Integrating patient movement and pathogen genomics to support hospital infection prevention with PathoPath: a method development study
1Child Health Research Foundation, Dhaka, Bangladesh
2Centres for Antimicrobial Optimisation Network, University of Liverpool, Liverpool, United Kingdom
3Department of Microbiology, Bangladesh Shishu Hospital and Institute, Dhaka, Bangladesh
4Department of Neonatology, Bangladesh Shishu Hospital and Institute, Dhaka, Bangladesh
5Department of Epidemiology & Research, Bangladesh Shishu Hospital & Institute, Dhaka, Bangladesh
6Bangladesh Medical University, Dhaka, Bangladesh
7Dr. MR Khan Shishu Hospital & Institute of Child Health, Dhaka, Bangladesh
8Department of Pediatrics, Bangladesh Shishu Hospital and Institute, Dhaka, Bangladesh
9David Price Evans Global Health and Infectious Diseases Research Group, University of Liverpool, Liverpool, United Kingdom
10Centres for Antimicrobial Optimisation Network, Imperial College London, London, United Kingdom
11The Fleming Initiative, Imperial College London and Imperial College Healthcare NHS Trust, London, UK
*Corresponding author information: Name: Yu Wan, Email: yu.wan@liverpool.ac.uk, Address: William Henry Duncan Building, University of Liverpool, 6 West Derby Street, Liverpool, United Kingdom, L7 8TX, Name: Yogesh Hooda Email: yhooda@chrfbd.org, Address: Child Health Research Foundation, 23/2, SEL HUQ SKYPARK, 5th Floor, Block-B Khilji Rd, Dhaka 1207, BangladeshSummary
Background
Healthcare-associated infections pose a major burden to neonatal health worldwide and remain difficult to track in low-resource hospitals because patient movement data and pathogen genomic data are rarely integrated into actionable transmission models. Existing approaches are often restricted to specific settings, highly structured electronic health records (EHRs), or analyses focused on either patient movements or pathogen characteristics alone. To address this gap, we developed PathoPath, an open-source integrative modelling platform, and evaluated its utility in a high burden paediatric hospital in Dhaka, Bangladesh.
Methods
PathoPath is an open-source R package that combines electronic health records with whole genome sequencing data to generate contact networks from direct and indirect contacts using minimal structured inputs. We retrospectively applied PathoPath to 373 cases of Klebsiella pneumoniae species complex (KpSC) infection identified in 2021 at the largest paediatric referral hospital in Dhaka, Bangladesh. Ward level patient movement trajectories were used to reconstruct contact networks, and genomic data from isolates from children <60 days were integrated to identify probable dissemination of bacterial clones and antimicrobial resistance plasmids.
Findings
PathoPath identified 750 direct contacts among 317 patients, forming 25 connected components, with the largest including 93 patients. KpSC infections were identified across 21 of 37 wards, with the neonatal intensive care unit accounting for 77.9% of all cases. Integration of genomic and network data distinguished sustained clustering of ST147 from multiple probable inter-clonal dissemination events involving IncFII plasmids carrying blaNDM-5 and/or blaOXA-181 within ST16. Four dominant sequence types accounted for 65.6% of sequenced isolates, and carbapenemase genes were detected in 95.8%.
Interpretation
PathoPath reconstructs hospital-wide contact networks and integrates them with pathogen genomics to map probable dissemination of pathogens and antimicrobial resistance using minimal structured clinical data. It could support more targeted infection prevention and control in hospitals where granular digital records are not available.
Funding
Gates Foundation, National Institute of Health Research, Wellcome Trust, and the David Price Evans Endowment.
Research in context
Evidence before this study
We searched PubMed (pubmed.ncbi.nlm.nih.gov) and Semantic Scholar (www.semanticscholar.org) databases in May 2025 for literature about existing methods for modelling healthcare-associated transmission of infectious diseases. We identified at least 3 Bayesian, 1 SEIR (Susceptible, Exposed, Infectious, Recovered) and 33 other models/approaches such as StEP and outbreaker2 to support these efforts. 19/37 of these methods have demonstrated their capability to reconstruct transmission or detect outbreak with varying level of accuracy depending on the context and pathogen. However, most of these tools frequently rely on dense patient metadata (e.g., Radio Frequency Identification sensors), specific types of electronic healthcare records that might not be readily available across settings, or a specific spatial structure. Also, most of these tools either focus on patient records or genomic data so there was a need for a generalised, integrative framework that can bridge the gap between clinical records with minimal digital input and pathogen genomic information.
