Wireless Sensor Network based System for the Prevention of Hospital Acquired Infections
Iuliana Bocicor, Maria Dascalu, Agnieszka Gaczowska, Sorin Hostiuc, Alin Moldoveanu, Antonio Molina, Arthur-Jozsef Molnar, Ionut Negoi, Vlad Racovita
WWireless Sensor Network based System for the Prevention of HospitalAcquired Infections
Iuliana Bocicor , Maria Dasc˘alu , Agnieszka Gaczowska , Sorin Hostiuc , Alin Moldoveanu ,Antonio Molina , Arthur-Jozsef Molnar , Ionut , Negoi and Vlad Racovit , ˘a SC Info World SRL, Bucharest, Romania NZOZ Eskulap, Skierniewice, Poland Carol Davila University of Medicine and Pharmacy, Bucharest Polytechnic University of Bucharest, Romania Innovatec Sensing&Communication, Alcoi, Spain { iuliana.bocicor,arthur.molnar,vlad.racovita } @infoworld.ro, [email protected], { agaczkowska,soraer,negoiionut } @gmail.com, [email protected], [email protected] Keywords: Hospital Acquired Infection, Nosocomial Infection, Clinical Workflow Monitoring, Cyber-physical System,Wireless Sensor Network.Abstract: Hospital acquired infections are a serious threat to the health and well-being of patients and medical staffwithin clinical units. Many of these infections arise as a consequence of medical personnel that come intocontact with contaminated persons, surfaces or equipment and then with patients, without following properhygiene procedures. In this paper we present our ongoing efforts in the development of a wireless sensornetwork based cyber-physical system which aims to prevent hospital infections by increasing compliance toestablished hygiene guidelines. The solution, currently developed under European Union funding integrates anetwork of sensors for monitoring clinical workflows and ambient conditions, a workflow engine that executesencoded workflow instances and monitoring software that provides real-time information in case of infectionrisk detection. As a motivating example, we employ the workflow in the general practitioner’s office in or-der to comprehensively present types of sensors and their positioning in the monitored location. Using theinformation collected by deployed sensors, the system is capable of immediately detecting infection risks andtaking action to prevent the spread of infections.
Hospital acquired infections (HAI) are a seriousthreat to the health and well-being of both patientsand medical staff within clinical units. Various micro-organisms, especially multi drug resistant bacteriamay lead to significant hospital related morbidity andmortality. These infections have high related costsand represent a direct occupational hazard for clin-ical personnel. Hospital infections are a worldwideproblem, regardless of geographical, political, so-cial or economic factors (World Health Organization,2002), (World Health Organization, 2010). Further-more, technological development and sophisticationof medical care does not automatically result in low-ered infection rates (Tikhomirov, 1987), (Coello et al.,1993), (World Health Organization, 2010), (EuropeanCentre for Disease Prevention and Control, 2015). According to the findings of the World Health Orga-nization, average HAI prevalence in Europe is 7.1%,in the United States it is 4.5%. In low- and middle-income countries infection rates vary between 5.7%and 19.1% (World Health Organization, 2011). Inintensive care units located in high-income countriesthe proportion of infected patients can be as high as51%, while in low- and middle-income countries itcan reach 88.9% (World Health Organization, 2011).Unfortunately, many of these infections lead to pa-tient deaths. Annually, infections are accountable for37 000 deaths in Europe and 99 000 deaths in the USA(World Health Organization, 2011). While measuresand precautions are being taken to successfully reducethese rates (Centers for Disease Control and Preven-tion, 2016), there is still much room for improvement.Many hospital infections arise as a consequenceof medical personnel that come into contact with con- a r X i v : . [ c s . C Y ] M a y aminated surfaces or equipment, relatives coming incontact with patients or as auto infections, which oc-cur by touching sensitive body parts, such as the face,with the hands. The most common target sites for hos-pital infection are the urinary and respiratory tracts,which are often involved in minimally-invasive pro-cedures such as catheter-related procedures.While the methodology for prevention exists, itis often ignored due to lack of time, unavailabilityof appropriate equipment or because of inadequatestaff training. Research shows that the most impor-tant transmission route are staff members who comeinto contact with patients or contaminated equipmentwithout following proper hygiene procedures (Ham-mer, 2013). More particularly, most often trans-mission is made by touching a patient or contami-nated equipment and then touching another patientwithout proper hand hygiene (Pittet, 2001). Recentguidelines, such as ”Five moments for hand-hygiene”(World Health Organization, 2015) provide conciseand well-structured information on efficient disinfec-tion means to significantly reduce the risk of infec-tion.In this paper we present our progress in the devel-opment of a cyber-physical system based on a wire-less sensor network (WSN) that is targeted towardsHAI prevention. Our solution employs a sensor net-work that will monitor clinical workflows and ambi-ent conditions, integrated with configurable softwareto