A Radio Frequency Identification (RFID)-based wireless sensor device for drug compliance measurement

Transcription

1 A Radio Frequency Identification (RFID)-based wireless sensor device for drug compliance measurement Alan Montefiore, Dave Parry, Ann Philpott AURA Laboratory and School of Computing and Mathematical Sciences Auckland University of Technology, Auckland Abstract PROBLEM: A high level of non-compliance relating to taking of medication can lead to compromised health benefits and wasted money. Improving patient compliance has the potential for improving issues related to Cardiovascular Disorder (CVD) and many other diseases. Although there are many telecare-based home monitoring schemes for clinical measurement, current solutions for medication tracking and reminding either lack needed functionality or are much too costly to be used in a home care environment. METHOD: The Intel Wireless Sensor Platform (WISPs) are compact microprocessor based systems that communicate with and are powered by RFID readers. This was used as an aid for developing a prototype system for monitoring medication dosages in a home health care environment. A number of prototypes were developed and tested based on various methods of pill dispensing. CONCLUSION: The combination of an RFID technology the Intel Wireless Sensor Platform(WISPs) and the construction of a specific pill dispensing container in this prototype demonstrated that it is possible to use RFID technology to effectively and ubiquitously monitor and track drug taking compliance. With further refinements on the dispensing unit and optimisations in the software, this product could be manufactured and released to home care patients to help increase compliance and reduce health related issues. RFID-based technology and an array of ubiquitous home health devices may provide both support for and monitoring information that improves healthcare outcomes for people with chronic conditions. 1. Introduction A high level of non-compliance (or non-adherance) in regards to taking prescription medicine is a cause of compromised health benefits and wasted money. A large review in 2001 [1] indicated that 30%-50% of patients fail to comply completely with treatment. Compliance issues have become a major focus for research and many interventions have taken place to help reduce the impact of non-compliance. Improving patient compliance has a great potential for improving issues related to Cardiovascular Disorder (CVD) and many other important chronic diseases. There are a number of ways that people can fail to comply with prescribed treatment, from whether they actually fill the prescription to whether they take the medication according to the prescribed regimen. This study is concerned with secondary non-compliance, that is when patients have physical possession of the medication but do not follow the instructions exactly. Studying secondary non-compliance is difficult as this generally occurs within the patient s own home and it is not reflected in statistics that are routinely collected e.g. in terms of prescriptions filled. There have been many developments in information technology intended to allow ubiquitous monitoring of a patient s health in their own homes, however demonstration of benefit is still elusive [2]. These approaches may be particularly beneficial for patients who have to live on their own and/or patients who can t make regular trips to the hospital [3]. Current demographic trends from Statistics NZ show that New Zealand s population in the 65+ age bracket is going to increase by 30,000 in the next five years. (Statistics NZ - The issue of population ageing prompts the need for improving home-based monitoring products as the elderly population often requires long-term care. These statistics along with health care being one of the highest expenditures of the gross domestic product (GDP) justify the need for more ubiquitous and more affordable solutions to health care issues [4].

