DIAGNOSIS OF SLEEP APNEA

Abstract

The present invention relates to methods and apparatuses for detecting sleep apnea by analyzing characteristic physiological oscillations of the heart rate variability (HRV). Starting from recorded ECG data of the patient, for example, as a long-term sequence of the changing RR intervals, the heart rate variability is examined using autocorrelation calculations for the occurrence of rhythmic oscillations of various frequencies. If oscillations typical for apnea occur having very long period durations in the range of 20 to 80 seconds, these are detected as a maximum of the autocorrelation function. If a pathological sleep apnea accordingly exists, individual apnea events may be identified by prompt analysis of short recorded RR sequences, e.g., in the minute interval.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention is explained in greater detail in the following on the basis of preferred exemplary embodiments with reference to the drawings and the symbols and reference numerals used therein.

FIG. 1 shows a definition of the RR interval in ECG recordings.

FIG. 2 shows an IEGM curve having intracardial R waves.

FIG. 3 shows fluctuations of the RR intervals.

FIG. 4 shows circadian oscillations of the median heartbeat rate.

FIG. 5 shows a scheme for determining the nocturnal range from RR data.

FIG. 6 shows a flowchart of the method according to the present invention for apnea recognition using autocorrelation function.

FIG. 7 shows an autocorrelation having a clear maximum in the apnea time range.

FIG. 8 shows an autocorrelation of a control dataset (apnea-free).

FIG. 9 shows an autocorrelation having weakly pronounced apnea maximum.

FIG. 10 shows a flowchart for prompt detection of individual apnea events.

FIG. 11 shows a flowchart for prompt detection of individual apnea events.

DETAILED DESCRIPTION OF THE INVENTION

The heartbeat sequence is determined using a surface electrocardiogram (ECG) or from an intracardial electrogram (IEGM), which is measured using electrodes by an implant. The interval between two sequential heartbeats is generally defined via the so-called RR interval. This results as a chronological interval between two significant beats (so-called R waves) in the ECG or IEGM. These characteristic signals represent the contraction of the cardiac muscle in the course of the beat cycle. FIG. 1 shows a detail of a typical signal curve in a surface ECG having a marked RR interval. FIG. 2 shows the corresponding curve derived via intracardial electrodes.

A continuous measurement of R waves over a longer period of time provides a sequence of RR intervals which is referred to as a time array. A detail of such a time array is shown in FIG. 3. The heartbeat sequence does not show an equidistant sequence of R waves even in healthy people in the rest state, i.e., the interval lengths vary irregularly around a mean value (which is in turn a function of various external factors, such as patient age, “fitness”, time of day, etc.).

For implants and in less complex external devices, e.g., transportable ECG measuring units carried by the patient, which may generally process smaller amounts of data than the complete ECG signal, the large quantities of recorded RR intervals occurring in long-term measurements are frequently too large for efficient data processing using the curves of the individual R waves resolved in morphological detail. The starting data for more extensive analysis are therefore frequently not the time arrays themselves, but rather the heart rate variability (HRV) as a chronological sequence of the occurring interval lengths. These data are shown in FIG. 3. The relatively small spontaneous fluctuations of the beat sequence having deviations of the interval duration of less than a tenth of a second around a mean chronological beat period of approximately 1.1 seconds are shown.

The mean beat period is known to be subjected to day/night oscillations, which are to be considered in an analysis of the HRV. FIG. 4 shows the change of the mean beat sequence of the patient over a 24-hour period of time. The HRV data range in the nocturnal phase is exclusively essential for detecting sleep apnea, the incorporation of the (much longer) diurnal phase would result in interference in the analysis in most cases due to concealment of the relevant information content. Therefore, upon subsequent analysis of HRV data, the nocturnal range is preferably identified before further analysis. A simple example of a flowchart for establishing the nocturnal range is shown in FIG. 5. All of the data (data set from physionet.org) is divided into one-third nocturnal phase and two-thirds diurnal phase with the aid of the circadian variations in such a way that the middle of the nocturnal phase is assumed as the maximum of the mean RR interval over all combinations of ⅓ to ⅔. If the HRV data is instantaneously recorded, the nocturnal phases may also be logged by additional sensor data (e.g., impedance, activity, etc.). Therefore, before further analysis, the precise position of the nocturnal or sleeping phase may be established individually for the particular patient and measurement procedure.