detect deviation from established hygiene prac-tices. The sensor network collects information in real-time about substance and material presence. Avail-ability of antimicrobial agents and sterile gloves, aswell as environmental conditions that affect pathogenspread, such as oxygen level, airflow, and temperaturewill be monitored. To complete the picture, the sys-tem will facilitate monitoring of complex processessuch as management of indwelling urinary catheters,postoperative care, intubation and mechanical venti-lation. Given the complexity of these processes, aswell as the diversity of hospital regulations, these pro-cesses will be described using software workflows.Each clinical process will be modelled using oneworkflow instance executed by a software workflowengine. When the sequence of transitions inferred bythe system from sensor data presents deviations fromthe expected flow, the system will alert responsiblepersonnel.The current stage of development represents a sys-tem proof of concept, including multifunctional smartsensors for monitoring the use of soap, antimicrobialgel and water sink together with a first clinical work-flow that describes the required hygiene procedures inthe general practitioner’s office. This paper details the smart devices employed, the hardware-software inte-gration as well as the software components that en-sure the cyber-physical system achieves its objectiveof lowering the number and severity of hospital infec-tions.The present paper is structured as follows. Thefollowing section presents some of the last decade’sadvancements, from an Information and Communica-tion Technologies (ICT) perspective. This includesseveral software-based or cyber-physical systems de-veloped to increase compliance to guidelines and de-crease infection rates. Section 2 offers a detailed de-scription of the workflow in the general practitioner’soffice, as well as the main challenges faced when en-coding this process using a software-based workflowmodel. The wireless sensor network custom designedfor the system is depicted in Section 3, where all as-pects related to its hardware and software componentsare covered. Section 4 focuses on presenting the soft-ware workflow engine that interprets events and mon-itors execution of configured workflows. Moreover,the means by which the hardware part of the systemis integrated with the software workflow engine, aswell as the communication infrastructure employedare illustrated in this section. Finally, we outline ourconclusions and further work in the last section. Because HAI have significant detrimental effects, inrecent decades health care facilities have started toimplement prevention programmes for patients andmedical staff. There are even practical guides devisedby specialised agencies (World Health Organization,2012), which can be used as starting points for the de-velopment of good practice plans concerning work-place and patient safety. Following the current tech-nological developments in all medical areas, technol-ogy is also present for monitoring and prevention ofHAI. This is illustrated by the development of varioussoftware-based or cyber-physical solutions that moni-tor and ensure compliance. In this section, we presentsome of the most popular such systems.
Inadequate hand hygiene is responsible for a largeproportion of infections (Pittet, 2001). There are sev-eral automated solutions to reduce infections causedby improper hand hygiene, most of which use con-tinuous surveillance and immediate notification incase non-compliance is detected (Shhedi et al., 2015).IntelligentM (Ryan, 2013) and Hyginex (Hyginex,2015) are two solutions that monitor employees us-ing bracelet-like devices equipped with Radio Fre-uency Identification (RFID) technology and motionsensors. Whenever a hygiene event has been omitted,the device alerts them either using vibration (Intelli-gentM) or coloured lights (Hyginex). Biovigil tech-nology (BIOVIGIL Healthcare Systems, Inc., 2015)and MedSense (General Sensing, 2014) are designedhaving the same purpose, only in these cases braceletsare replaced with badges worn by healthcare workers.The Biovigil device uses chemical sensors to detectwhether hand hygiene is observed according to estab-lished standards. The system can be configured to re-mind clinicians to disinfect their hands before enter-ing patient wards, or before administering proceduressuch as intravenous drips or catheter insertion. Fur-thermore, these systems record hygiene events, cen-tralise them and enable analysis, visualisation and re-port generation. SwipeSense (Swipe Sense, 2015)employs small, alcohol-based devices and wearablegel dispensers. This allows medical personnel to per-form hand hygiene without interrupting their activi-ties to go to a sink or disinfectant dispenser (Simon-ette, 2013). In opposition to the systems mentionedso far, which use sensors placed at patient ward en-trances, UltraClenz’s Patient Safeguard System (Ul-traClenz, 2016) is ”bed-centric” and prompts work-ers to sanitize before and after every patient contact.The DebMed system (DebMed - The Hand HygieneCompliance and Skin Care Experts, 2016) does notuse RFID technology, nor any devices for the medicalpersonnel, but instead estimates the number of handhygiene opportunities per patient-day and comparesthis number with the actual hand hygiene events thatwere performed, which are determined using a net-work of wireless-enabled dispensers.