2 2. Enabling Technology This section deals with some of the technology being used in the pilot device RFID Technology Radio Frequency Identification (RFID) has become an integral part of our lives, in many application areas [5], perhaps the most familiar being the proximity access card. RFID is a short-range radio technology that is best suited for communicating digital information about specific objects. In a typical scenario there are multiple RFID Tags/Transponders which are small and inexpensive and only one RFID Reader/Interrogator. Information can be sent from the tags by generating, modulating and transmitting a radio signal the reader can detect. A number of modulation techniques can be used to enable read-only or read-write communication. The use of passive RFID tags provides many benefits. They are inexpensive, disposable, and durable. These benefits allow the tags to be placed in many objects for gathering information for structural, medical and monitoring purposes. The major limitation of traditional Ultra High Frequency (UHF) tags is that they require a close proximity to a RFID reader to send data. This limits their functionality in scenarios such as supply chains or other applications where a reader is not always present. A common alternative to passive tags are active tags which use a battery to enable processing of sensor data when away from the reader. This however introduces new issues such as battery lifespan, recharging capabilities and the use of hazardous chemicals such lithium, complicating disposal Wireless Identification and Sensing Platform (WISP) WISP [6] is a new RFID tag technology that introduces a power model that is different to standard passive or active tags. WISPs are similar to passive tags in the sense that they only use power harvested from the reader s RF signals, and like active tags they can continue collecting data away from readers. This ability is achieved by storing the harvested energy in a capacitor for later use. This is particularly useful for applications that have limited contact with readers. WISPs have been designed to run on the standardised EPC Gen 2 infrastructure. ( This allows the use of commercially available readers that are compatible with EPC Gen 2 tags to collect tag data from the WISP (Figure 1). By allowing data capture and processing to continue while not within range of the reader, these tags are potentially an extremely flexible means of collecting summary data from the environment. The WISP has three in-built sensors - temperature, capacitance and a 3D accelerometer. Data from these sensors and a general purpose voltage sensor can be read when within range of an interrogator/reader via the normal EPC gen2 air interface, after loading a simple command program into the WISP onboard memory. The reader used was an IMPINJ Speedway reader, a standard UHF Reader, connected to a PC via Ethernet. This reader has a normal operational range of around 1 metre with a 30cm square antenna. As with many other readers, multiple antennae can be attached to a single reader. Figure 1. WISP mode of action

3 2.3. Prototype development The prototype was designed to allow reliable recording of the number of pills dispensed from a container. The prototype pill counter was built based on the following requirements: Each pill being dispensed would be detected, and movement of the container that did not result in a pill being dispensed would not result in a false count Because of limitations in the development environment for the WISP on-board computer, all data collection will occurr within range of the RFID reader, with minimal on-board processing. Pill dispensing should be based on a simple operation or set of operations. The rest of this section is divided into subsections dealing with the physical design of the pill container and the software design Physical Design The development of a physical container is required to demonstrate the ability to dispense pills in such a way that it can be monitored and tracked using the WISP technology. The 3D accelerometer sensor was used to detect movement in the container whether it be a specific pattern or a difference in values from two or more WISPs. There was also the possibility of using a combination of the sensors such as using the temperature sensor to detect a hand and the 3D accelerometer to detect movement. The following sections will discuss the evaluation of a series of prototype containers First Approach The first design approach aimed to produce a container that could detect dosages by recognising a specific movement using one WISP tag. The aim was to require a maze like movement before a pill was released such as tilt down, then up, then down. This could be easily detected by the software. The initial approach was to construct an alternative lid that had an inbuilt maze trap as seen in Figure 2. The first noticeable flaw found when constructing this was it would be difficult to ensure only one pill was released as the design was relying the corner to only allow one pill to go through. This design also had the capability of inconsistencies between actual pill count and what the software detected. If two pills managed to get into the second compartment in the maze the software would not be able to detect it. One advantage of this design if it could be engineered correctly would be that it be very easy to include a WISP chip and develop the software to detect the specific movement. However due to this simplicity it would also be prone to errors in detection when the container was just being shaken around, etc. Figure 2. Pill Container with maze type lid

4 Figure 3. Pill container with catchment The second design was similar to the first design in regards to the detection of a specific movement; however the design of the pill dispensing mechanism was much more advanced and reliable. This design incorporated a catchment area that could only fit one pill. The sequence of how this idea works can be seen in Figure 3. The idea was that when the container was turned on end all the pills would fall into the top section, then when turned back up-right they would all fall back apart from one which would be trapped in the catchment, then when inverted again the trapped pill would be free to fall out of the cap. The major problem with this design was the inability to construct it as precisely as needed for it to function correctly. Often the pills would not get trapped in the catchment, or more than one pill would escape when the container was inverted the second time. This design was again using the WISP tag to detect a specific movement (down, up, down) Second Approach Both these designs were unsatisfactory. However, another approach based on a lolly dispenser that had a more complex mechanism yet was still low cost seemed promising. We used the Smint dispenser as seen in Figure 4. This design could not utilise a single WISP for movement. The incorporation of two WISP tags allows the application to detect differences between their values, so when the button is pressed one of the WISPs rotates with respect to the other. The difference in values is detected by the system and if the difference is great enough it can be inferred that a pill was dispensed. The angle noise of the accelerometer is +/- 0.6 of a degree when stationary so the amount of rotation has to be greater than 2 degrees to reliably determine that the button was pushed. Any gross movement of the container will be picked up by both WISPS and hence will be cancelled out in the processing phase. To achieve the ability to rotate the rotating WISP we built an internal mechanism that forces the WISP to rotate around the point specified in Figure 5. The spring ensures the WISP locks back into place when the button is released; the bar is used for holding one side of the WISP up while the other drops. A second WISP tag is placed anywhere on the container and calibrated to return the same values during gross movement of the whole devices as the rotating tag. This design can handle being dropped and shaken around as the detection is based on differences rather than individual movements. Some considerations taken into account were isolating the sensors from elements that might affect reliability of results and distinguishing between actual dispensing of pills and general handling of the pill container. Figure 4. Smint dispenser. (

