Design and validation of an intelligent patient monitoring and alarm system based on a fuzzy logic process model

https://doi.org/10.1016/S0933-3657(97)00020-1Get rights and content

Abstract

The process of patient care performed by an anaesthesiologist during high invasive surgery requires fundamental knowledge of the physiologic processes and a long standing experience in patient management to cope with the inter-individual variability of the patients. Biomedical engineering research improves the patient monitoring task by providing technical devices to measure a large number of a patient's vital parameters. These measurements improve the safety of the patient during the surgical procedure, because pathological states can be recognised earlier, but may also lead to an increased cognitive load of the physician. In order to reduce cognitive strain and to support intra-operative monitoring for the anaesthesiologist an intelligent patient monitoring and alarm system has been proposed and implemented which evaluates a patient's haemodynamic state on the basis of a current vital parameter constellation with a knowledge-based approach. In this paper general design aspects and evaluation of the intelligent patient monitoring and alarm system in the operating theatre are described. The validation of the inference engine of the intelligent patient monitoring and alarm system was performed in two steps. Firstly, the knowledge base was validated with real patient data which was acquired online in the operating theatre. Secondly, a research prototype of the whole system was implemented in the operating theatre. In the first step, the anaesthetists were asked to enter a state variable evaluation before a drug application or any other intervention on the patient into a recording system. These state variable evaluations were compared to those generated by the intelligent alarm system on the same vital parameter constellations. Altogether 641 state variable evaluations were entered by six different physicians. In total, the sensitivity of alarm recognition is 99.3%, the specificity is 66% and the predictability is 45%. The second step was performed using a research prototype of the system in anaesthesiological routine. The evaluation of 684 events yielded a sensitivity, specificity and predictability of the alarm recognition of more than 99%.

Introduction

Biological systems consist of countless self-organised structures, and the underlying characteristics of their interaction are often not completely understood. Most of the physiological systems are described quantitatively by techniques that were originally developed for linear systems and control theory. These quantitative, analytical models can be used to improve the general understanding of the functionality of a system and to simulate general dynamic system behaviour, but for the diagnosis of an individual patient's disease these classic quantitative methods are not sufficient. An adaptation of these models for diagnostic purposes in an individual patient is very difficult to perform because there are inter- and intra-individual differences between patients, and most of the state variables that are needed for these models cannot be measured directly.

The field of artificial intelligence in medicine can provide new modelling techniques to describe and handle these complex systems in a more generalised way. In medical domains, experienced physicians diagnose the patient state by applying their expert knowledge and inferring information and data from different sources. This complex diagnostic task can be supported by modern computer systems on which formal reasoning methods or soft computing techniques like fuzzy logic and artificial neural networks are implemented. These techniques provide a framework to observe and handle biological processes which are difficult to describe with a precision that is needed for conventional quantitative models [30].

Kulikowski divides the evolution of artificial intelligence in medicine into four phases [23]. During the first phase from 1968 to 1976 it was proposed that automatic decision-making would help formalise the as yet informal rules of medical practice and also help disseminate rare and costly special expertise far beyond the narrow confines of academic medical centres to the general practice settings [33]. First experiences with knowledge representations for disease processes were made with CASNET [22], MYCIN [35], DIALOG/INTERNIST [31] and PIP [29].

The second phase is dated from 1977 to 1983. During this time general frameworks for building expert knowledge bases like EMYCIN [41] and EXPERT [44] were developed and applied to medical problems like the expert system ventilator manager (VM) [14]. The system provides a model to interpret quantitative data collected from the patient and the medical devices in the intensive care unit. The knowledge is represented in form of production rules similar to those which are used in MYCIN.

In the third stage from 1982 to 1987, the researchers explored the complexity of medical reasoning, for example qualitative explanatory models of clinical reasoning and the underlying physiologic processes [21].

Since then, the fourth stage—characterised by the continuing development of knowledge-based representations using greater structure and generic task dependencies—has led to so-called `second generation expert systems', which emphasise the multilevel nature of problem solving and related interacting knowledge structures [9].