Dangerous pathogens can remain in the air or on dif-ferent types of surfaces in a hospital room for longperiods after the infection source was removed. Totackle this issue, which cannot always be resolved us-ing traditional cleaning and disinfection procedures,several types of disinfection robots have been devel-oped. Generally, they are able to perform thoroughdisinfection using ultraviolet (UV) light or chemi-cal substances. The Xenex ”Germ-Zapping Robot” (Xenex, 2015) can disinfect a room using pulsesof high-intensity, high-energy ultraviolet light. Therobot must be taken inside the room to be disinfectedand in most cases, the deactivation of pathogens takesplace in five minutes. Tru-D Smart UVC (Tru-DSmart UVC, 2016) scans the room to be disinfectedusing eight sensors and computes the optimal shortwavelength ultraviolet light dose required for disin-fection according to the size, geometry, surface reflec- tivity as well as the amount and location of equipmentfound in the room. The robot performs disinfectionof the entire room, from top to bottom in one cycleand from one location, ensuring that the ultravioletlight reaches even shadowed areas. The Bioquell Q-10 robots (Bioquell, 2016) emit a powerful antibacte-rial bleaching agent, called hydrogen peroxide to killmulti-drug resistant organisms. As hydrogen perox-ide is toxic to humans, after disinfection the Q-10 usesanother solution to ensure that it is safe for humans toenter the room.
A different procedure in the fight against infectionis implemented by the Protocol Watch decision sup-port system for prevention and management of sepsis(Philips, 2015). Protocol Watch regularly checks cer-tain medical parameters of patients, to reduce the timeelapsed between the moment sepsis is first detectedand beginning of treatment. If the system detects thatcertain conditions indicative of sepsis are met, it alertsmedical staff and indicates which tests, observationsand interventions must be performed, according to es-tablished prevention and treatment protocols.Another goal pursued by clinicians when dealingwith hospital infection is the identification of controlpolicies and optimal treatment in infection outbreaks.A comprehensive approach that uses electronic healthrecords to build healthcare worker contact networks isdescribed in (Curtis et al., 2013). Its main goal con-cerns putting efficient vaccination policies into placein case of infection outbreaks.Among other relevant software systems developedto enhance treatment policy in case of infection out-break or epidemics are RL6:Infection (RL Solutions,2015) and Accreditrack (Excelion Technology Inc.,2013). RL6:Infection is a software solution devel-oped to assist hospitals in the processes of controllingand monitoring infections and outbreaks, while Ac-creditrack is designed to ensure compliance with handhygiene guidelines, verify infection management pro-cesses as well as to provide procedural visibility andtransparency.
The proposed platform is developed within the Hospi-tal Acquired Infection and Outbreak Prevention Sys-tem (HAI-OPS) research project (HAI-OPS, 2017).Its main objective is to decrease overall mortality andmorbidity associated with hospital infection. It is de-signed to handle both singular infection cases as wellas outbreaks, by targeting most common sources andtransmission pathways. Operationally, the platformill leverage advances in computing power and avail-ability of custom-developed, affordable hardware thatwill be combined with a configurable, workflow-based software system (Bocicor et al., 2016).Existing solutions, such as those detailed inthe section above (Ryan, 2013; Hyginex, 2015;BIOVIGIL Healthcare Systems, Inc., 2015; GeneralSensing, 2014) can be successfully employed to mon-itor a single process, such as hand hygiene, or equip-ment and room disinfection (Xenex, 2015; Tru-DSmart UVC, 2016; Bioquell, 2016). While these pro-cesses are important for keeping patients and staffsafe from infection, there are many other processesthat can lead to hospital infection. Among the mostprevalent, we mention catheter management, mechan-ical ventilation, invasive procedures and surgical sitecare (World Health Organization, 2002; Coello et al.,1993). One solution for monitoring multiple pro-cesses would be to deploy several such systems inparallel. However, given that eHealth interoperabil-ity is currently an open issue, this is not only cost-ineffective, but technologically infeasible. We be-lieve that monitoring several clinical and maintenanceworkflows can be successfully addressed using a sin-gle system. Such a system must be configurable sothat it covers differences between clinical unit loca-tion and layout, differences in types and specifics ofundertaken procedures, as well as variation betweenhygiene guidelines that must be observed by staff.The HAI-OPS platform is designed to address theseissues in both hardware as well as software. First ofall, using customized, but affordable hardware allowssensors to be deployed in key locations in cost effec-tive manner. Workflow engines allow researchers tocreate custom BPMN-encoded (Object ManagementGroup, 2015) workflows that encode key events inmonitored processes. Furthermore, implementationof a user interface for workflow management will al-low epidemiologists to further customize the moni-tored workflows. To the best of our knowledge, ourproposed system is the first of its kind to combine asensor network and software in a cyber-physical sys-tem of the proposed versatility.
