Genomic inference of sites of transmission during regional spread of blaNDM ST219 Klebsiella pneumoniae in Michigan
1Departments of Microbiology & Immunology, University of Michigan Medical School, Ann Arbor, MI, USA
2Department of Computation Medicine and Bioinformatics, University of Michigan Medical School, Ann Arbor, MI, USA
3Internal Medicine Division of Infectious Diseases, University of Michigan Medical School, Ann Arbor, MI, USA
4Michigan Department of Health and Human Services Bureau of Infectious Disease Prevention, Lansing, MI, USA
5Michigan Department of Health and Human Services Bureau of Laboratories, Lansing, MI, USA
*Corresponding author: Evan Snitkin, PhD, 1520D MSRB I, 1150 W. Medical Center Dr. Ann Arbor, MI, 48109-5680, Tel- (734) 647-6472, Fax-734-615-5534, Email- esnitkin@umich.eduAbstract
Carbapenem-resistant Klebsiella pneumoniae (CRKP) is recognized as an urgent public health threat due to its multidrug resistance and ability to spread rapidly in healthcare facilities. The frequent movement of colonized and infected patients between different facilities makes it challenging to discern where individual patients acquire CRKP, and in turn, identify facilities making the greatest contributions to regional spread. While high-intensity active surveillance combined with whole-genome sequencing has previously shown success in identifying sites of CRKP transmission, high costs and logistical barriers to regional coordination make this level of surveillance infeasible in most settings. Here, we developed an approach using genomic and healthcare exposure histories from passively collected regional isolates to identify the putative facility source for each new isolate by analyzing shared healthcare exposures with earlier case patients whose isolates shared a most recent common ancestor. As a proof of principle, we applied this approach to data collected by a state health department during a suspected regional CRKP outbreak involving 72 patients exposed to 47 healthcare facilities in Michigan from October 2019 to May 2022. Integration of genomic and healthcare exposure data enabled inference of a single putative source facility for 66/70 of non-index cases, with 35 and 31 cases attributed to a single facility with intra- and inter-facility transmission, respectively. Examination of transmission linkages over time supported a sustained role played by a single focal facility, with several other facilities inferred to be sources of a smaller number of cases. Importantly, in several instances, facilities were implicated as sites of transmission prior to cases being detected there based on overlapping exposures among genomically linked cases, with four facilities being inferred as sites of transmission despite no cases ever being reported. The ability to infer facility sources of transmission using passively collected regional isolates, in some cases ahead of case identification, supports the potential for real-time genome-informed surveillance to enable timely targeting of infection prevention interventions to interrupt regional spread of CRKP and other healthcare-associated pathogens.
Article notes
Competing Interest Statement
The authors have declared no competing interest.
Funding Statement
Funding was provided as part of the Michigan Sequencing and Academic Partnerships for Public Health InnovaHon and Response (MI-SAPPIRE) iniHaHve at the Michigan Department of Health and Human Services (MDHHS) which is supported with funds from the Centers for Disease Control and PrevenHon through the Epidemiology and Laboratory Capacity for Prevention and Control of Emerging InfecHous Disease Enhancing DetecHon Expansion (6NU50CK000510-02-07).
Introduction
Carbapenem-resistant Enterobacterales (CRE) has been recognized as a significant threat to vulnerable patients in healthcare settings, with mortality rates upwards of 50% for patients with CRE infections 1–6. Among CREs, New Delhi metallo-β-lactamase (NDM) producing Klebsiella pneumoniae outbreaks have been increasingly reported across regional healthcare networks in the U.S.3,7–9, with limited treatment options for NDM patients10. To prevent NDM-producing strains from reaching levels of endemicity, it is critical to identify and implement more effective strategies to prevent the spread of NDM in healthcare settings.
In addition to many CRE outbreaks that have been investigated at a single facility level 11–15, more recent studies have also tracked the spread of CRE across regional healthcare networks 16–20. Regional spread presents challenges for individual facilities, as it can be unclear whether a patient newly diagnosed with a CRE infection acquired it during their current hospitalization, or from a facility to which they were previously exposed due to the movement of CRE colonized patients 21,22. Knowledge of where CRE transmission is occurring across regional networks is critical to both early intervention for emerging threats like NDM and decreasing rates for endemic organisms 17,23,24.
