|Home | About | Journals | Submit | Contact Us | Français|
The use of telemedicine for diabetes care has evolved over time, proving that it contributes to patient self-monitoring, improves glycemic control, and provides analysis tools for decision support. The timely development of a safe and robust ambulatory artificial pancreas should rely on a telemedicine architecture complemented with automatic data analysis tools able to manage all the possible high-risk situations and to guarantee the patient's safety.
The Intelligent Control Assistant system (INCA) telemedical artificial pancreas architecture is based on a mobile personal assistant integrated into a telemedicine system. The INCA supports four control strategies and implements an automatic data processing system for risk management (ADP-RM) providing short-term and medium-term risk analyses. The system validation comprises data from 10 type 1 pump-treated diabetic patients who participated in two randomized crossover studies, and it also includes in silico simulation and retrospective data analysis.
The ADP-RM short-term risk analysis prevents hypoglycemic events by interrupting insulin infusion. The pump interruption has been implemented in silico and tested for a closed-loop simulation over 30 hours. For medium-term risk management, analysis of capillary blood glucose notified the physician with a total of 62 alarms during a clinical experiment (56% for hyperglycemic events). The ADP-RM system is able to filter anomalous continuous glucose records and to detect abnormal administration of insulin doses with the pump.
Automatic data analysis procedures have been tested as an essential tool to achieve a safe ambulatory telemedical artificial pancreas, showing their ability to manage short-term and medium-term risk situations.
The aim of an artificial pancreas is to calculate the optimum insulin delivery to maintain the patient in a normoglycemic state, taking the blood glucose (BG) concentration as the main input of the algorithm. An ambulatory artificial pancreas requires using the subcutaneous route, both for glucose measurement and for insulin delivery, due to the invasiveness of the intravenous route. However, the major difficulties facing closed-loop systems based on the subcutaneous route are the insulin absorption time and delays associated with subcutaneous glucose with respect to the glucose concentration in the blood.
In recent decades, two main types of closed-loop control algorithms have been employed in clinical studies: the classical feedback control method known as proportional-integral-derivative controller1,2 and model predictive control (MPC).3,4 Other techniques include adaptive control,5 adaptive inverse control,6 fuzzy control,7 neural predictive control,8 and robust control.9,10 Glucose control after meals is usually poor mainly because of the delays associated with glucose measurement and insulin action, but the control is improved when used at night under fasting conditions.3
Up until now, all the clinical studies with closed-loop algorithms have been conducted in a hospital setting under tight supervision. During clinical experiments, risk management becomes crucial in order to minimize health hazards for patients. Several mechanisms have to be considered to avoid data loss, device malfunction, or wrong decision making that might affect the patient's health. These circumstances lead to the development of appropriate automatic procedures to manage all the possible high-risk situations to guarantee patients' safety.
The complexity of a fully automated artificial pancreas makes the in-hospital evaluation scenario the only possible option over the coming years. However, research must be carried out to apply some of the artificial pancreas benefits to ambulatory scenarios, while managing all the possible high-risk situations and guaranteeing the patient's safety. The first step is to postpone the idea of having a 24-hour fully automated ambulatory artificial pancreas and to start looking for hybrid solutions that combine closed-loop algorithms with the prediction of hypo- and hyperglycemia, decision support tools, and hand-held terminals to provide patients with mobility, decision support, reminders, and feedback from health care providers.
The use of telemedicine systems for diabetes care enables assessment of the patient's condition and presents relevant clinical data for physicians to detect the need for therapy changes. The contribution of telemedicine systems to diabetes care has evolved over time. Earlier experiences with telemedicine were aimed at facilitating remote monitoring of a patient's BG levels from home through their transmission to the hospital.11–13 Most interactive telemedicine services have been developed using a distributed approach to integrate patient applications, implemented on a personal computer or a hand-held device, and medical workstations used by physicians and nurses at the hospital.14–16
Telemedicine provides an integrated approach to information technology tools, which enhances cooperation between users and information and knowledge sharing, and is able to support the infrastructure required to build a safe ambulatory artificial pancreas. The architectural design has to guarantee the interoperability between patient and professional environments and between different devices of the platform. It is recommended that open source platforms and plug-and-play hardware and software connectivity systems be included to make middleware development easier.17
The latest generation of telemedicine platforms implements distributed architectures that spread the users' interaction mechanisms and integrates advanced systems based on more powerful, portable, and easy-to-use terminals and applications for patients, such as electronic diaries implemented in Web applications,18–20 mobile phones,21 or smart personal assistants22 to register BG, insulin, diet, physical exercise, and so on.
