Closed-loop non-invasive oxygen saturation control system.

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to automatic control of blood oxygen saturation (S_aO₂) in a patient by adjustment of the fractional amount of oxygen inspired (FIO₂) by the patient, and more particularly, to a feedback control loop for a mechanical ventilator including a non-invasive S_aO₂ monitor in the feedback path for developing an adaptive control signal which controls the inspired gas blender of the ventilator.

2. Description of the Prior Art

Devices for controlling the oxygen content of the blood by controlling the fraction of oxygen breathed by a patient are well known. For example, U.S. Patent No. 2,414,747 issued to Harry M. Kirschbaum on January 21, 1947 shows a method and apparatus for controlling the oxygen content of the blood of living animals which discloses control of blood oxygen content by the use of an ear oximeter which produces a signal used to control the proportion of inspired oxygen. As the oximeters direct a beam of light through a capillary bed in the ear, the characteristics of the light become modified by the color of the blood that intercepts its path. Thus, the change in oxygen levels of the blood are detected non-invasively and electrical signals are generated, amplified and used to control the oxygen supply delivered to a patient.

Numerous improvements have been made since that time wherein better matching of oxygen delivery to the needs of the patient have been made such as shown in U.S. Patent No. 3,734,091 to Ronald H. Taplin issued on May 22, 1973. Taplin discloses an optical oximeter and a temporary oxygen-deficient mixture (anoxic) to control blood oxygen saturation. Thus, to prevent excessive oxygen levels, Taplin discloses limiting the inspired oxygen by intermittently providing the anoxic mixture each time the oxygen saturation of the blood reaches a predetermined percentage level.

U.S. Patent 4,889,116 issued to Taube on December 26, 1989 discloses one type of adaptive controller for adjusting the fraction of oxygen inspired by a patient. The controller utilizes a pulse oximeter connected by an optical sensor to the patient for measuring the patient's blood hemoglobin saturation (S_pO₂) and pulse rate. These signals from the oximeter are used by a calculator for determining the fractional amount of oxygen to be inspired by the patient. The calculated percentage of oxygen is provided to the patient so that the gas taken in by the patient automatically causes the blood in the patient to reach a predetermined hemoglobin saturation level in response to the patient's requirements as determined by changes in the S_pO₂ signal. However, the calculator is programmed to determine when there is an excess deviation of the patient's pulse rate, thereby indicating patient movement and the probability that the pulse oximeter will provide false S_pO₂ values during such patient movement. When an excess deviation in pulse rate is detected, the fractional amount of inspired oxygen is no longer responsive to the measured S_pO₂ value, but instead held constant until the excess deviation of the pulse rate has been terminated. Furthermore, a low S_pO₂ value, indicative of a depressed respiration (apnea) is also detected, and used to cause a preset higher percentage of inspired oxygen to be supplied to the patient until the depressed respiration of the patient has been terminated. Thus, responsive FIO₂ adjustment is suspended during patient movement and apnea, and during this time fixed FIO₂ levels are set.

In 1980, H. Katsuya and Y. Sakanashi published an article in the Journal of Clinical Monitoring 1989; 5:82-86 describing a method for evaluating pulmonary gas exchange using a pulse oximeter. They developed the concept of FI_9x (where x is a single digit number) which is the fraction of inspired oxygen necessary to achieve a measured blood oxygen saturation equal to the value of 9x% (e.g. 98%). This experiment was carried out by periodically manually increasing or decreasing the FIO₂ control of a gas blender portion of a ventilator until the S_pO₂ measurement reached the target percentage (e.g., 98%). The purpose of this experiment was to develop a diagnostic method to evaluate pulmonary gas exchange impairment. A high value of FI_9x was associated with poor pulmonary gas exchange. In this publication, no mention was made of feedback of S_pO₂ values for automatic adjustment of FIO₂. The present invention recognizes that the pulmonary impairment of a patient can often change during treatment, thereby requiring a change or adaptation of the responsiveness of the FIO₂ control loop.

