Neonatal sepsis caused by K. pneumoniae infection is a significant health challenge in Africa and Asia with disproportionately high incidence and mortality rates. Our analysis estimated that over half of all neonatal sepsis cases at the included sites were within transmission clusters, with a median 33.5% of all cases attributable to nosocomial transmission (Figure 1).

Transmission analyses typically require detailed data on patient admission, ward movements, and recent healthcare exposures in addition to pathogen sequence data; these are hard to standardise across studies and thus between-study comparisons or meta-analyses are challenging. Here, however, we used studies in which all consecutive neonates with sepsis in a defined physical unit were sampled and the resulting isolates sequenced, allowing us to undertake a standardised clustering analysis based on genetic and temporal distances within each unit, applied to data collected across 10 unique studies.

Furthermore, as our goal was to estimate the fraction of cases linked to others in transmission clusters, rather than trying to resolve specific transmission chains or pathways, our approach was robust to changes in distance thresholds (see S4 Figure, S1 Appendix). This is because lowering clustering thresholds tends to break larger single-linkage clusters into smaller ones (which has no impact on the overall fraction included in clusters), rather than breaking clusters into completely un-linked singleton cases. Variations in genetic and temporal distance thresholds have a greater impact in real-time outbreak investigations, where the goal is to identify specific transmission events and sources. In that context, the choice of thresholds can be adjusted to achieve a desired preference for sensitivity or specificity in cluster detection. Interestingly, previous studies have proposed whole-genome SNV cutoffs of between 21 and 25 for identifying K. pneumoniae nosocomial transmission, which is equivalent to the 10 Pathogenwatch core-gene SNV threshold used in our analysis [18,49,50].

As with other opportunistic pathogens frequently implicated in hospital-acquired infections, reservoirs and transmission routes for K. pneumoniae infection are abundant in the hospital environment, and genomics has been used to resolve specific sources in several studies. These include transmission through direct contact with contaminated surfaces, sinks, cleaning supplies, medical devices, and healthcare personnel [51–55]. Neonates are at high risk of hospital-acquired infections due to their naive or compromised immune systems, exposure through invasive procedures in critical care settings, repeated handling by multiple caregivers, and long hospital admissions [56,57]. While our study indicates a high impact of IPC deficiencies in neonatal units in LMICs, identifying suitable solutions will require better understanding of risk factors, transmission routes, and sources of contamination. The solutions will also require appropriate testing in clinical trials or complex interventions and subsequently funding for implementation at scale.

We observed a high diversity of K. pneumoniae lineages (172 unique STs) implicated in neonatal sepsis cases across hospitals in Asia and Africa (S5 Table), consistent with previous reports of a highly diverse etiology of K. pneumoniae-associated neonatal sepsis [5,22]. A third of these STs were part of at least one transmission cluster, among which 14 STs were identified as commonly transmitted, occurring in ≥3 clusters and ≥2 sites. However, most of these were also common among singleton isolates and accordingly our logistic regression analysis indicated only two of these STs were significantly associated with transmission clusters, compared with other STs. This finding is consistent with the hypothesis that many different STs can become common causes of transmission clusters, either because they possess traits that render them particularly fit for transmission in the hospital setting, or simply because they are more prevalent (e.g. as gut colonisers in the community) and therefore have greater stochastic opportunity due to more frequent introductions to the hospital setting. Importantly, the signal of transmission we are exploring in this study – detection as a cluster of clinical infections – is the product of multiple factors including colonisation potential, virulence, and environmental survival, and we are unable to tease apart the relative contributions of these different factors in determining which strains are identified frequently in clusters. Notably, many of the 14 commonly transmitted STs identified here, including ST11, ST14, ST15, ST17, ST29, ST101 and ST307 are recognised high-risk clones, widely implicated in multidrug-resistant (MDR) hospital-acquired infections worldwide [58–60], consistent with our observations. ST39 has been reported less frequently, but is also disseminating rapidly worldwide, including in Africa, where it has been implicated in MDR neonatal outbreaks in The Gambia and Malawi [22,28,61].