Although computer-based medical decision systems were originally introduced to relieve the physician from `encyclopaedic' aspects of medicine and to enable him to concentrate on patient care, it turned out that `consultation systems' were in fact problematic, because an experienced physician often knows the advice the system will give before all the necessary input parameters have been entered into the computer. This induced the development of knowledge-based systems which evaluate physiological parameters online and allow patient state monitoring without a special consultation or data entry step.

In the field of anaesthesia, a number of approaches to knowledge-based alarms and informative data presentation have been published [37]. Some of these intelligent monitoring systems have been validated in the clinical set-up. Most of these systems have been built as proof-of principle systems and describe approaches for ICU monitoring, for example Guardian [16], and ventilator management, or COMPAS [36], or SIMON, which was implemented in YAQ, an ontology for model-based reasoning in physiologic domains [38]. The question is whether these approaches are sufficient to support a medical practitioner and if they will be accepted. Recent studies show that especially in the case of acute respiratory distress syndrome (ARDS), complex ventilation protocols have to be performed, such as the pressure control inverse ratio ventilation (PCIRV) 8, 13, 24.

Van der Aa developed the first real-time intelligent alarm system for the detection of ventilator malfunctions [40]. The system is based on a rule-based approach and implemented with the expert system shell SIMPLEXYS [7]. Three prototypes of the system were evaluated in animal models. Orr and Westenskow implemented and evaluated an intelligent alarm system based on a neural network for the same task [28]. Both approaches to intelligent alarming, as well as other contemporary monitoring systems, are reviewed by Uckun [39].

There are some other approaches to intelligent alarm systems described in the literature. One of them is based on the determination of trends in physiological signals which are gathered from patients who are recovering from cardiac operations in the intensive care unit (ICU) [18]. These trends are generated using median filters and evaluated by a rule-based approach. This knowledge-based alarm system was evaluated at the bedside in the ICU [19].

Biomedical engineering research improved the patient monitoring task by providing technical devices to measure a large number of a patient's vital parameters. These measurements improve the safety of the patient during the surgical procedure because pathological states can be recognised earlier, but may also lead to an increased cognitive load of the physician. This was in part the result of several studies on human error in anaesthesia and intensive care 12, 34, 10.

In order to reduce cognitive strain of the anaesthesiologist and to support intra-operative monitoring of the patient, we have proposed [32] and implemented [5] an intelligent patient monitoring and alarm system which presents an automatic evaluation of the patient's haemodynamic state during cardiac surgery.

The approach described here is a problem oriented, ergonomic solution to the problem of information overload in the operation theatre based on a qualitative fuzzy process model. The process of patient care performed by an anaesthesiologist during highly invasive surgery requires a fundamental knowledge of the physiological processes and a long standing experience in patient management to cope with the inter-individual variability of the patients.

Section snippets

Materials and methods

To implement the intelligent patient monitoring and alarm system, an anaesthesiological mental model for the evaluation of the haemodynamic state of a patient undergoing cardiac surgery is necessary. To derive such a model, the decision-making process of the anaesthesiologist has to be analysed and the most important parameters and relations have to be identified.

In the daily routine, the anaesthesiologist does not consider the exact values of some input parameters, like a systemic arterial

Validation, results and discussion

The validation of the inference engine of the intelligent patient monitoring and alarm system was performed in two steps. Firstly, the knowledge base was validated with real patient data which was acquired online in the operation theatre. Secondly, a research prototype of the whole system was implemented in the operating theatre. As a part of the second step, the knowledge base was validated with respect to the anaesthesiological routine.

Conclusion

The validation of the knowledge base of the intelligent patient monitoring and alarm system showed good results in both validation steps. Compared to commercially available monitoring systems with threshold alarm functionality the concept of continuous alarms seems to be very promising.

The intelligent patient monitoring and alarm system is designed as a `self-referential' system for the anaesthetist. For legal reasons, the system must not make therapy decisions by itself, but only make

Acknowledgements

Parts of the research were supported by the German research foundation (Deutsche Forschungsgemeinschaft Ka 251/3-1)

References (45)

  • K. Becker, H. Käsmacher, K. Juffernbruch, G. Rau, G. Kalff, H.J. Zimmermann, Acquisition of a `fuzzy' knowledge base of...
  • K. Becker, H. Käsmacher, G. Rau, G. Kalff, H.J. Zimmermann, A fuzzy logic approach to intelligent alarms in...
  • K. Becker et al.

