US 9,775,570 B2
`
`1
`ADAPTIVE ALARM SYSTEM
`
`PRIORITY CLAIM TO RELATED
`PROVISIONAL APPLICATIONS
`
`15
`
`20
`
`30
`
`40
`
`45
`
`The present application claims priority benefit under 35
`US.C. §119(e) to U.S. patent application Ser. No. 13/037,
`184, filed Feb. 18, 2011 titled Adaptive Alarm System;
`Provisional Patent Application Ser. No. 61/309,419, filed
`Mar. 1, 2010 titled Adaptive Threshold Alarm System; and
`U.S. Provisional Patent Application Ser. No. 61/328,630,
`filed Apr. 27, 2010 titled Adaptive Alarm System;all of the
`above-cited provisional patent applications are hereby incor-
`porated by reference herein.
`
`BACKGROUND OF THE INVENTION
`
`Pulse oximetry systems for measuring constituents of
`circulating blood have gained rapid acceptance in a wide
`variety of medical applications, including surgical wards,
`intensive care and neonatal units, general wards, homecare,
`physical
`training, and virtually all
`types of monitoring
`scenarios. A pulse oximetry system generally includes an
`optical sensor applied to a patient, a monitor for processing
`sensor signals and displaying results and a patient cable
`electrically interconnecting the sensor and the monitor. A
`pulse oximetry sensor has light emitting diodes (LEDs),
`typically one emitting a red wavelength and one emitting an
`infrared (IR) wavelength, and a photodiode detector. The
`emitters and detector are typically attached to a finger, and
`the patient cable transmits drive signals to these emitters
`from the monitor. The emitters respondto the drive signals
`to transmit light into the fleshy fingertip tissue. The detector
`generates a signal responsive to the emitted light after
`attenuation by pulsatile blood flow within the fingertip. The
`patient cable transmits the detector signal to the monitor,
`which processesthe signal to provide a numerical readout of
`physiological parameters such as oxygen saturation (SpO,)
`and pulse rate.
`
`SUMMARY OF THE INVENTION
`
`Conventional pulse oximetry assumesthat arterial blood
`is the only pulsatile blood flow in the measurementsite.
`During patient motion, venous blood also moves, which
`causes errors in conventional pulse oximetry. Advanced
`pulse oximetry processes the venous blood signal so as to
`report true arterial oxygen saturation and pulse rate under
`conditions of patient movement. Advanced pulse oximetry
`also functions under conditions of low perfusion (small
`signal amplitude), intense ambient light (artificial or sun-
`light) and electrosurgical instrumentinterference, which are
`scenarios where conventional pulse oximetry tendsto fail.
`Advanced pulse oximetry is described in at least U.S. Pat.
`Nos. 6,770,028; 6,658,276; 6,157,850; 6,002,952; 5,769,785
`and 5,758,644, which are assigned to Masimo Corporation
`(“Masimo”) of Irvine, Calif. and are incorporated by refer-
`ence herein. Corresponding low noise optical sensors are
`disclosed in at least U.S. Pat. Nos. 6,985,764; 6,813,511;
`6,792,300; 6,256,523; 6,088,607; 5,782,757 and 5,638,818,
`whichare also assigned to Masimoandare also incorporated
`by reference herein. Advanced pulse oximetry systems
`including Masimo SET® lownoise optical sensors and read
`through motion pulse oximetry monitors for measuring
`SpO,, pulse rate (PR) and perfusion index (PI) are available
`from Masimo. Optical sensors include any of Masimo
`LNOP®, LNCS®, SofTouch™ and Blue™ adhesive or
`
`2
`reusable sensors. Pulse oximetry monitors include any of
`Masimo Rad-8®, Rad-5®, Rad®-5v or SatShare® moni-
`tors.
`
`Advanced blood parameter measurement systems are
`described in at least U.S. Pat. No. 7,647,083, filed Mar. 1,
`2006, titled Multiple Wavelength Sensor Equalization; U.S.
`Pat. No. 7,729,733, filed Mar. 1, 2006, titled Configurable
`Physiological Measurement System; U.S. Pat. Pub. No.
`2006/0211925,
`filed Mar.
`1, 2006,
`titled Physiological
`Parameter Confidence Measure and U.S. Pat. Pub. No.