The overall high rates of AMR among the K. pneumoniae isolates (clustered and singleton) in this study (see S6 Table) is concerning, as a high degree of resistance is known to be associated with higher rates of nosocomial spread [18,62], a finding that was corroborated in our analysis. We found that cluster cases had significantly higher rates of ESBL and carbapenemase gene carriage compared to single cases, and isolates belonging to the 14 most commonly transmitted STs more frequently carried ESBL genes compared to other STs (S6 Table, S8 Table). Carbapenemase and ESBL gene carriage were also independently associated with transmission in the multivariable regression analysis (S7 Table). It is likely that these AMR phenotypes also contribute to case detection, as strains resistant to empirical therapy are more likely to be present in the blood at sufficient quantities to be detected by blood culture, making it more likely for transmitted AMR strains to result in cluster detection than transmitted susceptible strains.

While AMR almost certainly plays a role in the successful persistence of K. pneumoniae clones within the hospital setting, other potential factors such as survival on environmental surfaces and colonisation ability are less well characterised [63]. For instance, ST17, the most commonly transmitted ST in this study, was recently demonstrated to colonise the gastrointestinal tracts of children for up to 2 years, a factor that may contribute to its increased transmissibility [64]. ST307 is frequently associated with outbreaks and believed to be adapted to the hospital environment where it can persist for extended periods and rapidly spread among patients [25,63,65]. In this study, ST307 was responsible for the largest identified cluster (involving 188 neonates over a 32-week period at a hospital in Zambia), and 14 other clusters across 8 sites in Kenya, Malawi, Rwanda, Zambia, South Africa and Ghana. However, ST307 was also detected in 43 singleton infections, and only 26% (n=15/58) of detected ST307 introductions in the neonatal units were associated with transmission clusters (similar to the background rate observed across 158 non-commonly transmitted STs). This suggests that the frequency of ST307 among hospital outbreaks may reflect a generally high prevalence (i.e., as a gut coloniser) resulting in greater opportunity, in combination with or instead of a specific propensity for transmission in the hospital setting. Further research is needed to determine if specific STs have enhanced risk for transmission once introduced into the hospital setting, and to elucidate potential factors that contribute to their success as hospital-acquired opportunistic pathogens.

In summary, our findings demonstrate the need to improve IPC practices in neonatal care settings in LMICs, and highlight the potential benefits of genomic surveillance of K. pneumoniae and other pathogens to assess IPC impacts and/or to support real-time outbreak detection and investigation. Limiting horizontal transmission through improved IPC measures is undoubtedly a crucial prevention strategy that could prevent a significant proportion of neonatal sepsis caused by K. pneumoniae and other pathogens, and curb the spread of AMR. Studies have shown that K. pneumoniae outbreaks in neonatal units can in principle be effectively controlled by implementing simple IPC measures like contact precautions and improved hand hygiene [21,28,31]. However, to be truly effective in reducing overall disease burden, these measures need to be proactive, rather than reactive, and be integrated into routine practice. The implementation of consistent and effective IPC measures in resource limited settings remains challenging due to resource constraints, inadequate infrastructure, limited cleaning supplies, understaffing, and limited access to training [66,67]. Whilst this analysis was not suited to probing the determinants of variation in transmission rates, we did find that the three facilities with limited access to piped water displayed high cluster proportions, highlighting the challenges faced in some settings and their impact on hygiene and patient safety. In contrast, the lowest cluster proportions and the lowest isolate counts were observed in consecutive studies at the same large tertiary care hospital in South Africa, which is comparatively well resourced in terms of infrastructure and staffing. Simplified IPC models tailored to resource-constrained settings and focused on training, improved cleaning, and hand hygiene have been proposed and demonstrated to reduce the incidence of neonatal sepsis and mortality [31]. However it is clear that to be effective, such measures must be supported by sustained investment in infrastructure, equipment, staff and training [68]. In addition, to maximise their effectiveness, IPC implementation should be periodically assessed using suitable assessment tools and reviewed as necessary [69–71]. IPC strategies must also be supported by expanding access to blood culture facilities [72,73], accurate diagnostics and robust clinical and microbiological surveillance [74], ideally incorporating whole-genome sequencing, to enable prompt detection of infections, precise transmission tracking, and timely institution of control measures.

Additionally, our findings also have implications for clinical management. WHO guidelines for empirical treatment of neonatal sepsis (third-generation cephalosporins plus gentamicin [75]) are likely to be ineffective against the vast majority of K. pneumoniae characterised in this study, which mostly carried ESBL genes, consistent with recent reports from NeoOBS and other studies [11,76]. Access to blood culture facilities, robust clinical diagnostics and susceptibility testing is essential to monitor AMR and update local syndromic guidelines, as is improving access to effective antibiotics in all settings.