Some electronic data management experiences that could be useful for risk management can be found in the literature: (1) the use of predictors of hypo- and hyperglycemic events that help the user anticipate his/her actions by predicting near-future BG values,23–25 (2) the automatic generation of alarms after the detection of anomalous situations,26 (3) decision support tools to help professionals in therapy planning,27 (4) decision support tools for patients, such as computer-assisted insulin delivery systems,28–33 and (5) clinical reminder systems that have been studied extensively34 showing positive effects in diabetes. Our approach combines several of these strategies through a telemedicine system. The aim is to hasten the implementation of a telemedical artificial pancreas.
This article describes the risk management and supervision procedures implemented in the Intelligent Control Assistant system (INCA) telemedical infrastructure to support a robust and safe artificial pancreas for ambulatory use. The preliminary results of short-term and medium-term data analyses are reported.
The INCA is a telemedical architecture that integrates a personal digital assistant (PDA)-based personal assistant for patients, which manages a continuous glucose monitoring (CGM) sensor and an insulin pump. The telemedicine system supports several loops of control and offers Web-based access to CGM and continuous insulin infusion data to diabetic patients and physicians.35
The INCA concept defines four control strategies, each of which is supported by a special setup of the personal assistant:
Implementation of these personal and remote control strategies led to development of a robust system provided with real-time bidirectional communication for remote interaction with the patient's medical devices from either the patient's personal network or long distance from the hospital.
The personal assistant runs in a PDA using a mobile network to access the remote loop and supports tele-monitoring, telecare, and remote information services.22 The personal assistant is able to work as a stand-alone system, supported by its own local application and data repository, and communicates with different medical devices through a personal wireless network that provides the patient with mobility and independence in his/her daily life.
The personal assistant is able to act upon patients' local requests for information retrieval and medical device operation and upon remote requests originated by physicians. It interacts with remote components of the telemedical system at the telemedicine central server without intervention of the patient.
Communications are activated on user demand in two scenarios: (1) to force bidirectional data exchange between the telemedical information system and the personal assistant and (2) to carry out remote control of medical devices through the telemedicine server when demanded by patients or physicians through Web access. The mobile personal assistant is the central user device for the control of patients' conditions, for adjustment of medical parameters, and for communication with physicians.
The INCA platform has been designed following a modular approach that makes the integration of different medical devices and/or control algorithms possible. The system has integrated communications with three different medical devices (glucose meter OneTouch Ultra®, LifeScan) that communicates via a serial cable or Bluetooth™, Disetronic D-TRON™ plus insulin pump (Disetronic, Burgdorf, Switzerland) that communicates with the personal assistant using the infrared port, and a continuous glucose sensor prototype based on micro-dialysis that communicates via Bluetooth. The personal assistant integrates a closed-loop control module that implements an algorithm based on a nonlinear MPC with Bayesian learning that has been tested previously.36
The four control strategies have been evaluated technically in the laboratory with prototype devices. Additionally, we completed two clinical experiments37,38 that tested control strategies #1 and #2. Their results provided useful information about difficulties in the use of personal assistant technology and the impact of CGM on patients' metabolic control. The clinical evaluation of strategies #3 and #4 remains to be completed due to the current reliability and availability of CGM sensors and insulin pumps with real-time reading and remote control in ambulatory conditions.
The architecture of the telemedical system implements an automatic data processing system for risk management and decision support that exploits available monitoring data. The ADP-RM system is very flexible, allowing the use of different communication channels with the users. In case of abnormal situations, alarms are notified through short message service messages, email messages, and/or Web-based messages. The mode of alarm reception is configured by the user according to his/her preferences or depending on the degree of importance.
In our case, data analysis is performed on two different timescales: (1) short-term risk analysis and (2) medium-term analysis of the patient's metabolic state.
Validation of the ADP-RM procedures was begun during the INCA clinical evaluation. Subsequently, validation was carried out retrospectively with experimental data collected from the previous clinical experiments. The two randomized crossover INCA studies (length: 4 weeks + 4 weeks for each experiment) included 10 type 1 pump-treated diabetic patients from Hospital de Sant Pau (Barcelona, Spain).