To discern potential facilities of CRE transmission across a region, patients’ histories of healthcare exposure leading up to their clinical CRE culture can be considered 16–19,25. In particular, putative sites of acquisition for a given case can be identified by inspecting overlapping exposure with prior cases. However, exposure data alone can be imprecise as circulating cases increase, with frequent overlaps among patients not linked by transmission, and further complicated by unobserved asymptomatic carriers, yielding transmission networks with missing links.
Whole-genome sequencing (WGS) allows for high-resolution delineation of linkage between isolates and has been increasingly applied in regional outbreak investigations 16–19. Genomic analysis can mitigate issues with incomplete case capture, as indirect transmission within and between facilities can be inferred using both distance-based or phylogenetic methods. However, identifying transmission from genomic data also has its challenges. Past regional genomic surveillance reports have relied on comprehensive epidemiologic data with high-intensity active surveillance, which is costly and logistically infeasible at regional levels in most settings26–29.
Additionally, the current gold standard is to define genetic linkages using genetic distance thresholds that are imprecise, leading to false negative linkages due to intra-patient variation accumulated during prolonged colonization and false positive linkages due to unsampled intermediates 19,28,30–35.
In this study, we sought to develop and apply a novel approach integrating WGS and exposure data to identify sites of transmission during a regional outbreak. By leveraging healthcare exposure data and a genetic distance threshold-free approach12, within the constraints of a largely passive sample collection with ad hoc surveillance conducted at only a few facilities, we sought to identify putative sources of transmission during regional outbreak of blaNDM-1-carrying K. pneumoniae ST219 in Michigan from 2019 to 2022. Taking a facility-centric approach and attempting to discern the origin of each case, we found support for the ability to use sample and data collections commonly available to regional public health labs to identify sites of transmission throughout the outbreak.
Methods
Sample collection
Our study was a retrospective analysis of cases of carbapenemase-producing NDM-carrying K. pneumoniae in Michigan among hospitalized patients. Cases were reported to the Michigan state public health laboratory and characterized. Specimens were confirmed using Bruker rapifleX® MALDI-TOF, carbapenemase was tested using modified carbapenem inactivation method (mCIM) and carbapenemase genes were identified using the Cepheid Xpert® Carba-R test.
Antimicrobial susceptibility was tested by Sensititre™ GNX2F or GN7F panels. Isolates were subjected to short-read whole genome sequencing using the Illumina MiSeq (See Supplementary Table 3 for BioSample identifiers and meta-data). Healthcare exposures in the 90 days preceding first case detection of a given patient was also collected using the state electronic surveillance system.
Genomic data processing
To identify previously sequenced strains that were closely related to outbreak isolates, annotated public genome assemblies listed as “Klebsiella pneumoniae” were downloaded from the PATRIC database 36, samples were annotated by RASTtk 37, and a core genome distance matrix was generated using the R package cognac 38. We then selected public genomes with core genome distances within 200 single nucleotide variants (SNVs) of at least one of the outbreak genomes. Antibiotic resistant genes in outbreak and public genomes were identified using AMRFinderPlus v3.11 39.
