IoT-Based Asset Management System for Healthcare-Related Industries

1.1 IoT Applications in the Healthcare Industry

The Internet of Things (IoT) is an emerging technology which is generally recognized as representing a revolution in Information and Communication Technology (ICT). It is expected to have a wide range of applications in various industrial sectors, including healthcare (Xu et al, 2014b; Yang et al, 2014). According to the latest Hype Cycle of newly emerging technologies, IoT was in one of the top three ‘innovation trigger positions’ in 2014, showing a tendency to grow towards the peak of the Hype Cycle (Burton & Willis, 2014). The Internet of Things Strategic Research Roadmap is based on the following definition: “The Internet of Things allows people and things to be connected anytime, anyplace, with anything and anyone, ideally using any path/network and any service” (Guillemin & Friess, 2009). The concept of IoT can be regarded as “an extension of the existing interaction between humans and applications through the new dimension of things' communication and integration” (Guillemin & Friess, 2009). There is a broad range of key opportunities for applications in different industries, such as healthcare, intelligent building (i.e. green building), product and brand management, retail and logistics management, people and goods transportation, and so on (Atkinson, 2014; Monares et al., 2014; European Commission, 2013; Istepanaian & Zhang, 2012; Guillemin & Friess, 2009). IoT is expected to play an important role across Europe (e.g. Horizon-2020) (European Commission, 2013; Sung & Chang, 2013) and Asia (e.g. China's 12th Five-Year Plan) (Atkinson, 2014; Sung & Chang, 2013) during the next decade. Ultimately, IoT is expected to include 26 billion connected units, and incremental revenue generated by suppliers of IoT products and services are expected to exceed US$ 300 billion, mostly in services, by 2020 (Gartner, 2013). This will result in US$ 1.9 trillion in global economic value-added through sales in diverse end markets (Gartner, 2013).

By adoption of the methodology of roadmapping (Cheng et al., 2014), a roadmap of developments in healthcare management by IoT technologies (e.g. Radio Frequency Identification, RFID and Wireless Sensor and Actuators Network, WSAN) in the period 2010–2020 was developed in this paper, as shown in Figure 1. Many researchers have been active in healthcare restructuring that leverages IoT technologies in medical asset management (Lee & Palaniappan, 2014; Ng et al., 2014), optimizing medical resources (Jara, 2014; Xu et al., 2014a; Jara et al., 2010), monitoring healthcare situations (Shahamabadi et al., 2014; Sung & Chang, 2013; Castellani et al., 2012; Istepanaian & Zhang, 2012; Luo et al., 2012; Sung & Chiang, 2012; Istepanian et al., 2011), and increasing the use of home healthcare (Monares et al., 2014; Sebestyen et al., 2014; Yang et al., 2014; Yang et al., 2014a; Pang et al., 2013; Tarouco et al., 2012). Lee and Palaniappan (2014) developed an RFID-based inventory management system (RFID-IMS) for tracking medical devices' utilization and managing their inventory levels using real-time data. Jara (2014) conducted a feasibility study for the use of RFID/NFC (Near Field Communication) technologies for improvement of quality assurance in drug identification. Shahamabadi et al. (2014) proposed a solution to establish a hospital wireless network (i.e. mobile network) using 6LoWPAN technology for healthcare monitoring in the IoT environment. Xu et al. (2014a) demonstrated a resource-based data model to store and access the IoT data to support decision-making in emergency medical services. Yang et al. (2014) proposed an IoT-based intelligent home-centric healthcare platform (iHome system) leveraging RFID as well as WSAN technologies for in-home healthcare services.

OPEN IN VIEWER

Figure 1.

Roadmap of Healthcare Industry by Internet-of-Things (IoT) Technologies (2010–2020)

1.2 Asset Management in Hospitals

Apart from using IoT to manage the inventory levels of medical assets, the role of demand forecasting in this context has increased significantly due to various innovative and effective concepts of forecasting and inventory management, which have helped greatly in keeping the cost of hospital operations under control (Kelle et al., 2009). Managing inventory levels is important in the operation and management of the hospital's assets. Hospital operators have to review in-patient flow constantly in order to make decisions about resource capacity. In-patient care is one of the main drivers of demand for resources in hospitals (Broyles et al., 2010). In-patient systems have very complex throughput systems that make medical inventory planning very complicated. Factors of the in-patient flow process, such as non-stationary arrival and varying medical processes, make current static forecasting models rather obsolete (Lee et al., 2011), as they do not capture the compound behaviour of the true in-patient system. However, mismanagement of resources has a considerable impact on the lives and wellbeing of the patients being served. Forecasting plays a critical role in medical inventory management. The challenge that most hospital managers face is the lack of visibility and integration of already-present data, i.e. data that are routinely collected but stored in different information systems, into useful demand forecasting that can help improve medical inventory management (Parker & DeLay, 2008). Current medical inventory management systems can be divided into four main conceptual components: physical infrastructure, inventory planning and control, information system, and organizational embedding (de Vries, 2011).

