The use of in-hospital GIS associated with the spatial statistics for endemic level monitoring and cluster detection of patients harbouring AMR bacteria is feasible. As far as we know, this is the first study to develop a surveillance methodology using GIS and spatial statistics to investigate and monitor target microorganisms causing infection/colonisation in patients from all sectors of the hospital for 3 consecutive years.
In spite of the literature recommending the use of GIS [18, 30], and spatial statistics for data analysis in hospitals and health units for almost two decades [17, 31], none of the studies used both tools for AMR surveillance and the number of in-hospital studies remains very incipient [17, 18, 30, 31]. Several sites and applications nowadays use GIS to gather and monitor infectious disease data worldwide (https://www.healthmap.org/en/; https://omictools.com/biocaster-tool; https://www.istm.org/geosentinel). These are bio-surveillance systems that capture, organise and analyse public health information that can be accessed by anyone on the internet, but none of them has applicability to in-hospital epidemiology. A new version of WHONET 2019 (http://www.whonet.org/index.html) uses Martin Kulldorf's trademark SaTScan™ to detect space–time clusters, including MDR bacteria in hospitals. The site reports the same statistical method used in our study in a 5-year project, but with no mention of in-hospital GIS.
The occurrence of MDR Gram-negative microorganisms in hospitals increases geometrically and surveillance manuals are regularly renewed [32,33,34,35], while methodologies for mapping events and detecting clusters in hospitals need to progress at the same pace . The spread of K. pneumoniae producing KPC carbapenemases has become an unparalleled public health crisis [7,8,9,10,11, 14]. In the studied hospital, the impact of evolving epidemiology with a global dimension was not different with an important increase in the resistance profiles of K. pneumoniae complex over the years .
Clusters of patients with CRKp complex in different hospital areas and floors, where most workflows are generally differentiated, would probably be overlooked by the surveillance methodologies currently used to control hospital-acquired infections. In the literature, we found no publications mentioning similar bacterial outbreak affecting adult and paediatric units located on different hospital floors. This partly explains why these were not noticed by the hospital surveillance system that used methodology without GIS and spatial statistics. Moreover, clusters affecting non-critical areas with lower frequencies were likely disregarded due to the higher occurrence in ICUs, where the higher-level resistance profile (concomitant resistance to polymyxin) probably attracted more attention from our HICC . The use of both technologies offered the advantage of monitoring a large number of data and different resistance patterns in hospital clinical areas with different endemic levels.
In this study, Additional file 1: Fig. S11 and S12 represent the traditional methodology for mapping occurrences without GIS and spatial statistics, while Figs. 1, 2 and 3 and Additional file 1: Figures S2, S9, S10 and Additional file 2: Figure S3 represent the new method for mapping the occurrences. The visual aspect of the new method contributes to understanding the complexity of in-hospital surveillance, whilst the detection of spatial–temporal clusters affecting diverse clinics on dissimilar floors, during the period of months, is likely only possible by using the space–time clustering method.
The agglomeration in time and space and flow of patients harbouring CRKp complex indicates the need to investigate the clonal aspects of isolates and the sources of dissemination. Therefore, molecular genetics of the microorganism is a necessary step for precise discrimination of K. pneumoniae from related phylogroups and genotyping . There are several potential sources of transmission of K. pneumoniae complex that needs to be further explored in the hospital . Radiology, operating rooms, support services such as hospital hygiene, laboratory, nutrition and even consulting professionals serving different areas of adults and paediatrics could serve as vehicles or sources of transmission in different floors and units. Silent colonisation is useful to explain some of our results [13, 37]. This becomes apparent when the resistant organism was detected in other clustered patients who never met physically but were hospitalised in the same wards.
The clusters last a few months in diverse neighbouring wards, indicating a continuous and widespread circulation of resistant bacteria between intensive and nonintensive care areas. The possibility of infection spread from critical wards to nearby health units throughout months is a clue to preventive measures . Although real-time spatial monitoring of AMR can provide early warnings for infection control from its origin, potential carriers (equipment, health professionals, etc.) and transmission routes, bacterial genotyping is required for investigating in-hospital bacterial transmission . The lack of genetic analysis of isolates is an important limitation in this study, especially for confirming outbreaks.
The addition of the use of hospital GIS and spatial statistics to other surveillance measures already used in hospitals, such as previous antimicrobial consumption, can be an important milestone for renewing the monitoring and control of diseases that kill thousands worldwide each year. The management of several infections and activities that take place within a hospital can benefit from this GIS methodology to ensure the safety of patients and healthcare professionals. Georeferencing patients, staff, visitors, or even hand hygiene locations and hours between or in contact with inpatients, among other factors that involve this issue, would help to understand and control the main sources and routes of transmission of microorganisms in hospitals [18, 30]. Further studies of this nature would improve the knowledge about the dynamics of microorganisms and thus contribute to the construction or remodelling in order to better meet workflows and prevent the circulation of infectious agents and the risk of transmission. This interactive and visual approach, specifically demonstrating which room, ward, corridor, floor, not just the clinic, and whenever events begin, would be performed prospectively in real-time and in 3D to proactively impact the health professionals’ adherence to infection prevention and control in the future.