For tracking the spread of outbreak strains, we performed variant calling using a custom reference-based pipeline (https://github.com/Snitkin-Lab-Umich/snpkit). The quality of sequencing reads was assessed using FastQC v0.11.0, and adapter sequences and low-quality bases removed using Trimmomatic v0.39. Single nucleotide variants (SNVs) were identified by first using Burrows-Wheeler short-read aligner (bwa v0.7.17) to map trimmed reads to the ST219 reference genome (SAMN26729713), then discarding polymerase chain reaction (PCR) duplicates with Picard v3.0.0, and calling variants with SAMtools and bcftools v1.9. Variants were filtered from using VariantFiltration from GATK v4.5.0.0 (QUAL > 100; MQ>50;>=10 reads supporting variant; and FQ< 0.025). We performed GATK HaplotypeCaller for indel calling only including those with root mean square quality (MQ) > 50.0, GATK QualbyDepth (QD) > 2.0, read depth (DP) > 9.0, and allele frequency (AF) > 0.9. We also excluded variants that were less than 5 base pairs in the proximity to indels, in recombinant regions identified by Gubbins v3.0.0, in a phage region identified by Phaster web tool, or resided in tandem repeats of length greater than 20 bp as determined using the exact-tandem program in MuMmer v3.23 using a custom Python script. The whole-genome masked variant alignment generated by the variant calling pipeline was then used to reconstruct a maximum likelihood phylogeny with IQ-TREE v2.0.3 using the general time reversible model GTR+G and ultrafasta bootstrap with 1000 replicates (-bb 1000). To estimate common ancestors of collected isolates, a dated phylogeny was generated using Bactdating “arc” model was chosen due to the lowest DIC and smallest root date confidence interval (Table S1) 40
Evaluation of putative transmission linkages
Several strategies were used to evaluate the veracity of putative transmission events. First, we evaluated whether using MSVs was linking patients with shared healthcare exposure history more than expected by random. To this end we performed a permutation analysis wherein each patient’s association with their isolate was maintained, but their history of healthcare exposures was randomly swapped with another patient. We compared the proportion of MSV linked pairs that had no shared healthcare exposure of our data with the 1000 permutated events using Z-test. Second, we evaluated the patient-sharing of facilities with predicted inter-facility transmissions between them. To accomplish this, we used inpatient billings from Center of Medicare and Medicaid services (CMS) and inferred patient transfer frequencies between facilities based on sequential billings. Using the raw patient transfer matrix for CMS patients, we calculated the patient flow between pairs of facilities using regentrans 41. For each case patient, we compared the patient flow between the case facility and predicted source facilities, to (i) patient flow between case facility and a facility where case was previously exposed to but not predicted as a source facility, and (ii) the flow between the case facility and other facilities that were shared patients in the CMS data, but were not predicted as linked to the case facility. We used Wilcoxin-signed rank test to compare the patient flows. Lastly, we evaluated the importance of each facility at time points when cases were collected using eigencentrality and pagerank algorithms (Figure S4).
Statistical analysis
All statistical analysis were done in R. Network algorithms were implemented using igraph42 in R. Figures were visualized in R using ggplot2 43, ggtree 44 and ggraph (https://github.com/thomasp85/ggraph).
Results
Regional spread of NDM-harboring K. pneumoniae ST219 was attributed to clonal expansion
From October 2019 to April 2022, 104 Klebsiella pneumoniae ST219 isolates harboring the blaNDM-1 gene were collected from 72 individuals across regional healthcare facilities in Michigan. For the first six months, cases were only recovered at a single acute care hospital (ACH10). During the next phase of the outbreak, cases were reported by 21 additional healthcare facilities, with the peak occurring from June 2020 to June 2021 (Figure 1a). Isolates were compared using whole-genome sequencing to understand whether the shift from a putative single facility outbreak to a broader regional spread was due to a single regional introduction or multiple introductions of NDM-carrying K. pneumoniae. Phylogenetic reconstruction of Michigan isolates, along with publicly available K. pneumoniae ST219 genomes showed Michigan isolates separate from public isolates in a distinct monophyletic cluster, consistent with a single introduction into the region (Figure S2). Further support for all cases deriving from a single recent introduction came from the small genetic distances among Michigan isolates, with a median of 8 pairwise SNVs (36 maximum, 0 minimum) (Figure S3). Of note, public K. pneumoniae ST219 genomes from diverse geographical locations showed variable association with blaNDM-1, indicating that blaNDM-1 gene had been acquired multiple times in this genetic background (Figure S2).
To further understand the temporal dynamics of clonal spread, we created a time-scaled phylogeny of Michigan isolates (Figure 1b). The time-scaled phylogenetic tree of Michigan isolates was also consistent with clonal spread, with the common ancestor of all but one isolate estimated to be in early 2019 (Confidence interval (CI): 2018.271, 2019.495). Although detected in 2020, the one outlier isolate had a predicted common ancestor with the other outbreak isolates that dated back to 2017 (CI: 2015.15, 2018.81). We noted that this outlier isolate was collected at a facility where cases from the dominant outbreak lineage were detected contemporaneously, consistent with independent acquisition of the blaNDM-1-encoding plasmid by a closely related K. pneumoniae ST219 strain. Combined with the observation that the most closely related public isolate to the Michigan outbreak lacked blaNDM-1, this supports the outbreak clone stemming from the acquisition of blaNDM-1 in a locally circulating strain of ST219.