Added value of this study
We developed PathoPath, an open-access integrative pathogen agnostic method that uses minimal electronic healthcare data and genome information to construct transmission pathways of pathogen and antimicrobial resistance within a healthcare facility. We have tested this package using 373 cases of Klebsiella pneumoniae species complex (KpSC) infection at Bangladesh Shishu Hospital and Institute (BSHI), the largest tertiary paediatric healthcare facility in Dhaka which serves as a referral centre in Bangladesh for the sick paediatric population. Using PathoPath, we successfully identified sustained K. pneumoniae ST147 clonal transmission and a few events of potential dissemination of carbapenemase-encoding plasmids among K. pneumoniae ST16, providing us a better understanding on how resistance is moving through the hospital network.
Implications of all the available evidence
PathoPath provides a scalable approach to generate high resolution data to inform infection prevention and control in settings that still lack well-defined electronic healthcare records. This could provide a greater level of understanding of transmission than that offered by standard practice and enable the implementation of targeted, cost-effective interventions to prevent further outbreaks or spread of antimicrobial resistance within hospitals.
Article notes
Competing Interest Statement
The authors have declared no competing interest.
Introduction
Healthcare-associated infections (HAIs) are a major cause of mortality and morbidity among neonates, especially in resource-constrained low-and-middle-income countries (LMICs).1 Limited staffing, overcrowding, and inadequate infection prevention and control (IPC) infrastructure allow sustained transmission of pathogens that are well adapted for such environment.2 The rapid emergence and dissemination of carbapenem resistance, particularly among Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacteriaceae species has further constrained effective therapeutic options, prompting the World Health Organisation (WHO) to classify these organisms as critical priority pathogens.3 Carbapenem resistant gram-negative bacterial clones and/or carbapenemase-encoding mobile genetic elements (MGEs) can spread rapidly within healthcare facilities through shared wards, equipment, healthcare worker or contaminated surfaces, making timely identification of transmission pathways essential for targeted IPC interventions.4
Several studies have utilised electronic healthcare records (EHRs), whole-genome sequencing (WGS), or both to reconstruct within/between healthcare facilities using Bayesian, agent-based or network modelling frameworks.5-9 These approaches have shown that combining patient movement data with pathogen genomics can substantially improve outbreak detection and transmission inference.7,10 However, many existing methods rely on highly structured digital health records, dense metadata, specific spatial structure, or advanced informatics systems6,8, which remain unavailable in many LMIC settings, where clinical documentation remains fragmented or paper-based.11 As a result, tools that operate with minimal structured clinical data remain scarce.
To address this gap, we developed PathoPath, an open-source pathogen agnostic R package that requires minimal EHR data and integrated them with WGS results to build a comprehensive contact network to help visualise potential transmission pathways of pathogen and/or MGEs to inform IPC strategies. As a proof-of-concept, we applied PathoPath to retrospective genomic and clinical data from Klebsiella pneumoniae species complex (KpSC) at a large paediatric referral hospital in Dhaka, Bangladesh.
Methods
Integrative modelling of contacts using PathoPath
We developed a generalised methodological framework to model spatiotemporal contacts between movement trajectories, expanding the target subjects from patients to include other individuals, and inanimate objects. The framework takes as input unique subject and pathway identifiers, along with date-stamped locations for each subject and pathway. For each patient in hospital settings, each pathway can be defined as a series of locations (bed, room, ward, etc, at the same granularity level) where the patient has stayed from admission to discharge. We have implemented this framework as an R package pathopath and released it on GitHub (github.com/wanyuac/pathopath). Here, we describe pathopath’s algorithm and analysis of contacts between the 373 inpatients admitted to the BSHI in 2021.