Due to the huge amount of medical items and humanintensive working processes, current systems cannot provide timely and accurate inventory management and forecasting. To improve this situation, the future of inventory management will involve building up an automated work-flow system that requires minimal manual interaction. Medical amenities' replenishment requirements will be aggregated and orders placed automatically. The usage data will also be recorded for use by the hospital management in predicting future demand.

1.3 Forecasting Methods for Asset Management

Improvements in statistical models and forecasting techniques have enabled investigation of complex throughput and the modelling of ideal inventories and inventory management by processing data inputs. Current time-series methodology first attempts to identify forecasting parameters such as trend cycle, seasonality and irregularity, and then extrapolates these components to come up with the forecasts. However, these trend-cycle and seasonal data components of a time forecast tend to evolve over time, and need to be continuously revised for higher accuracy in forecasting. In addition, a key assumption in the time-series forecasting model is that the activities responsible for influencing the past will continue to influence the future. This is often a valid assumption in forecasting short-term demand, but falls short when attempting to forecast for a long-term analysis (Chase, 2013).

The artificial neural network (ANN) is an analytical learning method inspired by biological nervous systems such as the human brain. A large number of interconnected neurons, each one only responsible for performing a simple task, allow for performance of much more complex tasks, such as voice and image recognition, with high speed and accuracy (Zou et al., 2008). The ANN is very closely based on the idea of the biological neural network in that it is formed of interconnected nodes analogous to neurons. Each neural network comprises three critical components: node connectivity, network topology, and learning rules (Hansen & Nelson, 2003). A neural network forecast is proposed to handle the deficiency. This uses analytical methodologies that make use of historical demand data as inputs and updates information over time as the number of training datasets provided is increased (Lee at al., 2011). The adaptive and learning abilities of this neural network improve the forecasting accuracy so that better decisions can be made.

Fuzzy logic (FL) is a form of logic used to formulate ‘approximate reasoning’, by which the true values represented by FL fall between the discrete false (0) and true (1) (Biacino & Gerla, 2002). An FL-based system consists of four main components: fuzzification processer, fuzzy rules, inference engine, and defuzzification processer. The fuzzification processer converts crisp inputs into fuzzy sets by using the linguistic variables and membership functions. Fuzzy rules are used to determine the relationship between the inputs and outputs of a fuzzy system. The rules are expressed in the form of IF-THEN rules and defined based on the knowledge of experts or experimental outcomes. The inference engine performs inference based on the fuzzy rules to generate fuzzy outputs. The fuzzy outputs are then converted to non-fuzzy or crisp outputs by the defuzzification processer (Mohd Adnan et al., 2015). FL has been applied in many applications in inventory management. Demand forecasting is one such application that has been of particular interest to many researchers. Petrovic et al. (2006) proposed an FL-based decision support system for product demand forecasting. This system incorporated subjective forecasts and statistical forecasts to deal with the issue of uncertain and imprecise historical data and improve the accuracy of the forecast result (Petrovic et al., 2006). Candan et al. (2014) presented a neuro-fuzzy model for product demand forecasting in the pharmaceutical industry, considering product sales statistics from the past six years. In the literature on demand forecasting, most of the forecasting methodologies are based on historical statistics or comments from experts, or both. Almost all the mentioned studies focus only on the amount of demand. Few researchers have rigorously considered other factors that may directly or indirectly affect the demand, and adopted these in their models.

1.4 Summary

On the whole, the existing methods offer considerable support for decision-makers implementing demand forecasting for asset management. Although many previous studies have demonstrated that medial asset management can be implemented successfully using ANN and FL approaches, little attention has been paid to medical asset management leveraging IoT that integrates ANN and FL approaches. In order to address the key issues found in the existing methods, this paper aims to present the design and development of an IoT-based healthcare asset management system (IoT-HAMS) which incorporates ANN and FL approaches for forecasting of demand for medical assets in hospitals. By taking advantage of ANN and FL approaches, the ANN forecasting approach aims to address asset demand forecasting in normal conditions (i.e. daily and regular operations), whereas the FL forecasting approach is concerned with asset demand forecasting in abnormal conditions (i.e. ad-hoc, unexpected or emergency operations).

