 2).

30The overall incidence of bacterial infections was 13·2 (95% CrI 11·2–15·6) per 1000 livebirths (table 5). For Ureaplasma spp, the overall incidence was 2·4 (95% CrI 1·6–3·2) per 1000 livebirths, with the highest incidence seen in Matiari. K pneumoniae was the next leading pathogen, for which the highest incidence was in Odisha. Incidence of E coli varied little across sites and S aureus was a rare cause except in the Indian sites, where incidence was greater than 2·0 per 1000 livebirths (table 5).

31The overall incidence of viral infections was 10·1 (95% CrI 9·4–11·6) per 1000 livebirths (range across sites 5·5–15·7 per 1000 livebirths, table 5). The incidence of respiratory syncytial virus was greater than 2·5 per 1000 livebirths in all sites, peaking in Sylhet (7·3 infections per 1000 livebirths). For all other viruses, the incidence was 1·1 per 1000 livebirths or lower, with site-specific variations (table 5).

32Overall, among 71 361 livebirths, 3061 (4%) babies died before age 60 days (range 11 per 1000 livebirths in Vellore to 57 per 1000 livebirths in Matiari; overall mortality 42·9 per 1000 livebirths). 1377 (45%) of these 3061 deaths occurred among non-registered babies (1284 within 7 days and 93 after 7 days) and 1684 (55%) among registered babies. Among 1377 babies who were not visited by a CHW before death, 941 (68%) died within 24 h of birth (median age 6·1 h, appendix). Mortality among registered babies was 26·7 deaths per 1000 livebirths, which was 7·4 (95% CrI 6·7–8·1) times higher among babies with pSBI than among healthy babies (125·0 vs 16·9 per 1000 livebirths). Among the 1684 registered babies who died, 689 (41%) were assessed by a physician before death and pSBI was confirmed. Samples were collected from 349 (49%) of these confirmed episodes in the 7 days before death (appendix). Among 102 babies with confirmed episodes who died within 24 h of birth, 37 (36%) had samples collected.

33Causal pathogen attributions were made in substantially more babies with pSBI episodes who died than among those who survived (46%, 95% CrI 37–56 vs 27%, 24–30). Of the deaths among babies with pSBI episodes with attributed pathogens, 92% were due to bacteria, led by E coli (causal proportion 8·7%, 95% CrI 5·2–13·4) and Ureaplasma spp (8·3%, 4·1–12·3; figure 3). Other pathogens that were over-represented among babies with pSBI who died compared with among those who survived were K pneumoniae, S pneumoniae, group B streptococcus, and Pseudomonas aeruginosa (figure 3). In contrast, the proportion attributed to respiratory syncytial virus was greater among babies who died than among those who survived (2·1%, 95% CrI 1·3–3·2 vs 6·7%, 5·9–7·8).

34Given these differences in pathogen distributions between babies who died and those who survived and the low capture of samples within 7 days of life (11·6% of deaths), we did a sensitivity analysis to ascertain the potential effect on the overall distribution of causal organisms. Specifically, we assumed that the pathogens among missed deaths were distributed equivalently to those among deaths with laboratory results in the most extreme scenario (100% of deaths fulfilling the pSBI definition). The overall pathogen prevalence distribution in ANISA did not change markedly, with the pathogen proportion attributed to Ureaplasma spp, E coli, and K pneumoniae being increased to about 3·0–4·0%, and those attributed to respiratory syncytial virus decreased from 6·5% to 5·3% (appendix).

35With neonatal deaths fast becoming predominant among deaths in children younger than 5 years, we aimed with the ANISA study to use state-of-the-art microbiology techniques and computational modelling in one of the largest community-based studies of its kind to bridge major knowledge gaps and improve estimates of incidence of neonatal infections and their causes in south Asia, where 36% of all neonatal deaths occur. Although causal attribution among babies with pSBI in our cohort of over 63 000 livebirths was low (28%) because of the highly sensitive clinical algorithm we used, we found that bacterial causes predominated over viral causes. Even more importantly, ANISA captured data on pathogens associated with pSBI deaths, indicating that a cause could be attributed in 46% of deaths, among which most (92%) were bacterial.

36The incidence of culture-confirmed bacterial pSBI episodes was 1·6 per 1000 livebirths, which was lower than previously reported incidence in south Asia (3·0 per 1000 livebirths in Mirzapur, Bangladesh, and 6·7 per 1000 livebirths in Odisha, India). A study in Africa reported incidence of 5·5 per 1000 livebirths for bacterial infections among neonates. The low frequency of culture-confirmed pSBI episodes in ANISA might have stemmed partly from efforts to minimise inclusion of isolates that were not clinically relevant, strategies to minimise contamination of samples, and expert review of positive cultures. Use of the molecular assays and modelling to attribute causes, however, indicated a higher proportion of bacterial causes than reported in previous studies.

37Among bacterial causes, Ureaplasma spp predominated, particularly during the first 3 days of life and among babies with pSBI who died. The role of this atypical bacterium as a cause of neonatal morbidity and mortality is not well understood because it does not grow under conventional culture methods. Previous associations of Ureaplasma spp with adverse outcomes in mothers (chorioamnionitis and extreme preterm delivery) and neonates (prematurity, sepsis or meningitis, and pulmonary and cerebral morbidity) have been documented, as have pSBI episodes in neonates due to U parvum and U urealyticum., Common empirical treatment with β-lactams or cephalosporins is not effective against Ureaplasma spp due to the lack of a cell wall. Future studies using various methods are needed to characterise the pathogenic role of Ureaplasma spp in babies and guide treatment policies.