Assuming subject s1 entered and left location l on days t1s and t1e (t1s ≤ t1e), respectively, and subject s2 entered and left the same location on days t2s and t2e (t2s ≤ t2e), respectively, then days of s1 and s2’s stay at this location were determined by L1 = t1e − t1s + 1 for s1 and L2 = t2e − t2s + 1 for s2. The maximum length of days when subjects s1 or s2 or both were continuously present at location / was calculated as Lmax = max (S1e, S2e) − min (S1s, S2s)+ 1. Then we calculated the time interval I12 = Lmax − (L1 + L2).A direct contact between s1 and s2 at location / was identified when I12 < 0, and this contact had a length of td = −I12. Given a custom parameter Δt that represented a number of days, an indirect contact was identified when 0 < I12 < ∆t, and this contact had a length of ti = ∆t − I12. The contact length was set to zero if neither direct nor indirect contact occurred between s1 and s2 at this location.
An undirected network was generated from all identified direct contacts. In this network, each node represented a patient, since each patient had a single movement pathway in our dataset, and each edge connected patients in at least one direct contact throughout their pathways, with the edge weight denoting the total length of direct contacts. Components were discovered in this contact network as subgraphs consisting of connected nodes that were not part of any larger connected subgraphs, and singletons were defined as individual nodes that did not connect to any other nodes. Bacterial species, genotypes, and SNP information were incorporated into the network as node attributes.
To track potential transmission, patients in the contact network were clustered using the same component-discovery algorithm but based on core-genome SNP distances between isolates recovered from these patients. Briefly, this method imports a matrix of pairwise core-genome SNP distances as an all-to-all undirected distance network, prunes edges with a configurable distance threshold, and identifies components of patients in the remaining distance network. Finally, such pathogen-based clusters were overlaid onto the contact network for comparison with the contact-based clusters. Readers are referred to Supplementary appendix 1, page 11 for details.
Study site and patient population
To assess the utility of PathoPath in a high-burden LMIC setting, this study was conducted at the BSHI which is the largest paediatric referral hospital in Dhaka, Bangladesh, with 37 unique wards having a total of 673-beds (Supplementary appendix 1, Table S1, page 5). We retrospectively retrieved EHRs and bacteriological stock cultures of all (n = 373) cases with culture-confirmed KpSC identified in 2021. While blood and/or cerebrospinal fluid (CSF) samples were routinely collected at BSHI for bacteriological culture, this study only included hospitalised children who met the WHO criteria for pneumonia, severe disease, or meningitis, assessed by trained physicians14. A detailed description of the bespoke EHR generation is described in Supplementary appendix 1, page 9.
Ethical approval
This work was conducted under the ethics approval No. BICH-ERC-1/2/2018 from the Ethical Review Board of the Bangladesh Shishu Hospital and Institute (BSHI), formerly Bangladesh Institute of Child Health.12,13 Written informed consent was obtained from each patient’s legal guardian prior to data collection and analyses.
Microbiological identification and whole-genome sequencing
The details of the microbiological identification are described in Supplementary appendix 1, page 9. Of the 373 culture-positive KpSC isolates identified in 2021, 314 isolates were from children aged <60 days. Of these, 311 (99.0%) isolates were successfully sequenced; three isolates from this age group could not be revived for sequencing. This age group was prioritised because it represents the population at highest risk for KpSC healthcare-associated infection in Bangladesh and accounted for the major burden of severe neonatal disease in this setting. Restricting sequencing to this high-risk subgroup also allowed more focused investigation of potential hospital transmission clusters. DNA extraction and 150-bp paired-end Illumina sequencing are described in Supplementary appendix 1, pages 9–10.
Data analysis
Genomic typing including KL, OC, and MLST assignment was done on all sequenced KpSC isolates (Supplementary appendix 1, page 10).15 Steps involved in reference-based SNP calling and genomic analysis are provided in Supplementary appendix 1, page 10. For comparative plasmid analysis, plasmid sequences were reconstructed from draft genome assemblies (Supplementary appendix 1, page 11) and carbapenemase-encoding plasmid clusters were identified through a protocol adapted from a previous study (Supplementary appendix 1, page 12).16
All statistical analyses and visualisations were performed in R v4.3.1. with packages dplyr, tidyr, stringr, lubridate, and readxl/openxlsx. Phylogenetic trees were visualised using iTOL v7 (itol.embl.de).