The cyber-physical system depicted in this pa-per employs pre-defined workflows that describe theprocesses that the system will monitor. They allowthe system to take real-time action in case an infec-tion risk is detected. Our development approach isbottom-up, and starts with modelling some of the less complex workflows, which involve only medical staffand patients. The more complex workflows, that alsoinvolve equipment, such as endoscopic or surgicalprocedures will be addressed at a later time. Thus,the first workflow we approach for the system proto-type, which is also the subject of the present paper’smotivating example, is the workflow of the generalpractitioner’s office.
The general practitioner (GP) is a medical doctorwhose practice is not limited to a certain specialityand who provides treatment and preventive care to pa-tients. As opposed to physicians working with inpa-tients admitted to hospital for certain procedures, thegeneral practitioner works with outpatients, who re-quire consultation or treatments which do not necessi-tate hospital admission. All information regarding theGP office, as well as the consultation workflow de-scribed were supplied by NZOZ ESKULAP (NZOZEskulap, 2016), an outpatient clinic from Poland thatis targeted for the first pilot deployment of our sys-tem. Figure 3 illustrates the general practitioner’soffice layout from the Polish clinic. The office con-tains a desk for the physician, a consultation bed and,most importantly for our use case, an area with sev-eral elements for ensuring hygienic conditions: a sink,a waste bin and an area dedicated to disinfectantsand disinfectant dispensers. The same figure also de-picts the planned layout of the wireless sensor net-work used for monitoring the workflow. These aredescribed in more detail in Section 3.In order to ensure compliance with the infection-prevention guidelines in the Polish clinic, the first stepwas identifying the hygiene practices to which thegeneral practitioner must adhere to before, during andafter patient consultation. The conventional workflowfor an outpatient consultation, including all requiredactions for ensuring conformity with hygiene stan-dards are depicted within the following sequence ofsteps:1. Patient enters the office.2. The GP starts a conversation with the patient, inorder to learn about their medical history, currenttreatment and reason for the visit. Generally, thephysician uses pen and paper or a hospital infor-mation system to record information to the patientfile.3. The GP prepares to examine the patient. Thepreparation process is crucial with regards to in-fection prevention. According to current regula-tions within the target clinic, the doctor must san-tize their hands according to 10 steps for effectivehygiene. These are:(a) Wet hands thoroughly.(b) Soap up, using the liquid soap dispenser. Theused tap must be elbow or wrist operated. Thephysician must rub palms.(c) Rub palms with finger interlaced.(d) Massage between fingers, right palm over theleft hand and then vice-versa.(e) Scrub with fingers locked, including fingertips.(f) Rub rotationally, with thumbs locked.(g) Rinse thoroughly.(h) Dry hands using a paper towel that must beplaced in proximity to the hand washing facil-ity.(i) Work towel between fingers.(j) Dry around and under the nails.4. The GP throws the wet towel to a special wastebin.5. The GP starts patient examination.6. After the examination, the GP uses an alcohol-based sanitizer for hand disinfection.7. The GP goes back to the desk and records exami-nation results using pen and paper or the hospitalinformation system.8. Patient leaves the office.The procedure described above concerns a regularexamination. However, for special cases such as ex-aminations involving the head, eyes, ears, nose andthroat (HEENT), or when the patient presents withskin infection, the doctor must also employ nitrile orlatex disposable gloves. Gloves should also be wornwhenever there might be contact with blood, body flu-ids, mucous membranes or non-intact skin. Glovesmust be put on immediately before the task to be per-formed, and removed and discarded as soon as theprocedure is completed.The BPMN workflow for the consultation processis illustrated in Figure 1. Sections 3 and 4 describehow the wireless sensor network is used for monitor-ing and how the workflow engine monitors the exe-cution of hygiene-relevant events. In the case a devi-ation from the expected steps of the workflow is de-tected, a real-time alert is generated and sent to theGP using a mobile device in their possession.