`2006/0238358, filed Mar. 1, 2006, titled Noninvasive Multi-
`Parameter Patient Monitor, all assigned to Masimo Labora-
`tories, Irvine, Calif. (Masimo Labs) andall incorporated by
`reference herein. An advanced parameter measurement sys-
`tem that includes acoustic monitoring is described in U.S.
`Pat. Pub. No. 2010/0274099, filed Dec. 21, 2009,
`titled
`Acoustic Sensor Assembly, assigned to Masimo and incor-
`porated by reference herein.
`Advanced blood parameter measurement systems include
`Masimo Rainbow® SET, which provides measurements in
`addition to SpO,, such as total hemoglobin (SpHb™),
`oxygen content (SpOC™), methemoglobin (SpMet®), car-
`boxyhemoglobin (SpCO®) and PVI®. Advanced blood
`parameter sensors include Masimo Rainbow® adhesive,
`ReSposable™and reusable sensors. Advanced blood param-
`eter monitors include Masimo Radical-7™, Rad-87™ and
`Rad-57™ monitors, all available from Masimo. Advanced
`parameter measurement systems may also include acoustic
`monitoring such as acoustic respiration rate (RRa™) using
`a Rainbow Acoustic Sensor™ and Rad-87™monitor, avail-
`able from Masimo. Such advanced pulse oximeters,
`low
`noise sensors and advanced physiological parameter mea-
`surement systems have also gained rapid acceptance in a
`wide variety of medical applications,
`including surgical
`wards,
`intensive care and neonatal units, general wards,
`home care, physical
`training, and virtually all
`types of
`monitoring scenarios.
`FIGS. 1-3 illustrate problems and issues associated with
`physiological parameter measurement systems having fixed
`threshold alarm schemas. FIG. 1 illustrates a lower-limit,
`fixed-threshold alarm schema with respect to an oxygen
`saturation (SpO,) parameter. Two alarm thresholds, D,
`(delay) and ND,(no delay), are defined. If oxygen saturation
`falls below D, for a time delay greater than TD, an alarm is
`triggered. If oxygen saturation falls below ND, an alarm is
`immediately triggered. D, 120 is typically set around or
`somewhat above 90% oxygen saturation and ND, 130 is
`typically set at 5% to 10% below D,. For example, say a
`person’s oxygen saturation 110 drops below D, 120 at t=t,
`162 and stays below D,for at least a time delay TD 163.
`This triggers a delayed alarm 140 at t=t, 164, where t,=t,+
`TD. The alarm 140 remains active until oxygen saturation
`110 rises above D, 120 at t=, 166. As another example, say
`that oxygen saturation 110 then drops below ND, 130,
`which triggers an immediate alarm 150 at t=t4 168. The
`alarm 150 remains active until oxygen saturation 110 rises
`above D, 120 at t=t, 169.
`FIG. 2 illustrates an upper-limit, fixed-threshold alarm
`schema with respect to an oxygen saturation (SpO,) param-
`eter. This alarm scenario is particularly applicable to the
`avoidance of ROP (retinopathy of prematurity). Again, two
`alarm thresholds, D,, (delay) and ND,, (no delay), are
`defined. D,, 220 might be set at or around 85% oxygen
`saturation and ND,, 230 might be set at or around 90%
`oxygen saturation. For example, a neonate’s oxygen satu-
`ration 210 rises above D,,220 at t=t, 262 and stays above D;,
`for at least a time delay TD 263. This triggers a delayed
`
`

`

`US 9,775,570 B2
`
`3
`alarm 240 at t=t, 264, where t,=t,+TD. The alarm 240
`remains active until oxygen saturation 210 falls below D,,
`220 at t=t, 166. Oxygen saturation 210 then rises above ND,,
`230, which triggers an immediate alarm 250 at t=, 268. The
`alarm 250 remains active until oxygen saturation 210 falls
`below D,, 220 at t=t, 269.