The irregular oscillations of the RR intervals occurring in HRV data sets carry information about the regulation mechanisms of the cardiovascular system, so that they are used according to the present invention to detect differences between regular physiological behavior and pathological changes. The most important mechanisms used as a basis allow classification according to respiratory sinus arrhythmia, blood-pressure oscillations, thermoregulation, and renin-angiotensin system, which are influenced by vagotonia, sympathicotonia, and catecholamine level.

In the framework of a spectral HRV analysis, it is typical to define an array of frequency bands. The most important influencing factors may be coarsely assigned empirically to these bands:

• • The HF band (0.15 Hz-0.40 Hz) occurs due to vagally mediated (physiological) respiratory influences.
  • The LF band (0.04 Hz-0.15 Hz) is influenced both vagally and sympathetically and is linked to blood-pressure oscillations
  • It is suspected of the typical VLF band (0.003 Hz-0.04 Hz) that its origin is partially physiologically caused by the renin-angiotensin-aldosterone system, but particularly requires a strong vagal mediation

The basis for the diagnosis according to the present invention of a sleep apnea is the fact that the rhythmic fluctuations of the sympathicotonia having period durations between 20 and 80 seconds result in a correlated oscillation of the RR interval in a corresponding spectral range (usually used: 0.005 Hz to 0.03 Hz). This long-wave fluctuation is theoretically always expected to be pronounced in the event of complete apnea, because it is a necessary result of temporary respiratory arrest. It is to be assumed that the amplitude and frequency of the oscillation may vary individually by at least a factor of two, but this oscillation is so generally present that it represents the approach according to the present invention for a sensitive detection of apnea.

Known methods for an analysis of the long-wave oscillation component are wavelet transformation, Hilbert transformation, and Fourier transformation. However, the interpretation is difficult, because either a discriminator must be connected downstream, fixed limits must be predefined, or derived parameters having arbitrary limits must be used.

To overcome these difficulties, in the method according to the present invention, features of the spectral power density are detected in the time range using the autocorrelation function

$C\left(\tau \right)=\sum _{i=0}^{N-\tau }\mathrm{RR}\left(i\right)\mathrm{RR}\left(i+\tau \right)$

N is the number of measured points, RR(i) is the standardized RR interval measured as the number i, and τ is a time shift, which the RR signal shifts in relation to itself. The variable τ passes through the range of integers from τ=0 up to a physiologically advisable upper limit of, for example τ=120 at step width τ. The step width τ does not necessarily have to be equal to one, but is selected in this way for the following description as a special case. Larger values of the step width reduce the computing effort.

The autocorrelation function has the essential property, in the event of a shift τ, which results in a relative congruence equality of the signal with itself, of having a maximum. In contrast, if the shift results in a mutual curve of shifted and unshifted signal, C(τ) assumes a minimum. The autocorrelation function is always maximum in the event of zero shift, i.e., identity.

Because the main identifier of apnea is the occurrence of long-wave oscillations of the RR interval in the time range between 20 and 120 beats, in the event of existing respiratory interference, a local maximum of C(τ) is expected in the event of shifts of τ in this range. Therefore, the sequence for apnea detection may be subdivided into the following steps:

• • determining the autocorrelation function,
  • rising search for the first minimum (e.g., from τ=3),
  • search for the position of the absolute maximum above the first minimum.

FIG. 6 shows a corresponding flowchart of the method according to the present invention, which may be used as an algorithm for automated apnea diagnosis. The criteria or threshold values used (identified in FIG. 6 as Δ and θ) for the position of first minimum and following maximum of the autocorrelation curve may be determined from general data (e.g., as empirical guidelines), or may be tailored individually to a patient, e.g., if this patient is not to be classified in the general pattern because of special physiology or clinical picture.

FIG. 7 shows an autocorrelation function thus obtained in the event of existing sleep apnea. The maximum is clearly pronounced at a long shift of τ=45. The large τ value of the position of the maximum provides the significant information about the presence of a sleep apnea here. The degree by which the maximum is pronounced may be used to evaluate the severity of the apnea.