Patients with genetically related isolates showed extensive overlap at regional healthcare facilities
To understand the pathways by which the major outbreak lineage spread within and between healthcare facilities, we overlaid a whole-genome maximum-likelihood phylogeny with patients’ healthcare transfer histories (Figures 2a and 2b). Visual inspection of Figure 2 showed numerous instances of temporal overlap in healthcare exposures among patients whose isolates clustered on the whole-genome phylogeny. These overlaps supported cases of intra-facility transmission (i.e. phylogenetically clustered cases detected at the same facility), direct inter-facility transmission (i.e. phylogenetically clustered cases with common exposure to one of the patient’s collection facilities), and indirect inter-facility transmission (i.e. phylogenetically clustered cases with common exposure to a third facility, distinct from the facilities where patient’s isolates were collected). Moreover, overlaying the phylogeny with healthcare exposures also revealed instances where cases detected contemporaneously at the same facility were clearly due to independent importation events. For example, in late 2020, strains collected concurrently through active surveillance at ACH10, were from two distinct genomic clusters sharing a more distant common ancestor.
Integrating genomic and healthcare exposure data revealed regional transmission pathways
Motivated by the overlapping healthcare exposures among phylogenetic neighbors, we sought to systematically extract intra-facility and inter-facility transmission events by considering both genomic and healthcare exposure data. To this end, we simulated the tracking of the NDM-carrying K. pneumoniae ST219 regional spread as a real-time event, by inferring transmission events based on linkages between “new cases” and cases identified prior. The putative source(s) for each case were inferred to be previously collected isolates with which their isolate had maximum shared variants (MSV) (Figure S1a). In contrast to the common approach of employing an SNV threshold to detect transmission linkages, MSV pair detection did not rely on a single threshold, rather identifying circulating strains with which they share a most recent common ancestor. Among pairs with maximum shared variants there was a median of 3 cgSNVs (IQR:1,5), while also permissive to larger variance (maximum: 22 cgSNVs, Figure 3a).
Using this simulated real-time approach, we identified 206 candidate source-to-case links among the 70 isolates included in the analysis, where cases could be linked to more than one source if they had MSV with more than one previously detected isolate. To determine pathways of intra- and inter-facility transmission, patterns of healthcare overlap were analyzed among MSV pairs. Our objective was to identify the most plausible facility where each case acquired their isolate, therefore, we assigned a single high-confidence source facility to each case when more than 50% of links to a case supported the same source facility (Figure S1b). Using this approach, 50% (35/70) of patients were predicted to acquire their isolate via intra-facility transmission events at 9 different facilities. Among the 35 putative intra-facility transmissions, 6 patients also had potential inter-facility transmission with shared exposure with another healthcare facility. The remaining 35 cases were attributed to inter-facility transmissions, with 60.0% (21/35) supported by both genomic and shared healthcare exposure, 28.6% (10/35) supported only by genomic linkages and 11.4% (4/35) of inter-facility cases assigned to more than one facility meeting the high-confidence source facility criteria (Figure 2c). These results show that by combining MSVs with healthcare exposure data we were able to effectively narrow down the putative source facilities, with 94.3% (66/70) of patients having a single predicted source facility for their isolate (Figure 3b).
Predicted transmissions are enriched in shared healthcare exposures and link facilities that share more patients than unlinked facilities
To evaluate the transmissions predicted using our approach, we conducted two additional analyses. First, we performed permutation analysis to evaluate whether healthcare-sharing among genomically linked pairs occurred more than expected. We created 1000 randomized data sets, where patients were assigned another patient’s history of healthcare exposures. In none of these permuted data sets were there more instances of shared healthcare exposure among MSV pairs than in the actual data (p < 0.001). This result supports the MSV approach identifying patients with shared exposures far more than expected by chance (Figure S6).