`FIG.3 illustrates a baseline drift problem with the fixed
`threshold alarm schema described above. A person’s oxygen
`saturation is plotted on an oxygen saturation (SpO.) versus
`time graph 300. In particular, during a first time interval T,
`362, a person has an oxygensaturation 310 with a relatively
`stable “baseline” 312 punctuated by a shallow, transient
`desaturation event 314. This scenario may occur after the
`person has been on oxygen so that baseline oxygen satura-
`tion is near 100%. Accordingly, with a fixed threshold alarm
`330 set at, say, 90%, the transient event 314 does nottrigger
`a nuisance alarm. However,the effects of oxygen treatments
`wear off over time and oxygen saturation levels drift down-
`ward 350. In particular, during a second timeinterval T, 364,
`a person has an oxygen saturation 320 with a relatively
`stable baseline 322. The later baseline 322 is established at
`
`a substantially lower oxygen saturation than the earlier
`baseline 312. In this scenario, a shallow, transient desatu-
`ration event 324 now exceeds the alarm threshold 330 and
`
`results in a nuisance alarm. After many such nuisance
`alarms, a caregiver may lower the alarm threshold 330 to
`unsafe levels or turn off alarms altogether, significantly
`hampering the effectiveness of monitoring oxygen satura-
`tion.
`A fixed threshold alarm schema is described above with
`
`respect to an oxygen saturation parameter, such as derived
`from a pulse oximeter. However, problematic fixed thresh-
`old alarm behavior may be exhibited in a variety of param-
`eter measurement
`systems that calculate physiological
`parameters related to circulatory, respiratory, neurological,
`gastrointestinal, urinary,
`immune, musculoskeletal, endo-
`crine or reproductive systems, such as the circulatory and
`respiratory parameters cited above, as but a few examples.
`An adaptive alarm system, as described in detail below,
`advantageously provides an adaptive threshold alarm to
`solve false alarm and missed true alarm problemsassociated
`with baseline drift among other issues. For example, for a
`lower limit embodiment, an adaptive alarm system adjusts
`an alarm threshold downwards when a parameterbaseline is
`established at lower values. Likewise, for an upper limit
`embodiment, the adaptive alarm system adjusts an alarm
`threshold upwards in accordance with baseline drift so as to
`avoid nuisance alarms.
`In an embodiment,
`the rate of
`baseline movement is limited so as to avoid masking of
`transients. In an embodiment, the baseline is established
`along upper or lowerportions of a parameter envelop so as
`to provide a margin of safety in lower limit or upper limit
`systems, respectively.
`One aspect of an adaptive alarm system is responsive to
`a physiological parameter so as to generate an alarm thresh-
`old that adapts to baseline drift in the parameter and reduce
`false alarms without a corresponding increase in missed true
`alarms. The adaptive alarm system has a parameter derived
`from a physiological measurement system using a sensor in
`communication with a living being. A baseline processor
`calculates a parameter baseline from an average value of the
`parameter. Parameter limits specify an allowable range of
`the parameter. An adaptive threshold processor calculates an
`adaptive threshold from the parameter baseline and the
`parameter limits. An alarm generator is responsive to the
`parameter and the adaptive threshold so as to trigger an
`alarm indicative of the parameter crossing the adaptive
`
`40
`
`45
`
`4
`threshold. The adaptive threshold is responsive to the param-
`eter baseline so as to increase in value as the parameter
`baseline drifts to a higher parameter value and to decrease in
`value as the parameter baseline drifts to a lower parameter
`value.
`
`In various embodiments, the baseline processor has a
`sliding window that identifies a time slice of parameter
`values. A trend calculator determines a trend from an aver-
`age of the parameter values in the time slice. A response
`limiter tracks only the relatively long-term transitions of the
`trend. A bias calculator deletes the highest parameter values
`in the time slice or the lowest parameter values in the time
`slice so as to adjust the baseline to either a lower value or a
`higher value, respectively. The adaptive threshold becomes
`less response to baseline drift as the baseline approaches a
`predefined parameter limit. A first adaptive threshold is
`responsive to lower parameter limits and a second adaptive
`threshold is responsive to upper parameter limits. The alarm
`generator is responsive to both positive and negative tran-
`sients from the baseline according to the first adaptive
`threshold and the second adaptive threshold. Thefirst adap-
`tive threshold is increasingly responsive to negative tran-
`sients and the second adaptive threshold is decreasingly
`responsive to positive transients as the baseline trends
`toward lower parameter values.