Since adult and neonatal patients in intensive care units suffering from respiratory distress are at risk for developing hypoxemia or oxygen toxicity, certain safety precautions should be taken to prevent O₂ under/overdose. In an attempt to maintain organ normoxia, appropriate clinical care often mandates ventilation with high FIO₂, sometimes for several days. Long exposure to high concentrations of oxygen has been implicated in complications including exacerbation of respiratory distress and various central nervous system symptoms. In neonatal patients, oxygen toxicity may result in blindness from retrolentalfibroplasia. Thus, care should be taken to minimize the FIO₂ exposure while maintaining adequate S_pO₂, so that the onset of these insidious complications can be delayed or avoided. Furthermore, artifacts (false output measurements) are commonly found in the pulse oximeter output due to patient movement and/or low blood profusion in the area where the patient contacts the pulse oximeter sensor. Additionally, it is difficult to actually know what the arterial blood oxygen saturation percentage is from the S_pO₂ (pulse oximeter) measurement. Thus, careful construction of the S_pO₂ feedback control system is required.

It is an object of the present invention to provide a method and apparatus which minimizes the FIO₂ of a patient while maintaining adequate S_aO₂ levels.

It is a further object of the invention to provide artifact rejection processing of the pulse oximeter which is tolerant of the expected false readings of S_pO₂, and furthermore, which is adaptive so as to gradually cause the FIO₂ to increase as the frequency of the artifacts increases.

It is still a further object of the invention to provide an FIO₂ feedback control loop which has a response (transfer characteristic) which is adaptive to the changing requirements of the patient.

It is an even further object of the invention to provide a safety sub-system for the FIO₂ control system in order to prevent a failure in the feedback control system from causing injury to the patient due to extremely inappropriate levels of FIO₂.

SUMMARY OF THE INVENTION

In accordance with the principles of the present invention, a method and apparatus is provided for evaluating the oximeter S_pO₂ measurements for artifact rejection purposes, and furthermore, causing a gradual increase in FIO₂ as the frequency of S_pO₂ artifacts increases.

In accordance with a further feature of the invention, a safety sub-system is provided which monitors the output signals from various portions of the FIO₂ control system for correlation and/or excursion beyond preset threshold values. Signals which exceed the thresholds or which do not correlate, cause safety sub-system to indicate an alarm condition to the physician.

In accordance with further features of the invention, a feedback control means of the FIO₂ control loop has an adaptively adjustable response. In the preferred embodiment, the feedback control means comprises a proportional, integral, derivative controller wherein the coefficients of the P, I, D, terms are selectively adjustable in order to provide appropriate respiration therapy to the patient. Additionally, these coefficients may be modified by input signals which are not part of the FIO₂ control loop, such as minute ventilation, tidal volume, etc. Finally, the controller response can be adaptive to the needs of the patient, by periodically disabling the controller and monitoring the response of the patient to random fluctuations of FIO₂. An evaluation of the response of the patient is used to adapt the response characteristic of the feedback controller.

Other features and advantages of the invention will be apparent from the description of the preferred embodiment and from the claims.

For a fuller understanding of the present invention, reference should now be made to the following detailed description of the preferred embodiment of the invention and to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

<ul id="ul0001" list-style="none"><li>Figure 1 illustrates a schematic representation of a closed-loop S_pO₂ control system constructed in accordance with the present invention for controlling the fractional amount of the inspired oxygen for a patient;</li><li>Figure 2 is a block diagram of the flow control algorithm used for artifact rejection in the output signal from the pulse oximeter shown in Figure 1;</li><li>Figure 3 is a block diagram of a digital PID controller used in the feedback control system of the present invention;</li><li>Figure 4 is a block diagram of the computational process of the PID controller of Figure 3.</li><li>Figures 5(a) to 5(e) are graphical representations of the non-linear operation of the PID controller of Fig. 3;</li><li>Figure 6(a) illustrates a block diagram of the PID controller transfer function in the Laplace domain and Figure 6(b) illustrates a range of values for K3, a term in the controller transfer function; and</li><li>Figures 7(a) and (b) illustrate a block diagram and waveform useful for understanding the RLS method.</li></ul>