Results
Study population for evaluating PathoPath
We evaluated PathoPath retrospectively using available data on KpSC from 2021, the major multidrug resistant cause of neonatal HAIs in Bangladesh. In 2021, a total of 21,301 patients were admitted at Bangladesh Shishu Hospital & Institute with an average of 1,775 admissions per month. Among 2,688 blood/CSF samples collected, 373 (13.9%) were culture-positive for KpSC and were recruited in this study. Patient demographics and outcomes are listed in Table 1. Most infections occurred early in life, with 47.2% (n = 177) presenting from primary clinics, birthing centres, and households across 27 districts in Bangladesh within the first week after birth, and 57.9% (n = 216) were hospitalised between 0 and 2 days before infection was identified (Table 1).
Hospital-wide contact network
Records of complete ward-level movement (from admission to discharge/death) were available for 317/373 (85%) patients and were used as input for Pathopath to reconstruct a hospital-wide network of direct contacts. This network comprised 750 direct contacts and 74 components (25 connected components and 49 singletons) (Figure 1A). The largest component consisted of 93 cases, accounting for 29.34% of the 317 KpSC-positive cases.
KpSC infections were distributed across 21/37 (57%) hospital wards, but contact density was concentrated within neonatal critical care areas. The Neonatal Intensive Care Unit (NICU) had the highest burden of KpSC infections with approximately 77.9% of all contacts (584/750) involving 163 patients, followed by General Ward 3 (GW3: 10.3 % contacts from 47 patients), Special Care Baby Unit (SCBU: 25 cases and 5.3% of all contacts), and Paediatric Intensive Care Unit (PICU: 5.5% and 39 patients) (Supplementary appendix 1, Table S4, page 7).
Genomic characterisation of KpSC isolates
Of the 373 archived KpSC isolates, 311 (83.4%) from children <60 days of age underwent WGS. These comprised 264 (85%) K. pneumoniae isolates of 24 MLST clones and 47 (15%) K. quasipneumoniae subsp. similipneumoniae isolates of 7 MLST clones (Figure 1 and Supplementary appendix 2, Sheet S1). Four dominant STs accounted for 65.6% of sequenced isolates: ST16 (24.4%, n= 76), ST147 (18.7%, n = 58), ST1998 (12.2%, n = 38), and ST11 (10.3%, n = 32) (Supplementary appendix 2, Sheet S2).
Carbapenemase genes were detected in 298 (95.8%) isolates and were dominated by blaNDM (n = 292) and blaOXA (n = 99) genes followed by blaKPC (n = 36). Notably, 42.1% (n = 130) isolates carried two carbapenemase genes. Specifically, 97.4% (74/76) of ST16 isolates harboured both blaNDM-5 and blaOXA-181, and only 2.6% (2/76) carried a single blaOXA-181. Approximately 98.3% (57/58) ST147 isolates carried blaNDM-5 and only one harboured the blaNDM-1 gene. For ST1998, the blaKPC-2 and blaNDM-1 combination was most frequent (86.8%; 33/38), followed by single blaKPC-2 (7.9%; 3/38), and blaNDM-1 (5.3%; 2/38). Carbapenemase genes in ST11 were mostly comprised of blaNDM-7 (19/32; 59·4%) and blaNDM-5 (13/32; 40·6%). Phenotypic resistance to Imipenem and meropenem were 98% for ST16, 100% for ST147 and 93.65% for ST1998 (Table 2). All the carbapenemase genes were harboured in plasmids, with IncFII (89.5% of ST16 and 93.1% of ST147), IncFIA(HI1)/IncFII (73.7% of ST1998), and IncFIB(K)/IncX3 (50% of ST11) being the most abundant vectors in our KpSC isolates (Supplementary appendix 2, Sheet S3).