The general practitioner workflow is shown usingBPMN specification in Figure 1. For the descriptionof the workflow we use both Figures 1 and 3, as the events specified in the workflow are detected by hard-ware devices placed in different locations in the of-fice. As soon as the patient enters the office, this isdetected by the infrared array sensor element placednear the entrance (Goga et al., 2016). The systemrecords and interprets the received data and a work-flow instance is started. As illustrated in Figure 1, thefirst steps required from the GP is to start the watersink, use the soap dispenser and then stop the sink.The system interprets this as hand hygiene being per-formed. These events are detected by the sensor el-ements in the sink and those in the disinfectant dis-penser area, which are all connected to smart nodesplaced near the physician’s desk. This enables trans-mitting the data to the software server via wirelessnetwork. While current regulations described in theprevious section require a specific sequence of actionsto be undertaken for hand hygiene to be consideredeffective, our system only checks that the sink anddisinfectants were operated. The main reason for thisis that the system is envisaged as an additional aid formedical personnel that ensures their safety from pos-sible infection. The system is designed on the princi-ple that medical personnel are responsible and awareof the detailed actions they must undertake to ensuretheir own, as well as their patients’ safety.The intermediate step of the workflow concern-ing patient examination starts when the system hasdetected that hand hygiene compliance is achieved.Otherwise, the system generates and stores a hygienealert, which is immediately sent to the general prac-titioner. The two activities are exclusive: if an alertis generated, the workflow instance is stopped andthe recorded hygiene breach is recorded. In case ofan alert, the GP must perform hand hygiene, afterwhich the system initiates a new workflow instance.In case initial hand hygiene and patient consultationare carried out according to the workflow, the GPmust disinfect their hands using antimicrobial gel af-ter the last contact with the patient. This event is againrecorded by the system using the same sensors situ-ated in the disinfectant dispenser area and the smartnodes near the physician’s desk. The workflow isthus completed. All the information related to patiententry/exit, hygiene compliance and alerts is saved topersistent storage for further reuse, including statis-tics and advanced analyses for finding the source orpropagation of an outbreak.
Although seemingly straightforward, the process de-scribed above can become quite complicated, mainlydue to various types of constraints and interferences igure 1: General Practitioner Workflow that may occur. Below we present the main challengesto the system, with regard to the GP office workflowand the methods we use to approach and overcomesome of them. Others are still open to discussion andsolutions are currently being investigated.First and foremost, one key aspect to consider isachieving minimal overhead on the clinical processand minimal intrusive interaction, from the user ex-perience point of view. It is important that the systemdoes not impose any constraints and does not restrictthe doctor’s movements. In many clinical units, in-cluding the one targeted for pilot deployment, hospi-tal regulations specify that personnel are not allowedto wear jewellery, watches or bracelets, as these canhamper their freedom of movement and spread bacte-ria, especially if these wearables are difficult to dis-infect. To tackle this, the proposed system does notrequire the use of additional wearables. Monitoringis done using the deployed wireless sensor networknodes, which are placed in key locations within theGP office, as described in Section 3. In addition, med-ical personnel already employ chest-mounted badgesto which radio-frequency tags can be easily added.The placement of the wireless sensors is an essen-tial challenge in itself, as locations must be chosen ina manner that allows a complete and preferably op-timal surveillance of monitored workflows. The ar-rangement of sensors in the office must be adjusted tothe process, but should also be sufficiently general inorder to allow monitoring several workflows: in thiscase, both the regular consultation workflow as wellas HEENT examinations. Thus, in addition to plac-ing wireless sensors at the office entrance, sink, soapor disinfectant dispenser, in order to ensure completeprocess monitoring, a device is also placed on thewaste bin, to detect when gloves are thrown away. De-vice positioning in the GP office is discussed in more detail within Section 3.One of the remaining challenges for cyber-physical systems such as the proposed one concernsshort-term human interactions that are difficult to de-tect. In the case of the GP workflow, how should thesystem detect and react to a person entering the GPoffice during an examination? In this case, the hy-giene event performed by the GP before patient ex-amination is considered cancelled, as the third personcan contaminate the physician or patient with micro-organisms. Medical staff wear badges that can beused to identify them using the sensors deployed nearthe entrance; however, if the person is not part of themedical staff, they cannot be identified. A potentialsolution is that once the system detects someone en-tering the office, regardless of whether the person ismedical staff or not, the system triggers the executionof a new workflow, including the necessary hygieneevents. In case this is not performed, the GP is alertedto take immediate corrective action.
A wireless sensor network consists of a group ofelectronic devices in which every node controls one ormore sensors that measure physical phenomena suchas light, heat or proper acceleration. All collectedmeasurements are sent using a wireless network pro-tocol to another device featuring more powerful pro-cessing capabilities. Depending on their functional-ity, nodes are classified into dummy and smart nodes.As the name suggests, dummy nodes consist of smalldevices that have to effectuate just one simple task:detect a generated event and pass the information toa smart node. A dummy node is particularly charac-terized by its small size (35x35mm) and low poweronsumption. Some sensors, like RFID readers, donot generate events by themselves and require a pre-processing stage, which can be exclusively carried outby a powered device. Smart nodes must be able tocollect key actions detected from dummy nodes andgenerate more complex events comprising informa-tion regarding four relative clauses: who is the personinvolved, what was the action generated, when it hap-pened and where it happened.