`Another aspect of an adaptive alarm system measures a
`physiological parameter, establishes a baseline for the
`parameter, adjusts an alarm threshold according to drift of
`the baseline and triggers an alarm in responseto the param-
`eter measurement crossing the alarm threshold. In various
`embodiments, the baseline is established by biasing a seg-
`ment of the parameter, calculating a biased trend from the
`biased segmentandrestricting the transient response of the
`biased trend. The alarm threshold is adjusted by setting a
`parameter limit and calculating a delta difference between
`the alarm threshold and the baseline as a linear function of
`
`the baseline according to the parameter limit. The delta
`difference is calculated by decreasing delta as the baseline
`drifts toward the parameter limit and increasing delta as the
`baseline drifts away from the parameter limit. A parameter
`limit is set by selecting a first parameter limit in relation to
`a delayed alarm and selecting a second parameter limit in
`relation to an un-delayed alarm. A segment of the parameter
`is biased by windowing the parameter measurements,
`removing a lower value portion of the windowed parameter
`measurements and averaging a remaining portion of the
`windowed parameter measurements. An upper delta differ-
`ence between an upper alarm threshold andthe baseline is
`calculated and a lower delta difference between a lower
`alarm threshold and the baseline is calculated.
`
`A further aspect of an adaptive alarm system has a
`baseline processor that inputs a parameter and outputs a
`baseline according to a trend of the parameter. An adaptive
`threshold processor establishes an alarm thresholdat a delta
`difference from the baseline. An alarm generatortriggers an
`alarm based upon a parameter transient from the baseline
`crossing the alarm threshold. In various embodiments, a
`trend calculator outputs a biased trend and the baseline is
`responsive to the biased trend so as to reduce the size of a
`transient that triggers the alarm. A response limiter reduces
`baseline movement due to parameter transients. The adap-
`tive threshold processor establishes a lower alarm threshold
`below the baseline and an upper alarm threshold above the
`baseline so that the alarm generator is responsive to both
`positive and negative transients from the baseline. The
`baseline processor establishes a lower baseline biased above
`the parameter trend and an upperbaseline biased below the
`
`

`

`US 9,775,570 B2
`
`5
`parameter trend. The lower alarm threshold is increasingly
`responsiveto negative transients and the upperalarm thresh-
`old is decreasingly responsive to positive transients as the
`baseline trends toward lower parameter values.
`
`DESCRIPTION OF THE DRAWINGS
`
`FIGS. 1-3 are exemplar graphs illustrating problems and
`issues associated with physiological parameter measurement
`systems having fixed threshold alarm schemas;
`FIGS. 4A-B are general block diagrams of an adaptive
`alarm system having lower parameter limits;
`FIGS. 5A-B are a graph of a physiological parameter
`versus delta space and a graph of delta versus baseline,
`respectively, illustrating the relationship betweena baseline,
`a lower-limit adaptive threshold and a variable difference
`delta between the baseline and the adaptive threshold;
`FIG. 6 is an exemplar graph of a physiological parameter
`versus timeillustrating an adaptive alarm system having a
`lower-limit adaptive threshold;
`FIG. 7 is a graph of oxygen saturation versus time
`illustrating a baseline for determining an adaptive threshold;
`FIG. 8 is a graph of oxygen saturation versus time
`comparing adaptive-threshold alarm performance with
`fixed-threshold alarm performance;
`FIGS. 9A-B are general block diagrams of an adaptive
`alarm system having upper parameterlimits;
`FIGS. 10A-B are a graph of a physiological parameter
`versus delta space and a graph of delta versus baseline,
`respectively, illustrating the relationship betweena baseline,
`an upper-limit adaptive threshold and a variable delta dif-
`ference between the baseline and the adaptive threshold;
`FIG.11 is an exemplar graph of a physiological parameter
`versus time illustrating an adaptive alarm system having an
`upper-limit adaptive threshold;
`FIGS. 12A-B are general block diagrams of an adaptive
`alarm system having both lower alarm limits and upper
`alarm limits;
`FIGS. 13A-E are physiological parameter versus delta
`space graphsillustrating a lower-limit adaptive threshold, an
`upper-limit adaptive threshold, and a combined lower- and
`upper-limit adaptive threshold in various delta spaces; and
`FIG. 14 is an exemplar graph of a physiological parameter
`versus time illustrating an adaptive alarm system having
`both lower and upper alarm limits.