2.1 Information Collection

In the perspective of system implementation, every asset should have an RFID tag, used as its identity, and some assets that require careful monitoring should also have wireless sensor nodes. In a building containing the assets, the RFID gateway should be deployed at each main entrance and there should be sufficient WSN base stations to cover the building. As the scope is different, the deployment strategy will not be discussed in this paper. When the infrastructure of the system is ready, the meta-data (e.g., access time, location and conditions) of the assets will be collected automatically through IoT middleware. The IoT middleware operates as a smart agent. On the one hand, the IoT devices connect and send the raw data to the middleware, and the middleware then passes the data to the back-end system after performing data pre-processing (e.g. aggregation, filtering and normalization). On the other hand, the back-end system is able to configure and control the IoT devices in a straightforward way through the middleware. Apart from the data provided by IoT devices, another information source is the professionals themselves, such as doctors, nurses and organization operators. In the healthcare industries, especially hospitals and nursing homes, unexpected or emergency conditions may occur at a high rate. A computer system therefore may not react to these conditions instantly or appropriately. Therefore, the proposed system provides a GUI for professionals to finetune the asset demand forecasting result based on real situations, using their professional knowledge and experience. Professional judgements are considered via editing and adding fuzzy rules into the system, and the new rules can be executed instantly. This approach helps to refine the result for decision support and enable better reaction to the unexpected conditions.

2.2 Information Persistence and Utilization

Through the Internet or Intranet, the information from the IoT devices and the professionals will finally reach the core management engine of the proposed system. The engine works as a server to interact with the main modules of the system. The collected information will then be transferred to the cloud database for storage. The cloud database allows practitioners to link other databases for, e.g. patient information, inventory information, supplier information and doctor information, in an efficient and effective way; finally, a data network will be established. Thus, more useful and relevant information can be further mined for decision support, specific analysis and forecasting.

In the back-end of the proposed system, the ANN and FL modules are designed to perform asset demand forecasting, which is an important feature of the proposed IoT-HAMS. The ANN module firstly collects the asset usage data in a certain period captured by the IoT devices from the cloud database, and then performs a preliminary calculation for demand forecasting. A detailed description of the ANN module can be found in our previous work (Lee & Palaniappan, 2014). Generally, the ANN forecasting is capable of providing better results than some other traditional forecasting techniques, such as time series and casual forecasting (Jones et al., 2008; Setzler, 2007). According to Jones et al. (2008), the SARIMA model can be used to forecast patient numbers at different facilities with competitive results compared to the ANN. Realizing the limitation of some forms of regression, Setzler (2007) also compared ANN with MEDIC and Navïe for temporal and spatial ambulance demand forecasting, and his results showed that the ANN includes mean squared errors of the four levels of granularity. However, as mentioned in section 2.1, the frequency of unexpected or emergency conditions is higher in hospitals and nursing homes than in other contexts, and the occurrence pattern may vary day by day. Since ANN training requires a long period of time and a large quantity of examples, ANN forecasting may not instantly or appropriately respond to unexpected or emergency conditions, and therefore may not provide accurate and satisfactory results in such situations. Considering this limitation of the ANN module, an FL module is proposed to refine the ANN forecasting results, in which the outputs from the ANN module are transformed into an input fuzzy set. The other input fuzzy sets include both internal and external factors that influence changes in demand, such as internal inventory level, supply lead time, and emergency level. This approach may reduce the dominance of the ANN module in the forecasting results and better take into account practical and instant conditions affecting asset demand. Through a Web-based GUI, professionals can easily modify the weight and range of these fuzzy sets. The system also allows professionals to add new input fuzzy sets and membership functions, and to define new rules when new associations between the identified or new factors and the asset demand are found. Finally, the forecasting results and the input fuzzy sets specified by professionals will be output to the GUI for decision support. Consequently, the decision-makers may not only consider the forecasting results for their decisions, but also the judgments of professionals.

By leveraging the capabilities of IoT technologies, the proposed system increases the visibility of asset location and status, thereby streamlining the process of asset allocation and improving the asset maintenance strategy. For some important assets such as blood bags and medicine, the system can provide a 24/7 monitoring service. It assists identification of items with high potential for deterioration, prompting managers to work out timely measures to remedy the situation. Furthermore, with the asset demand forecasting of the ANN module, managers can accurately control the supply requests in the manner of either purchase or borrowing. At the same time, with the adjustment of the FL module, the system automatically reserves an inventory buffer to dilute the effect of unexpected or emergency conditions. The proposed system helps save time, money and other resources so nurses and physicians can concentrate on their business of delivering high-quality services to the public.