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

Figure 1 illustrates in functional block diagram form a preferred embodiment of the present invention. A substantial portion of functions described below are actually carried out by the control system computer, which will be described later on with respect to the user interface. A patient 8 requiring mechanical ventilation is connected to a mechanical ventilator 10 via a breathing circuit 11. Breathing circuit 11 includes an inspiration line 12 by which ventilation gases are unidirectionally delivered to patient 8 and an expiration line 14 for causing unidirectional flow of expired gases from patient 8 to mechanical ventilator 10. Ventilator 10 may comprise any of several well known types of commercially available mechanical ventilators, such as the Model 900C manufactured by Siemens Corporation and available from Siemens Medical Systems, Iselin, NJ. A pulse oximeter 16 of conventional design includes an optical sensor 18 coupled to the finger of patient 8 for non-invasively (e.g., optically) providing a measurement S_pO₂, in a well known manner, of the blood oxygen saturation level of patient 8. Pulse oximeter 16 may be, for example, of the type manufactured by Siemens Corporation and comprise the Model 961 monitor system including a pulse oximeter cartridge unit therein. Pulse oximeter 16 includes an output for providing a high speed serial digital signal output S_pO₂ representative of the measured S_aO₂ level.

Before the measured S_pO₂ level can be used for controlling the FIO₂, certain precautions must be taken. The present inventor has recognized the importance of providing artifact rejection filtering techniques to the measured S_pO₂ level, in order to prevent erroneous operation of the FIO₂ feedback control system. That is, it is normal during the operation of a pulse oximeter, that artifact (false) output information will occasionally be generated in response to e.g., physical movement of patient 8 and/or low profusion in the area of sensor 18. This factor was noted in the forenoted Taube patent, however, responsive FIO₂ control was suspended during patient movement. However, in the present invention, an S_pO₂ artifact filter 20 is provided for processing the output of pulse oximeter 16. In the preferred embodiment, the function of filter 20 is provided completely by software processing in the control system microprocessor. An algorithm is used which rejects artifactual measurements of S_pO₂ based on the statistics of past measurements of S_pO₂. The algorithm replaces measured values of S_pO₂ which fall substantially outside the variance of previous values of S_pO₂ with pseudo values. These pseudo values are related to the average S_pO₂ values and, in accordance with a further aspect of the present invention, the frequency of the artifactual S_pO₂ values. As the frequency of artifacts increases the pseudo values are made increasingly less than the mean value, thereby causing the FIO₂ to be gradually increased. Thus, the algorithm provides an adaptive filtering of the S_pO₂ values and gradual increase in the FIO₂, based on the frequency of the detected artifacts. Details of filter 20 will be described later on in conjunction with Figure 2.

After appropriate artifact rejection filtering, the S_pO₂ value is then compared with a target S_pO₂ value in a comparator 21 and the difference signal E(t) (error) is then applied to a feedback controller module 22. The target S_pO₂ value is provided by a user interface 24 which may comprise a personal computer of conventional design programmed in the C software language and which includes a keyboard, display, microprocessor and data storage means. The system user will input, via the keyboard, the desired set-point and limit values for the blood oxygen saturation levels as well as other patient data. The user interface display provides data to the physician informing him/her of the status of the system, the S_pO₂ set-point, the measured S_pO₂ and the measured FIO₂.

Feedback controller 22 may comprise a PID (Proportional-Integral-Derivative) controller, such as described in greater detail later on. Feedback controller 22 receives the error signal provided by comparator 21 and produces a stable control signal in response thereto, having a zero steady state error and limited overshoot. In a preferred embodiment of the invention, the proportional, integral and/or derivative terms are selectively controllable for adapting the response of controller 22. Feedback controller 22 is also adaptive as a function of external input signals, such as minute ventilation (the amount of gas breathed by a patient in one minute). Additional novel features of feedback controller 22 will also be described later on relating to the periodic introduction of random FIO₂ level changes, which are used to adapt the response of the feedback compensation. Feedback controller 22 provides an output which indicates what the FIO₂ level should be, e.g., 30%.

An FIO₂ controller 26, comprised partly of software and partly of hardware including a stepper motor, calculates the present set-point of the FIO₂ being delivered to the patient, compares this with the instructions received from feedback controller 22, and then mechanically controls a gas blender included in mechanical ventilator 10 for causing the FIO₂ level to be adjusted. The gas blender, not specifically shown, is a conventional part added to or built-in to mechanical ventilator 10 which controls the mixing of gas from an air source and an O₂ source, respectively, to the gas input of ventilator 10. For example, assume that feedback controller 22 provides an output (in software) to FIO₂ controller 26 demanding that the FIO₂ be increased to 30%. The software portion of FIO₂ controller 26 will then calculate a number of pulses, e.g., 30 to the right, that must be applied to the stepper motor in order to cause the gas blender of mechanical ventilator 10 to cause a 30% mixture between O₂ and air. This mechanical connection is indicated by the output from controller 26 to mechanical ventilator 10, although an electronic control could also be used.

