From: AAAI Technical Report SS-94-01. Compilation copyright © 1994, AAAI (www.aaai.org). All rights reserved. NEONATAL WORKSTATION FOR INSPIRED OXYGEN CONTROL AND CLINICAL MONITORING Yao Sun MD,Isaac Kohane MD,Phi) Children’s Hospital, Harvard University Medical School, Boston, MA designed to advise clinicians about ventilator management. Some recent examples include: VentPlan, a ventilator managementadvisor that interprets patient data within a physiological model to predict the effect of proposedventilator changes; (11) ESTER, a program which assesses the patient’s pathophysiological state using modified APACHE-II criteria, then offers suggestions for weaning from intermittent mandatoryventilation; (6) WEANPRO, a program designed to help wean postoperative patients from ventilators; (15) and KUSIVAR, a program which describes a comprehensive system for respiratory management during all phases of pulmonary disease. (10) Background Although many such expert systems have been Oxygentoxicity plays a role in the developmentof described, few have been tested in clinical patient chronic lung disease in newborninfants requiring care. mechanical ventilation. (12,16,17) In premature infants, oxygentoxicity is also implicated in the Other investigators have studied direct computer developmentof retinopathy of prematurity. (1,7,8) control of specialized aspects of ventilator Therefore, minimizing oxygen exposure of management.(14,18) For example, studies ventilated newborninfants has becomea priority in computer-controlled optimization of positive endneonatal intensive care. expiratory pressure, and computerizedprotocols for managementof adult respiratory distress syndrome By constantly monitoring the signals from cardiac have been explored by East. (3,4) A computerized and pulse oximetry monitors and correlating these ventilator weaning system for post-operative signals, a computer system has been designed to patients has beentested by Strickland. (13) control the inspired oxygen concentration (FIO2) delivered to mechanically ventilated newborn Experience in computer controlled ventilation in infants. This computer-controlled FIO2 delivery infants, however, is limited. In one of the few system can continuously adjust oxygendelivery to reports available in the literature, Morozoffand the minimumamount needed to achieve a target Evans showedthat a computerized FIO2 controller oxygensaturation. could maintain the oxygensaturation of a ventilated newborn infant at least as well, and sometimes physiological data received by the computerfrom a better than manual FIO2 control. (9) They patient monitoring system is readily available for described their FIO2controller as a "differentialselective storage and analysis. Thesedata could be feedback" controller. This description matches the utilized in several ways. The information for an general class of controllers knownas "proportional individual patient could be analyzed and plotted to integral derivative" (PID) controllers. For best need to be detect trends in clinical improvement or response, most PID controllers deterioration. By comparing data from many optimized for the system that they will be used in. neonates, we might also elucidate physiological This maylead to degradation of performanceif the system changes. trends and patterns associated with particular pulmonary diseases. Displaying the data in a selective and informative waywill also benefit the Clinical monitoring systems have also been clinician by facilitating the analysis of enormous described previously. Examplesinclude Guardian, quantities of raw information. a "proof of concept"of an intelligent agent designed to coordinate a range of reasoningtasks in intensive Investigation into computer-controlledor computer- care monitoring, (5) and Simon, a series of semiassisted mechanicalventilation is expanding. One independent modules designed to collect and form of computerassistance is an "expert system" analyzeclinical data whileinterpreting patient status Introduction The regulation of oxygen delivery to acutely ill infants requiring mechanicalventilation is a vital process in neonatal intensive care medicine. We have designed a microcomputer based system to automatically and continuously control the inspired oxygeff-" concentration (FI02) delivered mechff~ically ventilated newborn infants. In addition, wewill utilize this computerplatform to collect and analyzeclinical data (such as heart rate, respiratory rate, arterial blood pressure, and oxygen saturation) in newborn infants with pulmonary disease. 156 and predicting parameters. (2) the evolution of monitored The present system incorporates a new FIO2 controller in the context of a system designed to collect, store, and display the physiological monitored data for mechanically ventilated newborn infants. System description The system consists of a microcomputer, a ventilator with electronic signal outputs, a clinical monitor with signal outputs, and the patient. Software currently exists that enable clinical monitors to accept input signals from a ventilator indicating ventilator parameters such as the FIO2. The clinical monitor also collects patient physiological data such as heart rate, respiratory rate, and oxygen saturation. Using these data inputs from the clinical monitor, the computerwill execute an algorithm to makechanges in the FIO2 to achieve a target oxygensaturation. The computer will also continuouslyrecord the clinical data input for later retrieval and analysis. that response. Thesystem’s user interface allows the clinician to enter identifying patient information and critical alarms limits for the oxygensaturation, heartrate, mean blood pressure and respiratory rate. The clinician also sets the target oxygensaturation for the systemto maintain. Although it is currently the only active control function, the FIO2controller is incorporated in a processing loop that includes data collection, diagnostic analysis, and data storage. This general architecture will enable expansionof the system to accommodatefuture diagnostic and therapeutic options, such as the implementation of total ventilator control and diagnosis of pulmonary disease. System Testing Thesystemhas been tested on the data set of patient physiological information provided for this symposium.For the purposes of FIO2 control, the only data that