`
`DETAILED DESCRIPTION OF THE
`PREFERRED EMBODIMENTS
`
`FIGS. 4A-B illustrate an adaptive alarm system 400
`embodiment having lower parameter limits L, and L,. As
`shown in FIG. 4A,
`the adaptive alarm system 400 has
`parameter 401, first limit (L,) 403, secondlimit (L,) 405 and
`maximum parameter value (Max) 406 inputs and generates
`a corresponding alarm 412 output. The parameter 401 input
`is generated by a physiological parameter processor, such as
`a pulse oximeter or an advanced blood parameter processor
`described above, as examples. The adaptive alarm system
`400 has an alarm generator 410, a baseline processor 420,
`and an adaptive threshold processor 440. The alarm genera-
`tor 410 has parameter 401 and adaptive threshold (AT) 442
`inputs and generates the alarm 412 output accordingly. A
`baseline processor 420 has the parameter 401 input and
`generates a parameter baseline (B) 422 output. The baseline
`processor 420, is described in detail with respect to FIG. 4B,
`below. An adaptive threshold processor 440 has parameter
`baseline (B) 422, L, 403, L, 405 and Max 406 inputs and
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`6
`generates the adaptive threshold (AT) 442. The adaptive
`threshold processor 440 is described in detail with respect to
`FIGS. 5A-B, below.
`As shown in FIG. 4A, in an embodiment L, 403 and L,
`405 may correspond to conventional fixed alarm thresholds
`with and without an alarm time delay, respectively. For an
`adaptive threshold schema, however, L, 403 and L, 405 do
`not determine an alarm threshold per se, but are reference
`levels for determining an adaptive threshold (AT) 442. In an
`embodiment, L, 403 is an upper limit of the adaptive alarm
`threshold AT whenthe baseline is near the maximum param-
`eter value (Max), and L, 405 is a lowerlimit of the adaptive
`alarm threshold, as described in detail with respect to FIGS.
`5A-B, below. In an exemplar embodiment when the param-
`eter is oxygen saturation, L, 403 is set at or around 90% and
`L, 405 is set at 5 to 10% below L,, ie. at 85% to 80%
`oxygen saturation. Manyother L, and L, values may be used
`for an adaptive threshold schemaas described herein.
`Also shown in FIG. 4A, in an embodiment the alarm 412
`output is triggered when the parameter 401 input falls below
`AT 442 and ends when the parameter 401 input rises above
`AT 442 or is otherwise cancelled. In an embodiment, the
`alarm 412 output is triggered after a time delay (TD), which
`maybefixed or variable. In an embodiment, the time delay
`(TD)is a function of the adaptive threshold (AT) 442. In an
`embodiment, the time delay (TD) is zero when the adaptive
`threshold (AT) is at the second lower limit (L,) 405.
`As shown in FIG. 4B, a baseline processor 420 embodi-
`menthas a sliding window 450, a bias calculator 460, a trend
`calculator 470 and a response limiter 480. The sliding
`window 450 inputs the parameter 401 and outputs a time
`segment 452 of the parameter 401. In an embodiment, each
`window incorporates a five minute span of parameter values.
`The bias calculator 460 advantageously provides an upward
`shift in the baseline (B) 422 for an additional margin oferror
`over missed true alarms. That is, a baseline 422 is generated
`that tracks a higher-than-average range of parameter values,
`effectively raising the adaptive threshold AT slightly above
`a threshold calculated based upon a true parameter average,
`as shown and described in detail with respect to FIGS. 7-8,
`below. In an embodiment, the bias calculator 460 rejects a
`lower range of parameter values from each time segment
`452 from the sliding windowso as to generate a biased time
`segment 462.
`Also shown in FIG. 4B, the trend calculator 470 outputs
`a biased trend 472 of the remaining higher range of param-
`eter values in each biased segment 462. In an embodiment,
`the biased trend 462 is an average of the values in the biased
`time segment 462. In other embodiments, the biased trend
`462 is a median or mode of the values in the biased time
`
`segment 462. The response limiter 480 advantageously
`limits the extent to which the baseline 422 outputtracks the
`biased trend 472. Accordingly, the baseline 422 tracks only
`relatively longer-lived transitions of the parameter, but does
`not
`track (and hence mask) physiologically significant
`parameter events, such as oxygen desaturations for a SpO,
`parameter to name but one example. In an embodiment, the
`response limiter 480 has a low pass transfer function. In an
`embodiment, the response limiter 480 is a slew rate limiter.