A final portion of the invention comprises safety sub-system 28. A series of safety features are implemented in hardware and software which comprise safety sub-system 28 to prevent hypoxic inspired mixtures and inadvertent errors in the operation of the FIO₂ control system. These include: <ul id="ul0002" list-style="none"><li>1. Limit (threshold) values of S_pO₂ and FIO₂ are entered into the system by the user via interface 24. Safety sub-system 28 monitors the artifact corrected S_pO₂ value via line 29, the FIO₂ output from controller 26 via line 31, and the actual O₂ level sensed in ventilator 10 and inspiration line 12 via lines 33 and 35, respectively. Excursions beyond the pre-set thresholds are detected by sub-system 28 and cause visual and audible alerts to be directed to the physician.</li><li>2. The system resets (see 4 below) based on missing signals from pulse oximeter 16 or ventilator 10.</li><li>3. The system detects error between desired FIO₂ (i.e., the output from controller 22 via line 37) and the measured FIO₂ at three different places. The first place is the output of controller 26 via line 31, the second place is the O₂ indication signal which is provided from mechanical ventilator 10, via line 33 and the third place is a redundant O₂ indication signal provided from an O₂ sensor in breathing circuit 11, via line 35. Each of these values should be correlated; if not, an alarm condition is indicated.</li><li>4. The system automatically sets the output of controllers 22 and 26 to 100% when any of the above signals indicate error, via set lines S1 and S2 which are coupled to controllers 22 and 26, respectively, from safety sub-system 28.</li><li>5. All data is stored on magnetic media for retrospective analysis.</li></ul>

Next, details of artifact filter 20 will be described in conjunction with Figure 2. This filter algorithm is intended to handle artifacts in a safe, i.e. conservative, manner. Firstly, it recognizes that the pulse oximeter is artifact prone because it is sensitive to both patient movement and blood perfusion in the area of sensor 18. Secondly, it assumes that when data is missing or suspiciously different (e.g., low) compared to the date immediately preceding it, it is probably artifactual. When artifacts are occurring, the worst scenario is that the S_pO₂ is actually falling, and increasing the patient FIO₂ would be the appropriate therapy. Thus, when artifacts are occurring, we want to balance our concern that the S_pO₂ may be falling with the knowledge that missing data is a routine occurrence. Furthermore, as the frequency of missing data increases, our concern that S_pO₂ might be falling should also increase. Thus, in accordance with a further aspect of the present invention, when S_pO₂ data is missing or suspiciously low (e.g., more than two standard deviations away from the mean of the previous values) we will initially assume the true S_pO₂ value is likely to be the mean value. Additionally, in accordance with still a further aspect of the invention, to be conservative we will subtract an adaptive factor from the mean value to recognize the potential that the actual S_pO₂ value may be falling. The factor is adaptively increased as the frequency of occurrence of the artifacts increases. This results in progressively increasing levels of FIO₂ for the patient to be set by the FIO₂ control system.

The artifact rejection algorithm of filter 20 is designed to insure that the FIO₂ control system operates normally. As artifacts occur, the control system will tend to increase FIO₂ gradually. Thus, as the frequency of artifacts increases, the controller will drive FIO₂ up harder. When the frequency of occurrence of the artifacts exceeds an intolerable limit, an alarm will sound and the controller will go open loop with FIO₂ = 100%.

As shown in Figure 2, filter 20 operates in accordance with the following process: <dl id="dl0001"><dt>Step 202:</dt><dd>Evaluate the S_pO₂ values for artifacts,</dd><dt>Step 204:</dt><dd>Determine the frequency of occurrence for the artifacts, and</dd><dt>Step 206:</dt><dd>Adapt the pseudo S_pO₂ control value in accordance with the artifact occurrence frequency.</dd></dl>

This process is accomplished as follows:

Step 202

As shown in Figure 1, S_n = current S_pO₂ value from pulse oximeter 16. S'_n = S_pO₂ value sent to comparator 21.