need processing are the oxygen saturation, the "arterial heartrate", the ECG Currently, the system is designed to operate by (cardiovascular monitor) heartrate, and the FIO2. displaying suggested FIO2changesto the clinician, The other patient information is gathered and rather than to control the ventilator directly. This stored. ensures medical safety until the system is fully tested for clinical efficacy. Since the data set represents static information, it did not test the FIO2controller’s response to a The algorithm to be tested in this system has two dynamically changing system. The system features that differ from a standard PIDcontroller. performsthe followingactions correctly for the data First, changes in the FIO2 will be based on the set: hemoglobin-oxygendissociation curve. This builds 1) For the target oxygen saturation set by the in a "predictive" componentto the controller based clinician, the controller will suggest FIO2 on the expected changes in oxygen saturation for changes in the appropriate direction (higher or given changes in partial pressure of oxygen. As lower) with appropriate adjustments in the patient oxygensaturation values are collected by the magnitude of FIO2 changes based on the system, these values are mappedto a corresponding difference between the measured oxygen portion of the hemoglobin-oxygen dissociation saturation and the target. curve. If the oxygensaturation mapsto a steeper 2) Validation of the oxygensaturation is performed portion of the curve, smaller changes in FIO2will by comparingthe "arterial heartrate" with the be made, and correspondingly larger changes in ECGheartrate. The arterial saturation is FIO2will occurfor flatter portions of the curve(i.e. presumedto be an artifact if the two heart rates at high oxygensaturation values). do not correlate within 5 beats/minute. 3) The system correctly displays alarm messages Secondly,the algorithm will dynamicallyadjust the whenthe physiological information represents magnitudeof the changes in FIO2 dependingon the valuesoutsidethe limits set by the clinician. individual patient’s responses. This will 4) All the patient data is stored in a file for later compensate for changes in a patient’s clinical retrieval and analysis. condition over time, and also allow the system to be used with different patients without the need to Subsequent clinical trials of the FIO2 control modifythe system for each patient. The adjustment system will test the effectiveness of the controller algorithm works by examining the patient’s past in maintainingpatient oxygensaturations. Data will response to FIO2 change and regulating the be analyzed to compare the oxygen exposure of magnitude of subsequent FIO2 changes based on patients during "open loop" computer control vs. 157 during manual FIO2 control. If the data analysis 6) Hernandez-Sande C, Moret-Bonillo V, reveals that the computer FIO2 control system Alonso-Betanzos A. provides lower patient oxygen exposure without ESTER:an expert system for management of compromise in tissue oxygenation,a clinical trial of respiratory weaningtherapy. direct computercontrol of ventilator FIO2will be IEEE Tl.ans Biomed Eng 1989; 36: 559. conducted. 7) Kinsey VE. Retrolental fibroplasia: cooperative study of Summary retrolental fibroplasia and the use of oxygen. The implementation of this system to control FIO2 Arch Ophthalmol 1956; 56: 481. in mechanically ventilated newborns may have 8) LenmanJT, GuyLP, Dancis J. important clinical implications for reducing the Retrolental fibroplasia and oxygen therapy. development of chronic lung disease and JAMA1954; 155: 223. retinopathy of prematurity. In addition, the data 9) Morozoff PE, Evans RW. collection functions of the system will allow Closed-loop control of Sa02 in the newborn analysis of physiological variables in multiple infant. ways. As the system evolves, concurrent selective Biomed Instl’um Technol 1992; 26:117. display of realtime data will also enable clinicians to 10) RudowskiR, Frostell C, Gill H. assimilate and process raw information in an A knowledge-based support system for efficient manner. mechanical ventilation of the lungs. The KUSIVARconcept and prototype. Currenttesting with static clinical data verifies that Comput Methods Programs Biomed the FIO2 control algorithms will produce 1989; 30: 59. appropriate recommendations to increase or 11) Rutledge G, ThomsenG, et al. decrease the FIO2 while observing the alarms and VentPlan: a ventilator-management advisor. limits imposed by the clinician user. Further Proc Annu Symp Comput Appl Med dynamictesting in the neonatalintensive care setting Care 1991: 869. will evaluate the ability to respond to changing 12) Saugstad OD. patient clinical conditions. Oxygentoxicity in the neonatal period. Acta Paediatr Stand 1990; 79: 881. 13) Strickland JH, Hasson JH. REFERENCES A computer-controlled ventilator weaning 1) Avery GB, Glass P. system. Retinopathy of prematurity: what causes it? Chest 1991; 100: 1096. Clin Pel.inatol 1988; 15: 917. 14) Tehrani FT. 2) DawantBM,UckunS, et al. A microcomputer oxygen control system for SIMON: a distributed computerarchitecture for ventilatory therapy. intelligent patient monitoring. In press: Ann Biomed Eng 1992 20: 547-58. Expert Systems with Applications. 15) Tong DA. 3) East TD, BohmSH, et al. Weaningpatients from mechanicalventilation. A successful computerizedprotocol for clinical A knowledge-based system approach. managementof pressure control inverse ratio Comput Methods Programs Biomed ventilation in ARDSpatients. 1991; 35: 267. Chest 1992; 101: 697. 16) Truog WE,Jackson JC. 4) East TD,in’t VeenJC, et al. Alternative modes of ventilation in the Computer-controlled positive end-expiratory prevention and treatment of bronchopulmonary pressure titration for effective oxygenation dysplasia. without frequent blood gases. Clin Pel.inatol 1992; 19: 621. Cl.it Care Meal1988; 16: 252. 17) WispeJR, Roberts RJ. 5) Hayes-Roth B, Washington R, et al. Molecularbasis of pulmonaryoxygen toxicity. Guardian: a prototype intelligent agent for Clin Pel.inatol 1987; 14: 651. intensive-care monitoring. 18) YuC, He WG,et al. Artificial Intelligence in Medicine 4 Improvementin arterial oxygen control using Elsevier Science Publishers B.V. 1992, pp. multiple-model adaptive control procedures. 165-185. IEEE Tl.ans Biomed Eng 1987; 34: 567. 158