`FIGS. 5A-B further illustrate an adaptive threshold pro-
`cessor 440 (FIG. 4A) having a baseline (B) 422 input and
`generating an adaptive threshold (AT) 442 output and a delta
`(A) 444 ancillary output according to parameter limits L,
`403, L, 405 and Max 406, as described above. As shown in
`FIG. 5A, as the baseline (B) 422 decreases (increases) the
`adaptive threshold (AT) 444 monotonically decreases (in-
`creases) between L, 403 and L, 405. Further, as the baseline
`
`

`

`US 9,775,570 B2
`
`7
`(B) 422 decreases (increases) the delta (A) 444 difference
`between the baseline (B) 422 and the adaptive threshold
`(AT) 442 monotonically decreases (increases) between
`Max-L, and zero.
`As shownin FIG.5B, the relationship between the delta
`(A) 444 and the baseline (B) 444 may be linear 550 (solid
`line), non-linear 560 (small-dash lines) or piecewise-linear
`(large-dash lines), to name a few. In an embodiment, the
`adaptive threshold processor 440 (FIG. 4A) calculates an
`adaptive threshold (AT) 442 output
`in response to the
`baseline (B) 422 input according to a linear relationship. In
`a linear embodiment, the adaptive threshold processor 440
`(FIG. 4A) calculates the adaptive threshold (AT) 442 accord-
`ing to EQS. 1-2:
`
`
`_ -(eMexaL, )oMax — B) + (Max — 1)
`AT=B-A
`
`()
`
`(2)
`
`where A=Max-L, @ B=Max; A=0 @ B=L,
`and where AT=L, @ B=Max; AT=L., @ B=L,, accordingly.
`FIG.6 illustrates the operational characteristics an adap-
`tive alarm system 400 (FIG. 4A) having parameter limits
`Max 612, L, 614 and L, 616 and an alarm responsive to a
`baseline (B) 622, 632, 642; an adaptive threshold (AT) 628,
`638, 648; and a corresponding A 626, 636, 646 according to
`EQS. 1-2, above. In particular, a physiological parameter
`610 is graphed versus time 690 for various time segmentst,,
`t,, t; 692-696. The parameter range (PR) 650 is:
`PR=Max-L>
`
`(3)
`
`5
`
`10
`
`15
`
`20
`
`25
`
`30
`
`and the adaptive threshold range (ATR) 660 is:
`ATR=L,-Ly
`
`35
`
`(4)
`
`As shown in FIG.6, during a first time period t, 692, a
`parameter segment 620 has a baseline (B) 622 at about Max
`612. As such, A 626=Max-L, and the adaptive threshold
`(AT) 628 is at about L, 614. Accordingly, a transient 624
`having a size less than A 626 does not trigger the alarm 412
`(FIG. 4A).
`Also shownin FIG.6, during a second time periodt, 694,
`a parameter segment 630 has a baseline (B) 632 at about L,
`614. As such, A 636 is less than Max-L, and the adaptive
`threshold (AT) 638 is between L, and L,. Accordingly, a
`smaller transient 634 will trigger the alarm as compared to
`a transient 624 in the first time segment.
`Further shownin FIG. 6, during a third time periodt, 696,
`a parameter segment 640 hasa baseline (B) 642 at about L,
`616. As such, A 646 is about zero and the adaptive threshold
`(AT) 648 is at about L,. Accordingly, even a small negative
`transient will trigger the alarm. As such, the behaviorof the
`alarm threshold AT 628, 638, 648 advantageously adapts to
`higher or lowerbaseline valuesso as to increase or decrease
`the size of negative transients that trigger or do not trigger
`the alarm 412 (FIG. 4A).
`FIG. 7 is a parameter versus time graph 700 illustrating
`the characteristics of an adaptive alarm system 400 (FIGS.
