Evaluation of the expert system for respiratory therapy of newborns on archival data.

The aim of this study was to evaluate the artificial ventilation expert system for neonates (AVES-N) using archival data. The recommendations of the system were compared to the decisions made by the expert-physician in the same clinical situation (patient condition, respirator settings). In our retrospective study we used data of 320 newborns which were ventilated in the Neonatal Intensive Care Unit of the Vanderbilt University Hospital in Nashville (USA). Best agreement between the recommendations of the system and the decisions of the experts was found for positive end expiratory pressure (PEEP), inspired oxygen fraction (FiO2) and peak inspiratory pressure (PIP)--about 70%. Worse agreement was found for time related parameters: respiratory frequency (f) - 54%, time of inspiration (ti) - 46%, time of next blood gas analysis - 15%. The expert system advised lower FiO2 PEEP and f. The differences were smaller in a group of patients who survived than in a group of patients who died. The overall agreement of the AVES-N advice and real therapeutic actions leads to the clinical evaluation of the expert system. The differences can be attributed to a) different therapeutic strategies at 2 NICU's, b) missing data regarding complications in the data base which were not taken into account by the expert system.

Current status of mechanical ventilation decision support systems: a review.

Objectives of computerized decision support systems for mechanical ventilation are discussed. Questions considered are: Why is computerized decision support for mechanical ventilation important? What parameter(s) should be optimized? What are the differences between a single attribute and a multiattribute value function used for optimization? How is it possible to achieve optimization in clinical practice with existing ventilators? How does one solve the problem of acquiring measurement of data needed for closed loop control? The possibilities and limitations of three existing decision support systems are discussed. 1) Computerized protocols from LDS Hospital in Salt Lake City, Utah, USA. 2) Optimization Program (OPTPROG) developed jointly at the Institute of Biocybernetics and Biomedical Engineering, Polish Academy of Sciences, Warsaw, Poland and Medical Intensive Care Unit, Department of Medicine at Karolinska Institute, South Hospital, Stockholm, Department of Medical Informatics Linkoping University, Sweden. 3) Ventilator Therapy Planner (VENT-PLAN) from the Section on Medical Informatics at Stanford University, Palo Alto, California, USA. Strategies leading to an optimal computerized decision support system are proposed. These strategies include development of better measurement methods for blood gases and cardiac output, improvement of man-machine and machine-machine interaction and the selection of optimization criteria. Finally, research directed towards building quantitative, dynamic patient models based on computerized databases of mechanically ventilated patients are discussed.

Statistical models for prediction of arterial oxygen and carbon dioxide tensions during mechanical ventilation.

The possibility of constructing statistical models for prediction of alveolar oxygen and carbon dioxide tensions has been investigated in 20 mechanically ventilated patients in acute respiratory failure (ARF). Linear multiple regression analysis using PaCO2 and PaO2 as dependent variables was used to construct (a) models for individual patients, (b) models for specific diagnostic groups and (c) general models (all patients). The coefficient of determination (R2) was highest for the individual patient models (0.38-0.99) and lowest for the general models (0.28-0.49). In order to achieve a high predictive accuracy, models matching individual patients should be constructed on the basis of initial invasive blood gas measurement. Statistically derived models may bring better understanding of the behaviour of factors influencing arterial gas tensions in ARF and may be of value in the management of patients on mechanical ventilation.

Multivariable optimization of mechanical ventilation. A linear programming approach.

The proposed method aims at improved ventilatory care with reduced morbidity. It combines two important aspects of mechanical ventilation: gas exchange and lung mechanics. A single criterion was selected as optimization index of lung trauma: peak respiratory power (PRP) defined as the maximum product of pressure times flow during inspiration. Arterial blood gases reflect gas exchange and constitute the constraints of the problem. The constraints as well as the optimization index are expressed as linear functions of the input variables (frequency of breathing, tidal volume, and positive end expiratory pressure). A linear programming approach can therefore be used to determine the values of input variables that minimize PRP and at the same time keep arterial blood gases within the prescribed limits. The coefficients of the constraints and the optimization index equation are found by manipulating input variables in order to obtain four different values of PaO2, PaCO2 and PRP (there are four coefficients in each equation). The coefficients can then be calculated and the optimization procedure run. In a pilot study 5 patients suffering from diseases of varying pulmonary pathology were investigated with this method. In 4 out of 5 the ventilator treatment improved in terms of blood gas values (mean increase in PaO2 was 4.7%) and reduction of mechanical load on the lungs (mean PRP reduction was 20%). Lower PRP is accompanied by lower mean power and pressure values, which results in increased cardiac output. Presently, the main problem is the time it takes to determine the patient coefficients (approx one hour), a procedure that needs to be simplified.