S_n-m , ..., S_n-3 , S_n-2 , S_n-1 , S_n is a series of m past S_pO₂ output values.

Compute a running average of S_n-m as: <chemistry id="chem0001" num="0001"><img file="EP0504725A2_D0001.tif" /></chemistry>

Compute a standard deviation of S_n-m as: <chemistry id="chem0002" num="0002"><img file="EP0504725A2_D0002.tif" /></chemistry>

Then, let M_n = a binary logical variable (possible artifact). We then set up four tests (windows) to determine if M_n is TRUE, i.e., an artifact, or FALSE, i.e., probably not an artifact but a good S_pO₂ value from oximeter 16. <ul id="ul0003" list-style="none"><li>1. If S_n &lt; (S_n - 2S_n) and S_n &gt; 2 then M_n = TRUE</li><li>2. If S_n &lt; (S_n - 4) and S_n ≦ 2 then M_n = TRUE</li><li>3. If S_n ≧ (S_n - 2S_n) and S_n &gt; 2 then M_n = FALSE</li><li>4. If S_n ≧ (S_n - 4) and S_n ≦ 2 then M_n = FALSE</li></ul>

As you can see from tests 1 and 3, if the standard deviation is greater than 2, it is relatively safe to say that if S_n is less than two standard deviations from the running average value, it's probably artifactual, and if S_n is not less than two standard deviations from the running average, it's probably valid. For example, if the running average = 96% and the standard deviation is 3%, if the current S_n is less than 90%, it's assumed to be an artifact and if equal to or greater than 90%, it is assumed to be valid.

Tests 2 and 4 are intended for the case where the standard deviation is small, i.e. less than or equal to 2%. Under these circumstances it would be difficult to use standard deviation as a measure of validity since even slight changes can be erroneously considered artifacts. Thus, under these circumstances, a fixed change in S_n from the running average is used as the validating criteria. In the preferred embodiment, the fixed amount is a 4% change. For example, if the running average was 96% and the current S_n was 93%, it would not be considered an artifact. It should be understood that the criteria given here are those currently being used by the inventor, and that future testing may result in modifications of these values and even the use of other criteria, such as the average absolute difference of each value from the mean value.

Step 204

Compute the frequency of artifact occurrences<maths id="math0001" num=""><img file="EP0504725A2_D0003.tif" /></maths> for M_i where i = M-m to n-1

Step 206

<ul id="ul0004" list-style="none"><li>1. If M_n = FALSE then S'_n = S_n</li><li>2. If M_n = TRUE then S'_n = S_n - Q, where Q is one of q₁, q₂ or q₃. Where, <dl id="dl0002"><dt>Q =</dt><dd>q₁ if 1/m &lt; F_n ≦ f_a;</dd><dt>Q =</dt><dd>q₂ if f_a &lt; F_n ≦ f_b; and</dd><dt>Q =</dt><dd>q₃ if f_b &lt; F_n ≦ f_c.</dd></dl> Also:</li><li>3. If F_n &gt; f_c then: a) <u>ALARM</u>, b) set FIO₂ = 100%; and c) set S'_n = 0.5% (an arbitrary small, non-zero amount)</li><li>4. If S_n &gt; 1/3 S_n then <u>ALARM</u></li></ul>

In the preferred embodiment, m = 10 sample S_pO₂ values, f_a = 3/10, f_b = 5/10, f_c = 6/10, and q₁ = 0.5%, q₂ = 1.0%, and q₃ = 2.0%.

Feedback controller 22 of the present invention is of the PID (Proportional-Integral-Derivative) controller type, and computes a feedback compensation response which is related to the difference between the desired (set-point) and measured S_pO₂ values i.e., the output of comparator 21. The flow of the O₂ gas which is added to the breathing mixture of the patient is controlled by an electronic signal from controller 22 which is used to adjust the stepper motor of controller 26. The electronic signal is a function of the sum of three terms:<maths id="math0002" num=""><img file="EP0504725A2_D0004.tif" /></maths> where Y(t) is the flow control signal, C₁, C₂ and C₃ are constants, E(t) is the Error (difference between the desired S_pO₂ and the measured S_pO₂ (S'_n) and dE(t)/dt is the time derivative of E(t). The output Y(t) may be adjusted by a constant of proportionality as circumstance may dictate.

