Patient monitoring systems collect and display a wide array of physiological data. The successful introduction of intelligent patient monitoring and automated control is the only practical solution to the problem of information overload produced by the increasing number of sensors developed for physiological monitoring.
by Dr Mark Ansermino
Physiological monitors have become ubiquitous in healthcare. Initially restricted to use in critical care environments, they are now increasingly deployed on wards and even at home. The modern critical care environment contains a sophisticated collection of electronics, sensors and computing systems. These monitoring systems collect and display a wide array of physiological data [Table 1]. In this chaotic arena of continuously competing noises, alarms, signals and monitors, clinicians must determine the status of the patient and identify and respond appropriately to critical events. This clinical setting is in urgent need of decision support. Events are typically life-threatening and require immediate attention. An enhanced monitoring system is required to reduce the cognitive workload of the stressed clinician and ease the decision-making process in critical care environments.
Even in less resource-intense environments, the monitor does not have the undivided attention of the expert clinician. On the ward the monitor is occasionally observed by a clinician with a lower level of expertise, who is distracted by the responsibility for the care of many more patients. At home the system may be monitored by a remote clinician, but more frequently by the family. The life-threatening critical events in such situations are identical to those seen in the critical care environments but are typically less common. The false alarm rates in the ward and home setting are usually very high, leading to a low compliance and reliance rate for alarms. A more intelligent monitoring system is required that will more precisely identify these less frequent but still important critical events.
Limitations of current monitors
Most current monitoring systems are not intelligent. Abnormal clinical conditions are signalled by a primitive alarm system, automatically triggered when a single parameter fluctuates beyond a preset threshold, or by visually tracking changes to a signal pattern over time. The alarm systems are based on upper and lower thresholds set with the assumption that each parameter remains within the same range for each patient over time. As physiological rhythms fluctuate and the acceptable range varies for each individual, false alarms are inevitable. When combined with clinical interventions and other artifacts, the high false-positive rate renders alarms ineffective so that they are frequently ignored.
The exponential growth in the number of monitored parameters, combined with the high probability of false alarms, means that the clinician is unable to simultaneously appreciate each variable and perform other essential tasks. The current visual and auditory displays have become ineffective monitoring systems that amplify demands on human attention. A small fraction of the information collected by the monitor is appreciated by the clinician; the remainder is unused and discarded.
Less than 20 years ago monitoring in most intensive care environments was via an ECG display with a numeric value for heart rate combined with intermittent manual measurements of blood pressure. Advances in technology have resulted in an exponential growth in the number of sensors. Eight are included in the current minimum standards for monitoring under anaesthesia but more then ten further sensors are routinely available [Table 1]. In the future physiological monitoring will rely on miniaturised, wireless sensors recording multiple physiological processes non-invasively in many settings including healthcare, work, and home. This physiological information will be combined with information located in computerised records of medical history, drug therapy, laboratory and radiological testing to provide early warning of health concerns and optimised management of diseases. Increased automation, especially for safety enhancement, will become widespread to improve monitoring, diagnosis and treatment of patients.
Smart sensors are created by the automated processing of the raw physiological data to identify patterns that represent specific clinical correlates. For example, scaled, processed electroencephalographic (EEG) data can be used to provide an indication of the level of unconsciousness and guide the administration of anesthaesia. The representation of variability in a signal, such as heart rate, blood pressure or the plethysmographic trace from pulse oximetery can provide valuable information about autonomic system function or the likelihood that cardiac output will increase with intravenous fluid administration.
Automated synthesis of data
Intelligent automated analysis is the solution to reducing the vast amount of data produced by current monitors. This can include processes such as artifact rejection, sensor fusion (combining information on the same measurement from different sources), feature extraction, trend detection and integrating pre-existing clinical knowledge [Figure 1].
Display of information
Current visual displays for patient monitoring follow the single-sensor, single indicator paradigm, showing one waveform and/or numeric for each sensor. The sonification (transfer of information through the use of sound) of heart rate, by a change in the interval between tones, and oxygen saturation, by a change in pitch, are routinely used to augment information transfer during monitoring. In the future, displays will integrate information from multiple sources and display only that information relevant to clinical decisions. The visual display will provide cues to additional detailed information or for navigation to further sources of information. Additional information will be communicated by other sensory pathways such as hearing  and touch .