Integration of genomic data into the contact network to infer transmission
Of the 317/373 cases with complete movement history, 263/317 were children <60 days age from whom sequencing data was available. The remaining 54 patients were either ≥60 days old or isolates could not be revived for sequencing. Once genotypical information was overlaid onto the network, >60% of cases on average within each of the three largest components (C1-C3) were dominated by a single ST (Figure 1A), K and O serotype (Supplementary appendix 1, Figure S3, page 4; Table S5, page 8). These major network clusters when combined with temporal trajectory of blood sample collection revealed concentrated seasonal waves of ST147 (C1), ST101 (C2) and ST16 (C3) mainly in NICU and GW3, the wards having the highest exchange of patients in between. These network patterns therefore prioritised ST147 and ST16, the largest two clusters (Figures 1A, 1B) for detailed analysis of clonal transmission and plasmid dissemination.
Identifying transmission clusters in ST147
All 58 ST147 KpSC cases at BSHI were detected between 18 March and 6 August 2021 (4.63 months), with an average of 2.88 cases per week (Figure 1B). Of these, 49/58 (84.5%) had complete ward-level movement records and therefore were used for the ST147-species network construction. PathoPath identified a total of 11 individual ST147 components linked by 110 edges (shared wards) and an average of 6.87 (± 5.82) pairwise contact days. Seven of the 11 components were singletons, with the largest network component having 29 patients connected (Figure 2A).
To identify DN clusters of potential transmission, genomic relatedness was assessed across a range of single nucleotide polymorphism (SNP) thresholds. Using a range of SNP distance cutoffs (0 to 100), between 3 and 21 distinct SNP clusters of ST147 were identified with an average of 5 partitions across all cutoffs. Overall number of partitions/clusters stabilised largely at >2 SNPs with no significant collapse for up to ≤100 SNP threshold tested (Supplementary appendix 2, Sheet S4). For further investigating potential transmission, we used ≤2 SNP as a threshold and found six distinct DN clusters. Notably, within Cluster 1, the largest of these six, 44/49 cases in the ST147 PathoPath network grouped together in just one compartment (DN Cluster 1), making this potential transmission lasting for about 141 days with an average number of 2.18 cases/week (Figure 2A).
Clinical timelines suggested mixed transmission timing within this dominant Cluster 1: 18 of 44 patients had blood cultures collected after at least 96 h of hospitalisation, whereas 16 had cultures drawn within 48 h of admission. Neonatal Intensive Care Unit (NICU), and General Ward 3 (GW3) were the only two wards where Cluster 1 infections were identified (NICU: 84 contacts involving 29 patients; GW3: 17 contacts involving 13 cases) (Figure 2). To account for environmental persistence and bridge any potential temporal gaps in transmission, a 28-day indirect contact window (Δt = 28) was applied. This created 255 additional contacts and included two singleton pairs, separated by 25 (Cluster 1) and 16 (Cluster 2) days. For both clusters, blood samples from the later patients were obtained on the day of admission.
Transmission of carbapenem resistance within ST16
Other than being the most abundant ST type (24.4.7%; 76/311) in our retrospective dataset, infections with ST16 at BSHI were continuously detected till the end of this study (from 6 April 2021 to 5 January 2022), with an average of 2.83 cases per week (Figure 1B). From this potentially ongoing transmission since 2021, 68/76 (89.5%) cases had complete movement history and thus were usable for constructing an ST16-specific network. Once built, a total of 12 components were identified with an overall 163 edges and approximately 3.52 (± 2.51) shared days on average between patients. There were nine singletons in the network plus three components with the largest SNP cluster of ST16 comprised of isolates from 47 cases (Figure 3A).
In total, 1 to 33 clusters of ST16 isolates were identified when varying the threshold of SNP distances from 0 to 100. The number and composition of clusters stabilised to less than 10 clusters when the distance threshold exceeded 2 SNPs (Supplementary appendix 2, Sheet S4). Therefore, the same ≤2 SNP threshold was used for clustering ST16 isolates as for the ST147, resulting in 10 clusters suggesting transmission of K. pneumoniae via direct contacts. Cluster 1, the largest continuous cluster had 69.1% (47/68) of all ST16 cases connected, spanning for 144 days with 2.92 cases/week until the end of this study.