Although the above is an analog implementation of feedback controller 22, it is also possible to use a sampled-data equivalent feedback controller, such as commonly used with microcomputers. The block diagram of a digital PID controller of the preferred embodiment is shown in Figure 3. The transfer function of the digital PID controller is as follows:<maths id="math0003" num=""><img file="EP0504725A2_D0005.tif" /></maths> The microcomputer of the user interface system samples the input variable (E(t))periodically (every T seconds). Computer software is used to mathematically implement the feedback algorithm of controller 22.

While the output of the digital controller is presented every T seconds, a circuit element called a zero-order-hold takes the periodically produced output e(t), holds it steady and implements the controlling output action.

In the block diagram of the digital PID controller in Figure 3, K₃, K₂ and K₁ are coefficients (gain constants) of the proportional, integral and derivative terms, respectively. T is the sample time interval, n is an integer number of sample intervals. Z is the Z-transform operator, E(nT) is the error signal e(t) sampled every T seconds, and Y(nT) is the desired FIO₂ value. Note that K₃, K₂ and K₁ serve the same purpose as C₁, C₂ and C₃ in the analog implementation discussed previously; however, the values of K₁, K₂ and K₃ may be different from C₁, C₂ and C₃.

A block diagram of this process is shown in Figure 4. The values of the coefficients as shown in Figure 4 are dependent on the characteristics of the design of the system which is being controlled. The increment (n), error (E), flow (F) and the output are functions of the volume and flow characteristics of the breathing circuit, the patient's lung and O₂ transfer characteristics, and the settings of the ventilator.

It is also possible to use the patient's minute ventilation (MV), title volume (TV), and/or respiratory rate (RR) to modify one or more of the terms in the above circuit equation. The patient's minute ventilation, tidal volume, or respiratory rate may be measured using any of several well-known techniques. For example, these measurements may be based on thermal dissipation, a pressure difference across a resistive element (pneumotachograph), the rotation rate of a vane, or the oscillation frequency of a fluid vortex. This type of modification of the invention will be described next in greater detail with respect to Figure 6. Furthermore, in accordance with another aspect of the invention, design goals such as tolerable overshoot, time to achieve control and accuracy of control are also used to determine the value of coefficients K₁, K₂ and K₃. Details relating to modifications of this type are described in greater detail with respect to Figure 7.

First, however, the initial set-up of controller 22 will be described. Referring to the above equation for the transfer function of the digital PID controller, the values of K₁, K₂ or K₃, or any selective combination of K₁, K₂ and K₃ are adjusted as a function of the error signal e(t), i.e., the output of comparator 21. The initial set-up of these values is accomplished in order to tailor the response of the FIO₂ control system in accordance with the clinical goals for given classes of patients. For example, there may be a first set of values for K₁, K₂, K₃ for adults and a second set of K values for neonatal patients.

In accordance with a further aspect of the invention, these K values of the PID controller are made non-linear in order to more appropriately tailor the response of the FIO₂ system in accordance with empirically determined desired values for specific classes of patients. Additionally, it is noted that the non-linear gain provided by PID controller 22 makes the response of the FIO₂ control loop more clinically appropriate. For example, Figure 5(a) illustrates a value for K₁ as a function of e(t). However, a limiting value is provided so that the integrator term does not get to large. Although the previously noted U.S. Patent 4,889,116 to Taube includes a PID controller, it is noted that no individual control of the proportional or integral or derivative terms is provided, such as specifically provided by the present invention.

Figure 5(b) illustrates an additional non-linear characteristics for the K values, which was used for the K₃ value specifically in the preferred embodiment of the present invention and Figure 5(c) is an alternative embodiment thereof. Thus, it is apparent from these figures that non-linear coefficients are provided for various ones of the terms of the PID controller which change in real time in accordance with the error signal provided at the output of comparator 21.

Additionally, it is noted that limiting values are also imposed upon the output of PID controller, as previously noted, which prevent the output FIO₂ values from being greater than 100% or less than 0%.

Figure 6(a) is illustrative of the way in which external inputs can be applied to control the coefficients of PID controller 22 in order to change the system response accordingly. The external inputs may comprise the minute ventilation (MV), the respiratory rate (RR) and/or the tidal volume (TV). For example, K₁ could be a function of minute ventilation or K₁, K₂, K₃ could be functions of RR, MV and TV, respectively. In fact, any combination of functional relationships between the K's and external inputs are possible. In the preferred embodiment, the value of K₃ is changed in response to the minute ventilation as shown in Figure 6(b).