    Fuzzy logic approaches to intelligent alarms

    IEEE Eng. Med. Biol. Mag.

    (1994)
  • K. Becker, Application of quantitative and qualitative models for the implementation and validation of an intelligent...
  • J.A. Blom, The SIMPLEXYS Experience. Real time expert system for patient monitoring, Ph.D. dissertation, Technical...
  • S. Böhm, B. Lachmann, Pressure-control ventilation—putting a mode into perspective, Int. J. Intensive Care 3...
  • B. Chandrasekaran, T.R. Johnson, Generic tasks and task structured history, critique and new directions, in: J.M. David...
  • J.B. Cooper et al.

    An analysis of major errors and equipment failures in anesthesia management: considerations for prevention and detection

    Anesthesiology

    (1984)
  • L.M. Fagan, VM: Representing time-dependant relations in a medical setting, Ph.D. dissertation, Department of Computer...
  • J.S. Gravenstein, D.A. Paulus, Praxis der Patientenüberwachung, Fisher Verlag, Stuttgart, 1985, (in...
  • H. Klocke et al.

    An anesthesia information system for monitoring and recordkeeping during surgical anesthesia

    J. Clin. Monit.

    (1986)
  • E.M.J. Koski et al.

    Development of an expert system for haemodynamic monitoring: computerized symbolization online monitoring data

    Int. J. Clin. Monit. Comput.

    (1991)
  • Cited by (63)

    • Artificial intelligence in cardiothoracic surgery: current applications and future perspectives

      2023, Artificial Intelligence in Clinical Practice: How AI Technologies Impact Medical Research and Clinics
    • Glucose-insulin model identified in free-living conditions for hypoglycaemia prevention

      2018, Journal of Process Control
      Citation Excerpt :

      The methods designed to generate the alarms can be divided in two categories: low-threshold detection or prediction. Hypoglycaemia alarms based on low-threshold detection notify the crossing of a critical BG level [10], while alarms based on prediction try to forsee this risk in advance to give the user the opportunity to act ahead of time in order to avoid the event [11–14]. The latter typically requires the use of patient models to perform glucose trend predictions.

    • Identifying the tool-tissue force in robotic laparoscopic surgery using neuro-evolutionary fuzzy systems and a synchronous self-learning hyper level supervisor

      2014, Applied Soft Computing Journal
      Citation Excerpt :

      These researches range from the applications of soft techniques in surgery and medical robotic to verifying the effects of chemical substances on human's health. Becker et al. [14] used fuzzy logic to design an intelligent patient monitoring and alarm system for high invasive surgery. It was reported that the developed intelligent tool can cope with the inter-individual variability of the patients.

    • Developing a continuous monitoring infrastructure for detection of inpatient deterioration

      2012, Joint Commission Journal on Quality and Patient Safety
      Citation Excerpt :

      In the longer term, the best improvements to alarm sensitivity and specificity are likely to come from algorithms that improve on the current static threshold approach. Machine-learning techniques have been successfully applied to clinical data in other areas,5,14,26,27 and the large datasets collected by inpatient surveillance systems are natural targets for statistical analysis and data exploration. Large physiologic databases allow the development of deterioration-detection algorithms and machine-learning analysis techniques on real, albeit retrospectively collected, physiologic data without interfering with patient care.

    • A review of current and emerging approaches to address failure-to-rescue

      2011, Anesthesiology
      Citation Excerpt :

      This class of analysis methods can extract time- and frequency-domain relationships between physiologic data and deterioration that are not necessarily evident in statistical approaches nor to standard clinical assessment. A wide range of machine-learning techniques has been applied to clinical data in the past, including fuzzy logic,68 neural networks,69,70 patient-specific trained classifiers,71 wavelets,72 syntactic analysis,73 Bayesian networks,74 and Markovian models.75 None of these studies had data on non-ICU general ward patients who underwent continuous monitoring.

    View all citing articles on Scopus
    View full text