Phenotypic resistance to carbapenems was nearly universal among ST16 as 97.4% (97.4% for blaOXA-181 and 97.3% for blaNDM-5) isolates were non-susceptible to imipenem and 98.7% (98.7% for blaOXA-181 and 98.6% for blaNDM-5) were resistant to meropenem (Table 2). Combining with the plasmid clusters, the contact network revealed potential inter-clonal transfer of carbapenemase-encoding plasmids (Figure 3). For example, an IncFII plasmid carrying blaNDM-5 (plasmid cluster 20) was observed to potentially move from a donor in the SNP Cluster 10 (patient 27) towards possibly a few SNP cluster 1 recipients (patients 31, 35, and 37) within the span of two weeks despite no recorded sharing of wards between these four patients, providing some insights gained to generate future research into environment or unrecognised transmission vectors. Similarly, an IncFII (cluster 5) plasmid carrying blaNDM-5 and blaOXA-181 genes potentially transferred from a group of SNP cluster 1 donors (patients 39, 40, 41, and 42) to a SNP cluster 9 recipient (patient 43) within the span of a week (Figure 3), suggesting horizontal dissemination of resistance determinants alongside clonal spread.
Discussion
This study developed and tested PathoPath, an open-access, pathogen-agnostic tool that constructs hospital-wide contact network using only date-stamped patient movement trajectories and integrates genomic data to infer potential pathogen and/or AMR pathways. Using retrospective neonatal KpSC surveillance data from BSHI, we showed that PathoPath can identify dominant transmission clusters, precise pathways, and probable dissemination of carbapenemase-encoding plasmids within a high-burden tertiary hospital setting. Further ST-specific investigation revealed the value of this integrative framework as PathoPath delineated one transmission cluster of ST147, lasting for about 141 days in Neonatal Intensive Care Unit and General Ward 3. It also helped identifying transmission of IncFII carbapenemase-encoding plasmids (carrying blaNDM-5 and/or blaOXA-181 genes) between multiple SNP clusters of KpSC ST16 isolates. Together, these findings show how clonal expansion and plasmid-mediated resistance dissemination can coexist within the same healthcare network and remain difficult to detect without integrated digital surveillance.
PathoPath is designed to function with minimal structured clinical input. Because EHR in different healthcare facilities are often designed differently, including various commercial (e.g., Oracle Health EHR) or in-house data collection and management tools. Therefore, to standardise input structure of the patient movement trajectories a simplified table requiring subject identifier, pathway identifier, location, start time, and end time have been utilised in PathoPath. To minimise data-entry related errors in pathway reconstruction, the current version also includes a few logic checks, for example it flags patient overlaps (e.g., two patients occupying the same bed for multiple days) that are implausible. By demonstrating what can already be inferred from sparse routine data, PathoPath can in future help identify which additional metadata would be most informative to collect in future implementations. For example, interpretation of putative transmission events will always benefit from contextual information on time, place, and mechanism of exposure. In this sense, PathoPath can guide data-prioritisation efforts by showing where enhanced metadata collection, such as more granular timestamps, shared equipment records, healthcare worker movements, environmental sampling, or laboratory measures of pathogen persistence, would most improve transmission inference.
This is particularly relevant for genomic interpretation. While genomic clustering can be dynamically integrated using MLST, cgMLST, or LIN codes, SNP-based methods remain the most widely used approach for investigating hospital outbreaks because they offer the highest resolution for distinguishing closely related isolates.15,17,18 However, there is no universally appropriate SNP threshold for defining nosocomial transmission. A few important considerations to define nosocomial transmission of a single clone using SNPs is the geographical, temporal (typically 48-72 hours)19 and SNP distance (2-10 SNPs) thresholds, as arbitrary cutoffs may detrimentally affect the sensitivity and specificity of the inference.
Meaningful interpretation of contacts in depends on the interaction between genomic distance, temporal proximity, healthcare setting, and pathogen biology. For example, a recent multi-country study on Klebsiella pneumoniae systematically tested a varying range of temporal plus genetic distance cutoffs to identify the most sensitive and context-appropriate threshold (≤10 SNPs and ≤4 weeks).17 However, in our study, 26/76 (34.2%) ST16 isolates coming from 13 geographically distant districts were 0 SNPs apart from each other with a median admission to blood culture duration of 24 hours (mean = 3, min = 0 and max = 15 days). Although this finding should be interpreted cautiously and will require validation in future studies, it raises several possibilities. These include a shorter incubation period for ST16, particularly in neonates with underlying comorbidities, a slower rate of mutation accumulation than previously appreciated, which could suggest that these infections are part of a broader transmission network across Bangladesh, or transmission through contaminated hospital supplies, ambulances, or transport cots shared across healthcare facilities.20 For these reasons, we did not impose an absolute definition of HAI in this study and instead used ≤2 SNPs to make DN clusters as a pragmatic placeholder to demonstrate how PathoPath can support exploratory transmission mapping in both ST147 and ST16.