A final control of the characteristics for controller 22 relates to an adaptiveness of its response to the changing needs of the patient. In this regard, it is noted that in the forenoted Taube U.S. Patent 4,889,116, the controller program described in Exhibit A initially asks the user to input a "lung time constant tl". The present inventor has recognized that the initial values of k which determine the controller response may not be appropriate during later periods of the ventilation therapy for the patient. In accordance with a further feature of the invention, the PID controller output is periodically and randomly varied (within prescribed safety limits) in order to adapt the responsiveness of controller 22 with the changing needs of the patient.

Figure 7(a) illustrates a functional block diagram of the FIO₂ control loop including a Recursive Least Squares (RLS) computation used for modifying the response of controller 22. When software switch A is in position one, the RLS algorithm 700 is dormant. Under these conditions, the values for the gains of the PID feedback compensation (K₁, K₂ and K₃) are set at their initial (default) values.

When RLS algorithm 700 is activated, switch A is in position two, and a sequence of random magnitude FIO₂ values are applied to the blender of ventilator 10 via line B. The sequence of random FIO₂ values are a percentage (±) of a predetermined FIO₂ value. The predetermined value is the last value of FIO₂ before the RLS algorithm was started (which is input to RLS algorithm 700 via line C). The initial value may also be an average of several previous FIO₂ values in order to eliminate the possibility that the last FIO₂ value is non-representative of the actual needs of the patient.

Figure 7(b) illustrates an example of what one such random sequence might look like. The random 10 second periods of FIO₂ ±m% occur approximately 20 to 60 times each 15 minutes, in the preferred embodiment. The range of amplitudes of m is set by clinical considerations and performance issues. A range of m equal to 10%, 15% and 20% has been used.

The response of the patient as measured by pulse oximeter 16 is input to RLS algorithm 700 via line D for processing. When the RLS algorithm has collected several values, in the preferred embodiment 30, it begins to process these values in accordance with the Recursive Least Squares (RLS) computation, such as known by those skilled in the art, and described for example in "Dynamic System Identification: Experiment, Design and Data Analysis" by Graham C. Goodwin and Robert L. Payne, Academic press, New York, 1977; and an article entitled "Digital Parameter-Adaptive Control of Processes with Unknown Dead Time" by Kurz et al. in Automatica, Vol. 17, No. 1, pp. 245-252, 1981, published by Permagon Press; and as also described in conjunction with control of arterial blood pressure in an article entitled "The Self-Tuning Controller: Comparison with Human Performance in the Control of Arterial Pressure" by Kenneth S. Stern et al., published in the Annals of Biomedical Engineering.

Briefly, what the RLS algorithm does is find a set of coefficients (a's and b's) of a linear combination of past and present FIO₂ values and past S_pO₂ values for determining an estimated S_pO₂ value, which set of coefficients minimizes the mean square difference between this estimated value of S_pO₂ and the actually measured value of S_pO₂. When values for a and b are obtained which cause the computed and measured values to be substantially equal, it can be said that the RLS algorithm has converged. If 30 samples of S_pO₂ at 10 second intervals was not enough to cause convergence, the RLS algorithm will continue acquiring new S_pO₂ values while dropping the oldest S_pO₂ values and then try to converge. Upon convergence, new values may be found for the time constant (T) and time delay (L) of the controller response, which leads to new values for the K₁, K₂, and K₃, coefficients, respectively. These new coefficients are coupled to controller 22 via line K. A more detailed description of the theory and process for accomplishing the RLS algorithm is described in greater detail in Exhibit A.

Thus, there has been shown and described a novel method and apparatus for controlling the amount of oxygen inspired by a patient. Many changes, modifications, variations and other uses and applications of the subject invention will, however, become apparent to those skilled in the art after considering this specification and the accompanying drawings, which disclose a preferred embodiment thereof. For example, the PID controller of the present invention could in fact be a PI controller or some other type of controller having adjustable response characteristics. Furthermore the feedback control loop could be integrated into the ventilator 10 or pulse oximeter 16. All such changes, modifications, variations and other uses and applications of the invention are deemed to be covered by the claims which follow.