Lung function analysis and optimization during artificial ventilation. A personal computer-based system.

In an intensive care unit a personal computer (PC) application for lung function analysis has been in use for 5 years. The PC system is applied to measure conventional and new parameters for diagnosis and therapy. The primary goal was to find parameters which could be used as optimization indices in optimal control systems for mechanical ventilation. Another clinical application of the PC system was as an automatic controller that stabilizes end-tidal CO2 concentration. The controller and the next application, the optimizer, could be integrated into an optimal control system. Such a system is described and a simulation trial of the integrated structure has demonstrated the potential.

Integrating knowledge-based technology into computer aided ventilation systems.

A knowledge-based decision support system for respirator treatment, the KUSIVAR system, has been designed in cooperation between hospital, university and industry. Changes in patient data from respirator and monitoring equipment trigger a computer program that generates advice to the staff concerning e.g. therapy modes and respirator settings using expert systems and process control technology. A prototype has been built on an advanced development workstation, the Unisys Explorer, using the software Knowledge Engineering Environment (KEE). The clinical version is implemented on an Intel 80396-based microcomputer connected on-line via a data-acquisition processor to the respirator. The decision support software is implemented as a module under the Microsoft Windows multitasking environment and communicates with modules for data acquisition, database handling and data presentation by means of message passing using the Windows Dynamic Data Exchange protocol. The modules present coherent user interfaces by conforming to Microsoft Windows standards. The knowledge base is being extensively validated by an expert group in the ICU and the system will be evaluated through animal experiments and clinical studies.

A knowledge-based support system for mechanical ventilation of the lungs. The KUSIVAR concept and prototype.

The KUSIVAR is an expert system for mechanical ventilation of adult patients suffering from respiratory insufficiency. Its main objective is to provide guidance in respirator management. The knowledge base includes both qualitative, rule-based knowledge and quantitative knowledge expressed in the form of mathematical models (expert control) which is used for prediction of arterial gas tensions and optimization purposes. The system is data driven and uses a forward chaining mechanism for rule invocation. The interaction with the user will be performed in advisory, critiquing, semi-automatic and automatic modes. The system is at present in an advanced prototype stage. Prototyping is performed using KEE (Knowledge Engineering Environment) on a Sperry Explorer workstation. For further development and clinical use the expert system will be downloaded to an advanced PC. The system is intended to support therapy with a Siemens-Elema Servoventilator 900 C.

Computer simulation of a patient end tidal CO2 controller system.

A computer model of the patient end tidal CO2 controller system has been developed and tested in simulation trials. It is intended to aid in finding the appropriate PI (proportional-integral) controller settings by means of computer simulation instead of real experiments with the system. The latter approach is costly, time consuming and sometimes impossible to perform. The simulator consists of two equations: the patient equation and the PI controller equation. The software has been written in the C language and can be run on an IBM-PC/XT. Some examples of the simulation trials, illustrating the choice of controller settings, are given.

A new method for calculating total respiratory system compliance. Theory and model experiments.

A new method for calculating total respiratory system compliance is described, based on simple modelling of a ventilator-respiratory system circuit that assumes linear characteristics of the circuit parameters compliances and resistances. The method requires only that flow measurement be conducted continuously to obtain compliance, if the internal compliance of the circuit is known beforehand. Model experiments showed that the compliance of a child test lung, calculated from the flow recording, differed at most by 10% from the compliance obtained by separate measurements of pressure and volume under static conditions, over a wide range of respiratory flows and airway resistances.