Required sensors were selected to enable monitoringthe clinical workflow detailed in Section 2. From thementioned steps, sensors in dummy nodes should beapplied mainly to detect key actions, such as the uti-lization of hygienic elements that can be found in theGPs office: water sink, soap dispenser, waste bin, al-cohol sanitizer and glove dispenser. The main sensortypes required to ensure effective monitoring of thegeneral practitioner office workflow are as follows:1.
Accelerometer . These sensors measure changesin gravitational acceleration on two or three axes,allowing to detect changes in motion and orien-tation. Accelerometers may be attached to watertaps, which regulate water flow on the vertical axisand temperature on the horizontal. They may alsobe applied to sanitizer or glove dispensers, wheredetected motion implies that they have been usedby a practitioner or checked by cleaning staff.2.
Proximity and light sensor . Proximity sensorsemit infrared radiation and look for changes in thereturn signal. This type of sensors are already ap-plied in some water sinks and soap containers, butto the best of our knowledge none of them havecommunication capabilities to report actions.3.
Switch detection . A switch is just an electroniccomponent that interrupts the flow of electric cur-rent from one conductor to another. It may be op-erated by a moving object, which makes it a greatchoice for applications such as detecting the useof waste bins or opening of a door. This is the leastenergy consuming element from the list, becauseit does not have to expend energy doing continu-ous measurement.Dummy nodes generate action events indicating,for instance, that someone used the soap dispenser orthe waste bin, but it is the smart node who has to fillin information and identify who generated the action.To achieve that goal, it is necessary to process andcombine the output from the following two sensors:1.
Infrared array sensor . It is a thermopyle type in-frared sensor which detects the amount of infrared
Figure 2: Infrared array sensor detecting people inside theGP office rays. It has a built-in lens with a 60 degree view-ing angle. The sensor offers output for thermalpresence, direction and temperature values.2.
RFID reader . Radio-frequency identificationworks using tag-based identification. Tags aresmall devices similar to stickers that may be car-ried by people, animals or objects. They can alsobe easily attached to wearables such as badges ormobile equipment. The frequency range and ap-plied antenna depend on the application and indi-rectly on the distance between readers and tags. Insome clinics, medical and cleaning staff are usedto carry a badge with an identification card basedon this principle.
Device positioning and calibration are crucial for theproper functioning of the system. In the case ofdummy sensors, the proximity sensor may detect falsepositives if the distance range is not correctly adjustedor if the sensor is incorrectly placed. When appliedto water sinks or gel dispensers, the proximity sensormust be tied to the tap pointing downwards. The sys-tem registers when someone places their hand underthe tap and when they stop using it. Figure 3 illus-trates the positioning of both smart and dummy nodeswithin the general practitioner’s office.In addition to the proximity sensor, the RFID an-tenna and passive infrared array sensor must also beplaced according to their detection range. RFID read-ers provide received signal strength indication (RSSI)levels for detected tags, a measure which is propor-tional to the distance between them. Patch antennasconsist of a planar dielectric substrate material witha radiating patch on one side and a ground plane onthe other. The radiating side must point to the GP’soffice where elements to be identified are located andthe ground plane must point to the corridor, ceiling orto an adjacent room. Radio frequency power output igure 3: Layout of the general practitioner office aug-mented with wireless sensor network must be configured to meet European Union regula-tions and to avoid false positive detections as much aspossible.Passive infrared array sensors complement the in-formation from RFID readers. If this information iscombined properly, the dummy node is able to lo-cate people inside the room, identify people wearingan RFID card (typically clinical staff) and detect peo-ple who are not wearing tags (typically patients). Thepassive infrared sensor must be placed on the ceiling,pointing downwards and centred in the room, as seenin Figure 2. If the sensor’s angle of view is not enoughand doesn’t fit the whole room, scalability is achievedusing several sensors.
As already stated, the main features of dummy nodesare their small size and low power consumption. Bothfeatures are very closely correlated, because in mostcases product size is determined by the battery. Inthis case, the power consumption in dummy nodes is so low, that they can be powered using coin batter-ies. The reduced power consumption is due to theintegration of a Bluetooth Low Energy (BLE) mod-ule (Bluetooth SIG, Inc., 2017). Compared to previ-ous Bluetooth standards, BLE is intended to provideconsiderably reduced power consumption and lowercost, while maintaining a communication range of upto 150 meters with connected devices.Smart nodes have more complex processing, com-munication and thus, higher power requirements thandummy nodes. They are continuously listening forinput BLE connections and when data is received, theresult is forwarded to a database server for persistentstorage and subsequent analyses. Connection withthis server is carried out using existent network infras-tructure, regardless of whether it is wired or wireless.Every smart device is identified within the networkusing a unique IP address and has a fixed location in-side the building.