In contrast, if the patient is free of apnea, the long-wave maximum is missing, in addition, a local maximum is to be expected during the oscillations of the blood-pressure variability (LF) or the respiratory sinus arrhythmia (HF). This results because the typical oscillations are undisturbed and, in addition, because the normal state is distinguished by a reduced sympathicotonia in relation to sleep apnea. These oscillations cause a maximum of the autocorrelation function at approximately τ<15.

FIGS. 8 and 9 are examples of these conditions, ascertained from control data. In both cases, a maximum in the decisive range (τ=40 through 50) is not or is only very weakly implemented. Small deviations in the monotonously decreasing curve are recognizable in the area of physiological blood-pressure variability (LF) and respiratory variability (HF)

In an expanded embodiment of the method according to the present invention, the significance of the oscillations established using autocorrelation in the apnea range may be increased for delimitation in relation to other events having influence on the HRV: the apnea band is embedded in the VLF band and may also partially overlap with the LF band, so that a small or also moderate increase of the power density in the apnea band, inter alia, may also be attributed to other physiological phenomena. This may occur, for example, if a pronounced vagotonia exists, which significantly strengthens the VLF component. In such a case, however, the fluctuation of the RR intervals as a whole is significantly increased, which is not the case in the event of an apnea. Thus, in the event of a slight increase of the power density in the apnea band, if a significant increase of the HRV overall fluctuation is detected in parallel, for which methods are known, apnea is to be precluded as the sole cause.

Establishing an apnea using the suggested autocorrelation method has the essential advantage that a diagnosis may be performed solely from RR data using only one parameter (localization of a maximum in the apnea band). For this purpose, only very simple data processing operations (multiplication and summation of integers) are necessary. This requires only little data processing resources and may therefore also be managed by devices having restricted functional scope. These particularly also include implants (cardiac pacemakers, defibrillators, etc.) and wearable devices, which are integrated in a diagnostic telemetry system, for example.

If a sleep apnea has been established using autocorrelation methods, in an expanded embodiment of the method according to the present invention, the diagnostic procedure may be refined to direct detection of individual apnea events. For such prompt sensing in the minute range, it is necessary to already prepare an analysis from a relatively small data field (RR log of a few minutes), which is then performed cyclically, e.g., in a minute cycle. To be able to detect long-wave oscillations correctly on such a short time scale, a two-stage algorithm which is intentionally kept simple is used according to the present invention, because the data processing required for this purpose may also be performed using simple means. Firstly, components of higher frequency oscillations are removed from the signal. This may be performed using a sliding mean-value filter which requires particularly few computing operations, for example. If sufficient resources are present, higher performance filter operations (e.g., recursive mean-value filter) may also be used to improve the selectivity of the low-pass filter. The total energy of the remaining signal is then a measure of the presence of long-wave oscillations. This total energy is ascertained, for example, by summation of the absolute value differences of sequential RR intervals, according to

$S=\sum _{i=2}^N\mathrm{RR}\left(i\right)-\mathrm{RR}\left(i-1\right)$

N describing the number of intervals in the period of time of a minute, for example. If the summation value S thus obtained is above a predefined threshold θ, this is evaluated as a signal for a current apnea. FIG. 10 shows a flowchart.

The threshold value to be input previously may typically not be specified from general experiential values, but rather is a function of the individual data of the particular patient. Because previously the diagnosis of a sleep apnea has been performed using analysis of long-term HRV data (at least 24 hours), the patient-related base intervals of the heart rate and their circadian changes (compare FIG. 4) are known and may be used to establish the threshold value. For an apnea patient, the nocturnal phase is composed of minute intervals in which apnea exists and of intervals in which normal breathing occurs. Therefore, the distribution density function of the absolute value summation values is composed of two ranges which more or less overlap, the higher values of the absolute value sum S being assigned to the apnea intervals. For an automatic, objective, and effective establishment of a threshold value between apnea and normal intervals, for example, the entropy method already developed for extrasystolic filters suggests itself. Alternatively, other methods may also be used, which provide an automatically determinable threshold value at sufficient sensitivity and specificity on an empirical basis, e.g., the simple mean value of the distribution, the median of the distribution, or suitable other percentiles. The threshold value determination may be performed depending on the method, outlay and resources in the implant, after telemedical data transfer, or in the scope of aftercare.