A similar principle applies to indirect contacts. PathoPath already allows users to model delayed or indirect interactions through a configurable Δt parameter, but in the present analysis this value was set to zero because the environmental persistence of KpSC under local hospital conditions remains uncertain. Experimental studies on environmental survival, ward-specific contamination, and clone-specific persistence could provide the empirical basis needed to calibrate indirect-contact windows more accurately.21 Likewise, prospective environmental sampling of high-contact surfaces, equipment, fluids, or shared clinical materials could clarify whether indirect transmission plays a substantial role in sustaining KpSC circulation in this setting.22,23 Such work would directly strengthen future implementations of PathoPath by enabling biologically informed parameterisation of contact networks.
While PathoPath is a scalable and versatile tool, it is important to mention a few limitations. First, the current network model is undirected and does not automatically infer transmission directionality. It is possible to utilise pathogen genome data and temporally infer directionality, our one-year sampling period did not support reliable evolutionary rate estimation. Second, although indirect contacts (Δt) can presently be modelled with PathoPath, uncertainty around environmental persistence prevented us from using non-zero Δt values in the retrospective BSHI analysis. Third, at BSHI, unique patient identifiers are not retained after discharge, which meant we could not link repeat admissions for the same individual across the study period. PathoPath can incorporate multiple pathways per patient, but this was not possible with the available dataset.
Despite these constraints, PathoPath offers a flexible digital framework where potential users can choose indirect interaction cutoffs based on their research context. Importantly, as open-source software, Pathopath enables local adaptation, providing users with method transparency and full control of code. While by integrating WGS information, PathoPath network can aid outbreak investigation, continuous genomic surveillance remains difficult to sustain. Therefore, future extensions of PathoPath may include the use of routinely available laboratory results in LMIC facilities (e.g., antibiotic zone diameters, minimum inhibitory concentration, and/or biochemical profiles), combined with machine learning tools to generate WGS-level inference and guide IPC practices in a more cost-effective and sustainable manner.
Supporting information
Data Availability
All raw paired-end Illumina sequencing reads (fastq.gz) analysed and presented in this study are available on NCBI under the study accession "ERP173558" (Supplementary appendix 2, Sheet S1). The PathoPath R package developed and used for modelling is openly accessible at: https://github.com/wanyuac/pathopath (Version 1.0.0; GPL-3.0 licence).
Data sharing
All raw paired-end Illumina sequencing reads (fastq.gz) analysed and presented in this study are available on NCBI under the study accession “ERP173558” (Supplementary appendix 2, Sheet S1).
The PathoPath R package developed and used for modelling is openly accessible at: https://github.com/wanyuac/pathopath (Version 1.0.0; GPL-3.0 licence).
Declaration of interests
We declare no competing interests.
Acknowledgments
The Authors acknowledge use of the High-Performance Computing platform provided by Liverpool Shared Research Facilities, Faculty of Health and Life Sciences, University of Liverpool.
Funding
This work was funded by the NIHR Global Health Research Development Award (YH; Grant number: NIHR208273), the Gates Foundation (SS; Grant number: INV073135), and Wellcome Trust Centres for Antimicrobial Optimisation Network (CAMO-Net) Research Fellowship (MSIS; Grant number: 226691/Z/22/Z). YW is a Research Fellow funded by the David Price Evans Endowment to the University of Liverpool (Grant number: UGG0057). AHH is the David Price Evans Chair in Global Health and Infectious Diseases at the University of Liverpool (Grant number: UGG0057). AHH and YW were supported by the CAMO-NET UK grant (Reference: 226691/Z/22/Z).
Supplementary materials
Supplementary appendix 1
Supplementary appendix 2