The software side of the system implements theclient-server paradigm and employs a software serverto which an arbitrary number of heterogeneous clientscan have simultaneous connection. The server con-tains components that receive sensor data from thenetwork, a persistence layer that manages sensorreadings as well as a workflow engine that executesworkflow instances in real-time, generating alertswhen sensor readings indicate deviations from ex-pected workflow transitions. The main software com-ponents of the system are as follows:1.
Connected Device Controller.
The connecteddevices, or smart nodes, constitute the principalhardware component of the platform. They are re-sponsible for monitoring the clinical environmentusing sensors and sending sensor readings to thesoftware server. By themselves, they cannot de-cide whether an infection risk is present. Eachconnected device includes a software controller,a generic software component that runs indepen-dently of the software server. Its objectives are toensure the correct functioning of smart nodes andto send sensor readings to the software server.2.
Data acquisition.
This is the software compo-nent that will be responsible with receiving sensorreadings. Received data is stored within the per-sistence layer, from which it is read and used byother components. This includes advanced analy-ses components yet to be developed which do notake the object of the present paper. The serveradopts a REST architecture (Fielding, 2000) to re-ceive sensor data acquired as presented in Sec-tion 3. The data interchange format employedis JavaScript Object Notation. Information con-tained in files received from sensors includes theevent’s timestamp, a Uniform Resource Locatorthat identifies which node generated the informa-tion as well as sensor reading values. For instance,a sensor monitoring temperature will transmit atemperature value in degrees Celsius. A sen-sor monitoring the presence of an individual willtransmit a boolean value, according to whetherpresence was detected. An RFID reader willtransmit the RFID tag identifier and the receivedsignal strength indicator value.3.
Workflow engine adapter.
This software com-ponent is a fac¸ade to the workflow engine imple-mentation used by the system. Its main purposeis to abstract the particularities of the workflowengine. This allows the system to operate withany major off the shelf workflow engine imple-mentation. This component provides the requiredfeatures that allow for the creation, update anddeletion of clinical workflows monitored by thesystem. The workflow engine interprets events,such as inputs from deployed sensors (e.g. handwashing detected), and acts upon them accord-ing to a predefined process. The actions are con-figurable and can vary from saving a new entryinto a database, sending an e-mail or emitting areal-time notification via an external applicationor short message service. Its input is representedby process descriptions. Processes are composedof activities connected with transitions. Processesrepresent an execution flow. Each execution of aprocess definition is called a process instance. Asan example, hand disinfection in the general prac-titioner’s office can be represented as a process.Each time a patient enters the office a new processinstance is started, managed by the business pro-cess management system. Some activities, suchas recording an event, or sending an alert are au-tomatic. Others involve waiting for an externalevent to occur, such as a sensing device reportingthe physician has disinfected hands. The work-flow engine keeps track of the state of processexecutions and manages creation and progress ofprocess executions.4.
Data store.
This component of the softwareserver acts as the system’s persistence layer forall system data. In addition to user, alert, con-nected devices and workflow data, it includes acomplete record of data received from connected sensors. This is required in order to facilitate ad-vanced analyses for outbreak prevention, identi-fication and monitoring. The data store will beimplemented using an SQL database.
This paper details the ongoing effort in the devel-opment of a cyber-physical system intended for pre-venting hospital infections. The HAI-OPS platformwill integrate a wireless sensor network and a soft-ware server that uses a workflow execution engine tomonitor key steps within various clinical processesthat were identified as responsible for a large pro-portion of hospital infection. When key steps to en-sure process hygiene are not taken, the system willgenerate real-time alerts. The present paper focuseson describing and modelling an initial clinical pro-cess used as motivating example: outpatient consul-tations within the general practitioner’s office. Thisprocess is the one selected for implementation dur-ing the system’s first pilot deployment within a Polishoutpatient clinic (NZOZ Eskulap, 2016). Althoughwe directed our attention specifically towards the mo-tivating workflow, the system is designed to allow fordeployment of diverse sensor network configurations,as well as facilitate creation and execution of manydifferent clinical workflows.Upcoming system development will build on ex-isting achievements. First of all, the system will beused to model more complex clinical workflows, in-cluding endoscopic and minor surgery procedures,which will be implemented in the pilot site location.Second of all, as part of the project a graphical com-ponent will be developed to allow management ofmonitored workflows. In addition, we aim to leverageavailable sensor readings by implementing advancedreporting and evaluation capabilities. These are ex-pected to help clinical epidemiologists in pinpointinginfection and outbreak sources using visualizationssuch as risk maps and healthcare worker contact net-works (Hladish et al., 2012).