Expert systems are computer applications with a built-in knowledge about a specific subject. The computer can combine inputs from observations in the real world with built in knowledge to solve problems, just like a human expert on a particular subject. An expert system requires knowledge that must initially be defined by one or more clinical experts. Once the expert knowledge has been captured, the knowledge base is integrated with real world observations (facts) using a real time inference engine. Expert systems are well established for the prevention of errors in the aviation and atomic energy industries. Human reliability has been extensively studied in these industries and significant gains in safety and reliability have been demonstrated with the use of decision support tools. The implementation of expert systems in clinical settings has been limited by the complexity of the environment and lack of validation in this setting. However, such a system presents an outcome to the clinician when a pattern of observations match a specific rule. The system may also request further information from the clinician. The matching of specific observations to a rule may also trigger warnings about adherence to clinical guidelines or even automatically initiate a rescue manoeuvre.
A number of schemes have been used to represent expert knowledge in a computer. The rule based approach is probably the most widespread and straightforward. The process requires the acquisition of expert knowledge and putting it into rules that the system can use. These rules for decision-making are specific for specialised domains of knowledge. They are limited by the fixed choices and may become redundant as new knowledge becomes available. To gain widespread acceptance rules will need to be clinically validated and sanctioned. The collaborative development of the knowledge base by local, national and international groups of experts is the only practical solution. This will facilitate the sanctioning of rules and avoid unnecessary duplication of effort in developing rule sets. The rules created in an expert system should promote better decision-making by less skilled or less experienced clinicians during training or in situations where advanced training is not available. The expert system should be designed to (a) allow clinicians to easily transfer their knowledge to a rule-based advice system; (b) be intuitive to use even for the most inexperience user; (c) make all decisions visible and sanctioned by the user and (d) to function as memory aids for information that is rarely used or frequently inaccurately recalled .
Integration and adoption into clinical environments
Intelligent monitoring systems have not been adopted into routine practice. Reasons for this include a lack of understanding of the systems, peer pressure, physical unavailability and other barriers such as reduced data linkages. Poor quality outputs, unconvincing information, inability to change practice and the belief that performance was already optimal have also been cited as common reasons for lack of desire to use expert systems . There is no replacement for the expert opinion of clinicians. At most, an intelligent patient monitor can be used as a supplement to expert human performance by increasing awareness and enhancing decision-making and recall. Clinical adoption will require that the clinician has an understanding of the processes that are used to detect these events and a trust that the information delivered has value in improving clinical care. Adoption by clinicians is essential for the success of any intelligent monitoring system.
An intelligent patient monitor with smart sensors, automated artifact rejection, trend detection and built-in expertise would enhance clinician performance. A computerised system is capable of monitoring and analysing physiological measurements faster and more carefully. It could provide a rich source of knowledge that could be used to ensure adherence to clinical guidelines with real-time alerts and provide cognitive aids for information that is complex or difficult to remember. This intelligent patient monitor could promote documentation or even trigger automated events (e.g., requesting a test). Combining data from more than one sensor into an intelligent framework could significantly reduce the false alert rate and improve performance.
The successful introduction of intelligent patient monitoring and automated control is the only practical solution to the problem of information overload produced by the increasing number of sensors developed for physiological monitoring. By linking the power of computers to the richness of human experience, intelligent patient monitors enhance the value of expert knowledge by making it readily and widely available. Intelligent patient monitors will help to overcome the limitations of human attention, enhance performance of the clinician in real time, reduce adverse outcomes and most importantly save lives.
1. Watson M, Sanderson P. Sonification supports eyes-free respiratory monitoring and task time-sharing. Human Factors 2004; 46 (3): 497-571.
2. Ford S, Daniels J, Lim J, Koval V, Dumont GA, Schwarz SKW, Ansermino JM. A novel vibrotactile display to improve the performance of anesthesiologists in a simulated critical incident. Anesthesia and Analgesia 2008; 106(4): 1182-1188.
3. Dunsmuir D, Daniels J, Brouse C, Ford S, Ansermino JM. A knowledge authoring tool for clinical decision support. Journal of Clinical Monitoring and Computing 2008; 22: 189-198.
4. Liu J, Wyatt J, Altman D. Decision tools in health care: focus on the problem, not the solution. BMC Medical Informatics and Decision Making 2006; 6: 4.
Dr Mark Ansermino
Director of Research
Department of Anesthesia
4480 Oak Street,VancouverV6H 3V4, Canada
Tel +1 604 875 2711