Next, we assessed support specifically for predicted inter-facility transmissions. To provide context to our predictions, we leveraged statewide patient transfer networks inferred by sequential billing events extracted from Centers for Medicare and Medicaid databases (CMS), which provides a quantitative summary of how patients move between regional healthcare facilities. We first focused on predicted transmissions with both genomic and epidemiologic support. For this set of predictions, we observed that source and case facilities linked by transmission had significantly higher patient flow with each other than did other facilities connected to the case facility in the CMS network (Figure 3c, p = 2.54 ξ 10−9), indicating that transmissions are predicted to preferentially occur between facilities that share the most patients. However, putative source facilities in linkages supported by genomic and epidemiologic data were did not have significantly higher patient flow to the case facility than the other facilities that the case patients were exposed to (Figure 3c, p = 0.89). This observation indicates that the genomic data can help differentiate among alternative epidemiologically plausible scenarios by helping identify the most likely source facility among multiple past exposures for a given case patient (Figure 3c, Supplemental Table 2).
Next, we examined predicted inter-facility transmissions supported only by genomic data. As with predictions with epidemiologic support, facilities with genomic-only transmissions had significantly higher patient flow than other facilities linked to the case facility in the CMS network (Figure 3c, p = 0.02). However, source and case facilities linked by transmission with only genomic support had significantly lower patient flow to the case facility than the other facilities case patients in these linkages were exposed to (Figure 3c, p =0.0001). Thus, genomic only prediction are linking facilities with an intermediate level of patient sharing. To further assess the plausibility of these linkages, we examined the genomic support more closely. Overall, the putative transmission pairs with only genomic support were highly clonal (IQR [1.5,5], min = 0, max = 12). Moreover, for four cases with healthcare overlap with other putative source patients without MSVs that were collected prior to the cases, the pairwise SNVs were significantly larger (IQR = (5.75,10.25), min =4, max = 18) compared to the genomically linked putative source (IQR = (0.75,3.25), min = 0, max =4). In addition, two of the ten patients with only genomic evidence were linked to a single source facility by more than one source patient, further supporting connectivity between the facilities despite the lack of direct healthcare exposure.
Regional genomic analysis enables identification of critical facilities over time
Having developed and validated an approach to track intra- and inter-facility transmission, we next examined patterns of transmission over the course of the outbreak at both patient- (Figure 4a) and healthcare facility-level (Figure 4b). Overall, we identified 64 independent inter-facility transmissions. ACH10, where the outbreak was first detected, was predicted to be a persistent site of both intra- and inter-facility transmission throughout the study period. With respect to inter-facility transmission, ACH10 was predicted to be a source for 26.6% (17/64) of inter-facility transmission events and the destination for 15.6% (10/64) (Figure 4a). Aside from links involving ACH10, most other inter-facility links were only observed once, with at most a small number of subsequent intra-facility transmissions detected after the introduction. Moreover, when considering exportations from facilities in aggregate and using the eigencentrality and pagerank algorithms 42 to calculate facility importance, ACH10 was consistently the most important facility throughout the outbreak (Figure 5 and Figure S5).
While analyses of the transmission network indicated that ACH10 was an important player throughout the outbreak, we were interested to see if there were more contributions from other facilities over the course of the outbreak. To this end, we examined the number of cases sourced to each facility over time. By comparing facility-level source curves to facility-level case curves, we were able to discern predicted intra-facility transmissions, importations and exportations from each facility over time (Figure 5). For example, by examining ACH10’s source and case curve, we were able to see the transition from initial predominance of intra-facility transmissions (source curve and case curve rising in sync) to exportation (source curve increases, while case curve remains flat). For most other facilities we see the case curve initially rises, indicating importation, followed by increases in the source and case curve during subsequent periods of intra-facility transmission and/or exportation (e.g. facilities ACH2, ACH5 and ACH11).
However, we noticed some facilities with unexpected patterns, whereby they were inferred as sources of transmission prior to any cases being detected. For SNF13, we observed the source curve rising before any cases were detected, with its role as a source being inferred based upon shared exposure of MSV pairs from other facilities. In total, there were three cases with genomic and epidemiologic support for SNF13 being a source, with two of these cases occurring before the first case was identified in SNF13. It is worth noting that while no cases were initially identified at SNF13, patients from ACH10 were often transferred from SNF13, supporting its potential role as an undetected reservoir early in the outbreak. In addition to SNF13, there were four other facilities with no cases ever reported, but for which shared exposures implicated them as a potential source of regional transmission. Support for cases being sourced to these facilities which never reported cases comes from many of these putative transmission pairs having extremely small genetic distances, consistent with recent shared exposures mediating transmission. Case NDM165 attributed to source facility ACH1 had 0 SNVs with its predicted source isolate. Case NDM130 and case NDM150 both had a putative source with 3 SNVs sharing healthcare exposure at ACH13. Case NDM89, NDM90, and NDM94 had 0 SNVs, 3 SNVs and 0 SNVs with shared health exposure at SNF11 respectively. Lastly, case NDM132 and case NDM156 were sourced at SNF4, with 6 SNVs and 20 SNVs with their respective putative source.