ACKNOWLEDGEMENT
This work was undertaken as part of the HAI-OPSproject funded by the European Union, under the Eu-rostars programme . EFERENCES
Bioquell (2016). Bioquell q-10. .BIOVIGIL Healthcare Systems, Inc. (2015). Biovigiland our team. .Bluetooth SIG, Inc. (2017). Bluetooth low energy. .Bocicor, M. I., Molnar, A.-J., and Taslitchi, C. (2016).Preventing hospital acquired infections through aworkflow-based cyber-physical system. In
Proceed-ings of the 11th International Conference on Evalua-tion of Novel Software Approaches to Software Engi-neering , pages 63–68.Centers for Disease Control and Prevention (2016). HAIdata and statistics. .Coello, R., Glenister, H., Fereres, J., Bartlett, C., Leigh, D.,Sedgwick, J., and Cooke, E. (1993). The cost of infec-tion in surgical patients: a case-control study.
Journalof Hospital Infections , 25:239–250.Curtis, D., Hlady, C., Kanade, G., Pemmaraju, S., Pol-green, P., and Segre, A. (2013). Healthcare workercontact networks and the prevention of hospital-acquired infections.
Plos One . DOI: 10.1371/jour-nal.pone.0079906.DebMed - The Hand Hygiene Compliance and Skin CareExperts (2016). A different approach to hand hygienecompliance. http://debmed.com/products/electronic-hand-hygiene-compliance-monitoring/a-different-approach/ .European Centre for Disease Prevention and Control(2015). Annual epidemiological report. antimicrobialresistance and healthcare-associated infections. 2014. http://ecdc.europa.eu/en/publications/Publications/antimicrobial-resistance-annual-epidemiological-report.pdf .Excelion Technology Inc. (2013). Accreditrack. .Fielding, R. T. (2000). Architectural styles and the design ofnetwork-based software architectures. Doctoral dis-sertation, University of California.General Sensing (2014). Medsense clear. handhygiene compliance monitoring. .Goga, N., Vasilateanu, A., Mihailescu, M. N., Guta, L.,Molnar, A.-J., Bocicor, I., Bolea, L., and Stoica, D.(2016). Evaluating indoor localization using wifi forpatient tracking. In
International Symposium on Fun-damentals of Electrical Engineering (ISFEE) .HAI-OPS (2017). home page. http://haiops.eu .Hammer, S. (2013). Hand washing: Reducing nosoco-mial infections. http://iwsp.human.cornell.edu/files/2013/09/Hand-Washing-Reducing-Nosocomial-Infections-2j1mlfb.pdf .Hladish, T., Melamud, E., Barrera, L. A., Galvani, A., andMeyers, L. A. (2012). Epifire: An open source c++ library and application for contact network epidemi-ology.
BMC Bioinformatics , 13(1):76.Hyginex (2015). Introducing hyginex generation 3. .NZOZ Eskulap (2016). NZOZ eskulap. .Object Management Group (2015). Business process modeland notation. .Philips (2015). Protocolwatch - SSC Sepsis. .Pittet, D. (2001). Improving adherence to hand hygienepractice: a multidisciplinary approach.
Emerging In-fectious Diseases , 7:234–240.RL Solutions (2015). The rl6 suite / infectionsurveillance. .Ryan, J. (2013). Medtech profiles: Intelligentm - a sim-ple yet powerful app to dramatically reduce hospital-acquired infections. https://medtechboston.medstro.com/profiles-intelligentm/ .Shhedi, Z. A., Moldoveanu, A., Moldoveanu, F., andTaslitchi, C. (2015). Real-time hand hygiene monitor-ing system for hai prevention. In
The 5th IEEE Inter-national Conference on E-Health and Bioengineering- EHB 2015 .Simonette, M. (2013). Tech solutions to hospital acquiredinfections. .Swipe Sense (2015). Hand hygiene. redefined. .Tikhomirov, E. (1987). Who programme for the control ofhospital infections.
Chemioterapia , 6:148–151.Tru-D Smart UVC (2016). About tru-d. http://tru-d.com/why-uvc-disinfection/ .UltraClenz (2016). Patient safeguard system. .World Health Organization (2002). Preventionof hospital-acquired infections - a practicalguide. .World Health Organization (2010). The burden ofhealth care-associated infection worldwide. .World Health Organization (2011). Health care-associatedinfections - fact sheet. .World Health Organization (2012). Preventionof hospital-acquired infections - a practicalguide. http://apps.who.int/medicinedocs/documents/s16355e/s16355e.pdf .World Health Organization (2015). Clean care is safer care- five moments for hand hygiene. .Xenex (2015). Xenex germ-zapping robots.