Although challenging, the methodologies using georeferencing software are on the internet. QGis software is free to download (https://qgis.org/en/site/forusers/download.html). Hospitals often have architects who support the construction and renovation of physical facilities. The application of spatial statistics would require skilled professionals, but more and more technologies that enable interactive learning or the use of this tool are becoming available on the internet (http://www.whonet.org), and hospitals could incorporate specialised personnel. Additionally, integrated real-time GIS components have been used for real-time surveillance . Similarly, the microbiology laboratory system can be integrated with QGis and R, since both statistical packages allow for many spatial analyses in an automated mode .
Since this is a retrospective study, several limitations are expected. First, our intention was to demonstrate a new approach to detect clusters, which can be simply defined as an aggregation of cases grouped in place and time . We did not aim to investigate outbreaks , because of the retrospective nature of the study. We cannot provide the analysis of infectious rates because we worked only with the hospital’s microbiology laboratory database. We investigated real-life data of K. pneumoniae complex isolates recovered from all clinical and surveillance material and excluded only isolates from the same biological sample collected on the same day. We did this at the beginning to make sure we would not introduce any bias into the study, since all laboratory results are important for spatial and temporal monitoring of agents’ AMR phenotypes.
Considerable variation in AMR data analysis still present and some of recommendations were made for reporting , but not for real-life surveillance as our purpose. Since we monitored different phenotypes of K. pneumoniae, we cannot include the first isolate from each patient only. This is especially important because different phenotypes can circulate concomitantly in the same patient, in blood and rectal swabs for example. In addition, the emergence of resistance during hospitalization in a given species in the same patient can only be suspected by following all isolates and phenotypes . We chose to use in this study the same and minimum data that the HICC works initially and daily to monitor bacterial agents. Additional file 1: Figure S2 and Additional file 2: Figure S3 represent georeferenced bacteria phenotypes considering all agents included, so one limitation of this approach is that we cannot counted each isolation as a different process of infection or colonization as a prior convention [42, 43]. While Figs. 1, 2 and 3 and Additional file 1: Figures S9 and S10 represent georeferenced patients considering the first isolate with the specific phenotype by the period of hospitalisation. Nevertheless, we probably investigated a smaller tip of the iceberg; sample collection was performed during the routine investigation of infectious processes in hospitalised patients, and carbapenem-resistant Enterobacteriaceae surveillance performed differently in critical and non-critical units. Thus, the results do not represent the actual occurrence.
The seasonal increase in CRKp complex carriers detected in time series analysis also deserves further investigation and may pose an additional challenge with regard to control outbreaks . A multivariate analysis considering other variables such as length of hospital stay, previous use of antimicrobials and age would be important to better estimate rates. In addition, our data point to a statistical analysis of the longitudinal study considering a nonlinear multivariate regression, based on recent literature , with different levels of correlation among the same or different patients, in the same or in different wards.
Although colonised/infected patients are the main reservoirs and sources of bacterial circulation, health-care workers also participate in the dynamics of infectious agents in hospitals . The transmission routes and contact network between clusters can also be investigated by space–time scan statistics . The proximity matrix can be improved in the future, using data on the flow of patients, health professionals and medical materials. The 1-month timescale was designed to mimic what is normally reported in-hospital surveillance, but a more refined scale will be necessary in real life .
Using the space–time permutation model without the use of denominators, such as total patient-days or total admissions, considering the stability in the number of samples collected and hospitalisations (data not shown), may facilitate and improve resistance surveillance locally in hospitals, especially in low-income countries, as the construction of denominators by infirmaries requires greater robustness of information, which is not always feasible. Although this is a limitation to compare rates between institutions, it is also a strong point, as it simplifies the monitoring of AMR, facilitating early detection in a hospital. Space–time permutation scanning statistic is a methodology designed for the early detection of events that uses only case numbers, with no need for population-at-risk data [27, 46]. But spatial statistics can be used with denominators improving comparability among institutions.
Nevertheless, we consider that none of these limitations interfered with our purpose to demonstrate a different approach for cluster detection with the substantial amount of data that HICC usually works with. The GIS methodology was designed to attend the real-life monitoring and can be applied using different AMR monitoring criteria. Our observation is that the amount of microbiological data, with the variety of target AMR phenotypes in hospitals, requires automated methods using spatial statistics for cluster detection. New and traditional methodologies for hospital surveillance are likely complementary, but more studies are needed to compare the benefit of each approach, the reliability of the mathematical model chosen and how it would do better for surveillance of AMR or any infectious agent in hospital. Finally, but not least, the rates found, although underestimated, are useful to demonstrate the importance of the theme in the hospital.