Discussion
Healthcare-associated infections caused by antibiotic resistant organisms are a major public health threat, associated with significant morbidity, mortality and economic impacts45,46. Most efforts to date to track and prevent the spread of infections have focused on individual healthcare settings. However, it is increasingly appreciated that a regional perspective is critical to identify the sites of transmission and pathways of inter-facility spread that initiate and sustain regional endemicity 26–29. While regional active surveillance combined with genomic analysis has been shown to enable regional transmission tracking in low prevalence regions, this strategy is costly and typically infeasible to implement at the state or province level. Here, we implemented and evaluated a regional approach using genomic and healthcare exposure data to identify putative source facilities where patients may have acquired their infection, without relying on an established active surveillance program. Applying this approach to a blaNDM-1 carrying K. pneumoniae ST219 regional outbreak in Michigan, we identified putative facility sources of patient’s isolates with high specificity, identified key facilities seeding cases in other facilities and showed the potential to identify facilities as sources of infection prior to their detection of cases. These findings support the potential for regional genomic surveillance programs using passive isolate collections and ad hoc surveillance to guide regional infection prevention action to target facilities driving regional spread.
Mapping patient transfer history on the whole-genome phylogeny of patients’ acquired strains showed that phylogenetically clustered patients tend to share healthcare exposure history. This observation, along with the success of prior studies using phylogeographic methods to track regional healthcare transmission 18,28,47–50, prompted us to use a maximal shared variant approach to infer the facility source of each patient’s isolate. The MSV approach leverages common ancestry but does not require phylogenetic reconstruction when a new isolate is sequenced, will scale to large numbers of strains, and is easily implementable and interpretable. Compared to the standard approach of imposing a genetic distance cutoff, MSV genomic links yielded isolate pairs with SNV distances comparable to those reported in prior K. pneumoniae transmission studies 19,28,30–35,51, while also allowing for larger genetic distances that may arise due to prolonged patient colonization and hypermutators12. Here, we showed that the MSV approach yielded links among patients that were significantly enriched in shared healthcare exposures, identified connections among facilities that are highly connected by patient transfer, and provided sufficient resolution to track the spread of a regional CRE outbreak.
Upon identification of MSV pairs, we implemented an approach to incorporate facility exposure data with an eye towards real-time surveillance, that focused on elucidating the contribution of individual facilities to regional spread. To simulate a real-time response, for each new case we only considered earlier cases when identifying putative source facilities. While imperfect, as a true source may only become apparent later due to the lack of active surveillance, only retrospective case data would be available in practice. To increase robustness to missing case capture caused by delayed case reporting and limitations in surveillance, we did not require temporal overlap in hospitalizations of genomically linked patients, but rather just that there was shared exposure prior to a case patient’s isolate collection. In this way, we could capture intra- or inter-facility transmission that may be mediated by one or more unsampled intermediate patients at the facility of interest. The viability of this approach was demonstrated by both the non-random concordance between genomic linkages and exposure data, as well as the specificity of predictions, with 94% (66/70) of patients being linked to one putative source facility. We note that to achieve this specificity, when greater than 50% of the shared exposures for a case patient with other patients linked by MSVs were from a single facility, this single facility was inferred to be the source. This specificity filter simplified the network and prioritized scenarios most supported by genomic and exposure data, aligning with our motivation to prioritize facilities for intervention, but in practice all putative source facilities could be retained and investigated. For the four cases with multiple putative source facilities after imposition of the specificity filter, this was often due to the frequent movement of patients back and forth among a common set of facilities. Thus, although a specific source facility was not identified in these cases, a cluster of connected facilities putatively exposed to carriers was identified.