`4A-B), as described with respect to FIGS. 4-6, above, where
`the parameter is oxygen saturation (SpO,). The graph 700
`has a SpO,trace 710 and a superimposedbaseline trace 720.
`The graph 700 also delineates tracking periods 730, where
`the baseline 720 follows the upper portions of SpO, values,
`and lagging periods 740, where the baseline 720 does not
`follow transient SpO, events. The tracking time periods 730
`illustrate that the baseline 720 advantageously tracks at the
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`8
`higher range of SpO, values 710 during relatively stable
`(flat) periods, as described above. Lagging time periods 740
`illustrate that the baseline 720 is advantageously limited in
`response to transient desaturation events so that significant
`desaturations fall below an adaptive threshold (not shown)
`and trigger an alarm accordingly.
`FIG. 8 is a parameter versus time graph 800 illustrating
`characteristics of an adaptive alarm system 400 (FIGS.
`4A-B), as described with respect to FIGS. 4-6, above, where
`the parameter is oxygen saturation (SpO,). Vertical axis
`(SpO,) resolution is 1%. The time interval 801 between
`vertical hash marks is five minutes. The graph 800 has a
`SpO,trace 810 and a baseline trace 820. The graph 800 also
`has a fixed threshold trace 830, a first adaptive threshold
`(AT) trace 840 and a second ATtrace 850. The graph 800
`further has a fixed threshold alarm trace 860, a first adaptive
`threshold alarm trace 870 and a second adaptive threshold
`alarm trace 880. In this example, L, is 90% and L, is 85%
`for the first AT trace 840 and first AT alarm trace 870. L, is
`80% for a second AT trace 850 and a second AT alarm trace
`880. The fixed threshold 830 results in many nuisance
`alarms 860. By comparison, the adaptive threshold alarm
`with L,=85% hasjust one time interval of alarms 872 during
`a roughly 6% desaturation period (from 92% to 86%). The
`adaptive threshold alarm with L,=80%, has no alarms dur-
`ing the 1 hour 25 minute monitoring period.
`FIGS. 9A-B illustrate an adaptive alarm system 900
`embodiment having upper parameter limits U, and U,. As
`shown in FIG. 9A,
`the adaptive alarm system 900 has
`parameter 901, first limit (U,) 903, second limit (U,) 905
`and minimum parameter value (Min) 906 inputs and gen-
`erates a corresponding alarm 912 output. The parameter 901
`input is generated by a physiological parameter processor,
`such as a pulse oximeter or an advanced blood parameter
`processor described above, as examples. The adaptive alarm
`system 900 has an alarm generator 910, a baseline processor
`920, and an adaptive threshold processor 940. The alarm
`generator 910 has parameter 901 and adaptive threshold
`(AT) 942 inputs and generates the alarm 912 output accord-
`ingly. A baseline processor 920 has the parameter 901 input
`and generates a parameter baseline (B) 922 output. The
`baseline processor 920, is described in detail with respect to
`FIG. 9B, below. An adaptive threshold processor 940 has
`parameter baseline (B) 922, U, 903, U, 905 and Min 906
`inputs and generates the adaptive threshold (AT) 942. The
`adaptive threshold processor 940 is described in detail with
`respect to FIGS. 10A-B, below.
`As shownin FIG. 9A, in an embodiment U, 903 and U,
`905 may correspond to conventional fixed alarm thresholds
`with and without an alarm time delay, respectively. For an
`adaptive threshold schema, however, U, 903 and U, 905 do
`not determine an alarm threshold per se, but are reference
`levels for determining an adaptive threshold (AT) 942. In an
`embodiment, U, 903 is a lower limit of the adaptive alarm
`threshold AT whenthe baseline is near the minimum param-
`eter value (Min), and U, 905is an upperlimit of the adaptive
`alarm threshold, as described in detail with respect to FIGS.
`10A-B, below. In an exemplar embodiment when the param-
`eter is oxygen saturation, U, 903 is set at or around 85% and
`U, 905 is set at or around 90% oxygen saturation. Many
`other U, and U, values may be used for an adaptive thresh-
`old schemaas described herein.