The first clinical experiment was devoted to comparing the use of the telemedicine system in supporting control strategies #1 and #2 versus traditional practice. Patients' decisions were based on BG self-monitoring and continuous insulin monitoring.37 The second clinical experiment evaluated the clinical utility of control strategies #1 and #2 combining real-time CGM and continuous insulin monitoring. The design of the control phase was similar to that of the intervention phase in the first clinical trial. In the intervention phase, the 10 patients used the personal assistant and the telemedicine platform. Additionally, the patients had to wear a CGM sensor (Guardian™, Medtronic, Northridge, CA) in ambulatory conditions for 3 days a week for a total period of 1 month. This second experiment demonstrated the clinical benefits of real-time CGM together with the INCA system.38
The second stage for ADP-RM validation comprised a retrospective generation of alarms using experimental data recorded during the clinical experiments. Validation was performed with 40 CGM files downloaded from 10 type 1 diabetic patients using the Guardian 3 days a week for a period of 4 weeks. Data regarding the insulin administered were downloaded automatically from D-TRON Plus insulin pumps (Disetronic, Burgdorf, Switzerland) using the personal assistant. Pump data reception was simulated with a time window of 2 weeks.
This section presents some validation results and examples for three automatic data processing methods: (a) assessment of patient's control, (b) detection of hypoglycemic events and pump interruption, and (c) detection of glucose sensor failure.
During the first INCA clinical experiment,37 the ADP-RM analyzed capillary BG measurements and generated 62 alarms for the physician, most of them because of hyperglycemia (above 300 mg/dl) in two consecutive readings (56%), with no differences between the two study periods. Those alarms contributed to 44 remote therapy changes prescribed through the telemedical platform. Reminders to send BG data allowed frequent data transfer from patients to the physician in charge (3.27 ± 1.1 weekly transmissions of glucose meter data per patient). After the experimental phase of the clinical experiment, fructosamine decreased significantly (393 ± 32 vs 366 ± 25 μmol/liter; p < 0.05) and HbA1c tended to decrease (8.0 ± 0.6 vs 7.78 ± 0.6; p = 0.073), while no changes were observed during the control phase.37
The following presents results for the retrospective analysis:
Figure 2 shows an example of pump interruption in a 30-hour simulation of closed-loop control.6 Insulin infusion is interrupted when the glucose level falls below a threshold (G ≤100 mg/dl) and is resumed when the glucose trend is positive and rises to the pump reactivation threshold (G ≥80 mg/dl). Infusion of an isolated microbolus can be observed when the pump is interrupted for more than 60 minutes in order to avoid obstruction of the catheter. Pump interruption prevents hypoglycemic events, although future work needs to address the aim of achieving normoglycemia after pump suspension.
Figure 3 shows, as an example, a simulation of how the detection of glucose sensor failure could be managed when the patient is wearing two glucose sensors. In two different situations, the glucose values of the CGM sensors differ by more than 25% and the patient is notified to get a capillary BG measurement. In the first event, the BG reading would confirm that sensor #1 is still working within the acceptable limits so no action would need to be taken. In the second event, it would be necessary to change to the measurements of sensor #2 while the patient is notified to recalibrate sensor #1. It would not be necessary to disrupt the closed-loop algorithm because of sensor failure.
This article focused on the role of automatic data processing methods that contribute to achieving a safe ambulatory telemedical artificial pancreas. Use of the INCA platform with always-on mobile networks enables the periodic update of data from the patient scenario to the telemedical server and the performance of short-term and medium-term risk analyses.
The ADP-RM system supports everyday decisions made by ambulatory patients and it also secures decisions during closed-loop experiments. The system can filter huge amounts of monitoring data and assesses the patient's metabolic control, improving physicians' decision making. The ADP-RM aids in the early diagnosis of anomalous situations by generating automatic alarms when a departure of the patient's parameters from predefined ranges is detected. Alarms for BG measurements have been tested successfully in clinical experiments, as well as retrospectively, on the basis of CGM records and continuous insulin infusion data.
The INCA platform can also serve as a complement for in-hospital closed-loop control experiments to manage critical situations, such as severe hypoglycemias or device malfunction that might pose a risk to the patient. It represents an alternative to make clinical experiments more flexible and safer, requiring less supervision and helping pave a safe path for development of the ambulatory artificial pancreas.
Further clinical evaluation of the ADP-RM methods presented is needed to demonstrate their impact on patient control. Our current efforts are focused on (1) validation of the run-to-run bolus calculator together with telemedicine to improve patients' daily decisions, (2) implementation and evaluation of prediction tools in order to determine their ability to avoid situations of hypoglycemia and hyperglycemia by using CGM measurements, and (3) implementation of multiparametric analysis to extract better conclusions about the patient's metabolic control by combining the information provided by different medical devices.