5 Figure 5. How the internal tilting mechanism works. (only one WISP shown for clarity) Software development Process The software development was fairly simple involving modules to calibrate the two WISPs, a module to read the unique tag IDs and record the angle as derived from the onboard sensors, and a calculation module to calculate the difference between the two angles and count pills dispensed. The pills dispensed algorithm is based on an angle difference of more than 6 degrees occurring, and then returning to less than this figure. 3. Results For testing the system as a whole (including container, static and dynamic tags, software and users) a number of tasks were used to ensure the system provided reliable results in all situations. The data collected from each test included a test number, software detected result (binary), actual result (binary), tilt value of static tag, tilt value of dynamic tag and the tilt difference. When evaluating the results the software detected result and the actual result pair were studied to see how closely the samples agree. The initial test was completed with a known number of pills using very deliberate movements when pushing the button. This was designed to test the functionality of the container and software when being used in ideal conditions. Each test was performed separately with values being recorded after each press. Results are shown in Table 1 for: Detected Result: Whether or not the software detected that a pill had been dispensed. Actual Result: If a pill was dispensed from the device. Tilt Static Chip: The tilt value returned from the WISP for the tag that does not move. Tilt Dynamic Chip: The tilt value returned from the WISP for the tag that rotates when the button is pressed. Difference Tilt: The difference between the tilt values. Used for determining if the button has been pushed. The second stage of testing the application and pill dispensing container was to get three people to use the device without giving them direct instructions on how it works. They were instructed to dispense five pills (lollies) at a random interval. The attempts from each user can be seen in Table 2. The results from the tests show that the software and container were performing as expected. With a 100% success in detection under general usage it can be seen that the device performs reliably. An interesting figure that was extracted from the test results was the mean of the tilt difference. Knowing the mean of the tilt difference allowed us to double check that the values that were coded into the software were correct. With a mean tilt difference of 38, we were confident that the detection range (30-50) was accurate and would provide reliable results.

6 Table 1. Initial calibration Test Detected Result Actual Result Tilt Static Chip Tilt Dynamic Chip Difference Tilt Table 2. Test user results Test Detected Result Actual Result Tilt Static Chip Tilt Dynamic Chip Difference Tilt Table 3. Limits to dispensing angle Test Detected Result Actual Result Tilt Static Chip Tilt Dynamic Chip Difference Tilt