`Also shown in FIG. 9A, in an embodiment the alarm 912
`output is triggered when the parameter 901 input rises above
`AT 942 and ends when the parameter 901 input falls below
`AT 942 or is otherwise cancelled. In an embodiment, the
`alarm 912 output is triggered after a time delay (TD), which
`
`

`

`US 9,775,570 B2
`
`10
`FIG.11 illustrates the operational characteristics an adap-
`tive alarm system 900 (FIG. 9A) having parameter limits
`Min 1112, U, 1114 and U, 1116 and an alarm responsive to
`a baseline (B) 1122, 1132, 1142; an adaptive threshold (AT)
`1128, 1138, 1148; and a corresponding A 1126, 1136, 1146
`according to EQS. 5-6, above. In particular, a physiological
`parameter 1110 is graphed versus time 1190 for various time
`segments t,, t,, t, 1192-1196. The parameter range (PR)
`1150 is:
`
`9
`maybefixed or variable. In an embodiment, the time delay
`(TD)is a function of the adaptive threshold (AT) 942. In an
`embodiment, the time delay (TD) is zero when the adaptive
`threshold (AT) is at the second upper limit (U,) 905.
`As shownin FIG.9B, a baseline processor 920 embodi-
`menthas a sliding window 950, a bias calculator 960, a trend
`calculator 970 and a response limiter 980. The sliding
`window 950 inputs the parameter 901 and outputs a time
`segment 952 of the parameter 901. In an embodiment, each
`window incorporatesa five minute span of parametervalues.
`The bias calculator 960 advantageously provides a down-
`ward shift in the baseline (B) 922 for an additional margin
`of error over missed true alarms. That is, a baseline 922 is
`15
`generated that tracks a lower-than-average range of param-
`As shownin FIG. 11, duringafirst time period t, 1192, a
`eter values, effectively lowering the adaptive threshold AT
`parameter segment 1120 has a baseline (B) 1122 at about
`slightly below a threshold calculated based upon a true
`Min 1112. As such, A 1126=U,-Min and the adaptive
`parameter average. In an embodiment, the bias calculator
`threshold (AT) 1128 is at about U, 1114. Accordingly, a
`960 rejects an upper range of parameter values from each
`transient 1124 havinga size less than A 1126 doesnot trigger
`time segment 952 from the sliding window so as to generate
`the alarm 912 (FIG. 9A).
`a biased time segment 962.
`Also shown in FIG. 11, during a second time period t,
`Also shown in FIG.9B, the trend calculator 970 outputs
`1194, a parameter segment 1130 has a baseline (B) 1132 at
`about U, 1114. As such, A 1136 is less than U, -Min and the
`a biased trend 972 ofthe remaining lower range of parameter
`values in each biased segment 962. In an embodiment, the
`adaptive threshold (AT) 1138 is between U, and U,. Accord-
`biased trend 962 is an average of the values in the biased
`ingly, a smaller transient 1134 will trigger the alarm as
`time segment 962. In other embodiments, the biased trend
`compared to a transient 1124 in the first time segment.
`962 is a median or mode of the values in the biased time
`Further shown in FIG. 11, during a third time period t,
`segment 962. The response limiter 980 advantageously
`1196, a parameter segment 1140 has a baseline (B) 1142 at
`limits the extent to which the baseline 922 output tracks the
`about U, 1116. As such, A 1146 is about zero and the
`adaptive threshold (AT) 1148 is at about U,. Accordingly,
`biased trend 972. Accordingly, the baseline 922 tracks only
`relatively longer-lived transitions of the parameter, but does
`even a small positive transient will trigger the alarm. As
`such, the behavior of the alarm threshold AT 1128, 1138,
`not
`track (and hence mask) physiologically significant
`parameter events, such as oxygen desaturations for a SpO,
`1148 advantageously adapts to higher or lower baseline
`parameter to name but one example. In an embodiment, the
`values so as to increase or decrease the size of positive
`response limiter 980 has a low pass transfer function. In an
`transients that trigger or do nottrigger the alarm 912 (FIG.
`embodiment, the response limiter 980 is a slew rate limiter.
`9A).
`FIGS. 10A-B further illustrate an adaptive threshold
`FIGS. 12A-B illustrate an adaptive alarm system 1200
`processor 940 (FIG. 9A) having a baseline (B) 922 input and
`embodiment having lower limits L,, L, 1203, such as
`generating an adaptive threshold (AT) 9