Analyzing the predicted patterns of regional transmission over the course of the outbreak revealed a single facility playing a central role. ACH10 was the first facility with cases identified, had the highest number of predicted intra-facility transmissions, and was predicted as the most common source of transmissions to other facilities. Our findings were concordant with previous studies of regional CRE outbreaks, which have consistently observed a small number of focal facilities either seeding initial regional spread, or sustaining it over time 17,18,48. In past reports where focal facilities were long-term care facilities, authors speculated that uncontrolled transmission at these facilities was critical for establishing a reservoir that subsequently spread to connected facilities 17,18. However, in cases where the focal facility is an ACH that shares patients with many other facilities, it is possible that one or more of these connected facilities is acting as a potentially unsampled reservoir 48 For example, patients from SNF13 were often later transferred to ACH10 in our study. Alternatively, the sustained presence of a strain at a single facility could be due to environmental or plumbing contamination, periodically seeding small clusters 52. Regardless, both this and prior studies show the potential of genomic surveillance to hone in on key facilities and help prioritize surveillance and intervention activities. In addition to identifying focal facilities, we also noted that integration of genomic and exposure data enabled early detection of facilities contributing to regional spread. This manifested in both the inference of facilities as sources of infections in other facilities prior to the source facility detecting cases, as well as inferring facilities as sources that never reported cases. This demonstrates the power of this integrated approach to help identify potential key facilities early in an outbreak before high case counts are reported.
Our study has several important limitations to consider. First, most facilities only contributed clinical isolates, with only a few facilities performing ad hoc surveillance to detect NDM carriers. For example, ACH7, had the most cases and performed the most surveillance over the study period. Given the known iceberg effect whereby asymptomatic carriers greatly outnumber patients with overt infection 4,53–57, we likely missed many cases due to this passive approach.
However, widespread active surveillance is likely to be infeasible in most settings. Therefore, we implemented our analysis to be focused on tracking routes of transmission within and between facilities, as opposed to constructing patient-level transmission networks. Here we show potential to support use genomic analysis to guide targeted active surveillance efforts, rather than regional active surveillance, to more efficiently utilize limited resources. The robustness of our approach was supported by non-random healthcare exposure overlap among genetically linked cases, as well as facilities with genomic linkages sharing significantly more patients’ according to CMS claims data. A second limitation is that we only had access to healthcare exposure information up until patients became a case (i.e. until blaNDM-1-carrying case was detected via surveillance or clinical culture). Therefore, we may be missing shared healthcare exposures that took place after case identification. However, despite this missing data, we were able to capture transmission linkages for most cases with high specificity, indicating that sufficient exposure data was available to link cases between facilities. Additionally, we only had healthcare exposure 90 days prior to CRE detection, while previous studies suggested CRE can colonize patients for months and years 58. Therefore, a longer period of healthcare exposure data may further inform inter-facility transmission. Third, our analysis focused on a region where NDM is not yet endemic, with cases likely originating from a single introduction and clonal spread, which may limit generalizability. Demonstrating the clonality of the outbreak enabled us to use the MSV approach without additional filtering by genetic distance. However, in a more complex setting including frequent outside importation to the region and/or plasmid transfer, distance thresholds and/or regional context may need to be considered.
Overall, we implemented a threshold-free approach to collate genomic and healthcare exposure data and applied it to elucidate the role of healthcare facilities as sources of transmission during a CRE outbreak in Michigan. Our approach has the potential to shed light on important healthcare facilities mediating transmission and capture missed facilities early on during the outbreak. As shown in previous studies, our results highlighted the value of using patient transfer history rather than solely relying on sites of case identification. By enabling early detection of key facilities of the outbreak, our approach could guide selection of facilities for enhanced genomic surveillance throughout the outbreak to prevent the spread of antibiotic resistance.
Supporting information
Data Availability
All whole genome sequence data are available online at NCBI All code used for data analysis are available online at github
Funding
Funding was provided as part of the Michigan Sequencing and Academic Partnerships for Public Health InnovaHon and Response (MI-SAPPIRE) iniHaHve at the Michigan Department of Health and Human Services (MDHHS) which is supported with funds from the Centers for Disease Control and PrevenHon through the Epidemiology and Laboratory Capacity for PrevenHon and Control of Emerging InfecHous Disease Enhancing DetecHon Expansion (6NU50CK000510-02-07).
Untitled section
Supplementary Table 3. Genome IDs and healthcare exposure data