The method may also be used to detect a current sequence of apnea events and thus to be able to engage with treatment immediately, for example. For this purpose, the analysis, as shown in FIG. 11 as a flowchart, may be supplemented by a further freely selectable threshold value Q, which predefines a critical number of apnea events. If more than Q apnea events are detected in sequence, a corresponding signal may be generated automatically.

The method presented here according to the present invention for diagnosing sleep apnea is particularly suitable for implementation in devices having restricted resources. It is particularly advantageous that the particular critical parameter may be adapted to the available technical resources. For example, the increment for the shift of the autocorrelation function may be increased, which causes a direct reduction of the processing outlay. Therefore, even in the event of restricted computing capacity, measurements may even be executed over multiple days and, in spite of fewer operations per cycle, a statistically relevant correlation statement may be generated.

The present invention allows the characterization of an apnea status and automatic recognition of changes which occur. For example, a rapid diagnostic opinion may be reached by telemetric transmission of little data. The amount of data to be externally transmitted is additionally very small and thus allows checking of the patient status without problems, e.g., in a 24-hour rhythm. Complex analysis of morphology criteria, which is susceptible to error, is dispensed with, because only the easily available RR interval is used.

An apparatus for implementing the method according to the present invention does not require any special effort, so that it is also suitable for wearable devices and particularly also for implants. The calculations to be performed for this purpose in an implant, for example, relate to a few simple numeric operations, the algorithm to be used for determining the autocorrelation function and the sum of the absolute value differences of sequential RR intervals is restricted to a calculation of interval values. The data storage in the implant requires few bytes. Functions of this type may already be performed by very simple data processing components. The energy consumption is correspondingly low.

In an expanded embodiment of the present invention, means for implementing the method according to the present invention are integrated as an algorithm on corresponding miniaturized data processing components in analysis devices, in particular implants, for example, as application-specific integrated circuits (ASICs) or microprocessors (μP). Therefore, the individual analysis sequences are also possible in the implants themselves. An existing apnea may then be detected by the implant itself and the corresponding information may be transmitted as a signal for further diagnosis.

In addition to the detection of sleep apnea, the improvement of continuous monitoring in the event of CHF is advantageous. The recognition of progressive or acute worsening of the CHF illness is the focus point, because progress of the illness may result in increased central sleep apnea, on the other hand, an existing sleep apnea may also have a negative effect on the CHF illness course.

Claims

1. A method for detecting sleep apnea from ECG data, heartbeats being detected as a sequence of R waves using electrodes and variations of RR interval lengths occurring in said sequence of R waves being recorded as heart rate variability (HRV protocol), wherein an HRV log is analyzed for existing HRV oscillations in a low-frequency range, oscillations in a time range of 20 to 80 seconds detected in this way being evaluated as a signal for existing sleep apnea.

2. The method according to claim 1, wherein said HRV protocol is analyzed for said existing HRV oscillations in said low frequency range using autocorrelation methods.

3. The method according to claim 2, wherein frequency position and/or intensity of oscillations detected using autocorrelation methods are used to evaluate a severity of said sleep apnea.

4. The method according to claim 1, wherein data of said HRV protocol used for analysis is restricted to data ranges of nocturnal and/or sleep phases.

5. The method according to claim 1, wherein a quantity of scattering of a recorded heart rate variability is analyzed and used to identify said sleep apnea and a severity of said sleep apnea.

6. The method according to claim 1, wherein individual apnea events are detected by analyzing predefined time sections of said HRV protocol for low frequency oscillations, as a sequence in a minute range, using absolute value summation of interval differences.

7. An apparatus for detecting sleep apnea from ECG data comprising: means for detecting heartbeats as a sequence of R waves;means for preparing an HRV protocol;means for transmitting or storing a corresponding data set;means for determining existing low frequency HRV oscillations using autocorrelation or absolute value summation; and,means for differentiating the values obtained according to predefined criteria.

8. The apparatus according to claim 7, wherein said means for preparing, means for storing, means for determining and means for differentiating are implemented as miniaturized data processing elements (ASIC, μP).

9. The apparatus according to claim 8, wherein said apparatus is at least partially integrated in a portable ECG device or in an implant.