7 Figure 6. Scatter plot showing relationship between the tilt values of the static and dynamic tags, The red dots represent the extreme values at which dispensing is physically possible After plotting a scatter graph of the static vs. dynamic tilt values as seen in Figure 6. It was observed that user three (tests ) was holding the device on a downwards tilt to dispense the pills where as other users all held it upright. It was reassuring to see that the tilt difference was still well in the range limits. These results prompted tests around the limits of when a pill would be dispensed, from the physical mechanism for example if the dispenser was held upside down a pill would not be dispensed. The following two tests (Table 3) show the extreme boundary conditions, for the angle at which the dispenser was held. These values were introduced into the software to ensure there are no false positives when the device is tilted to far before the button is pushed. Test 14 represents the backwards tilt limit, and test 15 for the forwards tilt limit. The boundary values have been added to the scatter graph to show the range in which all recorded results should fall (points in red). The one issue surrounding the boundary values is the 0.6 degree variance in the data returned from the WISP. The boundary conditions in the software have been set to the inner most value of each boundary because actually falling within the 0.6 degree variance in both tags is very slim. To check the degree of the relationship between the static and dynamic tilt values, i.e. whether the sensors would still work within the range in which dispensing was possible, a regression plot was constructed (Figure 6). After viewing the plot it was possible to see that the relationship was between the variables was positive as higher static values pair with higher dynamic values and vice-versa. A t-test and linear regression were performed. The results showed a p< This gave us confidence that the devices would record dispensing accurately during any physically reasonable dispensing process. 4. Conclusions The results from the tests show a completely reliable software detection result. This result was obtained from a series of four users including one of the authors. After plotting and analysing the data it was clear to see that all the results form a linear pattern that is determined by the degree of tilt of the device. The correlation result of 0.97 showed a very strong positive relationship between the static and dynamic tilt variables, during dispensing. The combination of RFID technology (WISPs) and the construction of a specific pill dispensing container in this prototype has proved to demonstrate that it is possible to use RFID technology to effectively and ubiquitously monitor and track drug taking compliance. With further refinements on the dispensing unit and optimisations in the software this product could be manufactured and released to home care patients to help increase compliance and reduce health related issues. Because the WISP devices can include other sensors, and record data, this approach would allow identification of which drug is being dispensed. This could be extended to identifying the user doing the dispensing, for example if the user was to wear one as a wristband, during the dispensing. As the WISP continues to record data even when it is outside the

8 detection range of the RFID reader, this is a potentially very flexible device. Other sensors such as heat or light detection could also be used to gain a better understanding of the storage conditions of the drugs. Because the RFID detector is a standard type, this could also be integrated into a drug reminder/identification system [7], or systems that might be used for other sorts of homecare such as those designed to support the visually impaired. Cost of the system depends on the cost of WISPs which are not currently commercially available, as well as the RFID detector and associated computer system. The RFID detector used (Impinj) is accessed via an Ethernet connection, so it is possible that wireless connection to this device could be established. These devices may become part of a home healthcare hub that would support both assisted living technology, and the recording of appropriate clinical data. Such systems are being advanced for particular user groups such as people with neurological impairment [8] By building devices that use common standards and that have low infrastructure requirements, the acceptance of these devices is more likely. A comparison can be made with home entertainment technology, where devices can be linked in a complex and expensive, but high performing system, home theatre, or as individual items, or somewhere in-between. It may be that considering home telecare as a branch of consumer electronics, rather than as the transplantation of hospital equipment, clinical transformation is more easily enabled. 5. Acknowledgments The authors would like to acknowledge the generous support of the INTEL WISP Challenge team and IMPINJ 6. References [1] Vermeire E, et al., Patient adherence to treatment: three decades of research. A comprehensive review. J Clin Pharm Ther, : p [2] Martin S, et al., Smart home technologies for health and social care support. Cochrane Database of Systematic Reviews:, (4). [3] Choi JM, et al. A System for Ubiquitous Health Monitoring in the Bedroom via a Bluetooth Network and Wireless LAN. in Engineering in Medicine and Biology Society, IEMBS '04. 26th Annual International Conference of the IEEE [4] Otto C, et al., System architecture of a wireless body area sensor network for ubiquitous health monitoring. Journal of Mobile Multimedia,, (4): p [5] Landt J, The history of RFID. Potentials, IEEE, (4): p [6] Sample AP, et al., Design of an RFID-Based Battery-Free Programmable Sensing Platform. Instrumentation and Measurement, IEEE Transactions on, (11): p [7] Houliston B, Merry A, and Parry DT, Sensors and Insensibility: Monitoring Anaesthetic Activity with RFID, in Health Informatics New Zealand Forum. 2008, HINZ: Rotorua. [8] Gentry T, Smart homes for people with neurological disability: State of the art. NeuroRehabilitation, (3): p