38Over half of bacteria identified as causing pSBI were Gram negative, particularly Klebsiella spp and E coli. These bacteria ranked among the top pathogens resulting in death among babies with pSBI episodes and were present in all study sites. Blood culture analysis revealed higher incidence of Gram-negative organisms among hospital-born babies than among community-born babies. These findings are consistent with those in previous studies from south Asia., The prevalence of Gram-negative organisms in the environment and on the body surface of neonates in south Asia might be the source of these infections.,

39In concordance with previous data from south Asia, the incidence of pSBI episodes caused by group B streptococcus was lower than that reported in Africa., Attribution was, however, more than 3·7 times more likely among babies with pSBI who died than among survivors, and 1·8 times more likely among babies with early-onset illness than among those with late-onset illness. These findings suggest that the burden of group B streptococcus might be underestimated because of the challenges in capturing samples after early death. A study done in Bangladesh at the same site as ANISA showed that mortality was increased by 6·6 times among neonates with group B streptococcus colonisation of the umbilical cord stump. Further investigations of early deaths and a maternal vaccine probe study will be crucial to understanding the true burden of group B streptococcus among babies in south Asia.

40We collected 349 samples from babies who died, which exceeds the number of samples obtained from babies after pSBI-associated deaths in previous studies, yet still represents only a small proportion of all deaths in the ANISA catchment areas, despite intensive efforts. This notable limitation stemmed from the high frequency of early deaths. 68% of deaths among non-registered babies occurred within 24 h of birth (median age 6·1 h), among whom over a quarter died in the first hour of life (appendix). The challenge of reaching babies before death also translates into challenges for the treatment and prevention of illnesses in at-risk babies. Among those who had pSBI episodes and died, the causative bacteria were diverse (>25 species within 19 genera), complicating development of empirical treatment and prevention strategies. Furthermore, antibiotic therapy might be given too late to be effective, particularly for births outside health facilities. Maternal immunisation has the potential to protect fetuses and neonates from infection, but maternal vaccines are not available for any of the leading organisms associated with pSBI deaths. Interventions such as use of a clean delivery kit at birth might prevent some early deaths due to infection.

41The predominant role of respiratory syncytial virus (5·4 per 1000 livebirths) among viral causes of pSBI in the first 2 months of life is consistent with previous studies in neonates in Bangladesh and Nepal., Among babies who had pSBI and died in this study, respiratory syncytial virus ranked sixth. Respiratory syncytial virus further stresses under-resourced health systems where the numbers of hospital beds are small and patients need continuous care for common sequelae, such as recurrent wheeze and asthma. The contribution in this study of respiratory syncytial virus to pSBI, even in the first days of life, suggests a specific opportunity to intervene and prevent deaths and morbidity through maternal immunisation, administration of monoclonal antibodies to neonates, or both, which are looking likely to become available options.,

42Our findings should be considered within the context of some limitations. First, despite using highly sensitive diagnostic methods, we could not identify causes in most (72%) pSBI episodes, and more than half (54%) of fatal pSBI episodes. The reasons for these outcomes are multifactorial. First, a subset of episodes could have had causative agents beyond the 28 prioritised by the expert panel. The unexpected finding of Ureaplasma spp as leading causes indicates that organisms that are not generally judged to be the most common might have important roles. Genomic analyses have shown that many bacterial species can only be detected by culture-independent approaches., With advances in metagenomics, the biobank of the ANISA study will provide an opportunity to probe for additional organisms. Second, ANISA only captured blood and respiratory samples. Although cerebrospinal fluid collection was attempted, very few samples were obtained. We did not try to collect samples from other sterile sites, such as lung aspirates or urine, or peripheral sites other than the nasopharynx and oropharynx. Third, collection of blood and respiratory samples from healthy babies in the first days of life was challenging due to difficulties in early birth capture and community acceptability. Thus, samples from healthy babies were skewed towards older babies who might have had more time to acquire organisms from the environment. Overestimation of background presence of potential pSBI organisms would, therefore, have underestimated causal attribution. Fourth, we used WHO's Integrated Management of Childhood Illness algorithm to define pSBI, but this approach emphasises sensitivity at the cost of specificity., The standard clinical signs of pSBI also overlap with those of intrapartum-related events and other non-infectious causes of neonatal mortality, making it unlikely that all pSBI episodes are of infectious origin. Our findings lend support to the prediction of Mulholland and colleagues that only a limited proportion of babies with pSBI actually have illnesses of bacterial origin. Moreover, poor specificity could explain the results of other studies in which standard and simplified antibiotic regimens (ranging from 14 to two injections and first-line oral antibiotics) were equally effective.,, We found that 83% of the true bacterial isolates from pSBI were sensitive to first-line antibiotics, which suggests that widespread antimicrobial resistance is not yet deeply rooted in the community.

43ANISA has opened several avenues for further exploration that might uncover new strategies for improving survival among babies and achieving Sustainable Development Goal 3. These include application of the latest metagenomic approaches for improved characterisation of causal organisms in samples; the role of Ureaplasma spp, for which studies should be designed to elucidate management policies; assessment of maternal immunisation against respiratory syncytial virus infection; identification of ways to differentiate infectious from non-infectious pSBI (eg, with biomarkers or distinct clinical signs, metagenomics approaches, and minimally invasive tissue sampling); and investigation of factors unrelated to infection that cause pSBI. Low antibiotic resistance across south Asian communities provides an incentive to limit unnecessary antibiotic use to prevent emergence of antimicrobial resistance.